Elizabeth Martin and Paul Siegel, Bureau of the Census
Prepared for NICHD Workshop "Improving Data on Male Fertility and Family Formation" at the Urban Institute, Washington, D.C., January 16-17, 1997
Errors made by respondents or interviewers in listing persons on household rosters are an important source of coverage errors in censuses and surveys. Within-household omissions account for about one-third of all census omissions, and are higher for males and minorities, and nonrelatives within households (Hogan, 1992; Ellis, 1994; Fay, 1989). Despite the evidence, the household roster has not been approached systematically as a survey measurement problem. Most surveys lack standardized questions and procedures to help interviewers decide whether to list persons whose residence is ambiguous, leaving these determinations to the interviewer's discretion and skill.
Research suggests several reasons why respondents may erroneously omit persons from household rosters. Persons may be concealed due to concerns about how the information is used by Government or others (Hainer et al., 1988; de la Puente, 1993; Tourangeau et al., forthcoming). Complicated living situations, transience, and tenuous attachments to households make it difficult to determine who should be counted as a household member. Mobility among multiple households contributes to residential ambiguity (Bates and Gerber, 1994). In ambiguous situations, respondents' judgments are influenced by intentions and agreements, financial contributions and permanence of attachment, and other criteria which may conflict with official residency rules (Gerber, 1990; 1994). Arcane terminology and counterintuitive instructions may confuse or mislead respondents (Gerber, 1994; Gerber, Wellens, and Keeley, 1996). Household respondents may lack information about persons in their household, and may assume that part-time residents have a home elsewhere, when they don't. There may be disagreements within households about who belongs there and who doesn't (Hainer, 1987).
In order to improve the coverage of tenuously attached persons, Census Bureau researchers devised an experimental strategy which cast a broad net in order to identify persons with any attachment to a household, no matter how weak or tenuous. The experimental rostering strategy was implemented in the Living Situation Survey, which was designed by Census Bureau researchers and conducted by RTI in 1993.
Step 1 in the survey was to interview household respondents and ask extensive roster probes and cues to list all persons with any attachment to the sample households. Extensive cues and probes were used to build rosters that included all persons with any attachment to the sample households, including (for example) persons who spent a night in the housing unit during the 2 month reference period, who received mail or messages there, had a key, contributed money for rent or bills, and so on. Cues also targeted undercounted categories, such as live-in employees, boarders, foster children, etc. The probes were developed based on evidence about undercounts, as well as cognitive and anthropological research on how people think about residency issues. The intent was to include on the roster everyone who had spent time in a household during the reference period, or who had other sorts of attachment to it.
Step 2 was to ask the household respondent questions to determine the residence of each person on the list. For example, household respondents were asked if this was the person's usual residence, "where he/she lives and sleeps most of the time."
Step 3 was to follow up a subsample of the rostered persons for individual interviews.
Step 4 was to determine (in the individual interview) all the places respondents had stayed during the reference period, and the nature of their attachment or participation in each household (e.g., did they help with chores, contribute money for rent, food, or bills, have children of their own who stayed in the household).
This design strategy offers the potential advantage of capturing information about persons in the gray area, who might otherwise be missed entirely. It also makes it possible to identify tenuously attached persons who have children staying in a household.
Interviews were conducted in 999 households (representing a 79.5 percent response rate) oversampled from areas with high concentrations of minorities and renters. A total of 3,549 people were listed on household rosters. The weighted mean of 3.62 persons listed per housing unit in the LSS is significantly greater than the mean of 2.63 persons per occupied housing unit in the 1990 census. The added probes in the LSS were especially effective at identifying more young minority males, who were less likely to be mentioned in response to more standard probes (Sweet, 1994).
More probing was needed to list persons with tenuous attachments than those who were more attached: Martin (1996) finds that on average 1.06 probes were needed to list the most strongly attached individuals, compared to 4.6 probes needed to elicit reports of persons with very weak attachments to sample households.
Based on the screening questions (Step 2), about three-quarters of the persons rostered using the new, inclusive procedure were residents of the sample households, and one-quarter lived somewhere else. When nonresidents were screened out, the mean number of usual residents per housing unit in the LSS was higher than the census for all race/ethnicity categories, but was significantly higher only for the total population (2.76) and for Hispanics (Sweet, 1994).
For a small but important group of marginal residents, household respondents' reports were often inconsistent with census rules and with reports of the individuals themselves. Nine percent of the persons rostered in the LSS (excluding casual visitors) had complex living situations, and household respondents' determinations of "usual residence" agreed with census rules for only 69 percent of them (Sweet and Alberti, 1994). The LSS followed up a sample of non-casual visitors rostered in the survey, and conducted individual interviews with them (or with proxies reporting for them). Sweet and Alberti (1994) find that in 95 percent of cases, the household respondent and the individual agreed on the individual's usual residence (proxy reports for the individual were excluded from their analysis). The 5 percent who disagreed tended to have complex living situations. Potential omissions due to inconsistent assessments of household membership were significantly higher for young, minority males compared to other groups (Schwede and Ellis, 1994).
To date, research based on the Living Situation Survey points to several conclusions. First, the expanded probing resulted in larger numbers of people listed on household rosters, with evidence of increases in undercounted categories (Hispanics, as well as young, minority males). Compared to the census, there was a 38 percent increase in the number of people rostered per household, but only a 5 percent increase in the number of usual residents per household. Second, household respondent reports of who lives in a household should not be taken as unproblematic. Third, people use different criteria and in many cases make different residency determinations than would be implied by the census residency rules. Fourth, living situations which are ambiguous and fluid are particularly vulnerable to misreporting and unreliable reporting. Fifth, more probing is needed to identify marginally attached persons than is customarily done in household surveys.
Identifying Tenuously Attached Fathers
Although the Living Situation Survey was not designed to investigate fathers' attachments to households where their children lived, the survey does suggest some avenues which might be worth exploring in future applications of this methodology. Once all of the places where a respondent had stayed during the 2-3 month reference period had been identified in the individual interview, respondents were asked for each, "Did you have children of your own who stayed at (PLACE)?" Thus, the survey provides preliminary information on parents' patterns of stay in households where their children also stayed. Table 1 shows, as one would expect, that the most common pattern is a stay in only one place where the R's children were living, and that place is the respondent's usual residence. This corresponds to a situation in which the children live with the respondent, and the respondent either has no other children, or didn't stay in the other children's residence during the reference period.
This pattern accounts for about 77 percent of fathers, and 85 percent of mothers, who stayed in households in which their children lived. The second most common pattern was respondents' staying in 2 places where their children lived, one their usual residence and the other not. This pattern was more common for fathers than mothers, but accounts for a sizable fraction of both. (It is important to note that a variety of situations may give rise to this pattern, including stays in two different households in which different children reside, or a trip to a place away from home accompanied by ones children.) Finally, there are small fractions (over 2 percent) of both men and women who report visits to multiple households where their children were staying. About the same fraction of men report visiting one place where their children lived but they did not.
These data suggest that multiple and frequent (occurring within 2-3 months) stays or visits by parents in households where their children live are fairly common in comparison to what might be the expected, normative pattern of staying in a single, usual residence, where children live with the parent(s). Multiple stays are more common for fathers than for mothers, and may contribute to men being left off household rosters, since their greater mobility may lead to their being regarded as more marginal in each household (Bates and Gerber, 1996). (It also may lead to double-counting in some cases.) Note that the parents who visit their children in households which are not the parents' usual residence are likely not to be counted there under less persistent rostering practices. To some extent, this will manifest itself as children with absent fathers, and to a somewhat lesser extent, children with absent mothers.
It is important to be careful in drawing any conclusions from Table 1, since several key pieces of information are missing. The survey did not collect information to identify parents per se, nor did it collect information on the number of households in which the respondent has children living. Nor did it identify the ages of the "children," who may be adult. (Table 1 is restricted to respondents 60 years or younger, to eliminate most visits to adult children.)
Nonetheless, the fact that the survey finds substantial numbers of parents visiting and staying in households where their children live, and that stays in multiple places where children live appear relative common, suggest that expanded roster probes may be effective in identifying parents (especially fathers) who have tenuous or multiple attachments to households in which their children live.
Bates, N. A., and Gerber, E., (1994), "A Typology of Temporary Mobility Patterns and Reporting of Usual Residence," Proc. of the Surv. Res. Meth. Sec., Am. Stat. Assoc.
Bounpane, P. A., (1991), Census Bureau Releases Refined Estimates from Post Enumeration Survey of 1990 Census Coverage. June 13, 1991, press release.
de la Puente, M. (1993) "Why Are People Missed or Erroneously Included by the Census: A Summary of Findings From Ethnographic Coverage Reports." Proceedings of the Conference on Undercounted Ethnic Populations. Washington, DC: Department of Commerce.
Ellis, Y., (1994), "Categorical Data Analysis of Census Omissions,"
Fay, R. E., (1989), "An Analysis of Within-Household Undercoverage in the Current Population Survey," Proceedings, Annual Research Conference. Washington, DC: Bureau of the Census.
Gerber, E. R., (1990), Calculating Residence: A Cognitive Approach to Household Membership Among Low Income Blacks. Unpublished report, Bureau of the Census.
Gerber, E. (1994) The Language of Residence: Respondent Understandings and Census Rules. Unpublished report, Bureau of the Census.
Gerber, E.R., and Bates, N.A., (1994), "Respondents' Understanding of Residence Terminology in Cognitive Research and in the Living Situation Survey," Proceedings, Annual Research Conference, Bureau of the Census.
Gerber, E. R., Wellens, T., and Keeley, C., (1996), "'Who Lives Here?: The Use of Vignettes in Household Roster Research," Proceedings, Am. Assoc. for Public Opinion Research.
Hainer, P., (1987), A Brief and Qualitative Study Exploring the Reasons for Census coverage Error among Low Income Black Households. Unpublished report, April 1987.
Hogan, H., (1992), "The 1990 Post-Enumeration Survey: Operations and New Estimates," Proc. of the Surv. Res. Meth. Sec., Am. Stat. Assoc.
Martin, E., (1996), "Household Attachment and Survey Coverage," Proceedings of the Survey Research Methods Section (American Statistical Association).
Robinson, J. G., Ahmed, B., Das Gupta, P., and Woodrow, K., (1993), "Estimation of Population Coverage in the 1990 United States Census Based on Demographic Analysis," Journal of the American Statistical Association 88: 1061-1079.
Schwede, L. and Ellis, Y., (1994), "Exploring Associations between Subjective and Objective Assessments of Household Membership: The Living Situation Survey." Proc. of the Surv. Res. Meth. Sec., Am. Stat. Assoc.
Sweet, E. M., (1994), "Roster Research Results from the Living Situation Survey," Proceedings, Annual Research Conference, Bureau of the Census.
Sweet, E. M., and Alberti, N. S., (1994), "Coverage Implications of Using the Term 'Usual Residence," DSSD 2000 Census Memorandum Series #H-17. Bureau of the Census.
Tourangeau, R., Shapiro, G., Kearney, A., and Ernst, L., (forthcoming), "Who Lives Here? Survey Undercoverage and Household Roster Questions," Journal of Official Statistics.