# SPSS regression imputation

I have a dataset of 45 observations (participants), with variables on demographic data and standardized tests. Two standardized test variables are such that they have missing values on only one participant. All other participants have values on them. I want to impute the missing two values for this one participant using the regression method. However I am unsure how SPSS does this, and thus I don't know if what I am reporting is the truth.

• The participant in question has missing values for variables TMT-A and TMT-B. I want to use regression imputation, with TMT-A and TMT-B predicted by education and age.

• I don't want to use any other variables in my data set to predict the imputation of the participant's score on these TMT-A and TMT-B values.

• So what I want to report is that "the TMT-A and TMT-B scores for one participant was imputed using the regression method, with age and education as predictors", basically.

My SPSS syntax is basically:

  MVA VARIABLES = Age Education TMTA TMTB
/MAXCAT=25
/CATEGORICAL=Education
/REGRESSION TMTA TMTB WITH Age(TOLERANCE=0.001 FLIMIT=4.0 ADDTYPE=RESIDUAL).


I got it from the Missing Value Analysis menus. I'm confused - why do I need to separate between categorical and quantitative in the main window of the menu?

In the Variables -window, why can't I set education (that I had set as a categorical in the main window) as a predictor? After all I want it to be a predictor don't I?

Is it a problem that I am imputing two variables at once? Are the other participants' TMT-A and TMT-B values being used as predictors for the missing TMT-A and TMT-B values for the participant that I am imputing?

SPSS web help and Tabachnick & Fidell 6th ed. do not answer my questions. Basically, I am still left unsure whether or not the imputation method now excludes all the other variables in my data (for example gender), and if it includes education at all?

Basically what I think I am doing is imputing by using the known TMT-A and TMT-B values in my data, while taking into account the possible "effect" of education and age also. How wrong am I?

Age and education are not correlated in my data. The reason I'm not just excluding the observation is that within this report supposed to use methods for the sake of using them. Also I am assuming that SPSS corrects for the problem of reduced variance by adding noise, isn't this correct?

1. As the help for MVA says: Data can be categorical or quantitative (scale or continuous). However, you can estimate statistics and impute missing data only for the quantitative variables.

I know - that's a bummer. However, you could create your own numeric dummy-coded variables in place of the categorical. Suppose your categorical variable had 3 levels (low, medium, high, or whatever). You could do this prior to MVA:

COMPUTE group1=0.
COMPUTE group2=0.
IF (MyCatVar eq "Low") group1=1.
IF (MyCatVar eq "Medium") group2=1.


Then the last group is the reference group (0 on both). Or you could do it the other way, so "Low" is reference. That's up to you. Enter those two variables as numerics in your MVA as predictors for the regression.

1. It is not a problem that you are imputing two variables at once. Indeed, MVA is written to do that.

2. If you do the dummy-coding I suggested then your statement:

"Basically what I think I am doing is imputing by using the known TMT-A and TMT-B values in my data, while taking into account the possible "effect" of education and age also."

is exactly right.

One further point. You could specify TMT-A and TMT-B as both to be predicted and predictors. That's up to you, too.