	New content
Sociological Methods & Research

Restricted access

Research article

First published May 1975

Estimation of Nonresponse Bias

Darnell F. HawkinsView all authors and affiliations

https://doi.org/10.1177/004912417500300406

Abstract

An imputation procedure is used to estimate the effects of nonresponse on issues of substantive interest in a social survey. Using this method, one can determine that nonresponse bias may have differential effects on variable means, depending on the combination of independent variables used in the ensuing substantive analysis of data from the survey.

Get full access to this article

View all access and purchase options for this article.

Get Access

1. For example, see the American Statistical Association Conference (1974). Participants in the conference reported that response rates of 60% to 65% are average completion rates for current surveys. In the Detroit Area Study from which the data for the current analysis is taken, response rates have dwindled from 85% or more in the 1950s to current rates of near 70% or less.

2. In sampling theory the terms precision and accuracy are used to refer to the effects of bias and variable errors. It can be shown that the total survey error is a function of variable errors due to sampling and the bias due to measurement errors. Total Error = √VE² + Bias² High rates of nonresponse will affect the latter term, and consequently increase the total survey error (Kish, 1965: 510). These terms are also roughly synonymous with the terms reliability and validity used in psychology and referred to by Campbell and Stanley (1963).

3. Eight variables chosen for this analysis are standard socioeconomic types (age, sex, race, occupation, family income, education, marital status, and religion). "City of residence" was added to investigate possible Detroit-suburban response differences. "Years in the Detroit Area" was added to check for the effects of geographical mobility and migration patterns. The "housing type" variable was added to investigate respondent accessibility effects.

4. The 2 additional census descriptions already referred to were also used to check on the accuracy of the racial designations for the 76 other blocks.

5. The imputed mean income for these refusals using the mean block income was approximately $11,000. With mean income for their age group, it was lowered to $6,000.

6. Because of the fact that there were 37 refusals with missing data that were not used in the nonresponse analysis, a final check of the racial characteristics of these refusals was made. Using the race imputation procedure outlined above, one finds that 30 of these would have been labeled "white" and 7 "black." If these 37 are then added to the 140 refusals with no missing data, the percentage of blacks in the sample of refusals would increase from 10.7 to 14.1. This small increase would not have seriously affected the results obtained in this analysis.

References

American Statistical Association Conference (1974 ) "Report on the ASA Conference on Surveys of Human Populations ". Amer. Statistician 28, 1 (February) 30-34.

Abstract

Get full access to this article

References

Cite article

Cite article

Download to reference manager

Share

Share this article

Share with email

Share on social media

Share access to this article

Information

Published In

Rights and permissions

Authors

Affiliations

Metrics

Journals metrics

Article usage*

Articles citing this one

Figures & Media

Tables

Get access

Access options

Sign in

View options

PDF/ePub

Also from Sage

Article usage^*