You are not currently logged in.
Access your personal account or get JSTOR access through your library or other institution:
Correcting for Survey Nonresponse Using Variable Response Propensity
Journal of the American Statistical Association
Vol. 105, No. 492 (December 2010), pp. 1418-1430
Stable URL: http://www.jstor.org/stable/27920175
Page Count: 13
Preview not available
All surveys with less than full response potentially suffer from nonresponse bias. Poststratification weights can only correct for selection into the sample based on observables whose distribution is known in the population. Variables such as gender, race, income, and region satisfy this requirement because they are available from the U.S. Census Bureau, but poststratification based on these variables may not eliminate nonresponse bias. I develop an approach for correcting for nonignorable nonresponse bias. Survey respondents can be classified by their "response propensity." Proxies for response propensity include the number of attempted phone calls, indicators of temporary refusal, and interviewer-coded measures of cooperativeness. We can then learn about the population of nonrespondents by extrapolating from the low-propensity respondents. I apply this new estimator to correct for unit nonresponse bias in the American National Election Study and in a CBS/New York Times preelection poll. I find that nonresponse bias can be a serious problem, particularly for items that relate to political participation. I find that my method is successful in substantially reducing nonresponse bias.
Journal of the American Statistical Association © 2010 American Statistical Association