The weight for a responding unit in a survey data set is an estimate of the number of units in the target population that the responding unit represents. In general, since population units may be sampled with different selection probabilities and since response rates and coverage rates may vary across subpopulations, different responding units represent different numbers of units in the population. The use of weights in survey analysis compensates for this differential representation, thus producing estimates that relate to the target population.
SIPP weights vary due to differential sampling rates as a result of oversampling and because response and coverage rates vary across subpopulations. For example, in Wave 1 of the 2004 Panel, the final person lower quartile weight is 1,682 and the upper quartile weight is 3,429 (the maximum weight is 16,482). A respondent with a final person weight of 1,682 represents 1,682 people in the U.S. population for the reference month, whereas a respondent with a weight of 3,429 represents 3,429 people. Because weights in SIPP vary over a sufficiently large range of values, performing unweighted analyses may produce appreciably biased estimates for the U.S. population.
The decision of which weight to use for a given analysis depends on the population of interest for that analysis. Useful guidance for choosing the correct set of weights is to consider to what population the results are intended to apply.
The weights in the SIPP files are constructed for sample cohorts defined by:
Users can choose to base their analyses on:
Monthly (cross-sectional) weights allow the use of all available data for a given month. For this type of analysis, users can choose among the following units of analysis:
Analysts can use longitudinal samples to follow the same people over time and hence study such issues as the dynamics of program participation, lengths of poverty spells, and changes in other circumstances (e.g., household composition). The longitudinal weights allow the inclusion of all people for whom data were collected for every month of the period involved (calendar year or full panel period), including those who left the target population through death or because they moved to an ineligible address (institution, foreign living quarters, military barracks), as well as those for whom data were imputed for missing months. The Census Bureau makes nonresponse adjustments to the longitudinal weights to compensate for panel attrition and poststratification adjustments to make the weighted sample totals conform to population totals for key variables.
This guide is intended as a reference for data users who need information about selecting the appropriate weight(s) for analyzing 2014 SIPP panel data. Please note this document contains guidance that may be inconsistent with the advice stated elsewhere on this page that applies to other SIPP panels.
This section describes how the weights are constructed. The basic components for all the different sets of weights are the same, namely:
Reference month final weights are provided on the SIPP core wave files for persons, households, families, and subfamilies. The special weights for persons are constructed first. The household, family, and related subfamily final weights are derived from the final person weights. This section summarizes the steps involved in constructing the various sets of weights, starting with the final person weights for a reference month. The reference and interview month weights1 for people on the core wave files are computed (i.e., are nonzero) for all responding sample members who are in scope. (i.e., a part of the survey's universe.the resident, noninstitutional population of the United States) in the specified month.2 A number of factors lead to fluctuations in sample size from month to month. They include births, deaths, immigration, and emigration from the population (and therefore from the sample). In addition to those population dynamics, people move into and out of the sample as a result of the changing household composition of sample members. In Wave 1, the weight for each sample person per month is a product of four components:
In subsequent waves, each person receives an initial weight that is carried over from the preceding wave. This weight is adjusted to compensate for changes in the sample between waves resulting from movers and nonresponse, and then it is realigned to match the population totals for the reference or interview month:
Final full panel and final calendar year weights are provided on the full panel files for eligible sample members. There is one set of final panel weights and generally more than one set of calendar year weights, one for each calendar year covered by the panel. Final panel weights are computed only for people who are in the sample at Wave 1 of the panel and for whom data are obtained (either reported or imputed) for every month of the panel for which they were in scope for the survey. Other people in the panel file are assigned weights of zero. Most people with nonzero final panel weights have provided data for all months of the panel. However, people who missed a wave and whose missing wave data were imputed and people who provided data up to the point that they left the survey (through death or because they moved to an ineligible address) are also assigned nonzero final panel weights. Final calendar year weights are computed only for people who had an interview covering the control date5 and for whom data are obtained (either reported or imputed) for every month of the calendar year for which they were in scope for the survey. Other people are assigned final calendar year weights of zero. Some people who joined the household of an original sample person after the start of the panel are assigned nonzero calendar year weights for the second calendar year, if data are obtained for that period.
The full panel weighting scheme does not assign weights to people who enter the sample universe after Wave 1. Similarly, the calendar year weighting scheme does not assign weights to people who do not have an interview covering the control date. This group consists of (a) people who enter the sample universe after the first wave of interviewing for the calendar year and (b) people who were in the sample universe in the first wave of interviewing in the calendar year but did not have an interview covering the control date. For example, newborn infants and people leaving institutions who are entering the sample universe after Wave 1 are assigned full panel and calendar year 1 weights of zero. Note that the same people will receive positive calendar year 2 (CY2) weights if they are in the sample universe in the first wave of interviewing for CY2 and have an interview covering the control date for CY2.
The final panel and calendar year weights are constructed from the following three components:
1 Interview month weights were not computed for the 1996 Panel.
2 Persons subjected to Type Z imputation receive weights, although they are not respondents.
3 This adjustment has been used since Wave 5 of the 1984 Panel.
4 Known as the control card information before the 1996 Panel, when computer-assisted interviewing (CAI) began.
5 The calendar year control dates are January 1 for the given calendar year. The exception is calendar year 1996 for the 1996 Panel. Its control date is currently March 1, 1996. This would change to January 1 should there be imputation for January and February data.