In Reply: Dr Cummings states incorrectly that we “used linear regression with the coronary mortality rate (deaths divided by the regional population estimate) as the outcome.” We in fact used age-adjusted mortality rates. He then says that division of the outcome and each predictor by the same common divisor will result in correlation of the resulting ratios even if prior to division by the common divisor there is no correlation. Cummings refers to a simple transformation that is the same for the outcome as well as the predictors. However, since our outcome has been age adjusted, it has undergone a different transformation from the predictors. Hence, any assumptions that might follow as an automatic consequence of equal transformations for outcome and predictors will not necessarily follow when the outcome and the predictors have undergone different transformations, as in our study. Also, if Cummings' point were valid, then our ethnic group variable would presumably have a significant univariable association with the outcome in all 3 years, but this was true only for 2008.
Levene LS, Baker R, Bankart MJG. Population and Primary Health Care Characteristics and Coronary Heart Disease Mortality—Reply. JAMA. 2011;305(8):778-779. doi:10.1001/jama.2011.156