ERIC Number: ED339747
Record Type: Non-Journal
Publication Date: 1991
Reference Count: N/A
An Analysis of Methods Used To Reduce Nonresponse Bias in Survey Research.
Johnson, Victoria A.
The effectiveness of five methods used to estimate the population parameters of a variable of interest from a random sample in the presence of non-response to mail surveys was tested in conditions that vary the return rate and the relationship of the variable of interest to the likelihood of response. Data from 125,092 adult Alabama residents in the 1980 United States Census were used to simulate the population of interest. Total income was the variable of interest, known only for respondents. Gender and age were the classification variables known for the entire population. From the population, 1,000 random samples were selected and divided into respondents and non-respondents based on 25 different conditions (5 assumptions of response times 5 response rates). The following five methods were used to estimate the population parameter (mean of total income) under each condition: (1) using data from respondents and ignoring non-response; (2) using analysis of waves techniques; (3) using double sampling; (4) weighting; and (5) using hot-deck imputation. Double sampling seems to be the most effective method overall, even though tested under ideal conditions. Ignoring non-response is the least effective method. The study illustrates the importance of determining any relationship between the tendency to respond and the variables of interest. Two tables of data and five graphs illustrate the study. A 54-item list of references is included. (SLD)
Publication Type: Reports - Evaluative
Education Level: N/A
Authoring Institution: N/A
Identifiers - Location: Alabama