Hello,
I am conducting a quantile regression on sentencing data using SAS 9.4. I conduct this regression twice, once with a restricted base model (Model 1) and another with several interaction terms (Model 2). Unfortunately the data is not public so I will have to describe it. The dependent variable is logged sentence length and there are a variety of independent variables such as plea (0=not guilty 1=guilty), gender (dichotomous male=0, female=1), logged number of guilty findings (nguilty). Model 1 has roughly 35 independent variables. Model 2 includes interaction terms such as plea*gender calculated previously. It has roughly 55 independent variables. Many of these predictors are dummy variables.
I have population-level data with approximately 950,000 observations.
My code is as follows:
PROC QUANT DATA=MYDATA ALGORITHM=SMOOTH (RRATIO=.5);
MODEL LOGGED_SENTENCE=PLEA GENDER NGUILTY/ quantile= .1 .25 .5 .75 .9 PLOT=QUANTPLOT;
TEST PLEA/QINTERACT;
RUN;
Model 2 is the same syntax except the interaction term PLEA*GENDER is placed in the model statement.
Model 1 works fine and as expected. Model 2 however gives me a warning concerning too many "fuzzy equations at the borders". I am not sure what this means nor what I should do to fix this issue. I cannot find any mention of this warning in any documentation.
I also attempted changing the RRATIO from 0.5 to 0.9 as suggested by SAS but this did not have any impact.
Any help is appreciated.
Best regards,
Brendyn
