Association for Public Policy Analysis & Management

The seam effect, sometimes known as seam bias or heaping, is often observed in longitudinal panel surveys. A seam effect occurs when administrative and cross-sectional data suggest that transition rates should be constant, or constantly changing, while a longitudinal survey estimates that a transition rate is higher around the time of the interview and lower at other times. Seams are a concern for data collection because they compromise the ability of researchers to generate high quality monthly estimates of coverage rates and to accurately measure transitions. Researchers have observed seam effects for topics such as labor force participation, program participation and health insurance coverage, and have attempted to handle these effects using a variety of instrument and post-hoc statistical strategies (For a review, see Callegaro, 2008). The 2014 Survey of Income and Program Participation (SIPP) was redesigned to include an event history calendar and dependent interviewing, two strategies designed to limit the seam effect; however, research has yet to examine whether changes to the SIPP survey produced desired consequences.

Objective:

Using health insurance coverage as a case study, this paper will examine three research questions: Is there evidence of a seam effect in the 2014 SIPP Panel? If found, how large is this seam effect relative to what others have found for past SIPP Panels? Finally, what characteristics (age, race, sex, race/ethnic origin, income, region, inconsistent reporting of demographic characteristics) are associated with having any seam effect and the magnitude of the effect?

Data and Measures:

Analysis will use data from Waves 1 and 2 of the 2014 SIPP Panel. Two outcomes will be studied—the hazard of transitioning into or out of any insurance coverage and the hazard of transitioning into or out of Medicaid. Model covariates include: the interview month, age (less than 18, between 18 and 64, above 65 years of age), race/ethnic origin (white non-Hispanic, black non-Hispanic, Hispanic, other race/ethnicity), sex (male/female), income (less than 100% of federal poverty level in either survey wave), region (South, Midwest, Northeast, Northwest), and inconsistent values of demographics across survey waves (change in age of more than 3 years, change in race/ethnicity, and change in sex).

Analytic Sample:

Persons in the analytic sample for all months of Waves 1 and 2 of the 2014 Survey of Income and Program Participation.

Methods:

This study will use a multi-state hazard model with two independent and exhaustive states: has coverage, does not have coverage. Models will start with Markov assumptions of time—future transitions depend only on the current state occupied. As needed, time covariates will be added to allow for period specific changes in health insurance coverage.

Results:

The key results to be shown are hazard ratios from relevant models and model estimates of monthly insurance coverage.

Discussion:

The seam effect has long been a concern in SIPP because it affects monthly estimates. This paper will inform understanding of whether seam effects are operating in the 2014 SIPP Panel and offer guidance for addressing any seam effect observed.