The Survey of Income and Program Participation (SIPP) is a national longitudinal survey, conducted by the U.S. Bureau of the Census, which collects detailed information on income and wealth. The information collected includes employment income and income received from government transfer programs at the person, family, and household levels. The survey uses a rotating panel design, with a new panel of sample households being introduced at the start of each calendar year. Each panel is divided into four approximately equal rotation groups. Each month households from a different rotation group are interviewed. During each interview, the respondent is asked to provide information for the preceding four months. The four-month cycle, in which all of the households of the panel are interviewed, is referred to as a wave. The number of waves in a panel is determined by the length of the panel. Through 1995, households in most of the SIPP panels had been interviewed once every four months over a period of 32 months. Starting with the 1996 panel, the length of the panel will increase to four years with a new panel being introduced once the old panel is completed.

A major problem in obtaining accurate estimates of income and program participation from the SIPP is nonresponse. In some cases, a person or an entire household does not respond for one or more interviews during the length of the panel, resulting in one or more waves of missing data. These persons are referred to as panel nonrespondents. Longitudinal weighting procedures adjust for this nonresponse by assigning a zero weight to the nonrespondents and multiplying the weights of the persons interviewed during the entire period by a nonresponse adjustment factor. Because of this adjustment, the information for panel nonrespondents from waves for which they are interviewed, is not used in the estimates. In an effort to include more of the available data in the estimates, a longitudinal imputation procedure is performed for some of the missing data, thereby changing panel nonrespondents to respondents before the longitudinal weighting adjustment takes place. For each nonrespondent, this procedure imputes data only for missing waves bounded on both sides by an interviewed wave. This procedure does not impute data for two or more consecutive missing waves or for the first or last wave of the panel.

This research expands on the work done at the Census Bureau by Antoinette Tremblay (Tremblay 1994), where she compared alternative longitudinal imputation methods for single missing waves. In the earlier research, longitudinal imputation was performed on simulated missing data for food stamp amounts based on data from the SIPP 1990 panel. The current imputation procedure was performed along with three alternative procedures. The evaluation of the procedures was primarily based on the estimates of the accuracy of the imputations. Comparisons were made between the food stamp estimates derived from the imputed amounts and those derived from the actual amounts for a selected subset of the panel. The general conclusion from the evaluation was that, for food stamp amounts using the SIPP 1990 panel data, none of the imputation methods appeared to be significantly better than the rest.

For this research, we evaluated the same four longitudinal imputation methods evaluated by Tremblay. In addition to food stamp amounts, imputation and evaluation were performed for wage and salary, social security, and Aid to Families with Dependent Children (AFDC) amounts. Wage and salary was chosen with the program amounts because it was a directly imputed, major component of the poverty estimates. The following sections describe the methodology and findings of our research. Section II. gives a brief discussion of the four imputation methods. The methodology used to perform the imputations is described in section III. and the methodology used to evaluate the imputations is described in section IV. Sections V. shows the results of the evaluation and section VI. provides a summary of our research and further recommendations.