Working Paper in Economics No. 808 
 
 
 
An App Call a Day Keeps the Patient Away? 
Substitution of Online and In-Person Doctor 
Consultations Among Young Adults  
 
Lina Maria Ellegård, Gustav Kjellsson and Linn Mattisson  
 
Department of Economics, June 2021 and rev. May 2022 
 
 
 
 
 
 
 
ISSN 1403-2473 (Print)  
ISSN 1403-2465 (Online) 
An App Call a Day Keeps the Patient Away? Substitution of Online and
In-Person Doctor Consultations Among Young Adults
Lina Maria Ellegård, Gustav Kjellsson and Linn Mattisson*
Abstract
The emergence of markets for on-demand online physician consultations
–direct-to-consumer telemedicine (DCT) – is currently transforming many health-
care settings. DCT may be a cost-effective substitute for ordinary consultations,
but the convenience of seeking DCT may increase demand and costs for health
insurers. To causally assess to which degree DCT consultations substitute for
in-person consultations, we exploit exogenous changes in patient fees in a fuzzy
difference-in-discontinuities analysis of young adults in Sweden. We estimate a
degree of substitution of 45%, implying an increase in the consultation volume.
Characteristics of the additional demand raise concerns related to healthcare eq-
uity, efficiency and costs.
*Ellegård: Department of Economics, Lund University & Kristianstad University, lina-
maria.ellegard@nek.lu.se. Kjellsson: Department of Economics, gustav.kjellsson@economics.gu.se. 
Mattisson: Department of Economics, Lund University, linn.mattisson@nek.lu.se. Acknowledgements: We 
are thankful to Otàvio Bartalotti, Jan Bietenbeck, Carl Bonander, Libertad González, Yang He, Annika Herr, 
Naimi Johansson, Filippo Tassinari and seminar participants at the EuHea seminar series, Universitat Pompeu 
Fabra, University of Gothenburg, and Lund University for helpful comments. Belit Ay provided excellent 
research assistance. This work was funded by Jan Wallanders och Tom Hedelius stiftelse and Tore Browaldhs 
stiftelse (grant no. P19-0064), the Swedish Research Council for Health, Working Life and Welfare (FORTE, 
grant no. 2019-00123), Adlerbert Research Foundation, and Foundation for Economic Research in West 
Sweden. None of the authors have any conflicts o f i nterest t o d isclose, a nd n o t hird p arty h ad t he r ight to 
review this work prior to its circulation. The study has been approved by the regional ethics committee (i.e. 
IRB) in Gothenburg, Sweden (Dnr 068-18).
1
1 Introduction
As in many parts of the economy, the outbreak of the Covid-19 pandemic greatly reinforced
the ongoing digitalisation of health care services (Mehrotra, Bhatia and Snoswell, 2021).
But even before the pandemic, the traditional way of providing care in-person at the physi-
cian’s office was challenged by fast-growing tech start-ups offering online consultations on-
demand, so-called direct-to-consumer telemedicine (DCT), in many countries.1
The usual virtues of online services, to which consumers have become accustomed from
experiences with online retail, carry over to DCT: Around the clock, DCT platforms are
ready to respond to demand by matching patients with available providers (physicians), who
may be located anywhere geographically. The online format virtually eliminates the need to
spend time travelling or waiting for appointments, time that in the in-person setting easily
may exceed the duration of the consultation itself. This indicates that there are substan-
tial economic gains to be made by substituting DCT for regular in-person consultations,2
whenever medically appropriate.
Yet, due to the novelty of DCT, little is known of how patients navigate the new landscape
enabled by digitalisation. One central question is to what extent the rapidly growing number
of DCT consultations represents substitution of in-person visits versus new demand, spurred
by the convenience of DCT.3 In settings where patients do not bear the full cost of DCT,4 the
1Examples of DCT companies include Teladoc and K-Health in the US, Babylon GP at Hand in the UK,
Ping An Good Doctor in China, and Kry in Sweden (Salisbury et al., 2020). Even before the pandemic outbreak,
Teladoc and Kry were doubling the number of consultations annually (SALAR, 2020; Li et al., 2021)
2For instance, the value of the time saved by Swedish patients if half of all primary care consultations were
to be provided online has been estimated at over 500 million USD (Ekman, 2018).
3DCT may induce new utilisation through several channels. First, the convenience of DCT may lower the
bar for when patients choose to seek care rather than to wait and see. Second, some DCT consultations will
inevitably result in referrals to in-person appointments, as patients lack all information needed to perfectly
judge the need for a physical examination. Third, the inability to perform physical examinations in DCT (or
remote care more generally) may have to be compensated for by more consultations to reach the same level of
clinical certainty an in-person setting.
4For instance, in the US, virtual care was covered by the health plans of a vast majority of large employers
already in 2019 (AHIP, 2019). In the UK and Sweden, the main DCT providers have access to public funding.
1
degree of substitution relates to issues of efficiency as well as health system sustainability
and equity. Efficiency-wise, the accessibility of DCT might aggravate already existing moral
hazard problems of insurance. From a health system perspective, the additional demand
might lead to increasing costs or a redistribution of health care resources away from patients
with more severe problems.
This study provides causal estimates of the degree of substitution, using detailed admin-
istrative data on a key group of early adopters – young adults – in the two largest regions
in Sweden. The institutional setting provides opportunities to obtain plausibly causal esti-
mates, due to the combination of a rapidly emerging DCT market and exogenous variation
in regulated user fees. We are able to exploit these features in a pre-Covid-19 context, as
the growth of the DCT market took off already in mid-2016. By 2018, DCT consultations
accounted for 5% of all primary care physician consultations, but as much as 20% in the age
groups we study (SALAR, 2020).
During our study period, patients could consult DCT providers for free until they turned
20; after the 20th birthday, the consultation fee was around EUR 25/USD 30. We use this
exogenous change to identify the effect of the fee on the demand for DCT and thereby infer
the degree of substitution. To account for other potential demand effects around the age
threshold, we use a counterfactual provided by the change in in-person visits around the
20-year birthday of older cohorts, who turned 20 before the DCT market emerged.5
We find that the demand for these services is very sensitive to price: After the onset of
the fee, the number of DCT consultations falls by half.6 Further, our estimate of the substitu-
tion rate implies that 45% of online consultations replace in-person physician visits. Conse-
5We obtain similar estimates with an alternative estimation strategy that does not rely on similarity of
different cohorts.
6This is a large decrease in comparison both to the literature on how co-payments affect care utilisation (e.g.,
Manning et al., 1987; Newhouse, Group and Staff, 1993; Bundorf, 2012; Chandra, Gruber and McKnight, 2014;
Brot-Goldberg et al., 2017), and in particular to studies using similar age-based threshold strategies to estimate
price sensitivity for in-person consultations in Sweden (Nilsson and Paul, 2018; Johansson, Jakobsson and
Svensson, 2019).
2
quently, around half of all DCT consultations represent additional demand, i.e., consultations
that would not have taken place, or would have been dealt with by other professions (e.g.,
nurses, midwives or psychotherapists), in the absence of DCT.
A decomposition exercise suggests that slightly less than half of the additional consulta-
tions are one-off encounters with no follow-ups within 14 days. The rest are thus related to
care episodes spanning more than one visit; of these consultations, one quarter initiate new
episodes, and three quarters represent increases of the number of follow-ups per episode.
Importantly, we do not find that the substitution rate is limited to a great extent by DCT con-
sultations that only function as an additional step before an in-person visit (i.e., cases when
physical examinations or lab tests are necessary). Thus, the additional consultations stem
primarily from the low barriers of access to DCT services.
Both demand- and supply-side factors may explain why the barriers are lower in DCT
than in regular care. On the demand side, the convenience of DCT may increase the propen-
sity to seek care for less severe conditions. That almost half of the additional consulta-
tions are one-off encounters is in line with such an interpretation. On the supply side, DCT
providers, which are funded fee-for-service, may be willing to meet care needs that regular
care providers would have assigned low priority or delegated to other professions. In this
regard, it is notable that we find particularly low substitution for diagnoses related to skin
conditions and reproductive health, which in the regular Swedish primary care setting typi-
cally would be handled by referring the patient to self-care, or by other professions (nurses
or midwives).
Our results speak to the present challenge of designing incentives for the continued use
of telemedicine in the post-pandemic era (Cutler, Nikpay and Huckman, 2020; Mehrotra,
Bhatia and Snoswell, 2021). Despite the additional demand, our back-of-the-envelope cal-
culations suggest that the production cost of a DCT consultation is low enough to make
the availability of DCT close to resource neutral. Although the estimates are surrounded
3
by uncertainty, this result is remarkable given the institutional structure, which gives DCT
providers very strong incentives to schedule additional consultations (i.e., to reduce the de-
gree of substitution). For health insurers considering to cover DCT services, our results thus
indicate that such coverage need not result in a cost explosion, provided that the reimburse-
ment rate is set at a sensible level.7
Nevertheless, our study setting highlights the importance of designing incentives to ad-
dress distributional concerns – how to distribute the gains from DCT. With a reimbursement
rate amounting to almost twice the marginal cost of a DCT consultation, third-party payers in
our setting are likely paying in excess given current patient volumes (over 2 million consul-
tations annually). The economic profits of DCT providers come at the expense of the insured
population (tax payers), who may need to pay higher premiums (taxes), or, to the extent that
the DCT profits are financed within a given health budget, patients whose health problems
cannot be dealt with in a DCT setting. Arguments related to both efficiency (deadweight
loss of taxes) and equity (larger needs of patients unsuitable for DCT (Roland, 2019)) may
justify another distribution of the gains from DCT, which at present fall exclusively on DCT
providers and on DCT users.
Our paper adds to a small number of previous studies on the substitution between DCT
and regular care. Survey data from the US indicate that the degree of substitution may be
even higher than 45% (Martinez et al., 2018; Nord et al., 2018). However, these estimates
may be exaggerated, if respondents rationalise their decision to seek care. Indeed, the few
previous studies looking at actual care utilisation have obtained smaller estimates. Ashwood
et al. (2017) studied acute respiratory infections in a sample of Californian public employ-
ees. Their matched difference-in-differences (DiD) analysis indicated that only 12% of DCT
consultations replaced in-person visits.8 Ellegård and Kjellsson (2019) found no evidence
7Coverage would imply large gains for the individuals in terms of time saved and decreased worry.
8Related, two studies of DCT visits for acute respiratory infections in the US find that the number of follow-
up visits were greater for patients whose first contact was in a DCT setting than for patients whose first contact
4
of substitution in a DiD analysis of a representative sample of the population in a Swedish
region (Skåne).
A limitation of the previous studies is that their estimation strategies may fail to account
for time-variant unobserved heterogeneity. DCT providers typically address sudden and tran-
sitory health problems, such as respiratory infections or skin conditions; i.e., issues that are
neither subsumed by time-invariant group-specific characteristics (fixed effects) nor possible
to account for by matching on previously observed conditions. Consequently, DiD analyses
may underestimate the degree of substitution. In line with this, the present study, which uses
an alternative identification strategy that avoids this limitation, yields a substantially higher
degree of substitution.
Outside the DCT setting, a related strand of literature considers telemedicine in tradi-
tional healthcare organisations, e.g., communication with the patient’s regular physician via
video calls or asynchronous messaging systems (North et al., 2014; Shah et al., 2018; Bavafa,
Hitt and Terwiesch, 2018; Zeltzer et al., 2021).9 Although the estimated magnitudes vary be-
tween telemedicine applications and contexts, the results from this literature are overall in
line with the evidence from DCT, indicating that telemedicine on net leads to more consul-
tations. Methodologically, most of the studies suffer from the same selection problem as the
earlier DCT literature. A recent exception is the study by Zeltzer et al. (2021), who exploit
variation in telemedicine adoption among Israeli GPs during the Covid-19 lockdown in a
DiD setting. They find that increased access to telemedicine leads to a small increase in
primary care visits, but 5% lower health care costs due to lower costs of drugs, lab tests, and
in- or outpatient services further downstream.
was an in-person visit (Shi et al., 2018; Li et al., 2021). In the regular (i.e., non-DCT) Swedish primary care
setting, Entezarjou et al. (2022) did not find differences between patients contacting a provider online first vs
in-person first with respect to the number of in-person follow-up visits within 2 weeks for respiratory or urinary
symptoms.
9Another small literature shows that states adopting telemedicine parity laws experience increases in pri-
mary care but reductions in hospital use (Dills and Chen, 2018; Grecu and Sharma, 2019). Due to data limita-
tions, these studies cannot definitively attribute these effects to changes in telemedicine use.
5
The paper proceeds as follows. The next sections provide a background to the insti-
tutional setting (Section 2) and describe the data and our empirical strategy (Section 3).
Section 4 provides results and Section 5 concludes.
2 Institutional background and conceptual framework
2.1 Traditional primary care
The financing and provision of health care services in Sweden is delegated to 21 independent
regions. Health care is mainly financed by regional proportional income taxes (71%), central
government grants (20%) and patient fees (5%) (SALAR, 2017). Primary care handles health
issues that can be treated outside hospitals, and is organised in group practices – primary care
centres (PCCs). PCCs may be public or private, and the staff (typically general practitioners
(GPs) and nurses) are salaried employees (Anell, Glenngård and Merkur, 2012). Patients are
free to contact any PCC, which typically are open during office hours. During evenings and
weekends, care is supplied by a subset of the PCCs.
The PCCs are reimbursed by a mix of (risk-adjusted) capitation, i.e., a fixed monthly sum
per registered patient, and fee-for-service based on the number of visits. In one of our study
regions (Region Västra Götaland), the reimbursement is almost entirely based on capitation,
whereas in the other region (Region Stockholm), capitation accounts for roughly 60% of the
reimbursement.
2.2 The Swedish DCT market
Since 2016, the PCCs face competition from private DCT providers. The emergence of the
Swedish DCT market was an unintended consequence of the Patient Right Law, enacted
in 2015, which gave patients the right to seek care outside their region of residence. DCT
6
entrepreneurs realised that they could locate a DCT company in one region, treat patients
in other regions, and then bill the patients’ home regions. Notably, this arrangement means
that the DCT providers operate outside the regular reimbursement systems, and are instead
subject to the regulation for inter-regional reimbursement. This regulatory framework is
negotiated by the Swedish Association of Local Governments and Regions (SALAR) and
designed as a fee-for-service scheme using visits as the base for payment.
Patients contact a DCT provider by describing their symptoms in a smartphone app or
on the company’s website, and are then contacted by the DCT physician.10 Around 80%
of the patients at the largest DCT provider are scheduled for an appointment within 30
minutes (Kry, 2019). The DCT consultation may be asynchronous (chat) or synchronous
(video call). Notably, few traditional primary care providers offered those care modes in our
study period, though telephone consultations were common (especially for follow-up vis-
its).11 DCT physicians have the same authorities as physicians working in PCCs; i.e., they
may set diagnoses, prescribe drugs, and write referrals to other providers.
The patient population and case-mix differs between DCT and traditional primary care.
Infants and adolescents are over-represented among DCT patients, and skin conditions and
respiratory infections account for a higher share of consultations in DCT settings (Dahlgren
et al., 2021). Common conditions in regular primary care such as hypertension are much
more rare.
2.3 User fees
Each regional health authority regulates the user fee for consultations supplied by providers
in the region. Throughout our study period, the fee level in Region Stockholm (RS) was
10The third largest company during our study period, Doktor.se, which only served a small fraction of the
market, had an initial nurse-led triage step.
11The traditional provider Capio and the public providers in Region Västra Götaland launched online plat-
forms in 2018, but the outreach was negligible.
7
EUR 20, while patients in Region Västra Götaland (RVG) paid EUR 10 for visits at the
PCC where they were registered and EUR 30 for visits at other providers. Children and
adolescents were (and still are) exempt from paying fees; the lower age limits for paying the
fee was 18 in Region Stockholm and 20 in Region Västra Götaland. In a given year, patients
never have to pay more than EUR 100 in user fees.
Since the user fee is based on the location of the provider, patients in our study regions did
not pay the same amount for DCT consultations as for visits in their region of residence. Dur-
ing our study period, the largest DCT providers Kry and Min Doktor, accounting for almost
90% of consultations in 2018, were located in Region Jönköping, where patients aged 20 or
above paid a user fee of EUR 25. A minor provider (Doktor.se) was located in Sörmland,
where there was no user fee.12 Table 1 summarises the user fee and reimbursement systems
in the traditional primary care and DCT sectors.
[Table 1 about here.]
2.4 Conceptual framework
To clarify the study setting, Figure 1 provides a framework illustrating the care paths for an
individual who experiences a care need. The individual might try do address the issue by
herself (self care). If she chooses not to, or if self care does not resolve the problem, she may
contact a DCT provider or a primary care centre. In the first case, the patient will be offered
an appointment with a physician contracting with the DCT company. In the second case,
a triage nurse will examine the need for further care and either schedule an appointment
with a physician or another health professional at the PCC, or deny access to further care
(typically combined with some self care advice). If, after this step, the patient perceives that
the problem has not been resolved, she may again choose between self care or to contact a
12In 2019, Kry and Min Doktor moved to Sörmland.
8
DCT provider or a PCC (possibly another one than the initial PCC) and so on.13
The framework is helpful to understand why the DCT providers may not only serve as
substitutes, but also increase the number of consultations. One example is that patients who
are dissatisfied with their initial consultation (be it a DCT or an in-person consultation) may
contact the other type of care provider for a second opinion. Further, DCT consultations may
substitute for self care – partly by virtue of its convenience, partly because DCT providers
may schedule appointments in situations where the PCC triage nurse would have denied the
patient access to further care.14 Additional consultations may also be due to differences
in the quality of the two care modes, relating to the impossibility of performing physical
examinations and lab tests in an online setting. In such cases, the DCT consultation may end
with the physician referring the patient to a PCC or scheduling a follow-up appointment to
monitor the development of the situation. In both cases, total consultations would increase.
The framework also illustrates that physician consultations in the DCT setting may be
substituting for consultations with other professions at PCCs. To give a relevant example
for our study population, contraceptive prescriptions are by default handled by midwives in
traditional primary care, but by physicians in DCT.
[Figure 1 about here.]
13The framework abstracts from care seeking outside the primary care setting, i.e., emergency unit atten-
dance, which is outside the scope of the paper due to data limitations.
14The opposite situation – DCT rejecting patients in situations when nurses would not – is less plausible, as
DCT providers but not PCCs obtain their entire funding from fee-for-service and are heavily profiled towards
providing accessible care.
9
3 Data and empirical strategy
3.1 Study population and data sources
Employed with data from the Swedish population register (held by Statistics Sweden, SCB),
we define a study population consisting of all individuals who belonged to the 19-20 year
age group in any of the years 2012-2018, who resided in either Region Stockholm or Region
Västra Götaland two consecutive New Year’s Eves,15 and who had lived in Sweden at least
since they were 15 years old.16
These data are linked to daily care utilisation from regional administrative registers for
2012-2018 covering the universe of consultations with primary care physicians and nurses
with providers within the patients’ region of residence (i.e. Region Stockholm or Region
Västra Götaland) and in the two regions where the DCT providers where located during
the study period.Consultations with providers in the region of residence include diagnosis
codes (ICD-10). To obtain diagnoses for DCT consultations the data are complemented with
information from registers from Region Jönköping, where the two main DCT providers were
located.17
These care data are also linked to annual data on demographic and socioeconomic char-
acteristics of the study population and their parents obtained from SCB. As the data include
the exact date of birth, we can construct a daily panel, where the time dimension is defined
relative to the 20th birthday.
15Our annual data on place of residence is measured on December 31st.
16We employ this restriction to avoid compositional changes driven by the immigration wave in 2014-15.
17These data are also linked to daily information on antibiotic prescriptions from pharmaceutical register
held by the National Board of Health and Welfare. This information is used in further analyses presented in the
appendix only.
10
3.2 Fuzzy difference-in-discontinuity design
Our objective is to estimate the causal effect of an online consultation, DCT , on the number
of in-person consultations, y. The identification problem is that an individual’s decision to
contact a DCT provider may correlate with unobservable characteristics that in turn influence
the decision to make an in-person visit at a regular PCC. To eliminate the influence of such
omitted variables, we need to find factors that exogenously alter the individual’s incentives
to contact DCT providers, while not directly changing the incentives to contact traditional
providers.
During our study period, the incentives to contact a DCT provider changed exogenously
at the 20th birthday, due to the onset of the DCT user fee. A natural starting point for the esti-
mation of the degree of substitution is therefore to consider a fuzzy regression discontinuity
(RD) design, using the discontinuity at the 20th birthday as an instrument for the number
of DCT consultations. The problem with such a strategy is that the incentives to contact
traditional health care may also change at the 20th birthday. As already noted, one of our
study regions (Region Västra Götaland) used the same age limit for in-person consultation
fees.18 To purge our estimates of other effects of turning 20, we use an older cohort to es-
timate discontinuities around the 20th birthday in the period before DCT was available – a
differences-in-discontinuities (diff-in-disc) strategy.
Before we introduce our fuzzy diff-in-disc estimand of the degree of substitution, it is
instructive to first express a sharp diff-in-disc estimand for any random variable Z:
τZ = (Z
+
1 −Z
− + −
1 )− (Z0 −Z0 ) (1)
Here, Z+c (Z
−
c ) denotes the upper (lower) limit of the regression function E(Zc|agec = 20)
18There may also be other changes at 20. For example, this is the lower age limit for being allowed to buy
strong liquor in Sweden, which has been shown to increase the risk of being hospitalised (Heckley, Gerdtham
and Jarl, 2018).
11
of cohort c ∈ (0,1) as it approaches the age threshold. Thus, the sharp diff-in-disc estimand
compares discontinuities at the 20th birthday of two cohorts: a young cohort with access to
DCT services both before and after they turned 20 (cohort 1), and an old cohort, who turned
20 before DCT emerged (cohort 0).19
Grembi, Nannicini and Troiano (2016) provide assumptions under which the sharp diff-
in-disc identifies the causal effects of the new treatment in a setting where other, pre-existing,
treatments are assigned at the same threshold. To identify the treatment effect on the cohort
affected by both the new and the confounding treatments, two assumptions have to be satis-
fied. The first is the standard RD assumption that the conditional expectations of all potential
outcomes must be continuous around the threshold (for both cohorts). Second, the effects
of the confounding treatments must be time-invariant. The assumption implies that the only
reason why the discontinuity at the 20th birthday would look different for the two cohorts is
that the younger cohort had access to DCT.
Under these assumptions, the sharp diff-in-disc identifies the effect of becoming subject
to the DCT consultation fee: When Z = y, Eq. (1) describes the effect of the DCT fee on the
number of in-person consultations, and when Z = DCT , the equation describes the effect of
the DCT fee on the number of DCT consultations. In principle, the second term of Eq. (1) is
zero when Z = DCT , as the DCT market did not yet exist for the older cohort.20
In order to estimate the degree of substitution between online and in-person consulta-
tions, we turn to a fuzzy diff-in-disc framework. Analogous with the standard fuzzy RD,
we construct the fuzzy diff-in-disc estimand as the ratio of the sharp diff-in-discs of y and
19Specifically, the old cohort comprises individuals turning 20 before July 1 2016.
20In practice, the data from the non-home regions used to identify the DCT consultations includes a small
number of in-person consultations and so the term differs from zero.
12
DCT :21
τ (y+− y−)− (y+− y−y 1 1 0 0 )θ = = (2)
τDCT (DCT+1 −DCT
−
1 )− (DCT
+
0 −DCT
−
0 )
The fuzzy diff-in-disc identifies a local average treatment effect that can be interpreted as the
degree of substitution for compliers – i.e., individuals who consult DCT providers less often
only because of the fee – under the assumption of monotonicity (Millán-Quijano, 2020). In
our context monotonocity implies an assumption of no one responding to the DCT fee by
consulting DCT providers more often. Monotonicity thus rules out that DCT services are
Giffen goods, which seems a plausible assumption to make.
The first assumption of continuous conditional expectations around the threshold war-
rants some extra discussion. On one hand, the assumption fits well to a context using age
as the running variable: Individuals will age and eventually be observed the other side of
the age threshold. On the other hand, individuals may anticipate the onset of user fees, and
adjust by scheduling care appointments before rather than after the 20th birthday. A strength
of the diff-in-disc approach is that, as seen from the nominator of Eq. (2), intertemporal
substitution of in-person consultations due to anticipation effects would be purged by the
difference of the two RDs, assuming that the incentives for intertemporal substitution are the
same for both cohorts (which is likely, as the user fees were constant).
For the online consultations, it is by definition impossible to use the old cohort to net out
”usual” intertemporal substitution. In section 4.1, we instead examine if intertemporal sub-
stitution is an issue by checking if the estimated τDCT is sensitive to removing observations
close to the 20th birthday. As seen from the denominator of Eq. (2), intertemporal substitu-
tion of online consultations would imply that we underestimate of the degree of substitution.
21Another example of a fuzzy diff in disc is Galindo-Silva, Some and Tchuente (2019). These authors
discuss a special case in which the treatment of interest – buying insurance – is affected by multiple policies in
a young cohort, but only by one policy in an old cohort. This setup differs from our setting, where the treatment
of interest – the number of DCT consultations – is affected by one policy in a young cohort, but not available
at all to the old cohort.
13
3.3 Estimation
It is standard to estimate the parameters of a RD model using a local linear (first order)
polynomial regression for a given bandwidth (e.g. Calonico, Cattaneo and Titiunik, 2014).
We follow the same route to estimate our diff-in-disc model. We apply a uniform kernel
throughout. Estimating a fuzzy diff-in-disc in this way is equivalent to estimating a two
stage least square model. The first-stage and the reduced form equations for the number of
DCT and in-person consultations made by observation i in age-bin j in cohort c is specified
as follows:
Z = β Zi jc 1 +β
Z
2 I(20)i jc +β
Z
3 C
Z Z
i j +β4 I(20)i jc ×Ci j + f (agei jc,I(20)i jc,Ci j)+ γi jc (3)
where I(20)i jc is a dummy for being at least 20 years old, Ci j ∈ 0,1 is a cohort dummy,
and γZi jc is an error term. f (agei jc,I(20)i jc,Ci j) is a function of the running variable agei jc
(normalised to 0 at the 20th birthday) and the age and cohort thresholds. In our main speci-
fication, this function equals
( )
f (agei jc,I(20)i jc,Ci j) = age β
Z
i jc 5 +β
Z
6 I(20)i jc +β
Z
7 C +β
Z
i j 8 I(20)i jc ×Ci j (4)
The coefficient of main interest in Eq. (3) is β Z4 , the diff-in-disc estimate. As we rescale all
care utilisation variables to reflect annual averages, the sharp diff-in-disc coefficient provides
an estimate of the effect of the DCT fee on the number of consultations per year (e.g., a value
of 1 implies one additional consultation annually per capita).
The second stage equation for the number of in-person visits yi jc, in which the endoge-
nous DCTi jc is replaced by the prediction from the first stage equation, can be expressed as
follows:
14
yi jc = α1 +α2I(20) ˆi jc +α3Ci j +α4DCT i jc + f (agei jc,I(20)i jc,Ci j)+ εi jc (5)
where εi jc is an error term and all other variables are defined as above. The fuzzy diff-in-
disc estimate α = β y/β DCT4 4 4 can be interpreted as the degree of substitution between online
and in-person consultations. α4 = −1 implies that each DCT consultation replaces exactly
one in-person visit. If α4 < −1, then each online consultation replaces more than one in-
person visit; this might occur for problems for which regular PCCs, but not DCT companies,
would provide both an initial and a follow-up consultation. α4 ∈ (−1,0) implies that each
DCT consultation offsets less than one in-person visit. In this case, the net effect of the
availability of DCT is an increase in the total number of physician consultations (DCT +
in-person).
In our main estimations we use a fixed bandwidth of 120 days each side of the 20th
birthday. To ensure that our results are not dependent on the bandwidth, we also estimate the
model across a range of alternative bandwidths as well as using an optimal bandwidth that
is chosen by a data-driven procedure minimising the mean square error of the reduced form
equation of in-person visits y (Calonico, Cattaneo and Titiunik, 2014).
We cluster standard errors on the running variable (age in days relative to 20th birthday),
using separate clusters for the young and old cohorts (Lee and Card, 2008).22 With standard
errors clustered at the daily level, we may greatly save computational time – without affecting
the point estimates or standard errors – by estimating the model on aggregated data. We
therefore collapse the individual-day-level data to cells defined by age (in days relative to
the 20th birthday), gender, region, and time period,23 and include frequency weights (=
22A recent literature discusses methods to obtain bias-corrected estimates and robust confidence intervals
for settings with data-driven bandwidth choices standard RD settings (Calonico, Cattaneo and Titiunik, 2014;
He and Bartalotti, 2020). Such methods are yet to be developed for the diff-in-disc setting. However, in a
robustness check we modify the wild bootstrap procedure of He and Bartalotti (2020) to fit our fuzzy diff-in-
disc setting.
23Time periods are equivalent to calendar year for the younger cohorts (who are 20 (± 365 days) years old at
any point in 2017 or 2018). For the old cohort (who are 20 (± 365 days) years old at any point in the pre-DCT
15
the number of individuals in each cell) in the estimations. To examine if auto-correlation
in the age dimension is a problem in our main specification, we also estimate a model on
individual-level data in which we cluster standard errors by individual.
3.4 Variable definitions
3.4.1 DCT consultations and in-person visits
Our main outcome variable counts the daily number of in-person physician consultations
(visits) at a PCC in the patient’s region of residence. To interpret the measure as the annual
number of visits, the number of consultations are multiplied by 365. To measure the number
of DCT consultations per day, we count the daily number of contacts with PCCs in the two
regions where DCT providers were located (Jönköping and Sörmland).24 This approach
slightly overestimates the number of online physician consultations, as the out-of-region
data does not allow us to distinguish between online and in-person consultations. Auxiliary
analyses in Appendix A show that the overestimation is completely inconsequential, which is
expected as the difference should by and large be purged out by the diff-in-disc estimator.25
In our main analysis, we include all consultations regardless of diagnosis. In sub-analyses,
we use data from Region Jönköping to look specifically at diagnoses that are commonly set
by DCT providers. Our definition of common DCT diagnoses cover roughly 90% of all
DCT consultations with a physician.26 We also divide the set of common diagnoses into
period July 1 2012 to June 30 2016), we define four 365-day time periods, each running from July 1 in year t
to June 31 in t +1 for t ∈ (2012 to 2015).
24The measure also includes a small number of DCT contacts with a provider that was located in a third
region (Region Skåne) before it moved to Jönköping, and with the public online service in Region Västra
Götaland. The age threshold for the user fee in these Regions are the same as in Region Jönköping.
25Using additional data obtained from Region Jönköping, where the age based user fee was applied, we note
that online consultations in this region account for almost 90% of the consultations in our preferred measure,
and that 9 out of 10 online consultations was with a physician (rather than a nurse etc). Our first stage estimates
are practically the same when we use data on out-of-region contacts from registers held by Region Stockholm
and Region Västra Götland and when we use data from Region Jönköping (see Appendix A)
26Common diagnoses = ICD-codes (on a three-digit level) that cover 80% of the registered diagnoses for
19– and 20– year-olds during online consultations with private providers, including diagnoses within the same
16
four subsets: upper respiratory infections, skin conditions, diagnoses related to genital and
reproductive organs, and a residual category (other), enabling us to estimate the degree of
substitution within diagnosis type.
In further analyses, we decompose the consultations by their stage in the care episode
(initial vs follow-up), and study outcomes such as the daily number of other in-person visits
(nurse visits at a PCC; visits at a midwife in primary care, or a visit at a youth clinic or a
clinic specialised in sexually transmitted diseases (STD-clinic)).
3.4.2 Background variables
We use the following predetermined background variables: Father (mother) with university
education (Y/N), father (mother) with income above national median (Y/N), at least one
parent born outside Scandinavia (Y/N), number of physician visits at age 18 (0/1/>1), ru-
rality of municipality of residence (sparsely populated / densely populated / metropolitan)27.
Descriptive statistics summarising these characteristics as well as the physician visits are
provided in Appendix Table C.1.
Individuals who used DCT in 2018 differ from persons that did not (non-users) in terms
of these background variables. For instance, when they were 18 years old, the DCT users
made almost twice as many in-person visits as non-users in their birth cohort. Women and
city residents are clearly overrepresented among DCT users. Further, a slightly larger share
of DCT users have parents that were born in Sweden, have high income or high education.
To say something about the generalisability of our analysis, we compare our study pop-
ulation to other parts of the age distribution in Appendix D. The 19-20 year-olds are quite
similar to other adolescents and young adults (<35) in terms of expected health care costs.
The share of individuals with a DCT-relevant diagnosis is similar for a much wider age range,
ICD-block with more than 10 registered episodes. See Appendix B.
27The rurality variable follows Statistics Sweden’s definition. The metropolitan category includes the city of
Stockholm, the city of Gothenburg, and municipalities close to these cities.
17
up to 49 years.
4 Results
4.1 Main results and robustness
4.1.1 Graphical evidence
The first step in the analysis is to examine if the use of DCT services changes discontinuously
at the 20th birthday, when the patient starts paying a user fee. Figure 2 shows the annual
number of DCT consultations per capita in different years,28 sorted by the day of the online
contact relative to the individual’s 20th birthday (day ”zero”). The subgraphs illustrate the
evolution of the DCT market. Starting from a situation with virtually no consumption prior
to July 2016, the market started to take off and expanded further in 2017 and 2018.29
[Figure 2 about here.]
Overall, the figures support that the onset of the DCT fee at the 20th-birthday reduced
the demand for DCT services. The black regression lines are estimated using 120 days on
each side of the user fee threshold (reflecting how the data portrayed in the figure are used
to obtain diff-in-disc estimates in our formal estimations). The regression lines illustrate a
small jump at the 20th birthday in both 2017 and 2018. In 2018, when the market had gained
some size, the drop at the threshold corresponded to about 50% of the annual average of .3
DCT consultations per capita for individuals aged below 20. The rapid growth of the market
makes it difficult to pool data for 2017 and 2018. In our further analyses, we therefore focus
28For each day relative to the 20th birthday, the number of consultations is multiplied by 365 to give an
annual interpretation.
29The definition of DCT consumption includes physicial visits at PCCs in Region Jönköping made by our
study population. Such visits account for the non-zero consumption in the preperiods.
18
on the comparison of the pooled pre-period and 2018. (An analysis for 2017 is presented in
Appendix E.)
Next, we examine if the reduction in DCT consultations at the 20th birthday is comple-
mented by an increase in in-person consultations, i.e., if there is evidence of substitution.
Figure 3 plots annual in-person physician visits per capita,30 by the day relative to the 20th
birthday. Subtracting the regression discontinuity in the left graph (pre-period) 31 from the
discontinuity in the right graph (post period, 2018) yields the sharp diff-in-disc of interest.
The graphs suggest that DCT partially substitutes for in-person consultations. In the pre-
period, the 20th birthday was associated with a drop in the number of in-person visits. In
2018, when the DCT market had exploded, there was no longer a notable drop in in-person
visits at the age threshold.32 Combined with the distinct drop in the number of DCT consulta-
tions documented in Figure 2, this difference in the discontinuities in 2018 and the pre-DCT
period suggests that the PCCs absorb some of the demand served by DCT companies in the
absence of a fee.
[Figure 3 about here.]
4.1.2 Main results: Fuzzy diff-in-disc
The first column in Table 2 presents the main results from the fuzzy diff-in-disc estimation
using a bandwidth of 120 days. In line with the graphical analysis, the precisely estimated
sharp diff-in-disc in Panel A confirms that the onset of the user fee reduces the number of
DCT consultations by .15 visits per year, corresponding to a decrease of about 50% compared
to the pre-20 mean. The F-statistic of 163.8 suggests that the effect of the user fee on DCT
30As before, the statistics are scaled by 365 to allow an annual interpretation.
31Appendix Figure H.5 shows that there were similar jumps in each of the years in the pooled pre-period.
32Appendix Figure E.1 shows that the decrease of in-person visits at the 20th birthday was similar in 2017,
when the DCT market still had limited outreach, as in the pre-period. In same the Appendix, we also provide a
formal analysis for 2017.
19
consultations is strong enough to allow us to use it to study the substitution between DCT
and in-person care.
Panel B captures the effect of the DCT user fee on the number of in-person visits at the
20th birthday (i.e., the reduced form estimate). The coefficient equals 0.067 and is statisti-
cally significant at the 5% level. The third row presents our fuzzy diff-in-disc (IV) estimate,
which can be interpreted as the degree of substitution. The estimate of -.45 suggests that
roughly every other DCT consultation replaces an in-person visit. The corresponding 95%
confidence interval covers neither zero (0) nor full (-1) substitution.
[Table 2 about here.]
4.1.3 Heterogeneity
The descriptive statistics (Appendix C) as well as further analyses of compliers and first
stage heterogeneity (Appendix F) show that the difference between women and men is the
most notable of all subgroups. The second and third columns of Table 2 show estimates by
gender. The first stage estimates (Panel A) suggest that women’s DCT use decreases much
more than men’s following the onset of the DCT user fee. This reflects that women use
DCT more in general; in relative terms, both genders react similarly to the fee. As the sharp
diff-in-discs for in-person visits (Panel B) are similar for both genders, the estimated degree
of substitution (Panel C) is considerably lower for women than for men. For women, we
cannot reject the null hypothesis of no substitution. For men, the results indicate that all
DCT consultations represent substitution from in-person visits, although the standard error
of the estimate is very large.
Apart from the gender differences, we find that the groups that are overrepresented
among DCT users — individuals residing in urban areas, and individuals who visited a
physician relatively often in the past – are also overrepresented among the compliers. That
is, these groups aggravate the (negative) estimate of the effect of the fee on the number of
20
DCT consultations. Notably though, all subgroups display similar relative decreases of DCT
consultations due to the fee (about 50%).
Appendix G presents analyses of the degree of substitution for the subgroups with larger
shares of compliers. The results suggest that there is no qualitative difference in the substitu-
tion for individuals with different levels of previous care utilisation. The degree of substitu-
tion is relatively higher (and more precisely estimated) for individuals living in urban areas
than for individuals in rural areas. Thus, the overall degree of substitution largely reflects the
degree of substitution among urban residents.33
4.1.4 Robustness and precision
Revisiting the scatter plots in Figure 2 and 3, we note that while the relationships between
age and the outcome variables are also almost flat over the two years surrounding the 20th
birthday, the black regression lines within the 120-day bandwidths may be influenced by
temporary fluctuations. We therefore examine the stability of the results with alternative
specifications using various bandwidths and a 0-degree polynomial in the running variable
(i.e., comparing difference in mean level of care utilisation each side of the threshold for the
two cohorts, see Appendix H.1).
[Figure 4 about here.]
Bandwidth: In line with the scatter plots in Figure 2, the baseline estimate of the effect
of the onset of the user fee on DCT consultations (first stage) is similar to the ones obtained
using other fixed bandwidths up to 365 days (see the upper left graph in Figure 4). The
baseline estimate of the substitution (IV) is very similar to the estimates obtained using
bandwidths between 90 and 180 days (shown in the lower left graph). Both the first stage
33An analysis by region (also in Appendix G) suggests that the degree of substitution is larger in Region
Stockholm than in Region Västra Götaland, but that it is mainly due to the large share of urban residents in
Region Stockholm.
21
and IV estimates are similar when using a data-driven procedure to choose an MSE-optimal
bandwidth (Calonico, Cattaneo and Titiunik, 2014). We allow the MSE-optimal bandwidth
to vary each side of the threshold and in the pre- and post period, but all lie within the range
of 105 and 122 (i.e., similar to the bandwidth in our main specification). The degree of
substitution is slightly smaller for bandwidths outside of this window.
Intertemporal substitution of care consumption is a potential threat to our identification
strategy. Individuals who anticipate the fee increase may decide to contact DCT providers
before their 20th birthday rather than afterwards. Such behaviour would lead us to overesti-
mate the effect of the user fee on DCT consultations, and consequently to underestimate the
degree of substitution. The upper right graph in Figure 4 presents coefficients from donut
estimations of the first-stage relationship, showing that removing up to 28 days on each side
of the threshold changes the coefficient little if at all. Given the dominance of sudden and
acute (though not necessarily severe) conditions among the DCT consultations, it is arguably
unrealistic that individuals to a large extent would reschedule physician appointments over a
period of more than four–eight weeks to avoid a fee.34 Thus, there is, at worst, very limited
intertemporal substitution of DCT.
For in-person visits, our diff-in-disc strategy accounts for intertemporal substitution if
such behavior is stable over time. The lower right subgraph in Figure 4 shows that removing
a donut of up to 28 days on each side of the threshold has only small effects on the size of
the fuzzy diff-in-disc estimate, although the precision decreases.35
Simple mean comparison: Our main estimates may be sensitive to the parametric spec-
34In Appendix H.2, we show that for larger outer bandwidths the first stage coefficient is also robust to
removing much larger donuts. Looking specifically at skin conditions, a conditions for which intertemporal
substitution would more be likely, the data also suggest that if intertemporal substitution occurs, it is limited to
the weeks closest to the threshold (see Appendix H.3)
35Extending the donut further within the 120-day bandwidth leads to small sample sizes and noisy estimates.
In Appendix H.2, we show that for larger bandwidths (180/365 days), the results are stable when removing
up to seven weeks on each side of the 20th birthday. Results are stable for even larger donuts when using a
zero-degree polynomial instead of a linear specification.
22
ification of the regression lines on each side of the threshold. Given the graphical evidence
suggesting no clear trends over the two years closest the 20th birthday, it seems most plausi-
ble to specify a zero-degree polynomial in the running variable (see Robust inference below
for higher-order polynomials). Across different bandwidths, this exercise produces point
estimates that are in the similar range as the main estimate, but more precisely estimated
(Appendix H.1). Notably, these estimates should be more robust to intertemporal substitu-
tion than our baseline estimates, as fluctuations within the bandwidth are smoothed out in
this specification.
Conditional expectations: As our sample consists of a panel of individuals followed
(up to) 240 days, we do not expect any discontinuities in the distribution of predetermined
background variables around the threshold. The balance test presented in Appendix H.4
confirms that there are no differences of meaningful size at the threshold. Including these
covariates in the diff-in-disc model does not affect neither the size nor the precision of the
estimates (see Appendix H.5).
Trends and time invariant confounding: As we use the discontinuity of the pre-DCT
cohorts to purge out effects of any confounding treatments, a crucial assumption is that the
effects of confounding treatments are time invariant, which in turn relies on the pre- and
post-cohorts being largely similar. A placebo analysis using other birthdays as the threshold
in a sharp diff-in-disc is a strong test of this assumption. In Appendix H.6, we show that
sharp diff-in-discs estimated around the 19th or 21st birthdays are zero for both DCT and
in-person consultations. This indicates that our main estimate is not a result of general
differences between the cohorts. Figure H.5 shows that the discontinuity at the 20th birthday
is of similar size when we split the pre-period into four 12 month windows, lending further
support to our strategy. In Appendix table H.4, we also re-estimate the fuzzy diff-in-disc
after subsequently deleting up to three out of the (baseline) four pre-cohorts. In all cases,
the estimate suggests a substitution rate of around 50%. Finally, in appendix H.8, we also
23
estimate the impact of the potential bias from the secular decrease in the number of in-person
visits between the pre- and the post-period. We find that if there is such bias, it would likely
be small (reducing the degree of substitution from 45 to 42%).
An alternative strategy that relaxes the assumption of time-invariant effects of confound-
ing policies would be to focus exclusively on Region Stockholm, where there was no con-
founding change of in-person user fees at age 20. Assuming there are no other confounding
policies at age 20, we can then use an ordinary fuzzy RD specification (instead of a diff-in-
disc). Appendix Table H.5 shows that such a specification yields a similar estimate (-.48).
Robust inference: A recent literature discusses methods to obtain bias-corrected esti-
mates and robust confidence intervals for settings with data-driven bandwidth choices in
standard RD settings (Calonico, Cattaneo and Titiunik, 2014; He and Bartalotti, 2020). Such
methods are yet to be developed for a fuzzy diff-in-disc setting. However, we obtain similar
confidence intervals for the optimal bandwidth when modifying the wild bootstrap procedure
of He and Bartalotti (2020) to fit our setting (Appendix H.10). Notably, this approach uses a
second-order polynomial to assess bias, and thus also mitigates concerns that the first-order
polynomial in our baseline specification is not flexible enough.
Autocorrelation: To address worries that the standard errors of the preferred model dis-
regard autocorrelation in the age dimension, we estimate the main model on individual-level
data with standard errors clustered by individual (Table H.7). This leads to a small increase
in the standard error of the point estimate (.22, compared to .20).
approximately half
In sum, our robustness checks support the conclusion that approximately half of all DCT
consultations reflect substitution from in-person visits in regular primary care.
24
4.2 Extensions
4.2.1 Substitution at various stages of the care episode
The estimated degree of substitution of 45% implies that the remaining 55% of the DCT con-
sultations in the no user fee scenario represent additional, induced consumption. To examine
where in the care episode these induced consultations take place, Figure 5 displays sharp
diff-in-disc estimates of the DCT user fee on DCT and in-person consultations, decomposed
by whether the consultation is the initial contact in a care episode or a follow-up. An initial
consultation is defined as a consultation with no DCT or in-person consultation during the
preceding 14 days.36 The remaining consultations are defined as follow-ups (although some
of these in practice may not belong to the same care episode). The onset of the user fee leads
to a reduction in initial DCT consultations – the effect on one-off encounters equals -.093,
and the effect on initial DCTs with a follow-up within 14 days equals -.021 – as well as in
DCT follow-ups (-.035). While these reductions are complemented by increases in initial
in-person visits (of 0.066 and 0.010), the effect on follow-up in-person visits is practically
zero.
Of the (approximately) 55% induced consultations, we thus find that 24 percentage points
are one-off encounters with no follow-up, and so the remaining part belong to multiple-
encounter episodes. Initial consultations in an episode (with a follow-up within 14 days)
account for 7 percentage points and follow-up consultations for the remaining 23 percentage
points.
Notably, the zero effect on in-person follow-ups is the net effect of an increase in follow-
ups that occur after an initial in-person visit and a decrease in follow-ups that occur after
an initial DCT.37 The decrease in the latter type of follow-ups implies that the overall de-
36See Appendix I for wash-out periods of 7 and 28 days.
37Appendix I show that the zero net effect is a result of heterogeneous effects across genders. The increase
in follow-ups after an initial in-person visit is driven by men, and the decrease in follow-ups after an initial
DCT consultation is driven by women.
25
gree of substitution is limited by initial DCT consultations that later are followed by one or
more in-person visits, potentially reflecting cases where a physical examination is necessary.
However, this effect is rather limited: such consultations account for about 8% of the addi-
tional consultations in the absence of a user fee. Thus, most of the additional consultations
do not reflect extra steps on the paths towards an unavoidable in-person visit. Consequently,
other explanations, such as the lower threshold for seeking care in the DCT setting, are more
important factors to explain why the total number of consultations increase.
[Figure 5 about here.]
4.2.2 Substitution within diagnosis types
We next focus on consultations with diagnoses that are commonly set during DCT con-
sultations. The 80% most common diagnoses cover 89% of all DCT consultations in the
estimation sample. In a specification including only these diagnoses, shown in the leftmost
column of Table 3, the estimated degree of substitution is very similar to our main estimate.
The other columns of Table 3 show estimates from models in which we relate DCT consul-
tations to in-person visits within the same diagnosis category: Upper respiratory infections
(Resp), skin related problems (Skin), genital and reproductive organs Gen/Rep, and a resid-
ual category capturing the Other common diagnoses set by DCT providers. These categories
cover 19% 24%, 21%, and 27% of all DCT consultations in the post period.38 The first stage
coefficients stand in proportion to each category’s share of all DCT consultations, but the
estimated degree of substitution varies between categories.39
The highest degree of substitution is obtained for upper respiratory infections. The es-
timate suggests that DCT consultations for respiratory infections (more than) fully replace
in-person consultations. The estimates for skin and genital/reproductive diagnoses are small
38The subsets are overlapping to some extent, because more than one diagnosis may be set during a given
consultation.
39These results are similar when using longer bandwidths, see Appendix J.
26
and not statistically significant.40 Although we cannot rule out partial substitution of sim-
ilar magnitudes as before, we believe it is plausible to find more induced consultations for
this type of diagnoses. First, the convenience of DCT may tip the balance for individuals
that would not have bothered to have a rash or eczema examined, if they would have had to
travel to the PCC for the examination. Further, for minor skin conditions in this category,
the triage nurse at the PCC might refuse to schedule a physician appointment (possibly of-
fering a nurse examination instead, or give self-care advice on the phone). Second, the gen-
ital/reproductive diagnoses include many contraceptive prescription renewals. In Sweden,
contraceptive management is normally handled by midwives in traditional primary care, so
there are few physician in-person visits that the DCT consultations may replace.
Notably, the gen/rep category mainly includes diagnoses that are only relevant for women.
For this reason, the table also presents estimates by gender. We note that, as expected, the
first stage estimate for men for conditions in the gen/rep category is very weak. We fur-
ther note that in the specification for the most relevant population (women), there is still no
evidence of substitution in this category.
For other DCT-relevant diagnoses (including, e.g., vague symptoms, mental problems,
and renewal of prescriptions) the degree of substitution is similar to that for the overall
category (Common), but the estimate is not statistically significant at conventional levels.
[Table 3 about here.]
40Given the prevalence of upper respiratory infections, skin conditions (acne) and genital and reproductive
health (cystitis) among DCT diagnoses, it is possible that DCT affects antibiotic use. Appendix K shows that
the onset of the user fee is not associated with a decrease in antibiotic prescriptions. Thus, DCT physicians
seem to be at least as restrictive as other physicians in terms of antibiotics prescriptions.This is consistent
with recent descriptive evidence (Entezarjou et al., 2021; Shi et al., 2018), though earlier studies indicated
otherwise (Uscher-Pines et al., 2015). Notably, the result holds also for the subset of antibiotics prescribed
against respiratory infections. This is consistent with the medical guidelines, that advocate against prescriptions
without a physical examination.
27
4.2.3 Other consultations and antibiotic prescriptions
Acknowleding the role of other professional categories (nurses and midwives) in Swedish
primary care, Table 4 presents estimates of the degree of substitution for various types of
other visits: Nurse visits at a primary care centre, visits at a midwife, youth or STD clinic,
and the sum of these two types of visits. To grasp the total degree of substitution across all
categories, we lastly sum of these two types with in-person physician visits at a PCC.
The results do not indicate that the additional DCT consultations found in the main anal-
ysis reflect visits that would otherwise have taken place in another part of the traditional
primary care setting. The estimated substitution is low for both nurse visits and visits at
midwife/youth/STD clinics. The specification including all professions together (All consul-
tations) indicates that the total substitution rate is slightly higher than in the main analysis.
Splitting the sample by gender, the point estimates suggest that women to some extent
substitute across professional categories (DCT physician vs midwife in primary care), al-
though the estimate is statistically significant. In Appendix L, we explore this further and
find that the estimate is closer to -1 when including only DCT consultations related to gen-
ital/reproductive issues. Another interesting patterns is that DCT consultations appear to
induce subsequent consultations at youth/STD clinics among men.41
[Table 4 about here.]
4.2.4 Cost analysis
Our main analysis suggests that the availability of DCT leads to more physician consulta-
tions, making it interesting to examine whether DCT also leads to increasing health care
costs.42 Given a substitution rate of approximately 45%, DCT would not increase costs if
41Although imprecise, these patterns are robust to using other bandwidths, see Appendix L.
42A full economic evaluation of the effects of DCT would not only include a comparison of costs, but also
of benefits. In particular, the two services may differ in quality. However, we have limited opportunities to
28
the unit costs in the DCT setting are at most 45% of those in the in-person setting.
In Appendix M, we compare the unit costs of a visit using the best available estimates
of costs and time use from the different settings. This implies lower wages and shorter
consultation times in the DCT setting, although we also examine costs under assumptions
of similar wages and consultation times.43 The cost analysis abstracts from all fixed costs
except for office space.44
Our baseline estimate indicates that the direct cost of a DCT consultation is 36% of that
for an in-person consultation, i.e., low enough not to add to total health care costs despite the
induced demand. The ratio reduces to 30% when we take into account the value of the time
saved by patients in DCT. In more conservative scenarios that assume similar wages and/or
consultation times in the two modes of care, the health care cost ratio rises to 42-52%. The
only scenario in which the additional consumption implies increasing health care costs is the
one with equal wages and consultation times. Notably, DCT is still resource saving from a
societal perspective (i.e., when adding the value of patients’ time savings) in this scenario.
We also note that in our study setting, the reimbursement per DCT consultation from the
public third-party payer is around twice as high as the cost for the physician time. Although
the reimbursement clearly must be high enough to cover fixed costs, this large differential
address the less tangible benefits, such as the reduced worry of DCT patients who would not otherwise see a
doctor at a PCC, or the potentially higher quality of in-person consultations for patient who would.
43The wage differential reflects that DCT firms employ less senior staff, which is not so interesting in a
general equilibrium perspective. Further, for patients who do not need to be physically examined (i.e., patients
whose care need is well suited for online treatment), there are no strong reasons to believe that the consultation
time would differ considerably between the two modes of care. Our choice of using a longer time in PCC in the
baseline calculation reflects that the in-person setting may by itself prompt physicians to physically examine
patients in borderline cases when it may not be strictly necessary, and to spend some minutes accompanying
patients to/from the waiting room.
44To realize savings from costs for management, lab equipment and a receptionist in the in-person setting,
the PCC would have to shut down. In our setting, it would be politically infeasible to realize such savings on
a large scale. It thus makes sense to abstract from these costs. When it comes to fixed costs that are specific
for the DCT setting, the key costs stem from the development and maintenance of the consultation platform.
For the purpose of comparing the costs to produce care in the two settings, it is less consequential than it may
seem to omit these costs, as PCCs nowadays also offer online consultations using similar platforms. Although,
of course, the overall costs would have been lower if there was only one platform.
29
suggests that the health care authorities would be able to reduce the reimbursement and still
make it worthwhile for the larger DCT companies to stay in the market.
5 Concluding remarks
We show that the demand for DCT consultations falls by half as individuals reach the age at
which they start to pay a consultation fee. Exploiting this exogenous variation in demand,
we estimate the degree to which online consultations substitute for in-person physician con-
sultations at the primary care practice. Our estimates imply that slightly less than 50% of
all online consultations replace in-person visits. Consequently, the availability of a direct-to-
consumer telemedicine market increases the total number of physician consultations (online
and in-person). This conclusion is robust to different bandwidths, functional form assump-
tions, and methods for estimating standard errors.
We do not find that the substitution rate is limited to a great extent by DCT consultations
that only function as an additional step before an in-person visit. This suggests that the ad-
ditional demand primarily stems from the lower barriers of access to DCT services. The fact
that half of the additional consultations are one-off encounters with no follow-up within 14
days aligns with the notion that the convenience of DCT may increase the propensity to seek
care for less severe conditions. The particularly low substitution for diagnoses related to skin
conditions and reproductive health, which would normally be dealt with by other professions
in Swedish primary care (or by referring the patient to self-care), is also indicative of lower
barriers to access.
Compared to previous related research, an important strength of the study is that our
identification strategy is better equipped to deal with time-variant unobserved heterogeneity
such as transitory health issues. In line with this, we also estimate a relatively high degree of
substitution. However, the differences may also relate to the external validity of our analysis.
30
Here, the study population and the study context are of particular concern.
With regards to the study population, our identification strategy captures causal effects
for individuals around 20 years of age that respond to the DCT user fee. A complier analysis
suggests that groups that are more frequent users of DCT in the first place (women, urban
residents, and users of in-person care), are also more likely to respond to the onset of the
user fee. As shown in Appendix D, the studied age group is similar to the (rest of the) 15-34
age group – a large fraction of DCT users – in terms of the proportion with a DCT-relevant
diagnosis and expected health care spending. It thus appears reasonable to generalise our
results at least to other young adults.
The Covid-19 pandemic has increased the tendency of both patients and health care
providers to adopt a ‘digital-first’ approach, and might thus have affected the degree of sub-
stitution. However, our results are likely relevant also in the post-pandemic era, as the young
individuals we study were among the most prominent early adopters already before the pan-
demic.
Across contexts, differences in incentives facing patients and provider may affect the
degree of substitution. On the patient side, the setting with heavily subsidised access to
online as well as traditional care is descriptive of many health systems. On the provider side,
the Swedish health care system is characterised by relatively few primary care physicians,
task shifting towards nurses, and a strong reliance on fixed payments rather than fee-for-
service. The scope for substitution might be greater in contexts where such factors, which
limit the access to physician consultations, are less pronounced. Notably though, we do not
find evidence of much more substitution when considering nurse and physician consultations
together, and the results are similar in our two study regions, which have different incentive
structures. These results suggest that our main findings are relevant outside the particular
study context.
Our preferred estimate of the substitution rate implies that if the unit cost of a DCT
31
consultation is at most 45% of the unit cost of an in-person visit, then the DCT market has
the potential to increase the care volume while leaving total costs unaffected. Our back-of-
the-envelope estimates suggest that the production cost of a DCT consultation is likely low
enough to enable this criterion to be fulfilled, and almost surely so when taking into account
the value of the time saved by patients.
However, due to the (positive) gap between the reimbursement and production costs in
our study setting, third-party payers likely pay more than they need for DCT consultations at
current care volumes. The overpayments to DCT providers have thus come at the expense of
individuals with greater health needs, for whom online consultations are inadequate (Roland,
2019), and, to the extent that total costs have been allowed to increase, at the expense of
the insured population (tax payers). Arguments related to both efficiency (deadweight loss
of taxes) and equity (larger needs of patients unsuitable for DCT (Roland, 2019)) might
justify another distribution of the gains from DCT, which at present fall exclusively on DCT
providers and on DCT users.
From a health policy perspective, it may also be a concern that about half of all spending
on DCT goes to patients who would not otherwise have seen a doctor.45 A potential policy
implication is that the regulation of the DCT market should incorporate similar incentives
for nurse triage as in traditional primary care. A further step in that direction might be to
apply higher reimbursement rates for consultations with higher priority, or with the cheapest
equivalent staff category.
References
AHIP. 2019. “Virtual Care Delivers Value.” https://www.ahip.org/virtual-care-delivers-
value/ Accessed: 2021-05-10.
45In the Swedish case, this would imply that 1.5% of the expenditure on primary care physician consultations
in 2019 was redirected in this fashion.
32
Anell, Anders, Anna H Glenngård, and Sherry M Merkur. 2012. “Sweden: Health sys-
tem review.” Health systems in transition, 14(5): 1–159.
Ashwood, J. Scott, Ateev Mehrotra, David Cowling, and Lori Uscher-Pines. 2017.
“Direct-To-Consumer Telehealth May Increase Access To Care But Does Not Decrease
Spending.” Health Affairs, 36(3): 485–491.
Bavafa, Hessam, Lorin M. Hitt, and Christian Terwiesch. 2018. “The Impact of E-Visits
on Visit Frequencies and Patient Health: Evidence from Primary Care.” Management Sci-
ence, 64(12): 5461–5480.
Brot-Goldberg, Zarek C., Amitabh Chandra, Benjamin R. Handel, and Jonathan T.
Kolstad. 2017. “What does a Deductible Do? The Impact of Cost-Sharing on Health
Care Prices, Quantities, and Spending Dynamics*.” The Quarterly Journal of Economics,
132(3): 1261–1318.
Bundorf, M Kate. 2012. “Consumer-directed health plans: Do they deliver?” The Robert
Wood Johnson Foundation Research Synthesis Report 24.
Calonico, Sebastian, Matias D Cattaneo, and Rocio Titiunik. 2014. “Robust nonparamet-
ric confidence intervals for regression-discontinuity designs.” Econometrica, 82(6): 2295–
2326.
Cattaneo, Matias D., Nicolás Idrobo, and Rocı́o Titiunik. 2019. A Practical Introduc-
tion to Regression Discontinuity Designs: Foundations. Cambridge:Cambridge University
Press.
Chandra, Amitabh, Jonathan Gruber, and Robin McKnight. 2014. “The impact of pa-
tient cost-sharing on low-income populations: Evidence from Massachusetts.” Journal of
Health Economics, 33: 57–66.
33
Cutler, David M., Sayeh Nikpay, and Robert S. Huckman. 2020. “The Business of
Medicine in the Era of COVID-19.” JAMA, 323(20): 2003.
Dahlgren, Cecilia, Margareta Dackehag, Per Wändell, and Clas Rehnberg. 2021. “De-
terminants for use of direct-to-consumer telemedicine consultations in primary health-
care—a registry based total population study from Stockholm, Sweden.” BMC family
practice, 22(1): 1–10.
Dills, Angela K, and Jiajia Chen. 2018. “Does Telemedicine Save Lives? Evidence on
the Effect of Telemedicine Parity Laws on Mortality Rates.” Working Paper, Available at
SSRN: https://ssrn.com/abstract=3272821.
Ekman, Björn. 2018. “Cost Analysis of a Digital Health Care Model in Sweden.” Pharma-
coEconomics - Open, 2(3): 347–354.
Ellegård, Lina Maria, and Gustav Kjellsson. 2019. “Nätvårdsanvändare i Skåne kontaktar
oftare vårdcentral och gör inte färre akutbesök.” Läkartidningen, 116.
Entezarjou, Artin, Maria Sjöbeck, Patrik Midlöv, Veronica Milos Nymberg, Lina Vi-
gren, Ashkan Labaf, Ulf Jakobsson, and Susanna Calling. 2022. “Health care utiliza-
tion following “digi-physical” assessment compared to physical assessment for infectious
symptoms in primary care.” BMC Primary Care, 23(1): 4.
Entezarjou, Artin, Susanna Calling, Tapomita Bhattacharyya, Veronica Milos
Nymberg, Lina Vigren, Ashkan Labaf, Ulf Jakobsson, and Patrik Midlöv. 2021. “An-
tibiotic Prescription Rates After eVisits Versus Office Visits in Primary Care: Observa-
tional Study.” JMIR Medical Informatics, 9(3): e25473.
Galindo-Silva, Hector, Nibene Habib Some, and Guy Tchuente. 2019. “Does Obamacare
Care? A Fuzzy Difference-in-discontinuities Approach.”
34
Grecu, Anca M, and Ghanshyam Sharma. 2019. “The effect of telehealth insurance man-
dates on health-care utilization and outcomes.” Applied Economics, 51(56): 5972–5985.
Grembi, Veronica, Tommaso Nannicini, and Ugo Troiano. 2016. “Do Fiscal Rules Mat-
ter?” American Economic Journal: Applied Economics, 8(3): 1–30.
Heckley, Gawain, Ulf-G. Gerdtham, and Johan Jarl. 2018. “Too Young to Die: Regres-
sion Discontinuity of a Two-Part Minimum Legal Drinking Age Policy and the Causal
Effect of Alcohol on Health.” Working Paper, Department of Economics, Lund University
2018:4.
He, Yang, and Otavio Bartalotti. 2020. “Wild Bootstrap for Fuzzy Regression Discon-
tinuity Designs: Obtaining Robust Bias-Corrected Confidence Intervals.” Econometrics
Journal, 23(2): 211–231.
Johansson, Naimi, Niklas Jakobsson, and Mikael Svensson. 2019. “Effects of primary
care cost-sharing among young adults: varying impact across income groups and gender.”
The European Journal of Health Economics, 20(8): 1271–1280.
Kry. 2019. “Kvalitetsrapport januari–september 2019 Sverige.”
https://www.kry.se/medicinsk-kvalitet/kvalitetsrapport-q1-q3-2019/ Last accessed:
2021-06-04.
Lee, David S., and David Card. 2008. “Regression discontinuity inference with specifica-
tion error.” Journal of Econometrics, 142(2): 655–674.
Li, Kathleen Yinran, Ziwei Zhu, Sophia Ng, and Chad Ellimoottil. 2021. “Direct-To-
Consumer Telemedicine Visits For Acute Respiratory Infections Linked To More Down-
stream Visits.” Health Affairs (Project Hope), 40(4): 596–602.
35
Manning, Willard G., Joseph P. Newhouse, Naihua Duan, Emmett B. Keeler, and Ar-
leen Leibowitz. 1987. “Health Insurance and the Demand for Medical Care: Evidence
from a Randomized Experiment.” The American Economic Review, 77(3): 251–277. Pub-
lisher: American Economic Association.
Martinez, Kathryn A., Mark Rood, Nikhyl Jhangiani, Lei Kou, Susannah Rose, Adri-
enne Boissy, and Michael B. Rothberg. 2018. “Patterns of Use and Correlates of Patient
Satisfaction with a Large Nationwide Direct to Consumer Telemedicine Service.” Journal
of General Internal Medicine, 33(10): 1768–1773.
Mehrotra, Ateev, R. Sacha Bhatia, and Centaine L. Snoswell. 2021. “Paying for
Telemedicine After the Pandemic.” JAMA, 325(5): 431.
Millán-Quijano, Jaime. 2020. “Fuzzy difference in discontinuities.” Applied Economics
Letters, 27(19): 1552–1555.
Newhouse, Joseph P., Rand Corporation Insurance Experiment Group, and Insurance
Experiment Group Staff. 1993. Free for All?: Lessons from the Rand Health Insurance
Experiment. Harvard University Press. Google-Books-ID: SVUJ4W9Lk5IC.
Nilsson, Anton, and Alexander Paul. 2018. “Patient cost-sharing, socioeconomic status,
and children’s health care utilization.” Journal of Health Economics, 59: 109–124.
Nord, Garrison, Kristin L. Rising, Roger A. Band, Brendan G. Carr, and Judd E.
Hollander. 2018. “On-demand synchronous audio video telemedicine visits are cost ef-
fective.” The American Journal of Emergency Medicine.
North, Frederick, Sarah J. Crane, Rajeev Chaudhry, Jon O. Ebbert, Karen Ytterberg,
Sidna M. Tulledge-Scheitel, and Robert J. Stroebel. 2014. “Impact of patient portal
secure messages and electronic visits on adult primary care office visits.” Telemedicine
Journal and E-Health, 20(3): 192–198.
36
Roland, Martin. 2019. “General practice by smartphone.” BMJ, 366.
SALAR, (Swedish Association of Local Authorities and Regions).
2017. “Sveriges Kommuner och Landsting – Sektorn i siffror.”
http://skl.se/ekonomijuridikstatistik/ekonomi/sektornisiffror.1821.html Last Accessed
2021-06-04.
SALAR, (Swedish Association of Local Authorities and Re-
gions). 2020. “Digitala vårdtjänster i primärvården.”
https://skr.se/halsasjukvard/ehalsa/digitalavardtjansteriprimarvarden.28301.html Last
Accessed: 2021-06-03.
Salisbury, Chris, Anna Quigley, Nick Hex, and Camille Aznar. 2020. “Private Video Con-
sultation Services and the Future of Primary Care.” Journal of Medical Internet Research,
22(10): e19415.
Shah, Sachin J., Lee H. Schwamm, Adam B. Cohen, Marcy R. Simoni, Juan Estrada,
Marcelo Matiello, Atheendar Venkataramani, and Sandhya K. Rao. 2018. “Virtual
Visits Partially Replaced In-Person Visits In An ACO-Based Medical Specialty Practice.”
Health Affairs, 37(12): 2045–2051.
Shi, Zhuo, Ateev Mehrotra, Courtney A. Gidengil, Sabrina J. Poon, Lori Uscher-Pines,
and Kristin N. Ray. 2018. “Quality Of Care For Acute Respiratory Infections During
Direct-To-Consumer Telemedicine Visits For Adults.” Health Affairs, 37(12): 2014–2023.
Uscher-Pines, Lori, Andrew Mulcahy, David Cowling, Gerald Hunter, Rachel Burns,
and Ateev Mehrotra. 2015. “Antibiotic Prescribing for Acute Respiratory Infections in
Direct-to-Consumer Telemedicine Visits.” JAMA internal medicine, 175(7): 1234–1235.
Zeltzer, Dan, Liran Einav, Joseph Rashba, and Ran D Balicer. 2021. “The Impact of
Increased Access to Telemedicine.” NBER Working Paper 28978.
37
Table 1: Summary of Institutional Setting
RS RVG
TRADITIONAL PCC
Reimbursement: Capitation + Per visit Capitation
User fee age limit: 18 20
User fee: C20 C10/30*
DCT (ONLINE) MARKET
Rembursement: Per visit Per visit
User fee age limit: 20 20
User fee: C25 C25
Note: The table describes the reimbursement system and user fees for in-person visits
and DCT consultations in the two study regions – Region Stockholm (RS) and Region
Västra Götaland (RVG) – during the study period. The DCT user fee was set by a third
region, Region Jönköping, where the two largest DCT providers were located during the
study period. * The lower fee applied if the patient visited the PCC at which (s)he was
registered as a patient.
38
Table 2: Fuzzy difference-in-discontinuity results
All Men Women
A. First stage (sharp diff-in-disc)
Online consultations -0.149*** -0.0700*** -0.234***
(0.0116) (0.0114) (0.0231)
F-stat 163.8 37.76 103.4
B. RF (sharp diff-in-disc)
In-person visits 0.0670** 0.0718* 0.0611
(0.0298) (0.0380) (0.0463)
C. IV (fuzzy diff-in-disc)
In-person visits -0.450** -1.026* -0.260
(0.200) (0.549) (0.196)
Avg. ind/day pre 151525 78130 73395
Avg. ind/day post 31490 16302 15188
Note: Variable names in left column represent the dependent
variable. Each model uses a fixed bandwidth of 120 days on ei-
ther side of the threshold (20th birthday). Avg. ind/day = average
number of individuals per day in cells, shown separately for the
pre cohorts (which include several birth cohorts) and the post co-
hort (which only includes individuals turning 20 in 2018). Data
are collapsed by region, gender, birth year and day relative to
20th birthday. Standard errors clustered by the running variable,
with separate clusters for pre-post periods. * P<0.1, ** P<0.05,
*** P<0.01.
39
Table 3: Decomposition by type of diagnosis, (bw=120)
A. DCT CONSULTATIONS (SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.131*** -0.0260*** -0.0424*** -0.0244*** -0.0457***
(0.0106) (0.00553) (0.00571) (0.00555) (0.00696)
Men -0.0603*** -0.00689 -0.0214*** -0.00347* -0.0308***
(0.00988) (0.00605) (0.00578) (0.00197) (0.00639)
Women -0.208*** -0.0467*** -0.0650*** -0.0470*** -0.0619***
(0.0198) (0.00980) (0.00949) (0.0113) (0.0119)
B. SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.436** -1.589*** -0.0426 0.0846 -0.504
(0.193) (0.553) (0.259) (0.339) (0.407)
Men -0.693 -2.336 -0.223 -1.418 -0.601
(0.526) (3.216) (0.656) (2.036) (0.716)
Women -0.351* -1.465*** 0.0223 0.208 -0.446
(0.186) (0.490) (0.241) (0.319) (0.462)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and the
IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all Com-
mon diagnoses set by DCT providers. In the second to fourth column, these com-
mon diagnoses are decomposed into subgroups: upper respiratory infections (Resp),
skin related diseases (Skin), genital and reproductive organs Gen/Rep, and Other
common diagnoses. Each row presents the diff-in-disc estimates for the diagno-
sis groups for the given estimation sample (All, Men, Women). Each model uses
a fixed bandwidth of 120 days each side of the age cutoff. Estimates using data
collapsed by region, gender, year and day relative to 20th birthday. Standard er-
rors clustered by the running variable, with separate clusters for pre-post periods. *
P<0.1, ** P<0.05, *** P<0.01.
40
Table 4: Visits to other health care professionals
FUZZY DIFF-IN-DISC
All Men Women
Nurse visits at a PCC -0.0545 -0.605* 0.124
(0.118) (0.312) (0.120)
Visits at a midwife/youth/STD clinic -0.129 0.352* -0.274
(0.176) (0.208) (0.219)
Nurse+midwife/youth/STD -0.183 -0.253 -0.150
(0.211) (0.350) (0.249)
All consultations -0.633** -1.279* -0.410
(0.320) (0.694) (0.339)
Note: The table shows fuzzy diff-in-discs estimates of the effect of online
consultations on in-person visits with other health care professionals than
physicians at primary care centers: nurse visits at a primary care center;
midwife/nurse/physician visits at a midwife/youth/STD clinic, the sum of
these two types of consultations; and the sum of all consultations, includ-
ing the physician visits at a primary care center we use in our main anal-
ysis. Each model uses a fixed bandwidth of 120 days each side of the
threshold. Standard errors clustered by the running variable, with sepa-
rate clusters for pre-post periods. * P<0.1, ** P<0.05, *** P<0.01.
41
Figure 1: Conceptual framework
Note: This figure illustrates the available care paths for an individual who experiences a care need (1). The
individual might try do address the issue by herself (self care). If she chooses not to, or if self care does not
resolve the problem, she may contact a DCT provider or a primary care centre (2). In the first case, the patient
will be offered an appointment with a physician contracting with the DCT company. In the second case, a
triage nurse will examine the need for further care and either schedule an appointment with a physician or
another health professional at the PCC, or deny access to further care (typically combined with some self care
advice) (3). If, after this step, the patient perceives that the problem has not been resolved, she may again
choose between self care or to contact a DCT provider or a PCC (possibly another one than the initial PCC) (4)
and so on.
42
Figure 2: DCT consultations over time
Note: The graphs display the average number of DCT consultations by age for four periods. The age variable
is measured as days in relation to the 20th birthday. The number of DCT consultation is rescaled to reflect
annual averages, and is plotted for 730 days grouped in 52 bins.
43
Figure 3: In-person physician visits
Note: The graphs display the average number of in-person physician visits by age. The age variable is measured
as days in relation to the 20th birthday. The number of in person visits is rescaled to reflect annual averages,
and is plotted for 730 days grouped in 52 bins.
44
Figure 4: Robustness analyses – bandwidth and donut
Note: The upper panel shows coefficients from sharp diff-in-disc models of the effect of the user fee on DCT
consultations (the first stage). The lower panel shows coefficients from fuzzy diff-in-disc models of the relation-
ship between DCT consultations and in-person visits (IV). The graphs on the left show how the estimates vary
over different bandwidhts (60–365 days + MSE-optimal bandwidth (Cattaneo, Idrobo and Titiunik, 2019)). The
coefficient at ”120” is equivalent to the main specification in Table 2). The graphs on the right shows how the
estimates vary when successively removing more observations around the 20th birthday (“ donut” estimations),
in all cases using a bandwidth of +/- 120 days relative to the 20th birthday. Here, the leftmost coefficients (0
days excluded) are equivalent to the main estimates in Table 2).
45
Figure 5: In-person physician visits
Note: The graph displays coefficients and 95% confidence intervals from sharp diff-in-disc models decompos-
ing the main effect. Each model uses a bandwidth of 120 days before and after the 20th birthday. Standard
errors clustered by the running variable, with separate clusters for pre-post periods. The outcome variables are
All DCT (in-person) consultations, All DCT (in-person) initial consultations and All DCT (in-person) follow-
up consultations. An initial consultation is a consultation for which the individual had no prior consultation
during the preceding 14 days. The remaining are defined as follow-ups. Initial consultations are decomposed
into consultations with no follow-ups and consultations with any follow-ups. Follow-up consultations are de-
composed by the mode of the initial consultation of the episode (same/other).
46
Online Appendix to An App Call a Day Keeps
the Patient Away?
Contents
A Comparison of DCT measures 3
B List of diagnosis categories 6
C Descriptive statistics 10
D External validity 12
E Results for 2017 cohorts 16
E.1 Graphical evidence of 2017 cohort . . . . . . . . . . . . . . . . . 16
E.2 Main results for 2017 cohort . . . . . . . . . . . . . . . . . . . . 17
F First stage heterogeneity and complier profiling 20
G Heterogeneity 24
H Robustness 27
H.1 Zero-degree polynomial estimates . . . . . . . . . . . . . . . . . 28
H.2 Longer donut specifications . . . . . . . . . . . . . . . . . . . . . 29
H.3 First stage graphs by diagnosis groups . . . . . . . . . . . . . . . 32
H.4 Balance test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
H.5 Including covariates . . . . . . . . . . . . . . . . . . . . . . . . . 33
H.6 Placebo tests using 19th and 21th birthdays . . . . . . . . . . . . 36
H.7 Sensitivity to inclusion of pre-periods of varying length . . . . . . 37
H.8 Sensitivity to secular decrease of in-person visits . . . . . . . . . 39
H.9 Fuzzy RD for Stockholm . . . . . . . . . . . . . . . . . . . . . . 40
1
H.10 Wild bootstrapping precision adjustment . . . . . . . . . . . . . . 42
H.11 Main specification on individual-level data . . . . . . . . . . . . . 45
I Decomposition by type of visits 46
J Robustness analysis of diagnosis data 49
K Prescriptions of antibiotics 55
L Other primary care consultations 59
M Cost analysis 61
M.1 Overview and sources . . . . . . . . . . . . . . . . . . . . . . . . 61
M.2 Cost scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2
Appendix A Comparison of DCT measures
The main data sources of our DCT measure are Stockholm and Västra Götaland’s
registers of billing information for their residents’ care consumption in other re-
gions. (Regions are financially responsible for their residents and get billed when
their residents are treated by providers in other regions.) The DCT variable in the
main analysis includes
• all primary care contacts in the Sörmland and Jönköping regions.
• contacts with private DCT providers in the Skåne region (Capio Go, un-
til May 2018) and a public DCT provider in Västra Götaland (Närhälsan
Online).
The vast majority of the DCT contacts come from Region Jönköping, where the
largest DCT providers were located at the time (Min Doktor, Kry, Doktor 24,
Medicoo, Accumbo, and from June 2018 Capio Go). Doktor.se was the only
provider located in Region Sörmland before 2019.
In Tables A.1, we examine how sensitive the estimated first stage sharp diff-
in-disc is to various definitions of the DCT variable. In column 1, the definition
of the DCT variable is the same as in the main analysis. This variable is based
on the billing information, and includes all primary care visits in Sörmland and
Jönköping. In the next two columns, we use the same definition as in column
1 but include only consultations registered in Region Sörmland (column 2) or
Region Jönköping (column 3). In column 4, we apply the same definition of DCT
contacts, but use information from register data from Region Jönköping (instead
of billing information from Västra Götaland and Stockholm). In column 5, the
DCT definition only includes registered remote contacts at the private PCCs in
Jönköping that have agreements with DCT providers.1 In column 6, we further
restrict this definition to only include remote consultations with physicians.
The estimates are similar across outcomes, except when we look at contacts in
Sörmland only (column 2). The similarity between the other coefficients implies
that the discontinuity at age 20 primarily affected the number of DCT contacts
with providers located Region Jönköping. This pattern is explained by the age
differentiated DCT user fee in Region Jönköping; in Region Sörmland the user
fee was zero for all ages groups. Notably, only one minor provider had an agree-
ment with a primary care center in Region Sörmland at this time period, and this
1It was the agreements with these PCCs that enabled the DCT get public funding via the inter-
regional agreement.
3
provider had a nurse triage system in place making the patient less likely to see
a physician. The positive coefficient in column 2 indeed suggests that the user
fee in Region Jönköping, if anything, had a reverse (although small) effect on
the number of DCT contacts in Region Sörmland. This is also supported by the
coefficients in column 3 and 4 being larger than in column 1.
The similarity between columns 1, 3, 4, and 5 suggests that the billing in-
formation from the home regions own registers provides an accurate measure of
the DCT contacts and a reliable estimate of the effect of the reduction of the DCT
user fee at age 20. (The estimated degree of substitution (fuzzy diff-in-disc) is also
very similar when using the data from Jönköping to measure DCT contacts) The
small difference between column 5 and 6 further suggests that only about 10% of
the first stage coefficient relates to changes in consultations with other health care
professionals than physicians.
4
Table A.1: Variation of DCT definitions
BOTH REGIONS (2018)
DCT DCT-Smld DCT-Jkpg1 DCT-Jkpg2 DCT-digpr DCT-phys
Both -0.149*** 0.00818** -0.158*** -0.163*** -0.161*** -0.151***
(0.0116) (0.00392) (0.0108) (0.0113) (0.0111) (0.0106)
Men -0.0700*** 0.00124 -0.0693*** -0.0674*** -0.0668*** -0.0622***
(0.0114) (0.00481) (0.0100) (0.0100) (0.0100) (0.0101)
Women -0.234*** 0.0156** -0.253*** -0.266*** -0.262*** -0.247***
(0.0231) (0.00757) (0.0213) (0.0220) (0.0215) (0.0206)
REGION STOCKHOLM (2018)
Both -0.158*** 0.0128** -0.170*** -0.177*** -0.176*** -0.163***
(0.0174) (0.00593) (0.0168) (0.0175) (0.0174) (0.0161)
Men -0.0783*** -0.000230 -0.0781*** -0.0773*** -0.0804*** -0.0710***
(0.0166) (0.00792) (0.0151) (0.0155) (0.0156) (0.0151)
Women -0.243*** 0.0266*** -0.270*** -0.284*** -0.278*** -0.262***
(0.0329) (0.0102) (0.0312) (0.0319) (0.0314) (0.0294)
REGION VÄSTRA GÖTALAND (2018)
Both -0.138*** 0.00240 -0.142*** -0.145*** -0.143*** -0.136***
(0.0174) (0.00512) (0.0152) (0.0158) (0.0154) (0.0136)
Men -0.0596*** 0.00308 -0.0585*** -0.0551*** -0.0501*** -0.0513***
(0.0163) (0.00584) (0.0138) (0.0139) (0.0137) (0.0137)
Women -0.224*** 0.00166 -0.232*** -0.244*** -0.243*** -0.228***
(0.0317) (0.00894) (0.0278) (0.0287) (0.0274) (0.0251)
Note: Table shows sharp diff-in-disc results for various definitions of DCT consultations using
a bandwidth of 120 days. Each column presents the sharp diff-in-disc estimates for the a given
outcome for each estimation sample (All, Men, Women) for both regions jointly, and separately.
Each model uses MSE-optimal bandwidths for in-person visits used in the main estimations for
both genders and regions jointly. Estimates using data collapsed by region, gender, year and day
relative to 20th birthday. Standard errors clustered by the running variable, with separate clusters
for pre-post periods. * P<0.1, ** P<0.05, *** P<0.01.
5
Appendix B List of diagnosis categories
We define the most common diagnoses using information on registered ICD-codes
from the local care register in Region Jönköping. These data exactly identify the
online-consultations with physicians at DCT providers in Jönköping, but have the
limitation that they do not include any consultations with DCT providers located
elsewhere. Each care contact may have up to five registered diagnoses classified
according to the ICD-10. The data include the complete ICD-code, but we define
the most common diagnoses categories on a three digit level.
To define the most common diagnoses, we first generate a list of all com-
plete ICD-codes and the corresponding number of recorded registrations at online
consultations from 2016 to 2018 (by each gender). We then sum the number of
registered diagnoses within each three digit ICD-code, and rank these three digit
codes by the number of registrations (again by gender). A three digit ICD-code
is defined as being among the most common diagnoses if it is included among
the top 80 % of all registered diagnoses for either male or female individuals.
For non-administrative ICD-codes (i.e., any ICD-code not in Z00-Z99), we also
include three-digit ICD-codes that belongs to the same ICD-block (subchapter)
as any of the most common ICD-codes and had been registered at least 10 times
during the study period.
We classify the set of common diagnoses into four subsets: upper respiratory
infections, skin conditions, diagnoses related to genital and reproductive organs,
and a residual category (other). Table B.1, B.2, B.3, and B.4 list the three digit
level ICD-codes in each subset. In 2018, 90 % of the online consultations with
a physician had at least one registered ICD-code that is included in the definition
of the most common diagnoses. The four categories respectively covered 19%
25%, 22%, and 28% of all online consultations in the same year. Because each
consultation may have more than one diagnosis, the contacts in these subsets are
not completely mutually exclusive.
6
Table B.1: Upper respiratory infections
icd icd-block Description
B27 B25 -B34 Infectious mononucleosis
B30 B25 -B34 Viral conjunctivitis
B34 B25 -B34 Viral infection of unspecified site
J00 J00 -J06 Acute nasopharyngitis [common cold]
J01 J00 -J06 Acute sinusitis
J02 J00 -J06 Acute pharyngitis
J03 J00 -J06 Acute tonsillitis
J06 J00 -J06 Acute upper respiratory infections of
multiple and unspecified sites
J30 J30 -J39 Vasomotor and allergic rhinitis
J31 J30 -J39 Chronic rhinitis, nasopharyngitis and pharyngitis
J35 J30 -J39 Chronic diseases of tonsils and adenoids
J45 J40 -J47 Asthma
R05 R00 -R09 Cough
Table B.2: Skin related conditions
icd icd block Description
B00 B00 -B09 Herpesviral [herpes simplex] infections
B02 B00 -B09 Zoster [herpes zoster]
B07 B00 -B09 Viral warts
B08 B00 -B09 Other viral infections characterized by skin
and mucous membrane lesions, not elsewhere classified
B35 B35 -B49 Dermatophytosis
B36 B35 -B49 Other superficial mycoses
B37 B35 -B49 Candidiasis (excluding B373, B373P, or B374)
L01 L00 -L08 Impetigo
L02 L00 -L08 Cutaneous abscess, furuncle and carbuncle
L03 L00 -L08 Cellulitis
L08 L00 -L08 Other local infections of skin and
subcutaneous tissue
L20 L20 -L30 Atopic dermatitis
L21 L20 -L30 Seborrhoeic dermatitis
L23 L20 -L30 Allergic contact dermatitis
L29 L20 -L30 Pruritus
L30 L20 -L30 Other dermatitis
L50 L50 -L54 Urticaria
L60 L60 -L75 Nail disorders
L63 L60 -L75 Alopecia areata
L64 L60 -L75 Androgenic alopecia
L65 L60 -L75 Other nonscarring hair loss
L70 L60 -L75 Acne
L71 L60 -L75 Rosacea
L73 L60 -L75 Other follicular disorders
R21 R20 -R23 Rash and other nonspecific skin eruption
R22 R20 -R23 Localized swelling, mass and lump of skin
and subcutaneous tissue
R23 R20 -R23 Other skin changes
7
Table B.3: Genital & reproductive organs
icd icd block Description
B37 B35 -B49 Candidiasis (only B373, B373P, or B374)
F52 F50 -F59 Sexual dysfunction, not caused by organic
disorder or disease
N30 N30 -N39 Cystitis
N39 N30 -N39 Other disorders of urinary system
N76 N70 -N77 Other inflammation of vagina and vulva
N77 N70 -N77 Vulvovaginal ulceration and inflammation
in diseases classified elsewhere
N92 N80 -N98 Excessive, frequent and irregular menstruation
N94 N80 -N98 Pain and other conditions associated with
female genital organs and menstrual cycle
R10 R10 -R19 Abdominal and pelvic pain
Y42 Y40 -Y59 Hormones and their synthetic substitutes
and antagonists, not elsewhere classified
Z30 Z Contraceptive management
Z92 Z Personal history of medical treatment
8
Table B.4: Other common diagnoses
icd icd-block Description
A08 A00 -A09 Viral and other specified intestinal infections
A09 A00 -A09 Other gastroenteritis and colitis of
infectious and unspecified origin
A69 A65 -A69 Other spirochaetal infections
B80 B65 -B83 Enterobiasis
F32 F30 -F39 Depressive episode
F33 F30 -F39 Recurrent depressive disorder
F40 F40 -F48 Phobic anxiety disorders
F41 F40 -F48 Other anxiety disorders
F42 F40 -F48 Obsessive-compulsive disorder
F43 F40 -F48 Reaction to severe stress, and adjustment disorders
F45 F40 -F48 Somatoform disorders
F51 F50 -F59 Nonorganic sleep disorders
G43 G40 -G47 Migraine
G44 G40 -G47 Other headache syndromes
G47 G40 -G47 Sleep disorders
H10 H10 -H13 Conjunctivitis
K12 K00 -K14 Stomatitis and related lesions
K13 K00 -K14 Other diseases of lip and oral mucosa
K14 K00 -K14 Diseases of tongue
K21 K20 -K31 Gastro-oesophageal reflux disease
K29 K20 -K31 Gastritis and duodenitis
K30 K20 -K31 Functional dyspepsia
M54 M50 -M54 Dorsalgia
M79 M70 -M79 Other soft tissue disorders, not elsewhere classified
R00 R00 -R09 Abnormalities of heart beat
R06 R00 -R09 Abnormalities of breathing
R07 R00 -R09 Pain in throat and chest
R11 R10 -R19 Nausea and vomiting
R19 R10 -R19 Other symptoms and signs involving
the digestive system and abdomen
R50 R50 -R69 Fever of other and unknown origin
R51 R50 -R69 Headache
R52 R50 -R69 Pain, not elsewhere classified
R53 R50 -R69 Malaise and fatigue
R61 R50 -R69 Hyperhidrosis
T14 T08 -T14 Injury of unspecified body region
T38 T36 -T50 Poisoning by hormones and their synthetic substitutes
and antagonists, not elsewhere classified
T78 T66 -T78 Adverse effects, not elsewhere classified
W57 W55 -W65 Bitten or stung by nonvenomous insect and
other nonvenomous arthropods
W64 W55 -W65 Exposure to other and unspecified animate mechanical forces
X58 X58 -X59 Exposure to other specified factors
Z00 Z General examination and investigation of persons
without complaint and reported diagnosis
Z02 Z Examination and encounter for administrative purposes
Z03 Z Medical observation and evaluation for suspected diseases
and conditions
Z71 Z Persons encountering health services for other counselling
and medical advice, not elsewhere classified
Z76 Z Persons encountering health services in other circumstances
9
Appendix C Descriptive statistics
Table C.1 shows descriptive statistics for three groups: the subset of individuals
who were 20 (± 365 days) years old at any point in 2018 and had at least one DCT
consultation in that year (DCT users 2018); individuals in the same cohort who
did not have a DCT consultation in 2018 (Non-users 2018); and individuals who
were 20 (± 365 days) years old at any point during the four year period before
the emergence of the DCT market, i.e., from July 2012 to June 2016 (cohorts
2012-2016).
The first row shows the average annual number of in-person physician vis-
its (our main outcome variable). DCT users visited a physician more often than
non-users, which may reflect a generally greater propensity to seek care or just a
worse health status.2 Further below in the table, we also note that the proportion
of individuals who did not visit a physician during their 19th life year (0 Phys vis
18) was lower among DCT users than in other groups. Women are very much
over-represented among DCT users, who also tend to have parents with higher
socioeconomic background than non-users in the same as well as the earlier co-
horts. A larger fraction of DCT users live in Region Stockholm (vs. Region Västra
Götaland), and a larger fraction of DCT users lives in a large city (Stockholm or
Gothenburg) rather than a town or a rural area.
2A similar pattern was also demonstrated for the study population in Ellegård and Kjellsson
(2019).
10
Table C.1: Descriptives across cohorts and DCT users
COHORT 2018 COHORT 2018 COHORTS 2012-2016
(DCT USERS) (NON-USERS) (NON-USERS)
Mean SD Mean SD Mean SD
In-person phys visits 1.462 23.160 0.889 18.063 1.053 19.659
0 Phys vis 18 0.381 0.486 0.521 0.500 0.475 0.499
1 Phys vis 18 0.264 0.441 0.246 0.431 0.251 0.434
≥ 2 Phys vis 18 0.356 0.479 0.233 0.423 0.274 0.446
Female 0.727 0.445 0.455 0.498 0.484 0.500
Share Sthlm 0.631 0.483 0.547 0.498 0.542 0.498
Resides Rural 0.071 0.256 0.103 0.303 0.109 0.311
Resides Town 0.171 0.377 0.231 0.422 0.238 0.426
Resides City 0.758 0.428 0.666 0.472 0.654 0.476
Dad uni 0.408 0.492 0.377 0.485 0.352 0.477
Mum uni 0.488 0.500 0.464 0.499 0.428 0.495
Mum inc > median 0.496 0.500 0.426 0.494 0.404 0.491
Dad inc > median 0.628 0.483 0.570 0.495 0.566 0.496
Parent non-nordic 0.242 0.429 0.263 0.440 0.228 0.419
Observations 2,366,695 20,666,406 110,131,616
Unique Indvs 8,058 86,844 236,601
Note: Descriptive statistics are presented here with sample split by DCT users 2018 and non-DCT
users. Non-users are also presented by cohort; those who were 20 (± 365 days) years old at any point
in the period before DCT being introduced (June 2012- July 2016), and those who were 20 (± 365
days) years old at any point in 2018. (These include individuals who potentially consumed DCT 2016
or 2017). Mean and standard deviation of each variable are based on the number of observations (in-
dividual and day) within the period. An individual is observed at most 730 days around the 20th birth-
day. ’In-person phys visits’ is our main outcome variable; any in-person visit to a physician which has
been scaled to a per capita annual basis. The 0/1/≥2 Phys vis 18 variables are dichotomous indicators
of the number of physician visits during the individual’s 19th life year.
11
Appendix D External validity
To obtain an idea of the generalisability of our results, we use care register data for
2015 to compare one of the pre-DCT cohorts in our study population with other
age groups with respect to a set of health measures. Specifically, we compare the
study cohort in period -1 – who turned 20 in the second half of 2015 or the first
half of 2016 – to individuals who resided in the study regions throughout 2015
(and did not move between the two regions). The reason why we use pre-DCT
data for this exercise is that DCT, to the extent that it affected care utilisation, may
have had different impact on the care utilisation of our post-DCT study cohort and
on other age groups.
Our first health measure is the individual’s predicted health care costs accord-
ing to the Johns Hopkins ACG(R) System (v 11). This software, which is used
for risk-adjustment in many settings (including Swedish primary care), uses diag-
noses recorded in care registers to group individuals into Adjusted Clinical Group
(ACG) with similar expected costs (similar to DRGs). The software produces an
value for each individual showing the expected costs relative to the average costs
in the region. Due to the presence of outliers, we recode the ACG values of indi-
viduals above the 95th percentile or below so that they get the ACG of the 95th
percentile (using the winsor2 package in Stata).
Figure D.1 compares the ACG values of the 2015 cohort (empty bars) to the
other residents in 2015 in various age groups. As seen from the figure, the study
cohort is very similar to the 21-34 age groups in terms of expected health care
costs, and quite similar to the 15-19 age group. The study cohort is less similar to
children <15 (in particular 0-4 year-olds) and to individuals above 35.
Secondly, we compare the groups with respect to the share of individuals who
had been diagnosed with at least one of the most common DCT-related diagnoses
(the ”All” category in Table J.5) in primary care in 2015. Notably, this group
only includes individuals who visited a (non-DCT) primary care provider in 2015.
Figure D.2 shows that roughly 40% of the study cohort received such a diagnosis
in 2015, which is similar to most age groups except the very youngest and oldest
age groups.
We then look at each of the four categories of common DCT diagnoses (resp,
gen/repr, skin, other). The proportions with a diagnosis are generally similar in
the study cohort and the 15-29 age group, and often also for other age groups.
12
Figure D.1: ACG distribution
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
ACG ACG ACG ACG
_below5 2015 cohort 5_9 2015 cohort 10_14 2015 cohort 15_19 2015 cohort
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
ACG ACG ACG ACG
21_24 2015 cohort 25_29 2015 cohort 30_34 2015 cohort 35_39 2015 cohort
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
ACG ACG ACG ACG
40_44 2015 cohort 45_49 2015 cohort 50_59 2015 cohort 60_plus 2015 cohort
Note: The figure shows the distribution of ACG values in a pre-DCT cohort (”2015 cohort”) and
the general population in 2015, by age group. ACG captures expected health care costs and is
based on diagnoses set during the year. Due to extreme values, the ACG variable is winsorised at
the 95th percentile.
13
0 .1 .2Fraction.3 .4 .5 0 .1 .2Fraction.3 .4 .5 0 .1 .2Fraction.3 .4 .5
0 .1 .2Fraction.3 .4 .5 0 .1 .2Fraction.3 .4 .5 0 .2 Fraction.4 .6
0 .1 .2Fraction.3 .4 .5 0 .1 .2Fraction.3 .4 .5 0 .2 Fraction .4 .6
0 .1 .2Fraction.3 .4 .5 0 .1 .2Fraction.3 .4 .5 0 .1 .2Fractio.n3 .4 .5
Figure D.2: Share with a DCT-relevant diagnosis
Note: The figure shows the proportion of individuals who were diagnosed with one of the most
common diagnoses in DCT in traditional primary care in 2015. The leftmost bar (”2015 cohort”)
shows the proportion for the youngest pre-DCT cohort.
14
Figure D.3: Share with a DCT-relevant diagnosis; by category
Note: The figure shows the proportion of individuals who were diagnosed with a diagnoses in our
resp, gen/rep, skin and other diagnosis categories in traditional primary care in 2015. The leftmost
bar (”2015 cohort”) shows the proportion for the youngest pre-DCT cohort.
15
Appendix E Results for 2017 cohorts
E.1 Graphical evidence of 2017 cohort
Figure E.1 plots annual in-person physician visits per capita, by the day relative
to the 20th birthday (i.e., the DCT user fee threshold). The first panel shows
the graph for the whole sample and the next two by gender. The middle and
rightmost sub-graphs illustrate the two years after the DCT market emerged (2017
and 2018), and the leftmost sub-graphs show the pooled pre-period (Figure H.5
shows that the pattern is similar for each of the years in the pre-period).
Figure E.1: In-person physician visits
Note: The graphs display the average number of in-person physician visits by age. The age vari-
able is measured as days in relation to the 20th birthday. The number of in person visits is rescaled
to reflect annual averages, and is plotted for 730 days grouped in 26 bins.
16
E.2 Main results for 2017 cohort
Table E.1 presents diff-in-disc estimates equivalent to the main results in Table 2
but compares individuals turning 20 in 2017 (instead of 2018) to the pre-DCT
periods. The estimates confirm the results from the graphical analysis in section
4.1: The first stage estimates are generally smaller and weaker compared to 2018,
although the negative direction is consistent with our expectation. The sharp diff-
in-disc for the full sample provides no indication that the onset of the user fee of
DCT increase consumption of in-person visits. The IV estimates are small and
statistically insignificant, although for the most precisely estimated model, for
men, the coefficient is negative as is consistent with substitution.
Figure E.2 shows estimates for 2017 corresponding to the results fort 2018 in
Figure 4 in the main document. The first stage is small but robust to varying the
size of both the donut and bandwidths. Comparable to the results from the main
specification, the degree of substitution is estimated to be close to zero for differ-
ent bandwidths, although with some increased precision for longer bandwidths.
17
Table E.1: Main result for 2017 cohort as post-period
All Men Women
A. First stage (sharp diff-in-disc)
Online consultations -0.0499*** -0.0195*** -0.0824***
(0.00706) (0.00701) (0.0114)
B. RF (sharp diff-in-disc)
In-person visits -0.00191 0.00279 -0.00679
(0.0352) (0.0397) (0.0563)
C. IV (fuzzy diff-in-disc)
In-person visits 0.0383 -0.143 0.0825
(0.704) (2.032) (0.681)
Avg. ind/day pre 151525 78130 73395
Avg. ind/day post 31929 16467 15462
Note: Variable names in left column represent the dependent vari-
able. Each model uses a fixed bandwidth of 120 days on either side
of the threshold (20th birthday). Avg. ind/day = average number
of individuals per day in cells, shown separately for the pre cohorts
(which include several birth cohorts) and the post cohort (which
only includes individuals turning 20 in 2017). Data are collapsed
by region, gender, birth year and day relative to 20th birthday. Stan-
dard errors clustered by the running variable, with separate clusters
for pre-post periods. * P<0.1, ** P<0.05, *** P<0.01.
18
Figure E.2: Robustness 2017 - bandwidth and donut
Note: The graphs show estimates for 2017 corresponding to Fig 4. The upper panel shows co-
efficients from sharp diff-in-disc models of the effect of the user fee on DCT consultations (the
first stage). The lower panel shows coefficients from fuzzy diff-in-disc models of the relationship
between DCT consultations and in-person visits (IV). The graphs on the left show how the esti-
mates vary over different bandwidhts (60–365 days + MSE-optimal bandwidth (Cattaneo, Idrobo
and Titiunik, 2019)). The graphs on the right shows how the estimates vary when successively
removing more observations around the 20th birthday (“ donut” estimations), in all cases using a
bandwidth of +/- 120 days relative to the 20th birthday.
19
Appendix F First stage heterogeneity and complier
profiling
To provide information on the type of individuals responding to the change in the
user fee, we examine the heterogeneity of the treatment effect of the user fee on
DCT use (i.e., the first stage) and characterise the compliers of the change in the
user fee using the logic from Angrist and Pischke (2008). In our setting a complier
is an individual that – on a given day – makes a DCT consultation under the ab-
sence of a DCT user fee but chooses not to make a DCT consultation when facing
a user fee. We define always-takers (i.e., individuals that would use DCT during
the given period independent of the presence of a fee) as the proportion of individ-
uals using DCT before the 20th birthday (when no fee applies) and never-takers
(i.e., individuals that would not use DCT during the given period independent of
the fee) as the proportion of individuals without any DCT contact in the period
after the 20th birthday (when the fee applies). We further assume that there are no
defiers (i.e. individuals who abstain from using DCT unless there is a fee).
Under the assumption of no defiers, the proportion of the compliers are equal
to the first stage estimate if the outcome is binary (Angrist and Pischke, 2008).
While the outcome variable in our application is adjusted to represent the yearly
number of DCT contacts, the relative size of the coefficients is still informative.
The proportion of compliers among the DCT users before the 20th birthday may
therefore be computed as the ratio between the effect of the user fee and the level
of consumption in the period with the higher user fee (before the 20th birthday).
That is, β2/(β4 +β1) in equation 3. That implies that the proportion of the users
that are compliers and always-takers are equal to 49.5% and 50.5%.3 Angrist
and Pischke (2008) further show that the ratio between the first stage coefficient
for a subpopulation based on a given characteristic and the total population is
informative of the relative likelihood that a complier has that characteristic. The
left subgraph of Figure F.1 displays sharp diff-in-discs coefficients estimating the
effect of the user fee on DCT consultations for various subpopulations compared
to the main estimate (depicted as the plotted line). These estimates correspond
to β3 in equation 3 in the main document. The center subgraph displays the
corresponding baseline utilisation per subpopulation in 2018 for individuals below
the 20th birthday. These estimates correspond to β1 +β4 in equation 3. The left
3To get an estimate of the absolute proportion of compliers and always-takers, these coeffi-
cients should be scaled back to the daily level. In a given day, there are 0.000412 always-takers,
and 0.000408 compliers using DCT.
20
subgraph displays the relative reduction (i.e., β3/β1 +β4).
The graphs shows that women, individuals residing in urban areas, and indi-
viduals with previous physician contacts have both larger first stage and a larger
baseline utilisation. These results suggest that these characteristics are relatively
more common among compliers than in the overall sample. However, the re-
sults also suggest that the size of the effect is very much related to the baseline
utilisation (and that these characteristics are relatively more common also among
always-takers). There are also minor indications that both first stage and baseline
utilisation are larger among individuals with high SES parents (fathers with in-
come below the median and no university education) and individuals having two
parents being born outside of the Nordic countries. Thus, the first stage is gener-
ally larger for subpopulations with a larger baseline utilisation. Notably, the left
subgraph show that the relative effect of the onset of the user fee in relation to
the utilisation before the 20th birthday is very similar between subpopulations,
indicating that price sensitivity is similar across subgroups.
Effect of user fee on DCT contacts (First stage) DCT contacts before 20th birthday Relative reduction
All
male
female
Sthlm
VGR
reside rural
reside town
reside city
Uni, dad
No uni, dad
Uni, mother
No uni, mother
income>median, father
income<median, father
income>median, mother
income<median, mother
Non-nordic parents
Nordic parents
0 Phys vis 18
1 Phys vis 18
>1 Phys vis 18
-.3 -.25 -.2 -.15 -.1 -.05 0 0 .05 .1 .15 .2 .25 .3 .35 .4 .45 .5 .55 .6 -.7 -.6 -.5 -.4 -.3 -.2
Coefficient: β4 Coefficients: β1+β3 Ratio:β4 / (β1+β3)
Figure F.1: Effect of user fee on DCT consultations, by subpopulations
21
Figures F.2 and F.3 display graphically the number of DCT and in-person con-
sultations around the age threshold, by gender.
Figure F.2: DCT consultations in the postperiod by gender
Note: The graphs display the average number of DCT consultations by age for each gender in
2018. The age variable is measured as days in relation to the 20th birthday. The number of DCT
consultation is rescaled to reflect annual averages, and is plotted for 730 days grouped in 26 bins.
22
Figure F.3: In-person physician visits
Note: The graphs display the average number of in-person physician visits by age for men and
women separately. The age variable is measured as days in relation to the 20th birthday. The
number of in person visits is rescaled to reflect annual averages, and is plotted for 730 days grouped
in 26 bins.
23
Appendix G Heterogeneity
Urban and previous care consumption We explore heterogeneity in the degree
of substitution in dimensions that are pointed out as important for the first stage
heterogeneity in Appendix F. The share of compliers are larger among residents
in urban areas and among individual that have consumed care between their 18th
and 19th birthday. Table G.1 displays the urban-rural dimension (columns 1-2),
as well as by the individual’s prior care consumption (columns 3-4). We find, as
expected from when splitting the sample by region, that the substitution is mainly
driven by the urban sample. In contrast, both the first stage and the substitution
effect found are largely comparable for individual with different levels of previous
care consumption (although the IV estimates are statistically insignificant).
Table G.1: Complier heterogeneity - Rurality and care consumption
Rural Urban 0 visits ≥ 1 visit
FS -0.101*** -0.172*** -0.124*** -0.176***
(0.0178) (0.0152) (0.0164) (0.0198)
IV -0.0490 -0.561*** -0.495 -0.389
(0.503) (0.212) (0.307) (0.257)
Note: The table shows estimates with sample split by first
urban-rural dimension and then by whether individual has
consumed care by 18 years old. Each model uses a 120
bandwidth with 2018 as the postperiod. Standard errors
in parentheses clustered by relative age (running variable),
with separate clusters for pre-post periods. * P<0.1, **
P<0.05, *** P<0.01.
Regional heterogeneity Our main analysis merges data from two independent
administrative regions – Region Stockholm and Region Västra Götaland. Given
the scope for institutional differences, it is of interest to examine heterogeneity
over these regions.
Table G.2 presents our main specification from models where regions are es-
timated separately for each region. As indicated by the graphical evidence, the
size of the Stockholm coefficient (-.77) is considerably larger than the coefficient
for Västra Götaland (.01), and it is only in Stockholm that we can reject the null
hypothesis of no substitution. The first stage estimates in the first panel shows that
the response to the DCT user fee is similar in both regions. Thus, the difference in
the degree of substitution is driven by differences in the reduced form estimates.
24
Table G.2: Fuzzy diff-in-disc results (IV), by region
All Men Women
A. FIRST STAGE (SHARP DIFF-IN-DISC)
Sthlm -0.158*** -0.0783*** -0.243***
(0.0173) (0.0165) (0.0328)
F-stat 82.49 22.38 55.04
VGR -0.138*** -0.0596*** -0.224***
(0.0174) (0.0163) (0.0316)
F-stat 63.23 13.41 50.35
B. REDUCED FORM (SHARP DIFF-IN-DISC)
Sthlm 0.121*** 0.108** 0.132*
(0.0441) (0.0518) (0.0677)
VGR -0.00189 0.0263 -0.0325
(0.0475) (0.0576) (0.0700)
C. IV (FUZZY DIFF-IN-DISC)
Sthlm -0.767*** -1.384** -0.543*
(0.288) (0.706) (0.287)
VGR 0.0137 -0.440 0.145
(0.343) (0.974) (0.313)
Sthlm: Avg. ind/day pre 82264 42380 39884
Sthlm: Avg. ind/day post 17502 9009 8492
VGR: Avg. ind/day pre 69283 35763 33520
VGR: Avg. ind/day post 13985 7291 6695
Note: The table shows estimates with the sample split by regions and sex. Each
model uses bandwidth of 120 days on either side of the 20th birthday. ’Aver-
age ind/day pre (post)’ refers to average individuals per cell in the pre (post)
cohort. Standard errors in parentheses clustered by the running variable, with
separate clusters for pre and post periods. * P<0.1, ** P<0.05, *** P<0.01.
We continue by further breaking down the results by region and rurality, which
is further motivated by urban residents being a large complier group. The esti-
mates in Table G.3 show that the estimated degree of substitution is more similar
when comparing the urban areas of the two regions (Stockholm -.631; Västra
Götaland -.391), although the estimate for Västra Götaland is not statistically sig-
nificant. This in turn implies that the differences in the degree of substitution
between the two regions is largely explained by differences in the share of urban
residents.
25
Table G.3: IV estimate by urban/rural heterogeneity
IV ESTIMATES
Rural Urban
RS -2.920 -0.631**
(2.291) (0.280)
RVG 0.495 -0.391
(0.558) (0.400)
AVG INDIVIDUALS/DAY (CELL)
Sthlm Pre: 40369 40369
Sthlm Post: 1869 1563
VGR Pre: 36815 36815
VGR Post: 8330 5654
Note: The table shows estimates with sam-
ple split by regions and urban/rural dimen-
sion, with a fixed 120-days bandwidth. ’Avg.
ind/day pre (post)’ refers to average individ-
uals per cell in the pre (post) cohort. Stan-
dard errors in parentheses clustered by relative
age (separate clusters for pre-post periods). *
P<0.1, ** P<0.05, *** P<0.01.
26
Figure H.1: Zero degree polynomial - bandwidth and donut
Note: The graphs show how removing 0 to 70 days affects the estimates using a bandwidth of +/-
180 (left) or +/- 365 (right) days relative to the 20th birthday. The upper panel shows coefficients
from sharp diff-in-disc models of the effect of the user fee on DCT consultations (the first stage).
The lower panel shows coefficients from fuzzy diff-in-disc models of the relationship between
DCT consultations and in-person visits (IV).
Appendix H Robustness
H.1 Zero-degree polynomial estimates
The graphs in Figure H.1 show first stage and IV estimates from a model using a
zero-degree polynomial for varying bandwidths, and a specification using a fixed
bandwidth of 120 days. The first stage estimates of the onset of the user fee on
DCT consultations are similar – but more stable than the estimates from the linear
specification. The estimates of the degree of substitution hover around the esti-
mate from our preferred specification (Table 2). Overall, these estimates are more
stable and generally more precisely estimated than those from the linear specifi-
cation. Thus, in difference to the linear specification, the zero-degree polynomial
provides similar estimates of degree of substitution also for larger bandwidths.
The graphs to the right hand side in Figure H.1 presents coefficients from donut
estimations removing 0 to 28 days, showing in line with the estimates in the main
document that these are stable for both the first stage relationship and the degree
of substitution.
27
H.2 Longer donut specifications
This section shows donut estimation results using longer outer bandwidths than in
the main document (for a linear and a zero degree polynomial specification). We
estimate these models with longer bandwidths to be able to further increase the
size of the donut hole without losing too many observations.
Figure H.2 shows estimates from donut specifications using longer bandwidths
of 180 days (left graphs) or 365 days (right graphs) using a linear specification.
The upper panel shows first stage estimates of the effect of the fee on the number
of DCT consultations and the lower panel shows IV estimates of the degree of
substitution. The horizontal axis indicates the number of days dropped on each
side of the 20th birthday (the donut hole).
For both bandwidths, the first stage estimates are rather stable. The IV esti-
mates are also stable up until we remove 56 days (8 weeks on each side of the 20th
birthday). Removing donuts of this size or larger eliminates the substitution effect
(for both bandwidths). Since the first stage estimates are stable or even decreas-
ing, the decrease in the estimated degree of substitution stems from variability in
the relationship between age and the number of in-person visits far away from the
threshold. Revisiting Figure 3 in the main text, it is easy to see that the slope of
a linear regression line may vary depending on which portion of the x-axis one
selects to fit the lines, but that the relationship looks close to flat when looking at
the period as a whole. Thus, to explore the sensitivity to the donut estimations to
the linear specification, we also estimate donut specifications using a zero-degree
polynomial.
Figure H.3 shows that for larger bandwidths (of 180 or 365 days) the specifi-
cation using a zero-degree polynomial yields results similar to the main estimates
even with the larger donuts of 56-70 days. The point estimates from the IV model
indicate that the degree of substitution is between 0.3 and 0.5. The difference to
the corresponding estimates from linear specification of the model suggest that for
larger donuts (+/-8 weeks) the estimated degree of substitution is sensitive to the
(linear) specification of the model. Notably, the estimates from the zero-degree
specification, which are less affected by volatility further away from the thresh-
old, are in line with the main estimates. Thus, our conclusions about the degree of
substitution are only affected if we rely on estimates that are clearly affected by
fluctuations far away from the threshold.
28
Figure H.2: Longer donut estimations
Note: The graphs show how removing 0 to 70 days affects the estimates using a bandwidth of +/-
180 (left) or +/- 365 (right) days relative to the 20th birthday. The upper panel shows coefficients
from sharp diff-in-disc models of the effect of the user fee on DCT consultations (the first stage).
The lower panel shows coefficients from fuzzy diff-in-disc models of the relationship between
DCT consultations and in-person visits (IV).
29
Figure H.3: Zero-degree polynomial - longer donut estimations
Note: The graphs show removing 0 to 70 days affects estimates from a specification using a zero
degree polynomial and a bandwidth of +/-180 (left) or +/- 365 (right) days relative to the 20th
birthday. The upper panel shows coefficients from sharp diff-in-disc models of the effect of the
user fee on DCT consultations (the first stage). The lower panel shows coefficients from fuzzy
diff-in-disc models of the relationship between DCT consultations and in-person visits (IV).
30
H.3 First stage graphs by diagnosis groups
Figure H.4 plots the annual number of DCT consultations per capita in 2018 rel-
ative to the individual’s 20th birthday, for four types of diagnoses common DCT-
diagnoses: upper respiratory infections, skin conditions, diagnoses related to gen-
ital and reproductive organs, and other common DCT-diagnoses.
Figure H.4: DCT consultations in 2018 by diagnosis group
Note: The graphs display the average number of DCT consultations by age for four subgroups
of diagnoses in 2018. The age variable is measured as days in relation to the 20th birthday. The
number of DCT consultation is rescaled to reflect annual averages, and is plotted for 730 days
grouped in 26 bins.
31
H.4 Balance test
Table H.1 shows balance estimation results for the dichotomous background vari-
ables displayed in the leftmost column. Each row presents the result of sharp
difference-in-discontinuity regressions (Eq. (1)) using the background character-
istic in question as the outcome variable. We estimate these models using individ-
ual level data and cluster standard errors on the individual.
In regression discontinuity studies not using age as the running variable, bal-
ance tests of this kind can spot suspected violations of the assumption of random
assignment due to, e.g., manipulation. In our case, such manipulation is ruled out
by definition, as we are studying a close to balanced panel of individuals observed
around their 20th birthday and all the outcomes variables in our regressions are
time-invariant (for instance, the variable Phys visit = 1 equals one if the individual
visited a primary care physician exactly once during his/her 19th life year). This
implies that all differences reflect sample composition changes (i.e., attrition) and
nothing else.
There are some statistically significant estimates, but the magnitudes of these
coefficients are negligible in relation to the means of the variables (c.f. Table C.1).
For our baseline sample (column All, 2018), the coefficient on the indicator for
having made exactly one primary care visit during one’s 18th life year is statis-
tically significant at the 10% level. Put in relation to the mean of the outcomes
variable, which is 0.251, it is clear that the coefficient of .0017 is not a meaningful
difference.
H.5 Including covariates
Table H.2 shows fuzzy diff-in-disc estimates with and without including covari-
ates, i.e., the pre-determined individual characteristics used in our balance test
(columns ”No Covs/With covs”), across a range of bandwidths. We note that, as
expected, covariates do not affect the point estimates nor the precision of the esti-
mates.
32
Table H.1: Balance test on individual data
ALL MEN WOMEN
2017 2018 2017 2018 2017 2018
Woman -0.00020 0.00010
( 0.00080) ( 0.00080)
Phys visits = 0 0.00195 -0.00187 0.00342** -0.000786 0.000325 -0.00273
(0.00119) (0.00117) (0.00164) (0.00161) (0.00170) (0.00168)
Phys visits = 1 -0.00219** 0.00170* -0.00292** 0.000920 -0.00140 0.00251*
(0.00103) (0.00102) (0.00141) (0.00139) (0.00151) (0.00148)
Phys visits ≥ 2 0.000241 0.000162 -0.000501 -0.000134 0.00108 0.000221
(0.00105) (0.00101) (0.00132) (0.00128) (0.00162) (0.00156)
Share Sthlm -0.00208* 0.00108 -0.00279* -0.000372 -0.00134 0.00262
(0.00118) (0.00116) (0.00164) (0.00162) (0.00171) (0.00168)
Resides Rural 0.00116 0.0000699 0.000313 0.000536 0.00207** -0.000417
(0.000716) (0.000693) (0.000992) (0.000963) (0.00103) (0.000997)
Resides Town -0.000217 -0.000993 0.00190 0.000436 -0.00248* -0.00253*
(0.00100) (0.000983) (0.00139) (0.00135) (0.00145) (0.00143)
Resides City -0.000945 0.000923 -0.00221 -0.000973 0.000405 0.00295*
(0.00112) (0.00110) (0.00155) (0.00152) (0.00162) (0.00159)
Dad uni -0.000909 -0.0000824 -0.00253 0.00129 0.000823 -0.00156
(0.00115) (0.00114) (0.00160) (0.00158) (0.00166) (0.00164)
Mum uni -0.00203* -0.000160 -0.00582*** 0.00128 0.00202 -0.00172
(0.00118) (0.00117) (0.00164) (0.00162) (0.00171) (0.00168)
Mum inc > median -0.000910 -0.000283 -0.00252 0.000346 0.000790 -0.000981
(0.00118) (0.00116) (0.00163) (0.00161) (0.00170) (0.00168)
Dad inc > median 0.00110 -0.000767 0.00126 -0.000625 0.000930 -0.000917
(0.00118) (0.00116) (0.00163) (0.00161) (0.00170) (0.00167)
Par non-nordic 0.000163 0.000414 0.00140 -0.000471 -0.00116 0.00135
(0.00105) (0.00104) (0.00145) (0.00144) (0.00151) (0.00149)
Note: Balance tests. Fixed bandwidth 120 days +/- either side of 20th birthday. Each table row represents an
outcome variable and each cell shows a sharp diff-in-disc estimate contrasting the changes at age 20 for the
cohort turning 20 in 2017 (2018) and the pre-DCT cohort. Robust standard errors clustered by individual in
parentheses. * P<0.1, ** P<0.05, *** P<0.01.
33
Table H.2: IV estimates over different bandwidths
ALL MEN WOMEN
No Covs With Covs No Covs With Covs No Covs With Covs
BW 30 -0.80 -0.79 -3.69 -3.69 0.04 0.04
(-2.06, 0.46) (-2.05, 0.47) (-7.32, -0.05) (-7.32, -0.05) (-1.16, 1.25) (-1.16, 1.24)
KP F-stat 34.00 34.25 5.55 5.61 19.17 19.40
BW 60 -0.23 -0.22 -0.75 -0.75 -0.05 -0.05
(-0.83, 0.37) (-0.82, 0.38) (-2.22, 0.72) (-2.22, 0.72) (-0.65, 0.55) (-0.64, 0.55)
KP F-stat 93.51 93.86 19.94 20.06 55.21 55.56
BW 90 -0.44 -0.43 -0.96 -0.96 -0.27 -0.26
(-0.88, 0.00) (-0.87, 0.02) (-2.14, 0.21) (-2.14, 0.22) (-0.71, 0.18) (-0.70, 0.19)
KP F-stat 126.15 126.54 29.36 29.47 81.67 82.07
BW 120 -0.45 -0.44 -1.03 -1.03 -0.26 -0.25
(-0.84, -0.06) (-0.83, -0.05) (-2.10, 0.05) (-2.10, 0.05) (-0.65, 0.12) (-0.64, 0.13)
KP F-stat 163.78 164.17 37.76 37.87 103.36 103.75
BW 150 -0.47 -0.46 -1.12 -1.13 -0.27 -0.26
(-0.85, -0.09) (-0.84, -0.09) (-2.20, -0.05) (-2.21, -0.05) (-0.63, 0.09) (-0.62, 0.10)
KP F-stat 176.89 177.27 37.17 37.25 117.75 118.12
BW 180 -0.37 -0.36 -0.84 -0.86 -0.22 -0.20
(-0.71, -0.03) (-0.70, -0.02) (-1.78, 0.10) (-1.80, 0.08) (-0.55, 0.12) (-0.54, 0.13)
KP F-stat 207.79 208.30 47.23 47.31 137.42 137.82
BW 365 -0.26 -0.26 -0.71 -0.74 -0.14 -0.14
(-0.51, -0.01) (-0.50, -0.01) (-1.56, 0.15) (-1.60, 0.12) (-0.36, 0.09) (-0.36, 0.09)
KP F-stat 348.22 349.27 57.53 57.46 279.83 280.72
BW opt -0.45 -0.40 -0.46 -0.56 -0.18 -0.13
(-0.86, -0.05) (-0.82, 0.01) (-1.63, 0.71) (-1.75, 0.63) (-0.59, 0.23) (-0.56, 0.29)
KP F-stat 154.77 146.39 29.31 28.57 89.28 81.53
Note: The table shows IV estimates for the full sample and for men and women separately. Each panel, other than
the last one, is associated with a fixed bandwidth on either side of the the threshold (and for both cohorts). The
last panel presents the results using a MSE-optimal bandwidth. 95% confidence intervals shown in parentheses,
derived from standard errors clustered by the running variable (days since 20th birthday with separate clusters for
pre- and post cohorts).
34
H.6 Placebo tests using 19th and 21th birthdays
Table H.3 displays a placebo analysis using alternative birthdays as thresholds in
sharp diff-in-discs for both DCT and in-person consultations. The thresholds used
are the 19th and the 21th birthday. For simplicity, these are coded as 365 days
from the 20th birthday.
Table H.3: Placebo:Alternative birthday
All Men Women
A. First stage (19th birthday)
Online consultations -0.00547 -0.00547 -0.00547
(0.00631) (0.00631) (0.00631)
B. Reduced form (19th birthday)
In-person visits -0.00234 -0.00627 0.00197
(0.0251) (0.0286) (0.0405)
C. First stage (21 birthday)
In-person visits -0.00329 -0.00329 -0.00329
(0.00724) (0.00724) (0.00724)
D. Reduced form (21 birthday)
In-person visits 0.00831 0.0131 0.00281
(0.0264) (0.0286) (0.0420)
Note: Variable names in left column represent the dependent
variable. Data are collapsed by region, gender, birth year and
day relative to 20th birthday. The first two panels display sharp
diff-in-discs using the 365th day before the 20th birthday as
the threshold. The last two panels display a sharp diff-in-disc
using the 365th day after the 20th birthday as the threshold.
Standard errors clustered by the running variable, with sepa-
rate clusters for pre-post periods. * P<0.1, ** P<0.05, ***
P<0.01.
35
H.7 Sensitivity to inclusion of pre-periods of varying length
Figure H.5 plots annual in-person physician visits per capita by the day relative
to the 20th birthday (i.e., the DCT user fee threshold) for 6 time periods, each
corresponding to one year. The first four sub-graphs show each of the four years
preceding the introduction of DCT services in mid-2016. The last two sub-graphs
present figures for 2017 and 2018 (when the DCT market emerged).
In the pre-period, the 20th birthday was associated with a drop in the number
of in-person visits. These graphs show that the drop at the 20th birthday was stable
during the pre-period (although somewhat smaller in the the first year, July 2012
– June 2013). The drop in 2017 is similar to the years before the DCT emerged,
while in 2018 we no longer observe a drop at the 20th birthday.
The thin black lines on each side of the user fee threshold reflect how the data
portrayed in the figure are used to estimate the diff-in-disc estimates. The length
of the lines corresponds to the 120 days bandwidth.
2012 Jul - 2013 Jun 2013 Jul - 2014 Jun 2014 Jul - 2015 Jun
-400 -200 0 200 400 -400 -200 0 200 400 -400 -200 0 200 400
2015 Jul - 2016 Jun 2017 2018
-400 -200 0 200 400 -400 -200 0 200 400 -400 -200 0 200 400
Figure H.5: Physician visits over time, separate graph for each pre-period
Table H.4 presents our main specification (fuzzy diff-in-disc) estimates for
different lengths of the pre-period. The first row presents estimates using only the
last 12 months before June 31 2016. Each row then adds another 12 months to the
pre-period. The last row presents the estimates from the main results in Table 2.
For correspondence with the main results, a 120-days bandwidths are used for
based on the sample using the full pre-period. Overall, the results are similar
across various length of the pre-period. Compared to our main estimate, which
includes all pre-periods, we would obtain a slightly higher degree of substitution
36
In-person visits In-person visits
.8 .9 1 1.1 1.2 .8 .9 1 1.1 1.2
In-person visits In-person visits
.8 .9 1 1.1 1.2 .8 .9 1 1.1 1.2
In-person visits In-person visits
.8 .9 1 1.1 1.2 .8 .9 1 1.1 1.2
Table H.4: Main IV result over different pre-periods
All Men Women
2015 Jul - 2016 Jun -0.584** -0.687 -0.544**
(0.246) (0.623) (0.262)
2014 Jul - 2016 Jun -0.507** -1.064* -0.324
(0.213) (0.577) (0.212)
2013 Jul - 2016 Jun -0.502** -1.127** -0.299
(0.198) (0.560) (0.195)
2012 Jul - 2016 Jun -0.450** -1.026* -0.260
(0.200) (0.549) (0.196)
Avg. ind/day pre 151525 78130 73395
Avg. ind/day post 31490 16302 15188
Note: The table present fuzzy diff-in-disc estimates using
various lengths of the pre-period. The first row presents es-
timates using only the last 12 months before June 31 2016.
Each row than adds another 12 months to the pre-period.
The last row presents the estimates from the main results in
Table 2. All estimates are from models using a fixed band-
width of 120-days. Avg. ind/day = average number of in-
dividuals per day in cells, shown separately for the pre co-
horts (which include several birth cohorts) and the post co-
hort (which only includes individuals turning 20 in 2018).
Estimates using data collapsed by region, gender, birth year
and day relative to 20th birthday. Standard errors clustered
by the running variable, with separate clusters for pre-post
periods. * P<0.1, ** P<0.05, *** P<0.01.
if we excluded the first period (for which the observed drop at the 20th birthday is
smaller). The estimated degree of substitution is noticeably higher if we remove
all but the last year in the pre-period (-.58, with a 95% confidence interval covering
-1 but not 0). We conclude that our main specification, which relies on more data
points, yields a conservative estimate of the degree of substitution.
37
H.8 Sensitivity to secular decrease of in-person visits
The graphs in Figure 3 suggest that there is a general decrease in the number of
in-person visits between the pre-period and the post period also among individuals
above 20 years old. The graphs indicate that this decrease is about 5 to 10%. Not
accounting for this overall reduction may bias the diff-in-disc estimates, as it is
possible that this secular decrease would have reduced the discontinuity at age 20
even in the absence of DCT user fee.
In section 3.2, we presented the diff-in-disc estimand as:
τ + − + −
θ y
(Y −Y )− (Y −Y )
= = 1 1 0 0 (1)
τDCT (DCT+−DCT−1 1 )− (DCT
+−DCT−0 0 )
To identify the degree of substitution this estimand relies on the assump-
tion that any effects of confounding treatments must be time-invariant. (Millán-
Quijano, 2020) suggests an estimand that relaxes this assumpion that in our con-
text corresponds to:
τ + −M y (Y1 −Y1 )− (1− γ)(Y
+ −
θ = = 0
−Y0 ) (2)
τDCT (DCT+1 −DCT
−
1 )− (DCT
+
0 −DCT
−
0 )
where γ is equal to the overall (proportional) reduction in the in-person con-
sultations between the two periods. Subtracting θ M from θ (i.e Eq. 1 from Eq. 2)
yields an expression of the bias of the standard diff-in-disc estimand under these
circumstances:
(Y+ −
bias = γ 0
−Y0 ) . (3)
τDCT
Thus, in order to get an estimate of the bias, we replace τDCT and (Y
+−Y−0 0 )
by estimates from the first stage regression and a standard RD-model in the pre-
period. Assuming that γ is equal to the proportional decrease in the in-person con-
sultations among individuals just above 20 (i.e. (Y+−Y+)/Y+1 0 0 ), we can use the
coefficients from our reduced form equation to compute γ . This exercise yields an
estimate of the bias of the main IV estimate amounting to .03 (which corresponds
to an overestimation of the degree of substitution of about 7 %, or 42% instead of
45%). The estimated bias for results split by gender is of similar size, or smaller.
Thus, we do not consider the general decrease in the number of in-person consul-
tations to be a major concern for our conclusions.
38
H.9 Fuzzy RD for Stockholm
The diff-in-disc assumes that any confounding policy has time-invariant effects.
An alternative estimation strategy that relaxes that assumption would be to fo-
cus exclusively on Region Stockholm, where there was no confounding change
of in-person visit fees at age 20. Assuming there are no other confounding poli-
cies at age 20, we can then use a standard fuzzy RD specification (instead of a
diff-in-disc). Table H.5 shows that such a specification yields similar estimates
at bandwidths of 90 days or more (including when we we use an MSE-optimal
bandwidth for the relevant estimation sample).
39
Table H.5: Fuzzy Regression Discontinuity, Sthlm 2018
All Men Women
BW 60 0.0844 -0.0211 0.123
(0.298) (0.736) (0.279)
KP F-stat 15.41 13.32 39.86
BW 90 -0.299 -0.528 -0.221
(0.258) (0.673) (0.251)
KP F-stat 19.86 16.65 54.03
BW 120 -0.480** -0.697 -0.398*
(0.244) (0.614) (0.239)
KP F-stat 22.63 21.59 57.11
BW 150 -0.463** -0.628 -0.399*
(0.233) (0.588) (0.223)
KP F-stat 27.68 22.92 70.60
BW 180 -0.403* -0.539 -0.343*
(0.213) (0.525) (0.206)
KP F-stat 31.74 26.73 82.35
BW 365 -0.282* -0.531 -0.199
(0.155) (0.504) (0.139)
KP F-stat 61.51 29.29 169.6
Opt BW -0.414* -0.626 -0.157
(0.243) (0.691) (0.239)
Left bw 115 78 80
Right bw 110 98 105
KP F-stat 22.72 16.62 51.63
Note: The table presents fuzzy regression dis-
continuity estimates for Stockholm in 2018
(that is, no differencing across cohort as in our
main specifications). The bandwidth is fixed
for all but the last panel, where the bandwidth
is chosen flexibly on either side of the thresh-
old. ’KP F-stat’ refers to the Kleibergen-Paap
F-statistic. Standard errors clustered by the
running variable. * P<0.1, ** P<0.05, ***
P<0.01.
40
H.10 Wild bootstrapping precision adjustment
In this section we examine the sensitivity of our results to the estimation of stan-
dard errors. A recent literature discusses methods to obtain bias-corrected esti-
mates and robust confidence intervals for standard RD settings with data-driven
bandwidth choices (Calonico, Cattaneo and Titiunik, 2014; He and Bartalotti,
2020). This literature has not yet developed methods adapted to a diff-in-disc
setting, so we modify the wild bootstrap procedure developed (and thoroughly
described in) He and Bartalotti (2020) to fit such a setting. In short, He and Bar-
talotti (2020) estimate the bias from choosing an optimal bandwidth, h, using a
higher order polynomial for a longer bandwidth b. Using a given set of h and
b this procedure consists of two algorithms (both h and b are allowed to vary
each side of the threshold). The first algorithm estimates the bias, and the sec-
ond algorithm estimates the distribution. Both algorithms rely on a higher order
polynomial with bandwidth b mimicking the data generated process, and the esti-
mation of linear polynomials with bandwidth h of a dataset obtained from the data
generating process. Using the notation from our study setting, the procedure inHe
and Bartalotti (2020) estimates Z+c and Z
−
c for Z ∈ (DCT,y) and a single cohort,
c, in order to obtain the bias and distribution of a fuzzy RD:
RD (y
+ −
c − yc )θ = (4)
(DCT+c −DCT
−
c )
By contrast, our modified procedure estimates Z+ and Z−c c for Z ∈ (DCT,y) and
c ∈ (0,1) in order to obtain the bias and distribution of:
τ +y (y1 − y
− +
1 )− (y0 − y
−
0 )θ = = (5)
τDCT (DCT+1 −DCT
−)− (DCT+1 0 −DCT
−
0 )
The first three rows in each of the three panels of Table H.6 reproduces the
coefficients, standard errors and 95% confidence intervals obtained in our main
specification using the MSE-optimal bandwidth for the relevant estimation sam-
ple (see Table 2). The table further displays the bias corrected estimates and con-
fidence interval obtained from the bootstrap procedure.
The bias-corrected bootstrap results are overall in line with the main results.
The bias corrected coefficient for the fuzzy diff-in-disc for the full sample equals -
.51 compared to the main estimate of -.45. The bootstrapped confidence interval is
only slightly broader, and excludes both zero and one. When splitting the sample
by gender, we observe that the bias corrected coefficients are larger for women,
41
but smaller for men, compared to corresponding standard coefficient. While the
bootstrapped confidence intervals are broader, they lead to the same conclusions
as the standard one.
Notably, both bias corrected coefficients and bootstrapped confidence intervals
for the first stage are very similar to the standard coefficients and confidence inter-
vals. Thus, the difference in the fuzzy diff-in-disc comes from the bias-correction
of the reduced form.
42
Table H.6: Bootstrapped standard errors comparison
All Men Women
FIRST STAGE
Coefficient -0.149 -0.0772 -0.238
SE (0.0119) (0.0143) (0.0252)
CI (-0.172, -0.125) (-0.105, -0.0493) (-0.287, -0.189)
Bias Corrected Coeff -0.151 -0.0828 -0.245
Bootstrapped CI [-0.176, -0.128] [-0.119, -0.0622] [-0.301, -0.201]
F-stat 154.8 29.31 89.28
REDUCED FORM
Coefficient 0.0673 0.0356 0.0437
SE (0.0307) (0.0464) (0.0500)
CI (0.00704, 0.128) (-0.0553, 0.126) (-0.0543, 0.142)
Bias Corrected Coeff 0.0767 0.0293 0.0617
Bootstrapped CI [0.0185, 0.148] [-0.0731, 0.115] [-0.0244, 0.180]
IV
Coefficient -0.453 -0.461 -0.184
SE (0.206) (0.596) (0.210)
CI (-0.857, -0.0496) (-1.629, 0.708) (-0.595, 0.227)
Bias Corrected Coeff -0.510 -0.272 -0.248
Bootstrapped CI [-0.989, -0.0800] [-1.599, 1.317] [-0.718, 0.112]
F-stat 35.01 6.656 24.30
Note: The table shows IV results with sample split by gender and two sets of coef-
ficients, standard errors and confidence inervals. Each model uses the MSE-optimal
bandwidth for In-person visits varying by the sex; see bandwidth in first three columns
in Table 2. In each panel, the first three rows (Coefficient, SE, CI) represent the results
as we have presented so far with the standard way of calculating standard errors. The
following two rows in each panel (Bias corrected coeff, Bootstrapped CI) present the
same IV coefficient with bootstrapped standard errors corrected for potential bias, and
the corrsponding confidence intervals.
43
H.11 Main specification on individual-level data
Table H.7 shows the main specification estimated on individual-level daily data
using a fixed bandwidth of 120 days. In this specification, we cluster standard
errors at the individual to account for autocorrelation. The coefficients differ only
on the third decimal. Just as for the collapsed data, the individual-level data gives
us a 95% confidence interval that excludes both 0 and -1 (CI, -.897, -.027), i.e. we
may rule out both zero and complete substitution.
Table H.7: Fuzzy diff-in-disc results; individual-level data
All Men Women
A. FIRST STAGE (SHARP DIFF-IN-DISC)
DCT consultations -0.149*** -0.0700*** -0.234***
(0.0148) (0.0131) (0.0273)
B. RF (SHARP DIFF-IN-DISC)
In-person visits 0.0670** 0.0718* 0.0611
(0.0320) (0.0394) (0.0511)
C. IV (FUZZY DIFF-IN-DISC)
In-person visists -0.450** -1.026* -0.260
(0.220) (0.598) (0.221)
Observations 43923493 22663627 21259866
Individuals 232657 119704 112953
Note: Estimations using individual-level daily data. Variable names in left column
represent the dependent variable. Each model uses a bandwidth of 120 days before
and after the 20th birthday. Standard errors in parentheses clustered by individual.
* P<0.1, ** P<0.05, *** P<0.01.
44
Appendix I Decomposition by type of visits
Table I.1 displays sharp diff-in-disc estimates of the DCT user fee on DCT and
in-person consultations decomposed by whether it is an initial contact in a care
episode or a follow-up consultation. An initial consultation is defined as a consul-
tation without either a DCT or an in-person consultation during the preceding 7,
14 or 28 days. (The wash-out period of 14 days is used in the analysis presented in
the main text.) The remaining consultations are defined as follow-ups (although
these may or may not be an actual part of the same care episode). Overall, the
decomposition results are similar across wash-out periods. The onset of the user
fee leads to a reduction in initial DCT consultations and DCT follow-ups. These
decreases are complemented by an increase in initial in-person visits, but not in
in-person follow-ups (in particular for wash-out periods longer than 7 days).
Table I.2 presents a further decomposition for the 14 days wash-out period
used in the main document. Initial consultations are decomposed into consulta-
tions with and without follow-ups (and by mode of follow-ups: only by the same
mode (DCT/in-person) or any follow-up by the other mode (DCT/in-person)).
Follow-up consultations are decomposed by the mode of the initial consultation
of the episode (DCT/in-person).
45
Table I.1: Decomposition of intitial and follow-up consultations
All Men Women
A. 7 DAYS WASH OUT
Initial DCT consultation -0.123*** -0.0578*** -0.194***
(0.0111) (0.0109) (0.0218)
DCT follow-ups -0.0258*** -0.0122*** -0.0405***
(0.00466) (0.00385) (0.00884)
Initial in-person visits 0.0558** 0.0488 0.0627
(0.0277) (0.0346) (0.0442)
Follow up in-person visits 0.0112 0.0230* -0.00162
(0.00941) (0.0118) (0.0146)
B. 14 DAYS WASH OUT
Initial DCT consultation -0.114*** -0.0563*** -0.176***
(0.0110) (0.0110) (0.0210)
DCT follow up 14 days -0.0352*** -0.0137*** -0.0584***
(0.00530) (0.00479) (0.00972)
Initial in-person visits 0.0662** 0.0503 0.0826**
(0.0263) (0.0338) (0.0415)
Follow up in-person visits 0.000861 0.0215 -0.0216
(0.0120) (0.0148) (0.0179)
C. 28 DAYS WASH OUT
Initial DCT consultation -0.0969*** -0.0528*** -0.145***
(0.0105) (0.0105) (0.0191)
DCT follow-ups -0.0520*** -0.0172*** -0.0897***
(0.00679) (0.00609) (0.0125)
Initial in-person visits 0.0692*** 0.0536 0.0855**
(0.0252) (0.0342) (0.0392)
In-person follow-ups -0.00217 0.0182 -0.0244
(0.0151) (0.0183) (0.0234)
Note: Estimations using individual-level daily data. Variable names in left column
represent the dependent variable. The analysis decompose DCT and in person visits
in initial and follow up visits (with a previous contact within 7/14/28 days). Each
model uses a bandwidth of 120 days before and after the 20th birthday. Standard
errors clustered by the running variable, with separate clusters for pre-post periods.
* P<0.1, ** P<0.05, *** P<0.01.
46
Table I.2: Further decomposition of consultations
All Men Women
A. DCT CONSULTATIONS
Initial DCT (All) -0.114*** -0.0563*** -0.176***
(0.0110) (0.0110) (0.0210)
-no follow-up -0.0938*** -0.0503*** -0.141***
(0.00965) (0.00981) (0.0178)
-any follow-up -0.0205*** -0.00562 -0.0365***
(0.00472) (0.00482) (0.00888)
-only DCT follow-ups -0.0101*** 0.0000182 -0.0211***
(0.00331) (0.00286) (0.00648)
-any in-person follow-ups -0.0103*** -0.00564 -0.0155**
(0.00376) (0.00373) (0.00689)
DCT follow up (All) -0.0352*** -0.0137*** -0.0584***
(0.00530) (0.00479) (0.00972)
-initial DCT -0.0300*** -0.00834* -0.0535***
(0.00490) (0.00425) (0.00901)
-initial in-person -0.00512* -0.00532* -0.00492
(0.00262) (0.00284) (0.00425)
B. IN-PERSON VISITS
Initial in-person (All) 0.0662** 0.0503 0.0826**
(0.0263) (0.0338) (0.0415)
-no follow-up 0.0576** 0.0431 0.0728**
(0.0242) (0.0320) (0.0370)
-any follow-up 0.00955 0.00690 0.0123
(0.00971) (0.0122) (0.0161)
-only in-person 0.0144 0.0125 0.0164
(0.00964) (0.0122) (0.0157)
-any DCT follow-up -0.00490** -0.00560** -0.00415
(0.00229) (0.00222) (0.00429)
In-person follow-ups (All) 0.000861 0.0215 -0.0216
(0.0120) (0.0148) (0.0179)
-initial in-person 0.0130 0.0224 0.00275
(0.0102) (0.0138) (0.0151)
-initial DCT -0.0122** -0.000916 -0.0243***
(0.00515) (0.00458) (0.00835)
Note: Estimations using individual-level daily data. Variable names in left column
represent the dependent variable. The analysis decompose DCT and in person visits
in initial and follow up visits (i.e. a consultation that followed a previous consul-
tation within 14 days). Each model uses a bandwidth of 120 days before and after
the 20th birthday. Standard errors clustered by the running variable, with separate
clusters for pre-post periods. * P<0.1, ** P<0.05, *** P<0.01.
47
Appendix J Robustness analysis of diagnosis data
Tables J.1, J.2, J.3, and J.4 present results using the same estimation approach as
in the main text but for a range of fixed bandwidths. Table J.5 shows estimations
similar to the corresponding table in the main text but with optimal bandwidth for
the outcome of interest.
The results from estimations using other bandwidths support the main pattern
of the substitution within diagnosis groups from table J.5. The results are overall
similar, suggesting an overall degree of substitution of about 30-45%, except for
the shortest bandwidth of 60 days. The result that consultations related to up-
per respiratory infections display the highest degree of substitution holds for all
bandwidths.
The overall pattern from Table 3 remains also when splitting the sample by
gender. Although there is more variation in size across bandwidths, consultations
related to upper respiratory infections display the highest degree of substitution
for both genders. The first stage estimates are similar across bandwidths for both
genders (and jointly). The variation comes from the reduced form (the sharp diff-
in-disc in-person visits).
48
Table J.1: Decomposition by type of diagnosis, (bw=60)
A. DCT CONSULTATIONS (SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.133*** -0.0218*** -0.0481*** -0.0242*** -0.0478***
(0.0154) (0.00808) (0.00845) (0.00736) (0.0112)
Men -0.0604*** -0.00183 -0.0329*** -0.00553** -0.0224**
(0.0141) (0.00794) (0.00843) (0.00271) (0.00900)
Women -0.212*** -0.0434*** -0.0646*** -0.0445*** -0.0753***
(0.0274) (0.0143) (0.0139) (0.0152) (0.0183)
B. SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.160 -1.424 -0.0804 0.147 0.385
(0.275) (0.956) (0.320) (0.474) (0.522)
Men -0.312 -13.72 -0.175 -2.393 1.152
(0.708) (61.82) (0.554) (2.004) (1.399)
Women -0.109 -0.860 -0.0269 0.491 0.144
(0.261) (0.650) (0.335) (0.484) (0.522)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and
the IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all
Common diagnoses set by DCT providers. In the second to fourth column, these
common diagnoses are decomposed into subgroups: upper respiratory infections
(Resp), skin related diseases (Skin), genital and reproductive organs Gen/Rep, and
Other common diagnoses. Each row presents the diff-in-disc estimates for the di-
agnosis groups for the given estimation sample (All, Men, Women). Each model
uses a fixed bandwidth of 60 days each side of the age cutoff. Estimates using data
collapsed by region, gender, year and day relative to 20th birthday. Standard er-
rors clustered by the running variable, with separate clusters for pre-post periods. *
P<0.1, ** P<0.05, *** P<0.01.
49
Table J.2: Decomposition by type of diagnosis, (bw=90)
A. DCT CONSULTATIONS (SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.132*** -0.0272*** -0.0383*** -0.0274*** -0.0480***
(0.0125) (0.00648) (0.00690) (0.00629) (0.00844)
Men -0.0601*** -0.00566 -0.0237*** -0.00466** -0.0277***
(0.0114) (0.00675) (0.00687) (0.00224) (0.00720)
Women -0.210*** -0.0504*** -0.0541*** -0.0521*** -0.0700***
(0.0227) (0.0114) (0.0114) (0.0129) (0.0142)
B. SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.378* -1.207** -0.203 -0.00501 -0.268
(0.220) (0.583) (0.324) (0.345) (0.437)
Men -0.630 -3.224 -0.384 -2.322 -0.136
(0.593) (5.282) (0.650) (1.916) (0.892)
Women -0.295 -0.957** -0.116 0.223 -0.320
(0.211) (0.468) (0.331) (0.328) (0.462)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and
the IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all
Common diagnoses set by DCT providers. In the second to fourth column, these
common diagnoses are decomposed into subgroups: upper respiratory infections
(Resp), skin related diseases (Skin), genital and reproductive organs Gen/Rep, and
Other common diagnoses. Each row presents the diff-in-disc estimates for the di-
agnosis groups for the given estimation sample (All, Men, Women). Each model
uses a fixed bandwidth of 90 days each side of the age cutoff. Estimates using data
collapsed by region, gender, year and day relative to 20th birthday. Standard er-
rors clustered by the running variable, with separate clusters for pre-post periods. *
P<0.1, ** P<0.05, *** P<0.01.
50
Table J.3: Decomposition by type of diagnosis, (bw=150)
A. DCT CONSULTATIONS (SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.129*** -0.0263*** -0.0376*** -0.0246*** -0.0457***
(0.00933) (0.00486) (0.00514) (0.00510) (0.00590)
Men -0.0584*** -0.00798 -0.0197*** -0.00178 -0.0308***
(0.00887) (0.00539) (0.00526) (0.00179) (0.00555)
Women -0.204*** -0.0460*** -0.0570*** -0.0493*** -0.0619***
(0.0175) (0.00860) (0.00850) (0.0104) (0.0102)
B. SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.422** -1.589*** 0.0381 0.167 -0.545
(0.178) (0.488) (0.262) (0.296) (0.365)
Men -0.773 -2.460 -0.257 -1.764 -0.797
(0.489) (2.518) (0.636) (3.694) (0.669)
Women -0.309* -1.420*** 0.149 0.247 -0.403
(0.171) (0.434) (0.254) (0.275) (0.404)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and the
IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all Com-
mon diagnoses set by DCT providers. In the second to fourth column, these com-
mon diagnoses are decomposed into subgroups: upper respiratory infections (Resp),
skin related diseases (Skin), genital and reproductive organs Gen/Rep, and Other
common diagnoses. Each row presents the diff-in-disc estimates for the diagno-
sis groups for the given estimation sample (All, Men, Women). Each model uses
a fixed bandwidth of 150 days each side of the age cutoff. Estimates using data
collapsed by region, gender, year and day relative to 20th birthday. Standard er-
rors clustered by the running variable, with separate clusters for pre-post periods. *
P<0.1, ** P<0.05, *** P<0.01.
51
Table J.4: Decomposition by type of diagnosis, (bw=180)
A. DCT CONSULTATIONS (SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.129*** -0.0297*** -0.0339*** -0.0257*** -0.0459***
(0.00834) (0.00432) (0.00483) (0.00454) (0.00542)
Men -0.0568*** -0.0109** -0.0168*** -0.000292 -0.0306***
(0.00808) (0.00494) (0.00490) (0.00169) (0.00499)
Women -0.208*** -0.0500*** -0.0526*** -0.0534*** -0.0626***
(0.0157) (0.00761) (0.00798) (0.00931) (0.00952)
B. SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.308* -0.882** -0.0191 0.0997 -0.397
(0.157) (0.349) (0.263) (0.263) (0.322)
Men -0.739 -1.193 -0.348 -24.77 -0.741
(0.454) (1.322) (0.685) (143.7) (0.607)
Women -0.174 -0.800** 0.0951 0.252 -0.205
(0.153) (0.322) (0.250) (0.238) (0.362)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and the
IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all Com-
mon diagnoses set by DCT providers. In the second to fourth column, these com-
mon diagnoses are decomposed into subgroups: upper respiratory infections (Resp),
skin related diseases (Skin), genital and reproductive organs Gen/Rep, and Other
common diagnoses. Each row presents the diff-in-disc estimates for the diagno-
sis groups for the given estimation sample (All, Men, Women). Each model uses
a fixed bandwidth of 180 days each side of the age cutoff. Estimates using data
collapsed by region, gender, year and day relative to 20th birthday. Standard er-
rors clustered by the running variable, with separate clusters for pre-post periods. *
P<0.1, ** P<0.05, *** P<0.01.
52
Table J.5: Decomposition by type of diagnosis (optimal bandwidth)
A:DCT CONSULTATIONS (FIRST STAGE: SHARP DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.129*** -0.0256*** -0.0379*** -0.0259*** -0.0468***
(0.0110) (0.00572) (0.00609) (0.00584) (0.00749)
Men -0.0600*** -0.00581 -0.0206*** -0.00484** -0.0298***
(0.0105) (0.00618) (0.00632) (0.00207) (0.00679)
Women -0.204*** -0.0468*** -0.0566*** -0.0487*** -0.0651***
(0.0208) (0.0104) (0.0103) (0.0119) (0.0129)
B: SUBSTITUTION (FUZZY DIFF-IN-DISC)
Common Resp Skin Gen/Rep Other
All -0.440** -1.447** -0.105 -0.0393 -0.472
(0.205) (0.575) (0.301) (0.346) (0.417)
Men -0.624 -2.253 -0.391 -1.169 -0.478
(0.546) (3.886) (0.695) (1.501) (0.771)
Women -0.378* -1.334*** 0.00721 0.0863 -0.464
(0.202) (0.504) (0.294) (0.329) (0.460)
Note: Table shows results from the first stage equation (sharp diff-in-disc) and the
IV-model (fuzzy-diff-in-disc) by diagnosis groups. The first column shows all Com-
mon diagnoses set by DCT providers. In the second to fourth column, these com-
mon diagnoses are decomposed into subgroups: upper respiratory infections (Resp),
skin related diseases (Skin), genital and reproductive organs Gen/Rep, and Other
common diagnoses. Each row presents the diff-in-disc estimates for the diagnosis
groups for the given estimation sample (All, Men, Women). All models apply the
bandwidths used for the main results in Table 2 (i.e., MSE-optimal bandwidths for
In-person visits for both regions and both genders). Estimates using data collapsed
by region, gender, year and day relative to 20th birthday. Standard errors clustered
by the running variable, with separate clusters for pre-post periods. * P<0.1, **
P<0.05, *** P<0.01.
53
Appendix K Prescriptions of antibiotics
Table K.1 shows sharp diff-in-disc estimates of the effects of the onset of the DCT
user fee on various outcomes related to antibiotic prescriptions.4 The positive and
significant coefficients in the first column indicate that the user fee if anything
has a small positive effect on the total number of antibiotic prescriptions. That is,
the larger use of DCT consultations among 19–year-olds relative to 20–year-olds
does not increase their antibiotics consumption, but rather decreases it.
The results by antibiotic type (columns 2-4) further suggest that the increase
is primarily driven by prescriptions of antibiotics related to respiratory infections
(rather than antibiotics related to skin conditions or cystitis). Together with the
results in section 4.2.2, which suggest that close to all DCT consultations for res-
piratory infection replace in-person visits, we interpret these results as indications
of physicians being more (or at least not less) restrictive during online consulta-
tions in terms of prescribing antibiotics for respiratory infections.
As for the main analysis, the diff-in-disc estimates are sensitive to general
trends that would affect the size of the drop at the 20th birthday even without
the onset of the DCT user fee. Indeed, antibiotic use has declined over the last
decade, and a proportional decrease in the number of prescriptions for individuals
each side of the cut-off would generate positive diff-in-disc estimates as observed
in Table K.1. We therefore study the components of the diff-in-disc estimates -
the pre and post RD estimates. We also specifically estimate a standard RD in
Stockholm, where there is no confounding user fee for in-person visits.
Table K.2 shows results from sharp RD before (panel A includes the com-
plete pre-period) and after (panel B includes 2018) the introduction of the DCT-
services. The 20th birthday is associated with a decrease in the number of pre-
scriptions and defined daily doses (DDD) before the DCT, likely driven by the
onset of the user fee in the region of Västra Götaland. In 2018, the same discon-
tinuity is associated with an insignificant increase in the number of prescriptions
and DDDs. Thus, these results implies that the estimates in table K.1 is not only
driven by a general decrease in the number of prescriptions but an actual change
in the sign of the effect at the discontinuity.
Table K.3 presents results for the same outcomes as in the previous tables
from a sharp diff-in-disc (panel A) and a RD in 2018 (panel B) for the region
4Antibiotics is defined as all at-codes within J01 except Metenamin J01XX05. We follow
the Public Health Agency of Sweden defining three categories of antibiotics relating to respira-
tory infections (J01AA02, J01CE02, J01CA04, J01CR02, J01DB, J01DC, J01DE, J01FA) cystitis
(J01CA08, J01EA01, J01MA02, J01MA06, J01XE01) Skin and soft tissues (J01FF01, J01CF05).
54
Table K.1: Antibiotic prescription, sharp diff-in-disc
A. SHARP DIFF-IN-DISC
All Resp Skin Cystit DDD
All 0.0368*** 0.0161 0.00724 0.00865 0.557
(0.0141) (0.0108) (0.00609) (0.00746) (0.442)
Men 0.0223 0.00680 0.00270 0.000913 0.733
(0.0173) (0.0130) (0.00904) (0.00381) (0.578)
Women 0.0519** 0.0259 0.0121 0.0167 0.365
(0.0230) (0.0172) (0.00867) (0.0149) (0.557)
Note: Table shows results from a sharp diff-in-disc (reduced form) on
antibiotic prescriptions. The first column presents diff-in-disc estimates
for the total number of antibiotic prescriptions, columns 2 to 4 present
estimates for the number of prescriptions of various types: respiratory
infections, skin conditions, and cystit. Column 5 presents estimates for
the total number of Defined Daily Doses (DDD). Standard errors clus-
tered by the running variable, with separate clusters for pre-post peri-
ods. Each model uses a fixed bandwidth of 120 days on each side of the
threshold (20th birthday). * P<0.1, ** P<0.05, *** P<0.01.
of Stockholm. Although the RD estimates are all insignificant and tend to be
smaller than the diff-in-disc estimates, the conclusion is still that physicians are
not less restrictive in terms of prescribing antibiotics during DCT consultations –
if anything they are more restrictive.
55
Table K.2: Sharp RD of antibiotic prescriptions, pre/post
A: SHARP RD OF ANTIBIOTIC PRESCRIPTIONS, PRE
All Resp Skin Cystit DDD
All -0.0177** -0.0141*** -0.000852 -0.00348 -0.317
(0.00699) (0.00469) (0.00268) (0.00356) (0.206)
Men -0.0112 -0.0119** 0.00000120 -0.000440 -0.178
(0.00825) (0.00587) (0.00351) (0.00160) (0.280)
Women -0.0245** -0.0164** -0.00176 -0.00668 -0.464
(0.0120) (0.00740) (0.00442) (0.00744) (0.295)
B: SHARP RD OF ANTIBIOTIC PRESCRIPTIONS, 2018
All Resp Skin Cystit DDD
All 0.0187 0.00426 0.00683 0.00473 0.211
(0.0133) (0.0104) (0.00592) (0.00721) (0.403)
Men 0.0111 -0.00509 0.00270 0.000473 0.554
(0.0153) (0.0116) (0.00836) (0.00346) (0.507)
Women 0.0274 0.00951 0.0104 0.0100 -0.0989
(0.0197) (0.0156) (0.00748) (0.0130) (0.474)
Note: Table shows results from a sharp regression discontinuity before (panel
A) and after (paned B) the introduction of DCT-services. In panel B, we show
results for 2018. The first row presents the RD-estimates for any type of pre-
scriptions, columns 2 to 4 present estimates for antibiotic types related to res-
piratory infections, skin conditions, and cystit. Column 5 presents the same
estimate for the total number of Defined Daily Doses (DDD). Each model uses
a fixed bandwidth of 120 days on each side of the threshold (20th birthday).
Data is collapsed by gender, year and day relative to 20th birthday. Standard
errors clustered by the running variable. * P<0.1, ** P<0.05, *** P<0.01.
56
Table K.3: Sharp RD of antibiotic prescriptions (Stockholm 2018)
A: DIFF-IN-DISC
All Resp Skin Cystit DDD
All 0.0428** 0.0238 0.00292 0.0124 0.826
(0.0196) (0.0150) (0.00877) (0.0101) (0.545)
Men 0.0154 0.0131 -0.00601 0.00251 0.732
(0.0239) (0.0181) (0.0127) (0.00523) (0.673)
Women 0.0711** 0.0348 0.0124 0.0224 0.919
(0.0321) (0.0239) (0.0114) (0.0202) (0.811)
B: REGRESSION DISCONTINUITY
All Resp Skin Cystit DDD
All 0.0279 0.0106 0.00117 0.00719 0.717
(0.0173) (0.0136) (0.00811) (0.00863) (0.465)
Men 0.0111 0.00649 -0.00673 0.00130 0.805
(0.0214) (0.0163) (0.0120) (0.00479) (0.562)
Women 0.0449 0.0147 0.00959 0.0130 0.617
(0.0281) (0.0220) (0.00994) (0.0171) (0.681)
Note: The table shows estimates from sharp diff-ind-disc in Stockholm
in panel A and a sharp regression discontinuity for the Stockholm re-
gion using data from 2018 in panel B. The first column presents esti-
mates for any type of prescriptions, columns 2 to 4 present estimates
for antibiotic types related to respiratory infections, skin conditions,
and cystit. Column 5 presents the same estimate for the total number
of Defined Daily Doses. Each model uses a fixed bandwidth of 120
days on each side of the threshold (20th birthday). Data is collapsed
by gender, year and day relative to 20th birthday. Standard errors clus-
tered by the running variable. * P<0.1, ** P<0.05, *** P<0.01.
57
Appendix L Other primary care consultations
Table L.1 displays fuzzy diff-in-disc results for primary care consultations with
other health care professionals than physicians (e.g., nurses) using other band-
widths than Table 4 in the main text. (Table L.2 displays the estimates for all
consultations (i.e., the sum of consultations with other health care professionals
and physicians) for other bandwidths than in the main text.) Table L.3 shows the
same estimations using a MSE-optimal bandwidth. The patterns are consistent
across tables and bandwidths. The results suggest that the degree of substitution
is slightly larger when including consultations with other health care profession-
als than physicians. The estimate suggests that there is partial substitution, but the
estimates are noisier than when we include physician consultations only and the
confidence intervals no longer exclude full substitution.
For women, there is a consistent negative (but insignificant) coefficient on con-
sultations at midwife/youth/STD clinics. To further explore this pattern, we retain
the same outcome variable but restrict the first stage to only include online con-
sultations with diagnoses related to the genital and reproductive organs. Table L.4
presents the fuzzy diff-in-disc estimates. For women, the results hover around -1,
which suggests that women substitute online consultations (with physicians) for
in-person midwife visits related to contraceptive management. In other words,
the observed the lack of substitution between in-person physician visits and on-
line consultations for these diagnoses (Table J.5) is explained by another type of
substitution. Note that the large positive coefficients among men is primarily due
to a very weak first stage (and likely relate to visits at a STD or youth clinic).
58
Table L.1: Visits to other health care professionals (fixed bandwidth)
FUZZY DIFF-IN-DISC
All Men Women
BANDWIDTH=60
Nurse visits at a PCC -0.217 -0.474 -0.130
(0.166) (0.416) (0.176)
Visits at midwife/youth/STD clinic 0.00387 0.457 -0.136
(0.260) (0.279) (0.328)
Nurse+midwife/youth/STD -0.213 -0.0166 -0.266
(0.311) (0.458) (0.391)
BANDWIDTH=90
Nurse visits at a PCC -0.0792 -0.721** 0.129
(0.132) (0.351) (0.140)
Visits at a midwife/youth/STD clinic -0.0717 0.361 -0.201
(0.205) (0.228) (0.258)
Nurse+midwife/youth/STD -0.151 -0.360 -0.0719
(0.243) (0.389) (0.295)
BANDWIDTH=150
Nurse visits at a PCC -0.0431 -0.632** 0.135
(0.110) (0.317) (0.109)
Visits at a midwife/youth/STD clinic -0.0734 0.477** -0.227
(0.163) (0.217) (0.199)
Nurse+midwife/youth/STD -0.117 -0.155 -0.0918
(0.197) (0.346) (0.226)
BANDWIDTH=180
Nurse visits at a PCC -0.0594 -0.608** 0.113
(0.104) (0.284) (0.104)
Visits at a midwife/youth/STD clinic -0.151 0.459** -0.323*
(0.150) (0.199) (0.184)
Nurse+midwife/youth/STD -0.211 -0.149 -0.209
(0.184) (0.321) (0.210)
Note: Table shows fuzzy diff-in-discs estimates of the effect of online con-
sultations on in-person consultations with other health care professionals
than physicians at primary care centers: consultations with a nurse at a pri-
mary care center; consultations with a midwife/nurse/physician at a mid-
wife/youth/STD clinic, and the sum of these two types of consultations. ;
and the sum of all consultations, , including physician consultations at a
primary care center. Each model uses a fixed bandwidth of 60/90/150/180
days each side of the threshold. Standard errors clustered by the running
variable, with separate clusters for pre-post periods. * P<0.1, ** P<0.05,
*** P<0.01.
59
Table L.2: Visits to other health care professionals, fixed bandwidth
FUZZY DIFF-IN-DISC
All Men Women
BANDWIDTH=60
All consultations -0.441 -0.769 -0.316
(0.497) (0.929) (0.552)
BANDWIDTH=90
All consultations -0.591 -1.325* -0.338
(0.376) (0.755) (0.412)
BANDWIDTH=150
All consultations -0.588** -1.277* -0.363
(0.299) (0.686) (0.308)
BANDWIDTH=180
All consultations -0.582** -0.991 -0.426
(0.273) (0.613) (0.282)
Note: Table shows fuzzy diff-in-discs estimates of
the effect of DCT consultations on in-person consul-
tations with physicians or other health care profes-
sionals at primary care centers. (consultations with
a nurse at a primary care center; consultations with
a midwife/nurse/physician at a midwife/youth/STD
clinic, and physician consultations at a primary care
center). Each model uses a fixed bandwidth of
60/90/150/180 days each side of the threshold. Stan-
dard errors clustered by the running variable, with
separate clusters for pre-post periods. * P<0.1, **
P<0.05, *** P<0.01.
Appendix M Cost analysis
M.1 Overview and sources
We lack information to perform a full cost-benefit analysis,5 but we are able to
calculate a rough estimate of the relative costs of providing a consultation in the
DCT and PCC settings, respectively.
5For instance, we cannot value the utility of patients’ decreased worry for minor conditions
that DCT but not PCCs would accept to take care of.
60
Table L.3: Visits to other health care professionals
FUZZY DIFF-IN-DISC
All Men Women
Nurse visits at a PCC -0.0268 -0.519 0.137
(0.123) (0.312) (0.127)
Visits at midwife/youth/STD clinic -0.0611 0.376 -0.197
(0.183) (0.212) (0.230)
Nurse+midwife/youth/STD -0.0879 -0.143 -0.0602
(0.220) (0.347) (0.265)
All consultations -0.541 -1.032 -0.366
(0.331) (0.691) (0.360)
Note: Table shows fuzzy diff-in-discs estimates of the effect of on-
line consultations on in-person consultations with other health care
professionals than physicians at primary care centers: consultations
with a nurse at a primary care center; consultations with a mid-
wife/nurse/physician at a midwife/youth/STD clinic, the sum of these
two types of consultations; and the sum of all consultations, including
physician consultations at a primary care center. Each model uses the
MSE-optimal bandwidth for main outcome variable (in-person con-
sultations) for both regions and both genders (to remove heterogeneity
due to changes in bandwidth). Standard errors clustered by the run-
ning variable, with separate clusters for pre-post periods. P<0.1, **
P<0.05, *** P<0.01.
Dahlstrand (2022) provides information about the labour cost per hour and
consultation times of the largest DCT provider in Sweden. DCT physicians on
average spend 5 minutes on the direct consultation and 11 minutes on administra-
tion.
To estimate the consultation time at PCCs, we use the standard consultation
length in the Swedish system for certain private-practising physicians (“Taxeläkare”),
i.e., 25 minutes including patient administration time (see §5 of SFS (1991:1121)).
We decompose the total time into the time on the actual consultation vs adminis-
tration time using estimates from a recent time use study in Swedish PCCs (See
Table 4.1 (p. 85) in Ivarsson Westerberg et al. (2021). Notably, the cited regulated
consultation length for private practitioners aims to capture a reasonable average
for a much more heterogeneous group of patients than those seeking DCT care
(including, e.g., elderly patients with multiple illnesses). It is therefore likely
that it provides an overestimate of the time that PCC physicians would spend on
61
Table L.4: Consultations at a midwife/youth/STD clinic
FUZZY DIFF-IN-DISC
INSTRUMENT: GEN/REP DCT
All Men Women
OPTIMAL BW, MAIN OUTCOME
Visits at a midwife/youth/STD clinic -0.350 5.522 -0.939
(1.048) (3.881) (1.114)
BANDWIDTH=60
Visits at a midwife/youth/STD clinic 0.0231 5.730 -0.691
(1.549) (4.424) (1.673)
BANDWIDTH=90
Visits at a midwife/youth/STD clinic -0.395 5.509 -0.917
(1.129) (4.316) (1.192)
BANDWIDTH=120
Visits at a midwife/youth/STD clinic -0.785 7.115 -1.366
(1.085) (5.721) (1.135)
BANDWIDTH=150
Visits at a midwife/youth/STD clinic -0.426 17.00 -1.052
(0.947) (18.52) (0.943)
BANDWIDTH=180
Visits at a midwife/youth/STD clinic -0.829 101.6 -1.356*
(0.829) (589.4) (0.798)
Note: Table shows fuzzy diff-in-discs estimates of the effect of online con-
sultations (with a registered diagnosis related to genital and reproductive
health) on in-person consultations with a midwife/nurse/physician at a mid-
wife/youth/STD clinic. That is, the excluded instrument is DCT consulta-
tions with a registered diagnosis related to genital and reproductive health.
See section 4.2.2. Each model uses a fixed bandwidth of 60/12/180 days
each side of the threshold. Standard errors clustered by the running vari-
able, with separate clusters for pre-post periods. * P<0.1, ** P<0.05, ***
P<0.01.
62
DCT-equivalent complaints. In some alternative scenarios, we therefore use the
same consultation lengths for PCCs and DCT. The main reason why we use the
25 minute length in the baseline scenario is that it is possible that the in-person
setting by itself prompts the physician to perform a physical examination, and to
accompany the patient from/to the waiting room.
Dahlstrand (2022) also provides information about the company’s labour costs
(75-90 USD per hour), and the shares of physicians that are specialists (0.31)
versus still in specialty training (0.36) or residents only (0.33). We assume that
the lower wage goes for the latter two categories and calculate an average wage
cost per hour as a weighted sum (i.e., 0.69*75+0.31*90), which gives us SEK734
per hour.
We set PCC physicians’ wages to 1,000 SEK per hour, an estimate sourced
from an administrator working with the costing model in one of the Swedish re-
gions.6 We also calculate costs under the assumption of similar physician wages
in DCT and PCC in some scenarios.7
Before the actual DCT appointment, the patient has to spend some time finding
the site or application and filling out a triage form. Similarly, the patient that
contacts a PCC first has to speak to a triage nurse. We lack good sources here
so we just assume that the amount of time is similar (10 minutes) in both cases.
Notably, it is only in the PCC setting that this tiem entails a cost for the health care
system, namely the cost for the triage nurse. We use an estimate of an hourly wage
using the official average wage statistics for nurses (296 SEK per hour8 including
income and payroll taxes) with basic or specialist education.
The in-person setting also includes some other activities that only involve the
patient (i.e., do not affect the costs of the health care system). The patient will
normally have to spend time in a telephone queue before talking to the nurse, and
then in the waiting room. We set this time to 30 minutes. We further assume that
patients spend 22 minutes travelling to/from the PCC. We derive this estimate us-
ing the figures provided in a report by the Swedish Competition Authority, which
described the shares of the population in the regions that can reach their PCC
within 5, 15, 30 or above 30 minutes) (Swedish Competition Authority, 2010).9
6András Borsodi, Östergötland
7The wage differential is not particularly interesting per se: The ability to hire junior physician
is not the interesting innovation in DCT and is not feasible in a general equilibrium.
8We divide the average annual wage times by the statutory working time of 2,080 hours/year.
9Specifically, we multiply the population share in each bracket with the upper bound
for the time limit in the bracket (e.g., 5 minutes), sum over all population shares,
63
In our study regions, the vast majority have a short distance from their home to
their closest PCC. We also assume that around half of the study population lives
in Stockholm. The direct cost for a bus ticket is estimated at 80 SEK (round trip).
Patient’s opportunity cost of time is set to 100 SEK in our baseline scenario.
Assuming an opportunity cost of 100 SEK is equivalent to assuming that compli-
ers are evenly spaced in the 0-200 SEK range. Our assumption about patients’
opportunity cost of time thus reflects the idea that the sensitivity to the user fee
should tell us something about the patients’ valuation of the visit. However, 100
SEK is likely an understatement of the value of time for many adults. In an alter-
native scenario, we therefore use average wage per hour (official wage statistics,
including income and payroll tax) instead.
Finally, we impose a cost for office space of 36 SEK/visit in DCT and 200
SEK/visit in PCC. The estimate for DCT comes from Ekman (2018) and is proba-
bly an overstatement given the subsequent growth in volume (which should led to
less than proportional increases in the need for space, especially since the medical
workforce works from home). To estimate the office space costs per visit in the
PCC setting, we use data from a costing model in a Swedish region (Östergötland),
which have a comparably sophisticated such model for primary care (cost data is
generally scant in Swedish primary care). To account for the higher rents in the
our study regions, we add 70 SEK to the base 130 SEK from the costing model.
All other costs are disregarded. In reality, one may imagine that DCT providers
have lower overhead costs (since they manage a less complex service, they might
need fewer managers and administrative personnel per clinician, and they do not
need a receptionist). However, and notably, to realize such cost savings at the
provider level, some PCCs would need to close. Closures are often politically
infeasible, and where they are not, they by themselves induce counterveiling cost
increases in terms of travel time for some patients.
One notable cost from which we abstract is the development and maintenance
of the virtual consultation software. Notably, though, PCCs too nowadays use
similar systems for remote communication, so the omission is more a theoretical
than a practical problem.
weigh by the population shares living in each region (assumed 50-50 in our study pop-
ulations), and multiply by 2 to get a round trip. 22 minutes = 0.36 of an hour =
=(2*(0.5*(0.72*5+0.24*15+0.04*30)+0.5*(0.41*5+0.43*15+0.16*30)))/60
64
Table M.1: Cost items and values in scenarios 1-6
Cost component Mode (1) (2) (3) (4) (5) (6)
Physician wage (SEK) DCT 734 734 734 734 734 734
PCC 1000 1000 734 734 734 1000
Consultation (share of hour) DCT 0.08 0.08 0.08 0.08 0.08 0.08
PCC 0.21 0.08 0.21 0.08 0.08 0.21
Admin (share of hour) DCT 0.18 0.18 0.18 0.18 0.18 0.18
PCC 0.2 0.18 0.2 0.18 0.18 0.2
Triage nurse wage (SEK) DCT N/A N/A N/A N/A N/A N/A
PCC 296 296 296 296 296 296
Booking and triage (share of hour) DCT 0.17 0.17 0.17 0.17 0.17 0.17
PCC 0.17 0.17 0.17 0.17 0.17 0.17
Office space (SEK) DCT 36 36 36 36 36 0
PCC 200 200 200 200 200 0
Patient’s opportunity cost (SEK) DCT 100 100 100 100 273 100
PCC 100 100 100 100 273 100
Patient’s travel time (share of hour) DCT 0 0 0 0 0 0
PCC 0.36 0.36 0.36 0.36 0.36 0.36
Patient’s travel cost (SEK) DCT 0 0 0 0 0 0
PCC 80 80 80 80 80 80
Patient’s waiting time (share of hour) DCT 0 0 0 0 0 0
PCC 0.5 0.5 0.5 0.5 0.5 0.5
The table shows the values used to compute relative unit costs for provid-
ing a physician consultation in the two modes of care (DCT and PCC). (1)
is our baseline scenario.
M.2 Cost scenarios
We estimate costs in six scenarios. Table M.1 gives an overview of the parameters
used in each scenario, where (1) is the baseline scenario.
Baseline: In our baseline scenario, we estimate that direct health care pro-
duction costs of a DCT consultation is 35% of an those for in-person visit. Hence,
each DCT consultation is resource-saving even from a health care budget perspec-
tive. Adding the value of saved time (indirect costs) of course increases the cost
savings even further, although the estimate is still in the same ball park (30% of
in-person costs).
Scenario 2: Like baseline, except assume same physician consultation times
as in DCT. When using the same physician time in the two modes of care, the
(pure health care) cost ratio increases to 45% (36% when including the value of
patients’ time savings). The nurse-led triage step, the office space and higher
65
wages in traditional care thus still makes DCT a cheaper alternative.
Scenario 3: Like baseline, except assume same labour costs as in DCT. Com-
pared to the baseline scenario, the cost ratio increases to 42% (34% when in-
cluding patients’ time costs). In this scenario, it is the nurse-led triage step, the
office space and the longer consultation times in traditional care that makes DCT
a cheaper alternative.
Scenario 4: Like baseline, except assume same labour costs AND consul-
tation times as in DCT). This scenario combines scenarios 2 and 3. In scenario,
4, physicians spend the same time on patients in DCT and PCCs, and the wage
levels are the same. Compared to baseline, the (pure health care) cost ratio in-
creases to 52% (40% when including patients’ time costs). In this case, it is only
the nurse-led triage step and the office space that makes DCT a cheaper alterna-
tive. Although this estimate, together with our estimated substitution rate, indicate
an increase in the health care budget, it is almost budget neutral (and still clearly
resource saving when including patients’ time savings).
Scenario 5: Like baseline, except assume same labour costs AND consultation
times as in DCT AND assume patient opportunity cost using mean wage in Swe-
den. In scenario 5, we use the mean wage (including income and payroll taxes) –
around 270 SEK – to see how much the estimate of the opportunity cost of time
matters. We use the same wage and time estimates as in scenario 4 and therefore
compare these two scenarios. As we only change parameters related to the patient,
this scenario only affects the cost ratio that includes the value of patients’ time,
which decreases from 40% in scenario 4 to 36% in scenario 5, implying that the
estimated opportunity cost of time is not crucial for the cost estimates.
Scenario 6: Like baseline, except assume no office costs. This scenario ab-
stracts from office space. It is equivalent to assuming that the existence or absence
of DCT does not affect PCCs’ office space decisions. Otherwise, everything else
is as in the baseline scenario, i.e., lower wages and shorter consultation times in
DCT. Compared to the baseline scenario, the cost ratio increases from 35% to
42% (30% to 33% when including patients’ time costs). Thus, office space plays
a role, but does not change the conclusion that the degree of substitution is high
enough to establish budget neutrality from a health care perspective and even save
resources in a broader perspective.
Comparison with previous estimates: An early cost analysis comparing one
of the DCT start-ups with standard primary care in 2016 (Ekman, 2018) arrived
at a substantially higher estimate, suggesting that the cost per visit in DCT was
58% of that in the in-person setting (even though the value of patients’ time saved
was included). One reason why Ekman arrived at a higher costs ratio was that he
66
aimed to include more costs, e.g., admin and support, management, equipment,
operations and – importantly – technical development, and write-offs. Since his
figures were from a very early phase of the DCT provider’s operations, it is likely
that these fixed costs are overestimated per visit in a setting with higher volumes
and (presumably) lower costs for developing the technical platform (which, as
already noted, nowadays exist for PCCs as well). If we subtract the costs for
equipment, operations and technical development from Ekman’s calculations, the
cost ratio is 42%, i.e., similar to our estimates.
To sum up, our cost calculations indicates that the availability of DCT does
not increase health care costs, i.e., the unit costs are low enough relative to those
in PCCs (35-52%) to make up for the additional demand. Adding the valuation
of patients’ saved time makes it even more likely that the availability of DCT is
resource-saving at the societal level.
67
References
Angrist, Joshua D, and Jörn-Steffen Pischke. 2008. Mostly harmless economet-
rics. Princeton university press.
Calonico, Sebastian, Matias D Cattaneo, and Rocio Titiunik. 2014. “Ro-
bust nonparametric confidence intervals for regression-discontinuity designs.”
Econometrica, 82(6): 2295–2326.
Cattaneo, Matias D., Nicolás Idrobo, and Rocı́o Titiunik. 2019. A Prac-
tical Introduction to Regression Discontinuity Designs: Foundations. Cam-
bridge:Cambridge University Press.
Dahlstrand, Amanda. 2022. “Defying Distance? The Provision of Services in
the Digital Age.” Unpublished manuscript.
Ekman, Björn. 2018. “Cost Analysis of a Digital Health Care Model in Sweden.”
PharmacoEconomics - Open, 2(3): 347–354.
Ellegård, Lina Maria, and Gustav Kjellsson. 2019. “Nätvårdsanvändare
i Skåne kontaktar oftare vårdcentral och gör inte färre akutbesök.”
Läkartidningen, 116.
He, Yang, and Otavio Bartalotti. 2020. “Wild Bootstrap for Fuzzy Regres-
sion Discontinuity Designs: Obtaining Robust Bias-Corrected Confidence In-
tervals.” Econometrics Journal, 23(2): 211–231.
Ivarsson Westerberg, Anders, Agneta Andersson, Eva Anskär, Daniel
Castillo, Magnus Falk, and Anders Forssell. 2021. Papper, pengar och pa-
tienter : primärvården i administrationssamhället. Lund:Studentlitteratur.
Millán-Quijano, Jaime. 2020. “Fuzzy difference in discontinuities.” Applied
Economics Letters, 27(19): 1552–1555.
SFS. 1991:1121. “Förordning om läkarvårdsersättning [Act on remuneration of
physician care].”
Swedish Competition Authority. 2010. “Uppföljning av vårdval i primärvården
– Valfrihet, mångfald och etableringsförutsättningar.” Konkurrensverkets rap-
portserie 2010.
68