*2.5. Statistical Analysis*

Descriptive statistics were performed according to states. Then, an ordinal hierarchical model was developed to verify if the states' variable affected the results. Two levels were considered, the individual and state subgroup. At the individual level, the level of impact by depression, anxiety, and stress and the psychological impact (avoidance, intrusion and hyperarousal) were considered as dependent variables. Sex, age, education, monthly income, previous mental and general health problems, sense of safety, number of people living with the participant, time of exposure to the news, frequency of socialization, knowing someone who tested positive for COVID-19, and change in mental health status after the start of the pandemic were independent variables. To verify the cluster effect, the intraclass correlation coefficient (ICC) was used.

The prevalence of psychological symptoms were calculated according to sex (reference category (rc) = male), age group (rc: ≥ 55 years), number of people living with the participant, economic level (rc: < 240.00 USD), education level (rc: complete graduate school), sense of security in relation to the pandemic (rc: unsafe), previous health problems (rc: no), frequency of socialization (rc: equal to or greater than usual), prior mental illness (rc: absent), change in mental state due to the pandemic (rc: no), knowing someone who tested positive for COVID-19 (rc: no), time spent with the news (rc: < 60 min). A multiple logistic regression model was constructed and odds ratio (OR) per point and 95% confidence interval were calculated. The dependent variables (psychological symptoms) were grouped into absent (normal category = 0) and present (symptom present in some level of impact); time spent with the news was categorized according to 25th, 50th, and 75th percentiles (1: < 60 min; 2: 60–90 min; 3: 90–150 min; 4: ≥ 150 min) and age by the 25th, 50th, 75th and 90th percentiles (1: < 24; 2: 24–33; 3: 33–43; 4: 43–55; 5: ≥ 55 years). The significance level was 5%. The analyses were performed using the IBM SPSS Statistics v.22 software (IBM Corp, Armonk, NY, USA).
