Monday, Feb 1 at noon, 6050 ISR-Thompson
a PSC Research Project [ARCHIVE DISPLAY]
Investigator: Roger Tourangeau
Every request to take part in a survey is framed in some way. It is important to understand how the presentation of the survey request affects nonresponse and measurement. We propose a set of randomized experiments that will affect both nonresponse and measurement error in survey estimates. The experiments are guided by a set of theoretical assertions arising from recent studies of survey participation and measurement design. Specifically, in two initial experimental designs the topic of the survey and the sponsor of the survey will be randomly varied, with hypothesized effects both on nonresponse propensities and reporting . In the third experiment, survey design features that can mediate or reduce the error-producing influences will be examined. Thus, the project experimentally tests mechanisms productive of nonignorable nonresponse and measurement errors and, once the mechanisms’ effects have been documented, provides guidance to the survey practitioner about how these harmful effects can be reduced.
Two different recruitment protocols will be used to test the theories — one, using requests for a second survey among respondents to a prior request; the other, random assignment of recrutiment and measurement protocols to subsamples. In some experiments, the sampling frame will contain information on known behaviors or interests of the sample person.
The key hypothesis is that topic interest and sponsorship act to produce nonignorable nonresponse when the questionnaire contains items relevant to those features of the design, and that those attributes serve also as an interpretive framework for the intent of the questions, thereby affecting measurement error.
While the research is theoretically motivated and experimentally controlled, there are large practical implications of the work for Federal statistical agencies and the larger survey community. Both of the effects on nonresponse error and measurement error, if experimentally demonstrated, will be important findings to interpret the lack of agreement among estimates of key societal attributes (e.g., prevalence of rape, use of handguns). The practical importance of this work is a) to help agencies conducting surveys anticipate when different sponsors may obtain different results, b) to provide evidence about potentially harmful effects on nonresponse error and measurement error of emphasizing single purposes of a survey, and c) to produce evidence regarding design features that reduce the effects of the presentation of the survey on nonresponse and measurement error.
|Funding:||National Science Foundation (SES 0550385)|
Funding Period: 04/01/2006 to 03/31/2012
This PSC Archive record is displayed for historical reference.