The SSM Procedure

Missing Values

For a variety of reasons the data might contain missing response and predictor values. Before starting the analysis of a particular BY group, SSM procedure makes an internal copy of the data. The actual analysis is done by using this copy. The data in the copy are first examined for missing values in the response, predictor, and the ID variables. No missing values are permitted in the ID variable (if it is specified). If all the missing values are associated with only the response variables, then the internal copy of the data is not altered. However, if any of the predictors in the observation equation—the elements of $\mb{X}$ matrix—are found to contain missing values, the internal copy of the data is altered as follows: any missing predictor value is replaced by 0, and the response values that are dependent on that predictor in the corresponding row are set to missing. These missing response values are called the induced missing values. The reported analysis is based on the (possibly altered) internal copy of the BY group.

Missing values are not permitted in any of the other system matrices that define the state space model. In particular, missing values are not permitted in $\mb{Z}, \mb{T}, \mb{W}$, and $\mb{Q}$ matrices. In some cases the elements of these matrices depend on the data values. In such cases, care must be taken to ensure that these data values are not missing.