1. Does dichotomizing the dependent variable (i.e. logistic regression) decrease power in relation to a continuous dependent variable? I noticed the logistic regression PPT states Cohen mentioned 75 cases per predictor to detect a large effect, which on average, is larger than multiple regression.
2. What are nonnormal error terms, nonconstant error variance, and constraints on the response function?
3. The table on page 124 indicates XRAY was the only variable input into the equation and the model’s overall percentage correct was 75.47%. On page 125, STAGE was entered into the equation, but it looks like the model still correctly classified 75.47%. I noticed the Nagelkerke R^2 went up, so we are explaining more variance, but why would the classification table be identical?