Multiple imputation involves making random draws from the posterior distribution of the variables. When you specify the MIN= and MAX= options, you restrict the possible values that can be drawn. If PROC MI draws a value outside the range, it will attempt to make another draw. If that continues beyond 100 times, it aborts the process and issues the following message.

You can specify the MINMAXITER= option to increase the number of attempts, although in practice that rarely works.

It might make sense to remove the MIN= and MAX= options since the restrictions they impose have been shown to decrease efficiency and increase bias because they alter the mean vectors and covariance matrices of the posterior distribution. Imputed values should not be viewed as equivalent to observed values and "serve only to help estimate the covariances between variables needed for inference" (Johnson and Young, 2011).

Nevertheless, if there is a need to have the imputed values fall within a certain range, then there are other options you can try. You could change the imputation method or try specifying your own prior distribution. Another option is to remove the MIN= and MAX= options and then impose those restrictions in a DATA step after performing multiple imputations. This process might require a trial-and-error approach and details of its implementation are highly data dependent.

Reference

Johnson, D., and R. Young (2011). "Imputing the Missing Y's: Implications for Survey Producers and Survey Users." Proceedings of the AAPOR Conference Abstracts, 6242–6248.

Operating System and Release Information