2.6. Statistical Analysis

For data clearance, we checked the “missing completely at random” (MCAR) assumption. The MCAR assumption was rejected according to the significant result of Little’s MCAR test. Therefore, assuming that the pattern of data missingness matched the “missing at random” (MAR), missing imputation was performed with the EM estimation. The software package IBM SPSS Statistics for Windows (version 25) (SPSS Inc., Chicago, IL, USA) was employed to carry out single imputations of variables with missing values.

Seroprevalence was estimated as the proportion of seropositive children with IgG SARS-CoV-2 antibodies. The prevalence of active SARS-CoV-2 infection was estimated as the proportion of children testing positive with RT-qPCR. For data weighting by age–gender–season (two categories for gender, three categories for age, and three categories for season), we used population data from the 2016 census, available from the Statistical Centre of Iran [26]. Frequency weights were calculated by dividing the size of the actual population in Tehran in each category by the number of participants in our study in the same categories. The survey data were analyzed using STATA version 14.2 (College Station, TX, USA). For descriptive data analysis, we report percentages with 95% confidence intervals (95% CI); however, because of intermittent epidemics over time, frequencies are reported by season. We used multivariable logistic regression analysis to evaluate associations between SARS-CoV-2 seropositivity and RT-qPCR positivity and demographics and potential risk factors. To adjust for confounding, we compared results from the univariable and multivariable analyses. A p-value less than 0.05 was considered significant.