To create the index, 130 metropolitan areas were examined in data from four sources:
In cases where BRFSS data divided MSAs into Metropolitan Divisions, responses taken from the primary city in the MSA were used. Hospital quality variables were also calculated at the Metropolitan Division, while provider specialty data were examined at the larger Metropolitan Statistical Area.
Data analysis was performed using the R statistical programming language. To explore the latent structure of the data, a principal component analysis (PCA) was performed on scaled data. Examination of a scree plot and component eigenvalues indicated that a four-component model best explained the underlying low dimensional structure of the data.
Consequently, a maximum-likelihood factor analysis fitted to four factors was produced on the data matrix using a varimax rotation. This model explained 51.2% of the cumulative variance. A chi-square test, evaluating the explanatory sufficiency of a four factor model, was statistically significant (χ2 = 328.59, df = 116, p < .0001).
To optimize the interpretability of the factor model, only each variable’s greatest positive factor weight above 0.4 was used to create the final factor scores. As a result, the following six variables were not retained in any final factor score.
The Factor Loading Table displays the factor loadings for each variable input into the analysis used to construct the National Health Index.
For each factor, a variable’s population value was multiplied against the factor weight and added to the product of all variables in a factor, creating a final factor score. Next, the four factor scores were cumulatively added to obtain the total factor score for each metropolitan area. Each metropolitan area was then ranked from highest to lowest total factor score to determine the final list of 100 cities for the National Health Index.