Missing values

The way to handle missing values is to replace each missing value by an estimate computed outside of Plaid. The estimate could be a neutral background value, or it could be taken from a simple model such as a global anova on the whole data set. In this case one might replace a missing value $X_{ij}$ by $\bar X_{i\cdot} + \bar X_{\cdot j} - \bar X_{\cdot\cdot}$. Here $\bar X_{i\cdot}$ is the average of non-missing $X_{ij}$ values in row $i$ $\bar X_{\cdot j}$ is defined similarly, and $\bar X_{\cdot\cdot}$ is the average of all non-missing data values in the array.