The results of logistic analyses when the variables are highly correlated: An empirical example using diet and CHD incidence