Page 1 of 1

All variables as responses

Posted: Mon Apr 21, 2014 7:29 pm
by shakespeare
It was suggested by a member of your staff that I set up my imputation model as "all responses" in order to mitigate error messages I received from Realcom. I presume that's that same as is alluded to here: http://www.jstatsoft.org/v45/i05/paper on p. 4, section 3. My data are are two level with subjects nested in locations, and all covariates at level one. Does making the imputation model all responses include just the explanatory and response variables, or should it also include the level 2 identifier and constant as well? A read back through the theory did not quite shed enough light to make it clear.

Re: All variables as responses

Posted: Thu Apr 24, 2014 10:17 am
by ChrisCharlton
The suggestion would have referred to just the explanatory variables and response variables, with the constant and level identifier used as in your original model.

Re: All variables as responses

Posted: Fri Apr 25, 2014 3:14 pm
by shakespeare
Thank you. I've changed everything except the constant and level 2 identifier to response variables in the Add/Remove responses dialog box. I'm wondering if any changes are necessary in the Add/Remove explanatory variables dialog. Seems like even though this is marking certain variables as explanatory, it is necessary to leave them as tagged as level 1 variables (all my covariates are at level 1), or Realcom will not know the proper structure to impute. Am I on the right track?

Re: All variables as responses

Posted: Fri Apr 25, 2014 3:22 pm
by ChrisCharlton
Once you have set all your variables to be responses the only explanatory variable specified should be the constant (which is set to be at level-1 and level-2).