how to generate residuals from multinomial logistic regressions

sandrawang — Mon, 13 Dec 2010 20:15:43 GMT

I used the following output options but only missing values are generated for the fields "peares" and "devres". Could any experience programmer tell what is wrong with that? Thanks very much.

proc logistic data = myDATA descending;

model y = xbeta1-xbeta5 /lackfit rsq;

output out = MODEL_OUTDATA p=prob_def reschi=peares resdev=devres xbeta=xbeta dfbetas=_ALL_ difdev = diff_dev difchisq = diff_chisq;

ods select ModelInfo ResponseProfile ConvergenceStatus FitStatistics RSquare
GlobalTests ParameterEstimates OddsRatios Association LackFitPartition LackFitChiSq;

run;

Re: how to generate residuals from multinomial logistic regressions

lvm — Tue, 14 Dec 2010 19:15:11 GMT

I think these residuals (and other measures of Influence) are only calculated for a response variable with two levels (binary variable). Try putting the INFLUENCE option on the model statement, and in the Log you will get a message about certain statistics only be determined for binary responses. But for a binary variable, you will get the two kinds of residual directly in the Output (and in the Output file you create).
LVM

topic how to generate residuals from multinomial logistic regressions in Statistical Procedures

how to generate residuals from multinomial logistic regressions

Re: how to generate residuals from multinomial logistic regressions