Table 6.
ARIMAa/ARIMAXb model performance (RMSEc) for Atlanta–Sandy Springs–Alpharetta, GA, Chicago-Naperville-Elgin, IL-IN-WI, Los Angeles–Long Beach–Anaheim, CA, and Miami–Fort Lauderdale–Pompano Beach, FL. Models that performed better than the baseline ARIMA are shown in italics.
| Variables | Atlanta–Sandy Springs–Alpharetta, GA, RMSE | Chicago-Naperville-Elgin, IL-IN-WI, RMSE | Los Angeles–Long Beach–Anaheim, CA, RMSE | Miami–Fort Lauderdale–Pompano Beach, FL, RMSE |
| (Baseline) percentage of individuals who have been administered at least 1 vaccine dose (7 day rolling average) | 4.0855 | 4.2182 | 4.1473 | 3.7109 |
| Number of users per 100,000 population | 1.6356 | 1.3900 | 0.7198 | 1.2971 |
| Number of tweets per 100,000 population | 1.6420 | 1.3700 | 0.7131 | 1.3492 |
| Average favorites | 2.1176 | 4.2100 | 0.6878 | 4.0045 |
| Average retweets | 5.3545 | 1.2414 | 4.1356 | 1.4062 |
| % Positive sentiment, | 1.6182 | 1.3000 | 0.7051 | 1.1691 |
| % Negative sentiment | 1.6238 | 1.3300 | 0.6915 | 1.2168 |
| % Neutral sentiment | 1.6236 | 1.1600 d | 0.7213 | 3.7217 |
| % Trust | 4.0854 | 4.2183 | 4.1471 | 1.1407 |
| % Surprise | 1.6522 | 1.3371 | 0.7078 | 1.1314 |
| % Sadness | 1.5826 d | 1.2400 | 0.7077 | 3.7117 |
| % Joy | 1.6243 | 1.3600 | 0.6865 d | 1.2322 |
| % Fear | 1.6751 | 4.1900 | 0.6973 | 3.7028 |
| % Disgust | 1.6401 | 1.3200 | 0.7054 | 3.7670 |
| % Anger | 4.6909 | 4.2000 | 0.7037 | 1.1006 d |
| % Anticipation | 1.6589 | 1.2800 | 0.7079 | 3.7115 |
| Best predictors: joy (%), negative sentiment (%), surprise (%), trust (%) | 1.7324 | 1.2878 | 0.6921 | 1.2901 |
aARIMA: autoregressive integrated moving average.
bARIMAX: autoregressive integrated moving average exogenous variable model.
cRMSE: root mean square error.
dBest-performing model for each metropolitan area.