Fig 1.
Timeline and participants of the longitudinal COVIMOD study.
A. Daily COVID-19 case counts in Germany (red bars), the OxCGRT Stringency Index (blue line), and COVIMOD survey administration periods (grey ribbons). B. Cumulative COVID-19-related deaths in Germany (red line), the OxCGRT Stringency Index (blue line), and COVIMOD survey administration periods (grey ribbons). C. Sample sizes and the proportion of people repeatedly sampled in the COVIMOD survey for which zero repeats indicate first-time participants.
Fig 2.
Pre-COVID19 scenario simulation experiments.
(Top left) Simulated social contact intensities for Male-Male contacts. (Top right) Simulated social contact counts for Male-Male contacts with a COVIMOD-like age aggregation scheme. (Bottom left) Posterior median estimates of social contact intensities from the age-age parameterised HSGP model. (Bottom right) Posterior median estimates of social contact intensities from the difference-in-age parameterised HSGP model.
Fig 3.
In-COVID19 scenario simulation experiments.
(Top left) Simulated social contact intensities for Male-Male contacts. (Top right) Simulated social contact counts for Male-Male contacts with a COVIMOD-like age aggregation scheme. (Bottom left) Posterior median estimates of social contact intensities from the age-age parameterised HSGP model. (Bottom right) Posterior median estimates of social contact intensities from the difference-in-age parameterised HSGP model.
Fig 4.
Empirical and estimated contact intensity patterns for POLYMOD using the Bayesian rate consistency model for age-granular contact data and age-stratified contact data.
(Top row) Crude empirical contact intensity patterns. (Middle row) Posterior median contact intensity estimates from the Bayesian rate consistency model applied to fine-age contact data meaning the age of contacts was not aggregated into large age bands. (Bottom row) Posterior median contact intensity estimates from the Bayesian rate consistency model applied to coarse-age contact data meaning the age of contacts was aggregated into larger age bands similarly to the COVIMOD study.
Fig 5.
Empirical and estimated contact intensity patterns for Zimbabwe using the Bayesian rate consistency model for age-granular contact data and age-stratified contact data.
(Top row) Crude empirical contact intensity patterns. (Middle row) Posterior median contact intensity estimates from the Bayesian rate consistency model applied to fine-age contact data (i.e., the age of contacts is not aggregated into large age bands). (Bottom row) Posterior median contact intensity estimates from the Bayesian rate consistency model applied to coarse-age contact data (i.e., the age of contacts is aggregated into large age bands in a similar fashion to COVIMOD).
Table 1.
Comparison of performance on simulated data for different scenarios, models, sample sizes, and parameterisations.
Fig 6.
Empirical and estimated contact intensity patterns for COVIMOD wave 1.
(Top row) Crude empirical social contact intensity patterns. There are some age groups with no participants, and they are represented by white vertical columns. (Middle row) Contact intensity patterns as estimated by the socialmixr R package [11]. (Bottom row) Posterior median contact intensity estimates from the Bayesian rate consistency model. For the crude estimates and socialmixr estimates, estimates are divided by the number of integer ages in the corresponding age group to make the estimates more comparable (e.g., if the age of contacts is 20 − 24 than the contact estimate is divided by 5 while if the age of contacts is 25 − 34 the contact estimate is divided by 10). The exact runtime arguments for this comparison are given in script figure-6.R on our accompanying GitHub repository.
Fig 7.
Adjustments on marginal social contact intensities.
Posterior median estimates (line) and 95% credible intervals (ribbon) of the number of social contacts in the last 24 hours are shown by age and sex of participants and survey wave in red under the Bayesian rate consistency model that adjusts for both missing & aggregate contact reports and reporting fatigue. For comparison, posterior median estimates are shown for the model incorporating no adjustments (long dashed blue), adjustments for reporting fatigue but not missing & aggregated contact reports (green dashed lines), and adjustments for missing & aggregated contact reports but not reporting fatigue (turquoise small dashed lines).
Fig 8.
Comparison of social contact intensities before and during the first months of the COVID-19 pandemic in Germany.
Solid lines represent posterior median estimates of the marginal contact intensities by age and gender of participants, and after adjusting for missing & aggregate contacts and reporting fatigue for the COVIMOD survey. Dashed lines represent posterior median estimates from the pre-pandemic POLYMOD study and, unlike the estimates reported in [1], also include missing & aggregate contacts. Shaded ribbons represent 95% credible intervals.
Fig 9.
Relative percentage change in marginal contact intensities relative to those in wave 1.
The red and blue lines represent posterior median estimates of the relative percentage change in marginal contact intensity estimates in women and men by the age of participants. Shaded ribbons represent 95% credible intervals.
Fig 10.
Time evolution in age-specific social contact intensities.
The posterior median estimates for conditional contact intensities (top row) and relative change in the conditional contact intensities in waves 2 to 5 relative to those in wave 1 (bottom row) for individuals aged 10, 20, 35 and 70, respectively. Conditional contact intensities were aggregated across men and women. The colours represent different COVIMOD survey waves, and the shaded ribbons represent 95% credible intervals. We only show credible intervals for wave 5 to reduce overlaps and ease interpretation.