How Skip Patterns Bias Disability Measurement: Evidence from SIPP and Cross-Survey Imputation Methods

• In Event: Survey Data Analysis, Issues in Measurement Error, and Latent Variable Modeling

Abstract

Differences in disability and health condition prevalence across major U.S. surveys are well documented but not well understood, with important implications for public policy design, program eligibility, and resource allocation. This study examines how questionnaire structure—specifically, skip patterns—contributes to these discrepancies. The Survey of Income and Program Participation (SIPP), a widely used data source for policy analysis, measures medical conditions using a hierarchical design in which respondents are asked about conditions only if they first report functional limitations. As a result, condition information is systematically unobserved for respondents who do not meet this threshold, creating a form of structural item nonresponse embedded in the survey instrument.

We quantify the impact of this skip pattern by leveraging the National Health Interview Survey (NHIS), which collects information on medical conditions without a functional gate. First, we simulate the SIPP questionnaire structure within NHIS to estimate the mechanical effect of the skip pattern on condition prevalence. We then estimate models of condition prevalence in NHIS and apply these models to SIPP respondents who are excluded from the condition module, implementing a cross-survey imputation approach to recover missing outcomes. This allows us to construct corrected population estimates that account for gate-induced missingness.

We define the gate effect as the difference between corrected estimates and inverse-probability-weighted estimates that represent the full population but omit condition information for gate-excluded respondents, thereby isolating the contribution of structurally induced outcome missingness. Results indicate that the skip pattern introduces substantial and heterogeneous distortions in prevalence estimates, with some conditions underestimated due to exclusion and others overrepresented due to selection into the observed sample.

These findings have direct implications for policy analysis. Because SIPP data are frequently used to study eligibility for disability-related programs, health disparities, and economic well-being, gate-induced measurement error may bias estimates of need, mischaracterize target populations, and affect the evaluation of program effectiveness. More broadly, this study highlights how survey design choices shape the evidence base used in policymaking and demonstrates a general approach for diagnosing and correcting structurally induced measurement bias in population data systems.

Author

• Hyun Ju Kim, University of New Hampshire
  Presenting Author