Measuring the impact of nearest-neighbour imputation.

I am working on survey data, with missing values imputed using a nearest neighbour method. I am currently interested in assessing the effect of imputation on the quality of data, to consider whether incorrect values may have been imputed. This would be based on relationships with non-missing data. So for example, someone might have a high amount of a particular asset, and this may be associated with high ownership of another asset.

My initial thoughts were that as the data relates to assets, it is very likely skewed and would violate parametric assumptions. Therefore some form of transformation may be needed, possibly Box-Cox. A possible method could be Logistic Regression on to the imputation flag variable. Precision or applicability of statistical significance is not important. An impartial/objective means to measure/approximate any effect is all I need. I am using SAS, and have access to SPSS v12.

I was wondering if anyone had any suggestions for possible appropriate statistical tests?

Many thanks :)
Last edited: