Chapter 2: Sample Design and Attrition (con't)

Table of Contents

2.5 Interview Methods

Before each survey period began, the Census Bureau generated lists of respondents to be interviewed and distributed them to 12 regional offices. Current addresses and contact information were generated from data collected during the last interview and through a postal check conducted by Census, and cases were assigned to interviewers who lived in the same geographic area as the respondent. Interviewers then received copies of the questionnaire (or a laptop computer for CAPI interviews) and (if used) respondents' Household Record Cards and flashcard and/or information booklets.

In each survey round, interviewers were responsible for contacting each respondent in their caseload and for using additional local level resources to locate those respondents who had moved since the last interview. Respondents who had moved outside the geographic district of their original interviewer were assigned to another unless there were no personnel nearby. In the latter event, an effort was made to interview the respondent by telephone.

Each respondent to be interviewed was sent various materials designed to encourage continued participation. Advance letters thanking respondents for taking part in the interviews and informing them of the coming survey were mailed prior to each interview period. Fact sheets highlighting recent research findings from each cohort's survey data were also provided. Respondents who had initially refused to participate in a survey were sent letters and some additional materials by the regional offices designed to encourage their continued participation and were once again contacted by local level interviewers to secure the interview.

While the type of survey, personal or telephone, determined the chief mode of contact, an alternate contact method was used for certain respondents. During a personal survey, for example, those respondents who lived long distances from the Census interviewer's base of operation or those for whom the Census supervisor has decided that another contact method was warranted were contacted by telephone. Although survey instruments are written in English only, multilingual interpreters were made available by the regional offices to interviewers who needed them.

Beginning in 1995, respondents in the two women's cohorts were interviewed during the same time period; a single computer-assisted personal interview (CAPI) replaced the separate paper-and-pencil interview (PAPI) instruments used during the previous interviews. While data were collected simultaneously for the two cohorts, they were released separately by cohort. This CAPI interview continued on a biennial basis through 2003.

The average length of an interview varied depending on the type conducted, with personal PAPI interviews lasting 50-60 minutes, telephone PAPI interviews averaging 20-25 minutes, and CAPI interviews lasting about 70 minutes. As a validation check, a percentage of the respondents were contacted following the interviews to ascertain that the interviews had taken place. No stipends were paid to Original Cohort respondents for their participation until 2003. During the 2003 fielding, a special incentive experiment was conducted to see how a subset of respondents responded to being offered money to participate. Variables associated with this incentive experiment are R65017.00, R76056.00-R76060.00, R76105.10, and R76105.20.

Return to top

2.6 Eligible Sample & Reasons for Noninterview

In general, respondents selected for interviewing each year were those who participated in the initial survey and who were alive, residing within the United States at the interview date, not members of the Armed Forces, and noninstitutionalized. (Exception: In 2003, an interview was conducted with an institutionalized respondent. See R65097.00-R65124.00.). However, the criteria used to select the eligible sample--respondents whom the Census Bureau attempts to interview in a given round--have varied somewhat over the years.

Beginning in 1969, any respondent who had refused to be interviewed during a previous round was dropped from the eligible sample. Beginning in 1971, respondents were also dropped from the eligible sample if they had not been interviewed in two consecutive surveys for reasons other than death or refusal (for example, respondents who could not be located or contacted during the field period--those with 'Reason for Noninterview' codes of 1, 2, 3, 4, 5, 6, 8, or 11). The User Notes after Table 2.6.3 describe how dropped respondents can be identified. In 1983, the Census Bureau ceased dropping individuals for these two reasons, and in 1985 attempts were made to reinterview some of the dropped individuals. The following selection criteria determined which respondents would be reinterviewed (see Table 2.6.1 for examples of each case):

1. If the respondent refused to be interviewed in 1972 or earlier, she was not eligible to be reinterviewed. See example respondent 1.
2. If the respondent missed her second consecutive survey in 1972 or earlier, she was not eligible to be reinterviewed. See example respondent 2.
3. If the respondent refused an interview in 1973 or later, survey staff attempted to reinterview her in 1985. If the respondent was interviewed in 1985, she rejoined the eligible sample for all subsequent surveys. If she was not interviewed in 1985, she remained ineligible for all subsequent surveys. See example respondent 3.
4. If the respondent was interviewed in 1971 and subsequently dropped due to two consecutive noninterviews, she was eligible for reinterview in 1985. If the respondent was interviewed in 1985, she rejoined the eligible sample for all subsequent surveys. If she was not interviewed in 1985, she remained ineligible for all subsequent surveys. See example respondent 4.
5. If the respondent first refused to participate in 1982, she was not eligible to participate in 1983 but rejoined the eligible sample in 1985. She remained in the eligible sample regardless of her interview status in 1985. See example respondent 5 in the table.
6. If the respondent was not interviewed in 1981 and 1982 for reasons other than death or refusal, she was never dropped from the sample. She remained eligible for all subsequent rounds regardless of prior participation. See example respondent 6.

Table 2.6.1  Selection of Respondents Eligible for 1985 Survey

Table 2.6.2 below depicts reasons for exclusion from the eligible sample and the years each applied; Tables 2.6.4 and 2.6.5 later in this section present reasons for noninterview across survey years.

Table 2.6.2  Reasons for Exclusion from the Eligible Samples

Each survey year, CHRR has created a cumulative 'Reason for Noninterview' variable for the full sample of respondents. This created variable is a combination of (1) the noninterview reasons provided by Census for the subset of respondents designated as eligible for interview in that survey year and (2) the reason for noninterview assigned during a previous survey to out-of-scope respondents. In 1982, CHRR began releasing an additional variable reflecting the reasons for noninterview for only those respondents with whom interviews were attempted that year (e.g., R07564.00). The number of respondents that Census designated as eligible for interviewing fluctuated by survey year. The eligible sample can be identified by adding together the interviewed respondents and the respondents with a noninterview code in the original 'Reason for Noninterview' variables, those not marked 'Revised Deceased.'

There was concern throughout the survey years that the number of respondents marked 'deceased' might be underreported and that some of the other 'Reason for Noninterview' categories might thus be overreported. In 2003, the Census Bureau, through the cooperation of the Social Security Administration, obtained SSA records for respondents who had shared their Social Security number with Census to verify whether or not respondents were living or deceased. As was previously suspected, a comparison of the SSA data with the survey data shows a substantially higher number of deceased respondents in the more recent surveys. To correct for past survey discrepancies, a revised 'Reason for Noninterview--Revised/Deceased' variable using the SSA data was created for each survey year from 1970 on. These revised variables are R01453.20, R02525.20, R03353.20, R04171.20, R05195.20, R05483.20, R05874.30, R07099.30, R07564.70, R08032.70, R09473.40, R10628.40, R11092.40, R12327.40, R13652.40, R16014.20, R34985.20, R42671.20, R63203.30, and R76154.00. Researchers are urged to use these updated variables rather than the original 'Reason for Noninterview' variables.

Instructions to interviewers on how to code a respondent's reason for noninterview appear within the Interviewer's Reference Manuals (or Field Representative's Manuals). The set of noninterview coding categories present during the initial survey years has been supplemented over the years with additional reasons for noninterview, and the meanings of existing categories have been refined. Table 2.6.3 presents the raw coding categories present on the public data files and specifies the survey years during which each category was utilized.

Table 2.6.3  Conceptual & Raw Coding Categories for the Reason for Noninterview Variables

The reason for noninterview coding categories depicted in Tables 2.6.4 and 2.6.5 were constructed from the raw coding categories as shown in Table 2.6.3. For example, the conceptual category "can't locate" is the sum of codes "1" and "4." Tables 2.6.4 and 2.6.5 depict the number of respondents not interviewed by survey year, reason, and race.

Table 2.6.4 Reasons for Noninterview: 1969-2003

Table 2.6.5  Reasons for Noninterview by Race: 1969-2003