Andrew Perrin
2007-May-11 19:05 UTC
[R] OT: Predicted probabilities from ordinal regressions
Sorry if this is too off-topic, as we may not implement this in R (although we may do so). A student of mine is looking at some public opinion data in which there appears to be a statistically significant difference between levels of support for a proposal based on which of two question wordings is used. That is, roughly half the sample was asked the question with wording 1, the other half with wording 2, and the difference is large enough to be of interest (approx. 6 percentage points different with an N of about 30,000). The question is how best to model this. In the past I have generated predicted probabilities based on the sample asked wording 1, used these to assign those asked wording 2 to predicted categories, and used a logistic regression to predict difference between predicted and actual response. In this case, though, the question of interest uses a four-level ordinal response, so ordinary predicted probabilities based on, e.g., and ordinal logistic regression generate literally probabilities of being in each of the four categories. Transforming this outcome into a prediction of membership in one of the given categories is not straightforward. Can anyone provide some insight into how to model predicted vs. actual outcomes on an ordinal scale? Thanks. ---------------------------------------------------------------------- Andrew J Perrin - andrew_perrin (at) unc.edu - http://perrin.socsci.unc.edu Assistant Professor of Sociology; Book Review Editor, _Social Forces_ University of North Carolina - CB#3210, Chapel Hill, NC 27599-3210 USA New Book: http://www.press.uchicago.edu/cgi-bin/hfs.cgi/00/178592.ctl
Frank E Harrell Jr
2007-May-11 21:03 UTC
[R] OT: Predicted probabilities from ordinal regressions
Andrew Perrin wrote:> Sorry if this is too off-topic, as we may not implement this in R > (although we may do so). > > A student of mine is looking at some public opinion data in which there > appears to be a statistically significant difference between levels of > support for a proposal based on which of two question wordings is used. > That is, roughly half the sample was asked the question with wording 1, > the other half with wording 2, and the difference is large enough to be of > interest (approx. 6 percentage points different with an N of about > 30,000). > > The question is how best to model this. In the past I have generated > predicted probabilities based on the sample asked wording 1, used these to > assign those asked wording 2 to predicted categories, and used a logistic > regression to predict difference between predicted and actual response. In > this case, though, the question of interest uses a four-level ordinal > response, so ordinary predicted probabilities based on, e.g., and ordinal > logistic regression generate literally probabilities of being in each of > the four categories. Transforming this outcome into a prediction of > membership in one of the given categories is not straightforward. Can > anyone provide some insight into how to model predicted vs. actual > outcomes on an ordinal scale? > > Thanks.With the lrm function in the Design package you can get predicted probabilities for each class as well as predicted mean scores. Frank> > ---------------------------------------------------------------------- > Andrew J Perrin - andrew_perrin (at) unc.edu - http://perrin.socsci.unc.edu > Assistant Professor of Sociology; Book Review Editor, _Social Forces_ > University of North Carolina - CB#3210, Chapel Hill, NC 27599-3210 USA > New Book: http://www.press.uchicago.edu/cgi-bin/hfs.cgi/00/178592.ctl > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >-- Frank E Harrell Jr Professor and Chair School of Medicine Department of Biostatistics Vanderbilt University