Missing Value Analysis, Little's test sig (so not MCAR), how to tell if MAR or MNAR?

[Statsallgood]

Hi Everyone,

I have a 7 variable dataset with 500+ participants, and 5.5% of values are missing for a single variable (V7). I'm using SPSS.

I have performed Little's test and the result is highly significant p<.001. So I will conclude that the data are not MCAR.

Now I have run separate variances t-test and I see that while the result is not significant for most other variables (V1-V5), there is one other variable (V6) where there is a significant relationship between missing values and scores on V6. This is in accord with the finding that data are not MCAR, and may suggest they are MAR or NMAR.

How do I tell whether they are MAR or MNAR? That is, where do i go from here, having performed little's test for MCAR, and separate variances t-tests?

Thanks very much.

 

[hlsmith]

Re: Missing Value Analysis, Little's test sig (so not MCAR), how to tell if MAR or MN

I am also interested in what others think about this.

Do you know of any systematic error that may theoretically make this question different than the others?