After saving model information to a data set with the OUTMODEL= option, if you then update the OUTMODEL= data set in any way and attempt to use it in the INMODEL= option of a subsequent PROC LOGISTIC step, the following error is issued and the step terminates:

ERROR: INMODEL=xxx.yyy is not of type LOGISMOD.

After creating the OUTMODEL= data set, you should not do any processing of it before using it in the INMODEL= option. The OUTMODEL= data set records all details of the fitted model and is not intended to be modified by the user because altering one aspect of it makes the overall stored model information incorrect. Using such an altered data set would then result in incorrect statistics. For instance, if you changed a parameter estimate, the parameter variance-covariance matrix would then be incorrect and any confidence intervals computed using the altered data set would be incorrect. Consequently, if the model needs to be altered you should fit the desired model in PROC LOGISTIC, save it with the OUTMODEL= option, and not alter the resulting data set.

Note that copying the OUTMODEL= data set in a DATA step to the same or a new name will remove the special type designation of the data set and cause this error. For example, these statements will result in the error when INMODEL=NEWNAME is specified in a subsequent PROC LOGISTIC step:

data newname;
   set outmodel;
   run;

If you only copy or rename the OUTMODEL= data set and do not edit any of its values or change its structure, then you can retain the special data set type by specifying the TYPE=LOGISMOD option in the DATA statement or in the PROC LOGISTIC statement. For example:

data newname(type=logismod);
   set outmodel;
   run;

or

proc logistic inmodel=newname(type=logismod);