This message has been cross posted to the following eGroups: Statistical Consulting Section and Statistical Computing Section .
-------------------------------------------
Dear All,
I have a data set at two time points, at t0 there is nothing missing but at t1 109 out of 442 patients have 4 variables (which I am including in mymodel) as missing
and the table below shows the pattern of missingness in our data. Looking at the pattern of missingness I believe that our data is monotone missing (please feel free to correct me if you think otherwise) v1 v2 v3 v4 freq
Percent
1
X
X
X
X
286
64.71
2
X
X
X
.
3
0.68
3
X
X
.
.
1
0.23
4
X
.
.
.
14
3.17
5
.
X
X
X
29
6.56
6
O
O
O
O
109
24.66
To impute missing values for a continuous variable in data sets with monotone missing patterns, we should use either a parametric method that assumes multivariate normality or a non-parametric method that uses propensity scores (Rubin 1987, pp. 124, 158; Lavori, Dawson, and Shera 1995). Parametric methods available include the regression method (Rubin 1987, pp. 166-167) and the predictive mean matching method (Heitjan and Little 1991; Schenker and Taylor 1996).
I am using default option of PROC MIANALYZE with variables v1 ,v2, v3, v4 as in the above table. PROC MI's default method for imputation of missing data is the Marko Chain Monte Carlo (MCMC) algorithm. As Allison (2005) notes, this method is based on the assumption of multivariate normality which implies valid imputations may be generated by linear regressions.
If SAS uses the multivariate normal approach via MCMC methods as default, what exactly is the default method doing internally to impute the values of variables selected in the proc MIANALYZE "VAR" statement and is it correct to use the default method for the missing patterns such as mine?
Thanks
Tasneem
-------------------------------------------
[Tasneem] [Zaihra]
[Post Doctoral Fellow]
[McGill University]
-------------------------------------------