The PDLREG procedure skips any observations at the beginning of the data set that have missing values. The procedure uses all observations with nonmissing values for all the independent and dependent variables such that the lag distribution has sufficient nonmissing lagged independent variables.