Licensure status and the extensiveness of regulation in the two groups of States were the independent variables against which facility and resident quality measures were compared. Consistent with the sample design and original study power calculations, we used a probability level of 0.05 as the determination of statistical significance.
To study the relationship between regulation, licensure, and quality of care constructs at both the facility and the resident level, we used multivariate modeling techniques that control for multiple explanatory variables. For continuous quality measures we used linear regressions. When the outcome was binary, logistic regression was the technique of choice. Because no single measure of quality of care summarized all aspects of care, we fit multiple models with dependent variables that characterized the different aspects of care. The results of these analyses are presented in a separate report.
All analyses were conducted using weighted estimates of the number of homes and residents within the group based on the sampling design. To account for the multistage, complex cluster sampling techniques used, we used software that adjusts the standard errors of estimate for the intercorrelation among sampled units within clusters (SUDAAN). SUDAAN produces unbiased variance estimators for linear (or nonlinear) statistics no matter how subsampling occurs within FSUs.