Concordance of sibling's recall of measures of childhood socioeconomic position

Data were from the National Survey of Midlife Development in the United States (MIDUS), a survey of the health, psychological well-being, and social and economic circumstances of adults in mid-life in the United States (U.S.), conducted in 1995-1996 [11]. The survey included four samples: the main sample, siblings of participants in the main sample, a sample of twins, and dedicated samples in five selected cities. Participants in the main sample were recruited by random-digit dialing of households in the contiguous U.S. from working telephone banks to provide a representative sample of noninstitutionalized English-speaking adults. One member of each household who was age 25 to 74 was randomly selected to participate. Participants completed a telephone interview and were then asked to complete a mailed self-administered questionnaire. The response rate for telephone interview was 70% and for the questionnaire was 87%, for a final sample of 3032 participants. Siblings were recruited from a random subsample of 529 participants in the main sample who completed the interview. Only full siblings age 25 to 74 were eligible, but more than one sibling per family could be enrolled. Interviews were completed by 950 of 1480 siblings identified. Twins were referred from other national surveys that screened approximately 50,000 households for the presence of twins, of which 60% were enrolled (n = 1914). Some families had more than one twin pair enrolled, and for some pairs, both twins did not enroll. The city-specific samples were not included in this study.

From these samples, we identified 529 sibling pairs and 885 twin pairs, selecting at random one twin pair per family in cases where more than one twin pair had been enrolled. For families with more than one sibling enrolled, we selected the sibling closest in age to the participant in the main sample. One sibling pair per family was selected to avoid non-independent observations. Of these 1414 pairs, we then excluded 134 pairs for which either the identified male head of household during most of their childhood (specified as before age 17) or the identified female head of household during most of their childhood differed between members of the pair. For example, one member of the pair may have identified the biological father as the male head of household while the other member of the pair identified a step-father as the male head of household. Because information on education and occupation were specifically asked about the head of household, we required both members of each pair to reference the same individual. Eighty-six pairs were deleted because they identified different male heads of household, 32 pairs were deleted because they identified different female heads of household, and 16 pairs were deleted because both heads of household differed, resulting in 1280 pairs (476 sibling pairs and 804 twin pairs).

Questions on measures of socioeconomic position before the age of 17 were asked during the telephone interview. Participants were asked to report the main job title of the male head of household (hereafter, father), which survey investigators then classified into one of nine categories of the U.S. census occupational classification system (professional, manager, technical worker, clerical, sales, craftsman, service worker, operative/laborer, farm worker) [12]. For analysis, the father's occupation was considered both as the 9-category classification and as a dichotomous variable representing professional occupation versus other. Data were only collected for a single main job title; if respondents reported their father changed jobs, they were instructed to report the main job he had during their adolescence. Participants were also asked if their father supervised others at work. Participants were asked their father's highest level of educational attainment in 12 categories, which for analysis was collapsed into 5 categories (grade school, some high school, high school graduate or General Educational Development qualification, some college, and college graduate). Educational attainment of the female head of household (hereafter mother) was similarly classified. Participants were asked if during their childhood or adolescence their family had received welfare or Aid to Dependent Children for at least 6 months. Lastly, participants were asked if they thought that while growing up, their family was better off or worse off financially than other families at the time, on a 7-category scale ranging from "a lot better off" to "a lot worse off". For analysis, responses were collapsed into three categories (better off, the same, and worse off). Only 28.6% of participants reported that their mother worked during most or all of their childhood, so mother's occupation was not analyzed.

Percent concordant responses between members of each pair were tabulated for each measure of childhood socioeconomic position, with 95% confidence intervals based on binomial proportions. Concordance measures only identical responses and does not account for chance. Agreement was therefore also estimated using weighted kappa, with exact 95% confidence intervals. Kappa provides a measure of agreement beyond that expected due to chance alone. Weighting of the kappa takes account of the degree of discrepancy between ordinal responses, with widely divergent responses discounted more than slightly divergent responses. However, kappa is sensitive to the prevalence of responses across categories [13]. Higher kappa indicates greater agreement. Data were missing for at least one member for father's occupation in 142 pairs (11.1%), for father's supervisory role at work in 126 pairs (9.8%), for father's education level in 174 pairs (13.6%), for mother's education level in 99 pairs (7.7%), for welfare during childhood in 20 pairs (1.5%), and for subjective appraisal of whether the family was better or worse off financially than others in 126 pairs (9.8%). These pairs were excluded from the corresponding concordance estimate because only non-missing responses are informative for concordance. Data were missing for both members of the pair for between 10% (welfare during childhood) and 38% (father's supervisory role at work) of pairs with missing data. No pairs had missing data on all measures.

To investigate if the degree of concordance was related to participant characteristics, we computed estimates for subgroups by age (younger or older than the group median of 46 years, and categorized based on the age of the younger member of the pair), sex, twin status, education level (less than high school, high school graduate, some college, or college graduate, based on the education level of the member of the pair with the lowest education level), and income (poor versus not poor). Pairs were classified as poor if either member reported an annual household income of less than $31,200, which was 200% of the 1996 federal poverty level for a family of four. Adjustment of income for household size was not possible because data on the number of members in the household was not available.

Analyses were performed using SAS programs (SAS Inc, Cary, NC).