topic Multivariate normality for imputation by linear regression (PROC MI) in Statistical Procedures

Multivariate normality for imputation by linear regression (PROC MI)

grep2638 — Mon, 28 Aug 2017 13:57:03 GMT

Hello everyone,

The SAS documentation about PROC MI (on this page) states that multivariate normality is required for imputation by linear regression (implemented in MONOTONE REG or FCS REG statement).

Is it an assumption for residuals and/or variables ?

Where does this assumption is needed in the regression algorithm described here ?

Thanks in advance

Re: Multivariate normality for imputation by linear regression (PROC MI)

SAS_Rob — Tue, 29 Aug 2017 12:37:49 GMT

The multivariate normality is related to the variables themselves--that is, those you include in the specific imputation model

Re: Multivariate normality for imputation by linear regression (PROC MI)

grep2638 — Mon, 18 Sep 2017 15:02:52 GMT

So, is it incorrect to include class variables in the model ?

Re: Multivariate normality for imputation by linear regression (PROC MI)

SAS_Rob — Tue, 19 Sep 2017 16:08:27 GMT

While it does assume multivariate normality, it is fairly robust to small departures from it such as when you have dummy variables.