Error message from multiple imputation: what does this mean?

[ziweiguan]

I have about 140 questions in my questionnaire, and I coded them as nominal variables. There are about <2% data missing and when I tried to do imputation, I got the error message:

The imputation model for question1 contains more than 100 parameters. No missing values will be imputed. Reducing the number of effects in the imputation model, by merging sparse categories of categorical variables, changing the measurement level of ordinal variables to scale, removing two-way interactions, or specifying constraints on the roles of some variables, may resolve the problem. Alternatively increase the maximum number of parameters allowed on the MAXMODELPARAM keyword of the IMPUTE subcommand.
Execution of this command stops.

does any one knows about this error message? Thanks.

[David Marso]

Perhaps if you pasted the generated syntax you could inspect what sort of monstrosity is being generated under the hood.  Have you tried following the suggested options?

In a nutshell it means:
You are attempting to estimate something using a model which is WAY TOO COMPLICATED.
Back off a bit and use your cranium instead of blindly clicking buttons.

ziweiguan wrote

I have about 140 questions in my questionnaire, and I coded them as nominal variables. There are about <2% data missing and when I tried to do imputation, I got the error message:

The imputation model for question1 contains more than 100 parameters. No missing values will be imputed. Reducing the number of effects in the imputation model, by merging sparse categories of categorical variables, changing the measurement level of ordinal variables to scale, removing two-way interactions, or specifying constraints on the roles of some variables, may resolve the problem. Alternatively increase the maximum number of parameters allowed on the MAXMODELPARAM keyword of the IMPUTE subcommand.
Execution of this command stops.

does any one knows about this error message? Thanks.

[ziweiguan]

Here are the syntax:

DATASET ACTIVATE DataSet1.
*Impute Missing Data Values.
DATASET DECLARE eww.
MULTIPLE IMPUTATION   *** my more than 100 variables omitted to save space ***

  /IMPUTE METHOD=AUTO NIMPUTATIONS=5 MAXPCTMISSING=10
  /MISSINGSUMMARIES NONE
  /IMPUTATIONSUMMARIES MODELS
  /OUTFILE IMPUTATIONS=eww .

[Alex Reutter]

I would say at this point you should try some of the suggestions in the error message.  

Have a look at the syntax chart.  
http://www-01.ibm.com/support/knowledgecenter/SSLVMB_23.0.0/spss/mva/syn_multiple_imputation.dita

One suggestion in the error message is to "Alternatively increase the maximum number of parameters allowed on the MAXMODELPARAM keyword of the IMPUTE subcommand. "  By default, it's 100.  You could specify syntax with a larger value and see if it runs.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Art Kendall]

Are there subsets of items with different roles?

Are these variables items intended to to be items in a summative scale?

Are some of them Independent variables (IVs) and some of them Dependent variables (DVs)?

I would be very leery of using scale items from different constructs.  E.g., imputing "sexual freedom LIBCON" from "Equalitarian LIBCON"   or of imputing spelling items from addition items.

I would strongly object to using the DV items to impute IV items.

If you have summative scale items, I would be surprised if you had perceptible difference if you were to scatterplot a scale score with something like
compute Equalitarian = mean.8(eual01 to equal10) .
vs
compute Equaitarian = mean(imp.equal01 to imp.equal10)

In the second compute imputed scale item scores would be used.

[ziweiguan]

I tried to increase the number to 1000
  /IMPUTE METHOD=AUTO NIMPUTATIONS=5 MAXPCTMISSING=10 maxmodelparam = 1000

and still got the same error message. I think the reason is I have too many variables (items), each variable has 6 levels (1 to 6 from strongly disagree to strongly agree).

I think the solution is to reduce the number of variables (by sum or mean for some of them).

Another possible solution is, to change the variables from ordinal to scale. This may not be good because these are not continuous variables anyway.