We Know Where you Live: The Impact of Movement on Differential Panel Survey Attrition Ned English, Cathy Haggerty, And Colm O’Muircheartaigh Presented AAPOR 2008 New Orleans, LA
Introduction Making Connections longitudinal, in-person study –Targets ten inner-city neighborhoods –Topics concerning “neighborhood” Programs to improve life for children Considerable movement between waves ~ 50% households left wave 1 housing unit 60% Households with Children
Introduction contd. Study followed child movers for interview –Eligible at any move destination Multi-phase locating with differential effort: –‘simple’, ‘moderate’, ‘challenging’ Found 66% of eligible movers –Interviewed 89% of them –Majority of non-response in locating “Movement”, “locating” thus key panel quality What are impacts of increased locating effort?
1.What households move vs. stay? 2.How do “simple” differ from “challenging”? Found vs. lost? 3.Can we predict qualities of movement? 4.Impact panel composition if less locating effort? Simulate by eliminating difficult cases Informs planning, understanding of any bias Summary
Background- Theory Loss of certain households affects panel composition (Matthews et al. 2006, Dennis and Li 2003) Particular categories more challenging (Uhrig 2008, Couper 1991) –Smaller households –Households without children –Those disconnected from labor market –Those detached from neighborhood itself
10 Making Connections Sites
Results- Whom Did we Field? n = 6743
Results- Who Stayed? ** * n = 6743n = 3006
Results- Who Moved? ** * n = 6743n = 3006n = 3737
Results- Who Moved n = 6743n = 3006n = 3737
Results- Predicting Who Moved EffectOdds Ratio African-American1.49 ** Drug-dealers problem1.2 ** Food stamp user1.31 ** Single-parent1.3 * Home Owner.29 ** n = 3737 How did we find movers with children (n = 1874)?
Locating Movers 1.All 1874 movers received “simple” treatment –LexisNexis “Accurint” search –Postcards –Consistent information found for 46% overall (n = 856) 2. Escalated to “moderate” –Additional locating tools –Telephone “contacts” –Consistent information found for 43% overall (n = 815) 3.Remainder escalated again “challenging” –In-person locating –11% overall (n = 203)
Results- Locating Rate by Treatment Simple (n = 856): 72% Moderate (n = 815): 68% Challenging (n = 203): 38% Locating imperfect What kinds of households fit categories?
Results- Who was hard to find? ** n = 1874n = 856n = 815n = 203
Results – Who Required “Challenging” Locating? n = 1874n = 856n = 815n = 203
Results- Predicting “Challenging” EffectOdds Ratio High-Income.178 ** African-American.39 ** Drug dealers problem.417 ** Own vehicle.646 * n = 203
Results cont.- Found vs. Lost * * n = 1874n = 1245n = 629
Results cont.- Found vs. Lost * n = 1874n = 1245n = 629
Results- Predicting Lost EffectOdds Ratio Savings Account.72 ** High Income.496 * n = 629
Results – official panel n = 2234n = 2173n = 1671
Results – official panel n = 2234n = 2173n = 1671
Results – official panel * * * * n = 2234n = 2173n = 1671
Discussion and Conclusions Demonstrated method to quantify locating effort –Some unexpected results e.g., “simple” cases not found – Locating comprehensive considering challenges –Found 66% of all movers Low income, African-American, single-parent households moved “Challenging” tended to be Latino, no bank account –We depend on visible financial transactions –Car ownership, high-income –Possibility of re-locating outside USA –Will examine length of time in US, country of origin
Discussion and Conclusions contd. Absence of locating affects panel composition Panel benefited from the locating efforts re “moderate” and “challenging” movers –If no locating would be less Latino, wealthier, more stable households –Locating effort ensured panel consistency, contiguity
Thanks...