In a research paper we recently presented the odds ratio obtained from a logistic regression analysis using the plot summary method available in rms. By default, the plot shows interquartile range effects for continuous variables. I know we can ask the summary method to show the results for a specified cutoff value. However, I am curious as to the reason behind the choice of the interquartile range effect in the plot. Why not any other metric ?
This would have been better posted under the RMS Discussion topic.
Inter-quartile range is a useful measure of dispersion when computed on raw variables, as it is robust and easy to interpret - an interval containing 1/2 of your data. So it was chosen as a default effect measure for a continuous predictor. Roughly it is e.g. an odds ratio over a range of X containing 1/2 of the values. But any choice that translates a whole curve into a single number is arbitrary, especially if the curve is not linear. For that reason I am using such summary measures less and less over time, and partial effect plots remain my favorite general-purpose effect display.
4 Likes