| dc.description.abstract | Spatio-temporal analysis of COVID-19 data with the two different statistical approaches is the main objective of this thesis. The first classical approach, the
Endemic-Epidemic framework (Held et al., 2005) is a class of multivariate time-series models for the incidence counts, obtained from the surveillance systems. In
this formulation, the conditional mean of the number of cases is partitioned into
endemic, autoregressive and spatio-temporal parts, representing different sources
of infection contribution. The second approach used in the thesis is INLA (Integrated Nested Laplace Approximation (Rue and Martino, 2007)), which performs
the approximate Bayesian inference for latent Gaussian models. The flexibility
of the both approaches allows for various extensions of the models. As the thesis
progresses, we search for the best model with different metrics used as a selection
criteria. Both frameworks allow for the inclusion of the socio-demographic covariates
in the analysis, as possible drivers of the disease spread. Guided by a previous
study of Söderberg et al. (2022), we chose the covariats of interest to be: Income,
Foreign background, Education, Overcrowding, Square meters per person, Employed, Care workers. Also, the age factor was added as two covariates: Young
and Older. It was shown that the Endemic-Epidemic approach with a complex seasonal
trend, random intercepts and the spatial weights, assigned according to the powerlaw principle, but without any socio-demographic covariate, achieved almost as
low metric values as the best model. Given the aforementioned extensions, the
best model included the following socio-demographic covariates: Education and
Foreigners in the endemic part, Young and Square meters per person in the autoregressive and Overcrowded, Foreigners, Older, Income in the spatio-temporal part.
All these covariates positively correlated with the number of counts. The model with the random walk time formulation applied within INLA technique showed on average a positive correlation of the case counts with Foreign
Background, Care workers, Overcrowded, Education and Income. A negative correlation with the case counts on average was shown by the Older, Young, Employed and Square meters per Person. The results suggest further research about
the impact of the socio-demographics on the case counts of viral diseases. | |
