The Technical Report on the Methodology of the MIDUS Survey
consists of three sections:
1. Field Procedures
2. The Response Rate
3. Weighting the MIDUS Data
View Field Procedures and The Response
Rate below, and/or download complete 24-page report
(includes Weighting the MIDUS Data which is only available through download).
The National Survey of Midlife Development in the U.S. (MIDUS) was based on a nationally representative random-digit-dial sample of noninstitutionalized, English-speaking adults, aged 25 to 74, selected from working telephone banks in the coterminous United States. Field procedures lasted approximately one year, with all data collected during 1995. Predesignated households were selected in random replicates, one-fourth of which included a special enhanced nonrespondent incentive component and the other not. Contact persons were informed that the survey was being carried out through Harvard Medical School and was designed to study health and well-being during the middle years of life. They were told that participation would entail completing a telephone interview and two mail questionnaires.
After explaining the study to the informant, a household listing was generated of people in the age range 25-74 and a random respondent was selected. Oversampling of older people and of men was achieved by varying the probability of carrying out the interview at this stage as a joint function of the age and sex of the randomly selected respondent. No other person in the household was selected if the selected respondent did not complete the interview.
Once it was determined that we wanted to include the random respondent in the survey, an attempt was made to talk with this person and recruit them to be a participant. A study fact brochure was mailed to respondents who asked for more information before deciding and a recontact telephone appointment was made after the time they received the brochure. Senior study staff were also made available to respondents requesting information not contained in the brochure before deciding to participate.
Once random respondents decided to participate we carried out a telephone interview that lasted an average of thirty minutes and mailed questionnaires that we estimated took an average of an additional two hours to complete. The questionnaire mailing also included a boxed pen and a check for $20. A reminder postcard was mailed to all respondents three days after the initial questionnaire. A second questionnaire with a cover letter urging respondents to return the questionnaire was mailed two weeks later to all respondents who had not returned the questionnaire by that time. Reminder telephone calls were made two weeks later to all respondents who had still not returned the questionnaire. In the subsample of replicates designated for participation in the nonrespondent survey, interviewers in these final calls offered respondents an additional financial incentive to complete and return the questionnaires.
The Response Rate
We cannot compute an exact response rate for MIDUS because we only wanted to interview about half of people we contacted, and it is only the latter who should be in the denominator of the response rate. We have no way of knowing how many of the refusers would have been selected for interview, which means that we cannot compute the denominator exactly. However, it is possible to make an estimate of the number of people in the denominator and, based on this, an estimate of the response rate. This estimate, the calculation of which is described below, is 70.0% for the telephone interview, 86.8% for the completion of the main questionnaire among the telephone respondents, and 60.8% for the overall response rate (.700 x .868).
The sample disposition (Table
1. MIDUS Sample Disposition in the National RDD Prior to Refusal Conversion)
shows that 3323 respondents completed telephone interviews in the national
random sample before we carried out the refusal conversion. In the refusal
conversion of one-in four refusers who were offered a financial incentive
($100) to complete the interview, 162 initial refusers gave us a telephone
interview based on this incentive. (Table
2. MIDUS Sample Disposition in the National RDD Prior to Refusal Conversion.)
We gave each of these 162 a weight of four to adjust for the fact that
we only made this offer to one-fourth of respondents. The weighted numerator
of the response rate for the telephone interview, then, is: 3323 + 4 x
162 = 3971.
3323 (initial interviews usable prior to refusal conversion)
2 (unusable interviews)
131 (selected for interview but appointment never kept)
1008 (selected for interview but refused)
people had appointments but never kept, so we do not know if we would
B (2087 people refused before we determined if we wanted to interview them)
The uncertainty in computing the response rate involves the number of people in A and B who we wanted to interview. It is impossible to know. However, we estimated from the people who were screened that the percent eligible is 56.3%. This estimate is based on the fact that the people for whom screening was completed prior to refusal conversion included 3323 + 2 + 131 + 1008 eligible and 3636 ineligible minus 71 with language problems and 104 circumstantial. The ratio of eligible to total here is 4464/(4464 + 3461) = 56.3%. If we apply this same percent to A and B, we estimate that 1212 of the people in A and B combined were eligible, which means that a total of 5676 people were eligible overall, for a telephone response rate of 3971/5676 = 70.0%.
Once we had the telephone interview completed we developed a weight that adjusted the telephone respondents to be representative of the population. This weight was applied to the data, and then we computed empirically the weighted percent of these people who also returned the mail questionnaire. This conditional response rate is 86.8%. The overall response rate was then computed by multiplying the telephone response rate (.700) by the conditional mail response rate (.868) to get an overall response rate of 60.8%.
Sample Weights: Representativeness of the sample was increased by using a series of weights that adjusted for differences in probability of selection and differential nonresponse. A total of six weights were developed and the product of these weights used to create a final summary weight that can be used in analysis.
Typically, MIDMAC analyses are made with weighted and also unweighted data; when outcomes differ, the analyst reports this fact and attempts to interpret the basis. Conventions by various social science disciplines prefer the use of unweighted or weighted data. MIDMAC's convention is to present both when results warrant.
A detailed description of the procedure used to create these weights is entitled Weighting the MIDUS Data.
Guest Register / Top / Main
/ Biographies / Books / Publications / Bulletins / Links /
/ Research / MIDI Questionnaires /