SIPP Home > Technical Information > SIPP Data Editing and Imputation > Types of Missing Data >
There are many instances in which data are missing for a person in one or two consecutive waves but are present for that same person in the two adjacent waves. For example, a person may be missing in Wave 5 but have complete data for Waves 4 and 6. Beginning with the 1991 Panel, the Census Bureau began imputing those missing waves in the full panel files. Missing wave imputation is performed only when one or two consecutive missing waves are bounded on both sides by waves in which the sample member was present. If a respondent has missing data for more than two consecutive waves, the imputation is not performed.
For missing waves that are bounded on each side by interviewed waves, data are interpolated using a random carryover procedure. A value r is randomly assigned to each nonrespondent.s household for each missing wave, where r = 0, 1, 2, 3, or 4. The first r reference months within the missing wave receive their imputed values from the fourth month of the preceding wave, and the remaining 4 r reference months receive their imputed amounts from the first month of the subsequent wave.
Although this procedure results in data conducive to many analytic purposes, the random carryover forces stability in responses for wave nonrespondents. That stability could result in underestimation of between-wave changes. The procedure also results in imputed waves that do not exhibit the seam effect common to waves of reported data (Chapter 6 of the SIPP Users' Guide). Williams and Bailey (1996) provide a complete account of the handling of missing wave data in SIPP.
|
Main |
Introduction to SIPP |
SIPP Survey Content |
Technical Information |
Using & Linking Files |
SIPP Publications |
|
Access SIPP Data |
SIPP Users' Guide |
SIPP Tutorial |
User Notes/ListServe/News |
SIPP Help |
Page Last Modified: May 9, 2006