150 likes | 263 Vues
This study investigates the relationship between nonresponse and measurement error in employment research through a panel study on labor market dynamics. It examines whether extra effort in recruiting respondents leads to lower quality data, and whether cooperators provide more accurate information. Using a dual-frame survey, the research includes analysis of recall biases among late respondents and the overall impact of additional efforts on total bias. By linking survey data with administrative records, this work aims to enhance understanding of measurement errors in employment surveys.
E N D
Nonresponse and Measurement Error in Employment Research Gerrit Müller (IAB) joint with: Frauke Kreuter (JPSM – U Maryland), Mark Trappmann (IAB)
Research Questions • Do survey respondents recruited with extra effort, provide answers of lower quality? • Are cooperators more motivated to provide accurate data? • Or, are late respondents hampered by recall deficits? • How does extra effort affect total bias?
Survey Data • Panel Study “Labor Market and Social Security” (PASS) • Dual frame survey (benefit recipients / residential population) • Wave1: 12,000 HH 20,000 P • RR1: 30.5% (within HH: 85%) • Mixed mode survey (sequential CATI -> CAPI)
Record Linkage I • Individual survey data linked with individual administrative data (80% of all Rs agreed; 72% successfully matched) • Administrative records on: employment, earnings, unemployment, labor market programs , • Contact data on HH-level only
Record Linkage II • Administrative data linked with paradata for the gross sample of recipients (from unemployment register) • Contact data on HH-level only • Indicator for Respondents / Nonrespondents on HH-level only ,
Hypotheses about Measurement Error(response process model, Tourangeau ’84) • Unemployment benefit (UBII) • July 2006 • Nov 2006 • at time of interview • Income in month prior to interview • Occupation • Educational degree Relationship between ME and response propensity (number of contact attempts)
Contact Quintiles and Follow-Up Efforts • Transfer CATI to CAPI • CATI NR follow-up of “soft refusals”
Measurement Error (in percent) by Contact Quintiles and Follow-up Efforts
Measurement Error by Contact Quintiles and Follow-up Efforts
5. NR-ME Bias Decomposition • for Recipient sample only • HH-level variables only (!) • UBII in Jul06 not feasible for bias decomposition • UBII in Nov • UBII at date of interview
“Pick your brains” • ME-Model for UBII in July06 (handout) Puzzle: high ME for the young? HH-interview by target head? • Administrative data not always „Gold Standard“ (error-free) Assumption: ME in register data unrelated to ME in survey reports and response propensity
“Pick your brains” • Decomposition findings statistic-specific • Extend analyses to P-level variables (e.g. employment, income) • Problem: unknown on individual level • How to go ahead?