Table 2.
Regression results from linear probability models: pooled and stratified analysis
| Pooled analyses | Stratified analyses | |||
|---|---|---|---|---|
| Model 1a | Model 2 b | Model 3 c (ADI<85) |
Model 4 c (ADI>=85) |
|
| Time trend (quarters) | −3.6d | −3.6 d | −3.6 d | −2.6 d |
| Individual race and ethnicity (White individuals as the reference) | ||||
| Black | −0.1 | −0.2 | −0.6 d | 2.4 d |
| Hispanic | 3.8 d | −0.1 | −0.7 d | 7.8 d |
| Community socioeconomic status (less-deprived community, ADI<85, as the reference) | ||||
| ADI>=85 | - | −2.5 d | - | - |
| Variations in time trends by subgroups: interactions between race, ethnic, and community socioeconomic status with time trend | ||||
| Black×Time | 0.1 | 0.0 | 0.1 d | −0.6 d |
| Hispanic×Time | 0.1 d | 0.1 d | 0.1 d | −0.8 d |
| ADI>=85×Time | - | 0.6 d | - | - |
The estimations were based on linear probability models with individual random effects and robust standard errors. The models were estimated in STATA 16 using command xtreg.
Numbers in the cells represent percentage-points changes, which were calculated as probabilities x100. We only presented the variables of interest in Table-2. The full model results can be found in supplementary table 2.
Symbol “-” in the cells means the estimation is not applicable because the variable is not in the model.
Model 1 accounted for individual-level characteristics, without community-level (zip-code-level or county-level) variables. Individual covariates included socio-demographic characteristics (e.g., age, gender, Medicare-Medicaid dual status, etc.), chronic conditions (e.g., hypertension, depression, schizophrenia, etc.), and prior quarter health care utilization (i.e., any ED visits, hospitalizations).
Model 2 accounted for the variables in Model 1 and additional community-level characteristics (e.g., percentage of Black individuals, percentage of Hispanic individuals, percentage of households with broadband subscriptions, etc.).
Model 3 and model 4 were stratified analysis by zip-code-level ADI and accounted for individual-level characteristics and other community-level (i.e., zip-code-level and county-level) characteristics.
P<0.001.