An Imputation Procedure to Extrapolate Data for Unobserved Panels in
Longitudinal Surveys


Jason Legg, Iowa State University

Many longitudinal monitoring programs use a partially overlapping panel 
observation scheme.  Missing data occur for units in a panel when the 
panel is unobserved.  Missing data are often imputed for the released 
dataset so that researchers can work on a complete dataset.  Interpolation 
between two observations on a unit often uses the natural structures of 
what is being monitored.  Extrapolation of missing data is more difficult 
than interpolation due to the lack of information about whether or not 
the unit has changed status.  It is desirable that the extrapolated 
dataset have estimates of changes computed from the observed dataset.  
We propose an extrapolation procedure that imputes two values, one 
representing change and one representing no change.  The resulting 
dataset will have the same weights on units as in previously released 
studies and estimates of change will match direct estimates from the 
real observations.  The procedure is described as an application to 
the National Resources Inventory.