Table 3.
Model 1: dynamic regression model predicting daily tweet frequency with ARIMAa error term (2, 0, 0).
| Measures | Tweet frequency | P value | |||
| Predictors of daily tweet frequency, estimate of effect (95% CI) |
|
|
|||
|
|
Intercept | 3143 (2837 to 3450) | <.001 | ||
|
|
Statutory holidays (1 for holidays, 0 for nonholidays) | –385 (–761 to –7.8) | .04 | ||
|
|
Business closure (increase in 1 log day) | 196 (121 to 271) | <.001 | ||
|
|
School closure | 130 (60.1 to 199) | <.001 | ||
|
|
Additional measures | 544 (178 to 910) | .003 | ||
|
|
New COVID-19 case counts (in hundreds of cases) | 391 (311 to 470) | <.001 | ||
|
|
New COVID-19 case counts in Canada, excluding Ontario (in hundreds of cases) | 46.20 (20.9 to 71.6) | <.001 | ||
|
|
Official COVID-19–related updates | 373 (95.4 to 650) | .008 | ||
|
|
Regional differences in lockdown | ||||
|
|
|
Province-wide lockdown: regions are in the same stage of lockdown | Reference group | N/Ab | |
|
|
|
Partial lockdown: regions are in different stages of lockdown | 140 (–343 to 624) | .53 | |
|
|
|
No lockdown: regions are not under lockdown | –440 (–1513 to 632) | .43 | |
|
|
|
Regions are in different stages of lockdown × new cases | –257 (–361 to –153) | <.001 | |
|
|
|
Regions are not in lockdown × new cases | 1219 (–2161 to 4599) | .49 | |
| Goodness of fit | |||||
|
|
With covariates, AICc | 3693.89 | N/A | ||
|
|
Without covariates, AIC | 3873.20 | N/A | ||
aARIMA: autoregressive integrated moving average.
bN/A: not applicable.
cAIC: Akaike information criterion.