When one wants to collect more open-ended data from each subject, it may be appropriate to draw a smaller random subsample of a survey population for longer interviews designed to elicit information on a wide range of topics. A simple random sample or a stratified random sample may be used (assuming the appropriate demographic categories can be identified--for example, groups defined by race, age, family status, or those with children of particular ages) and can be interviewed in situ or in a central location. On the other hand, there may be situations for which it is helpful to select purposeful samples (that may or may not be selected randomly) for in-depth interviews. For example, among those leaving the welfare rolls, we may want to learn more about respondents who have never worked or who have not worked in many years. Pulling a subsample of this kind for an in-depth interview study can yield important insights. Of course, among respondents with literacy issues, using mail questionnaires is problematic anyways.
Studies of either kind can explore in some detail the experience "informants" are having in seeking a job, adjusting to employment, managing children's needs, coping with new expenses, finding transportation to work, relying on neighbors, and a host of other areas that may shed light on the TANF and post-TANF experience. As long as the subsample is representative, the researcher can extrapolate from it to the experience of the universe in the same way one would generalize from any representative group.
The advantage of the smaller subsample is that it solicits greater depth of knowledge on a larger number of subjects, yielding a more well-rounded perspective than is possible with only one or two open-ended questions. Such a methodology is appropriate when the study aims to understand the intricacies of subjective perspectives or the intertwined nature of family behavior when policy change impacts directly on one household member, but indirectly on other household members. Problems of this complexity can be understood only with a great deal of qualitative information.
The longitudinal study of the Milwaukee New Hope experiment is a good example of the value of this kind of research. New Hope provided low-income families in the experimental group with generous childcare, insurance supports, and earnings supplements to bring them above the poverty line to make it easier to remain in the labor force if they work at least 30 hours a week. Under the direction of Greg Duncan at Northwestern University and Tom Weisner at the University of California, Los Angeles (UCLA), the New Hope research team developed both a longitudinal panel survey and an embedded ethnographic study(1) that drew mainly on (1) repeated interviews with a representative sample of participants and controls, as well as "outliers" chosen because they appeared to deviate from patterns observable in their data and (2) classical fieldwork (discussed in a later section). From Duncan's perspective, the blending of "hard" and "soft" data has been critical in understanding program impacts:
New Hope's qualitative data proved indispensable for understanding the nature and meaning of program impacts. As simple as an experimental design may seem, analyses of experimental impacts are complicated by needs to quantify the key outcomes and isolate program impacts within important sample subgroups. Qualitative data are very helpful in both of these tasks.
One of the most important--and initially puzzling--impacts of the New Hope experiment was on teacher-reported improvements in the behavior of preadolescent boys, but not girls. Boys but not girls in the experimental group were 0.3 to 0.5 standard deviations better behaved and higher achieving than their control-group counterparts. Based on the survey data alone, however, we were unable to account for this gender difference.
Qualitative interviews suggested that interviewed mothers felt that gangs and other neighborhood pressures were much more threatening to their boys than girls. As a result, experimental group mothers channeled more of the program's resources (e.g. childcare subsidies for extended-day programs) to their boys than girls. Further quantitative analyses of both New Hope and national-sample survey data support this interpretation (Romich, 1999). It is unlikely that this important finding about family strategies in dangerous neighborhoods would have been discovered from the quantitative data alone (Greg Duncan, personal communication, 11/29/99).
The New Hope project also has provided useful analyses that separate the experiences of subgroups of participants who have responded differently to the same program opportunities. Because New Hope mirrors what some of the more generous states have tried to accomplish in their welfare-to-work programs, its experience is useful in parsing the differential impact of these supports for working families. As Duncan suggests in his comments on labor supply and earnings, without the qualitative component, it would have been harder to "unpack" the behavioral differences that distinguish subgroups:
It was clear from the beginning of our quantitative work that program effects on work and earnings were heterogeneous. Roughly one-third of the families attracted to New Hope were already working more than 30 hours and viewed the program's benefits as a way of making work and family demands more manageable. If anything, experimental/control differences in the labor supply of these families were negative. In contrast, families not working full time at the start viewed New Hope as a way of facilitating a transition to full-time work. On balance, experimental/control impacts on labor supply were positive for these families, although stronger in the first than second year of the program.
Qualitative interviews pointed to important heterogeneity among this latter set of families. Some, perhaps one-fifth, had multiple problems (e.g., drug dependence, children with severe behavior problems, relatives in ill health) that New Hope's package of benefits were not designed to address. Others had no such apparent problems and, in these cases, both experimental and control families could be expected to do well in Milwaukee's job-rich environment.
But a third group, who were only one or two barriers away from making it, profited the most from the New Hope package of benefits (Weisner et al., 1999). Program impacts on the labor supply of families with a small number of barriers were large, and larger in the second than the first year. This key set of findings simply would not have been discovered were it not for the qualitative work (ibid.).
A popular technique for exploratory research involves the use of focus groups, small gatherings of individuals selected for their demographic characteristics who engage in collective discussion following questions or prompts issued by a researcher acting as a facilitator. Focus groups operate in the native language of the participants and can last as long as 2 hours, providing an in-depth discussion of a topic. They can be used for a variety of purposes. Some researchers rely on focus groups as a means of generating questions they expect to ask in surveys. Others use focus groups as a primary means of data collection. Here the appeal usually lies in the modest expense involved: This is a "quick and dirty" method of gathering data on the subjective responses of program participants.(2) As a result, focus group studies can often be done on an ad hoc basis if they are not part of an initial evaluation design. A wide range of interested parties--from politicians to business firms--utilize focus groups as a means of "testing the market," particularly where public opinion is at issue.
Of course, the focus group approach has limitations. The contamination of opinion that occurs when individuals are exposed to the views of others can render the data hard to interpret. When particularly forceful individuals dominate the discussion, the views of more passive participants can be easily squelched or brought into conformity in ways that distort their true reactions. Some people understandably are hesitant to air their opinions on sensitive subjects (e.g. domestic violence, employer misbehavior, criminal behavior) in these types of settings.
Moreover, it is hard to make focus groups representative of a population in any meaningful sense. They must therefore be used purposively or with caution. Focus groups are not a good tool for producing data that will withstand scrutiny for representativeness. What they do provide is a relatively inexpensive and rapid means of learning about underlying attitudes and reactions, an approach that may be informative for officials or scholars looking to design more nuanced research instruments. They are often used as an exploratory tool to help design survey or interview studies because they help to expose important problems that should be subjected to more systematic study. These are important goals for researchers. For program administrators looking for ways to give their staff members insight into the lives of those they may see only in "numerical form," focus groups can be a means of putting a human face on administrative records.
Some of the limitations of focus groups can be addressed to a modest degree through the careful selection of focus group members. Sensitive subjects may best be addressed by drawing together people who are as similar as possible, who have experienced a common dilemma, in the hopes that the similarities between them will lessen any discomfort. Hence investigators often construct focus groups along the lines of racial or ethnic groups, gender or age groups, or neighborhood groups. The "contamination" of forceful individuals can be limited by the guiding hand of a highly skilled facilitator who makes sure that others have a chance to participate. However, none of these approaches eliminates the difficulties inherent in public discussions of this kind.
Focus groups are therefore probably best used to gather data on community experience with and opinions toward public assistance programs rather than to gather systematic data on individual perspectives. For example, the problems associated with enrollment in children's health insurance systems probably could be well understood by convening focus groups. Indeed, one of the strengths of the method is that it prompts individuals who may not be able to express themselves easily in a one-on-one setting to recall and describe difficulties they have encountered. Information of this kind is far more textured and complete than fixed-choice questionnaires and can help public officials to address the deficiencies in outreach programs, for example.
Qualitative Longitudinal Studies
Welfare reform is a process unfolding over a number of years, where the before and the after may be widely separated and the "in between" states of at least as much interest as the ultimate outcomes. We have good reason to believe that families pass through stages of adaptation as their children age, new members arrive, people marry, jobs are won and lost, and the hold of new requirements (work hours, mandated job searches) exert their influences. For this reason, it will be critical that at least some of the nation's implementation research follow individuals and families over a period of years, rather than rest easy with cross-sectional studies. Indeed, one need only look at how the Panel Study of Income Dynamics, the National Longitudinal Study of Youth, or the Survey of Income and Program Participation have altered and enhanced our understanding of income over the lifespan or movements in and out of poverty over time to recognize the value of panel studies of this kind.
These longitudinal studies contain very little qualitative data. The number of sample members and broad coverage of information is expensive so that cost containment often means depth has been sacrificed in favor of coverage. However, anthropologists and sociologists have developed longitudinal interview studies in which the same participants are interviewed in an open-ended fashion at intervals over a long course of time. I have two studies in the field at the moment--one on the long-range careers of workers who entered the labor market in minimum-wage jobs in poor neighborhoods and the other on a sample of working poor families, intended to assess the impact of welfare reform on those who were not the targets of policy change--that utilize this approach. In each case, representative samples of approximately 100 subjects were drawn from larger samples of subjects who completed face-to-face surveys. Thereafter, the smaller subsamples were interviewed at 3-to 4-year intervals, for a total of 6-to 8-years' worth of data collection. Here it has proven possible to capture changes in perceptions of opportunity, detailed accounts of changing household composition, the interaction between children's lives and parents' lives, and the impact of neighborhood change on the fate of individual families. Although the samples are very small by the standards of survey research, the depth and nuance of the data that emerges from such an approach are of great value in opening the "black box" that may resist interpretation in studies based solely on administrative records or fixed-choice instruments.
Qualitative panel studies are, however, labor intensive and expensive for the number of respondents they generate. They ask a great deal from participants who typically have to give up several hours of their time for each wave. Given these high demands, providing honoraria of $50-100 to ensure participation in interviews is generally important to generate adequate response rates. Such generous honoraria would bankrupt a larger study. Longitudinal interview studies are typically done via the use of tape-recorded interviews, which must be transcribed and possibly translated. Given the nature of the data that studies of this kind are seeking, it is often helpful to employ interviewers who are matched by age, race, gender, and class. This process is not simple. For example, I have developed research teams that were closely matched along race and gender lines, only to discover that vast class differences became quite apparent between respondents who were poor and living in rundown neighborhoods and students who are clearly middle class in origin and living in far better circumstances.(3) Indeed class was often at least as important as race in making a match. The gulf between a professor in her forties and informants in their twenties can be quite substantial just because of the different worlds they inhabit because of their ages.
Not all studies attempt the matching process, and the question of whether it is necessary to find counterparts who are sociologically similar is controversial. For example, Edin and Lein, both white professional women, have done exquisite interview work with women of color on welfare and in low-wage jobs. Other white researchers (myself and some members of my research teams) have had good success despite racial differences. Indeed, it is sometimes easier for informants to reveal sensitive information to outsiders who are perceived as less likely to "spread their personal business" around town (Kathryn Edin, personal communication).
My experience has shown that long-term relationships are easier to develop when racial barriers are minimized and a comfort zone is reached based on perceived similarities.(4) It is imperative to have staff fluent in the languages of the subjects. It is even more important to invest in training the members of a research team: All the matching in the world will not make up for lack of training, and one should never assume that sharing skin color or gender is sufficient. These requirements add to the costs involved in research.
The quality of the data obtained through well-designed and well-executed qualitative, longitudinal studies can make them well worth the effort. This may be particularly true when one wants to go beyond a scholarly or policy audience to engage either the public or political figures in the exploration of welfare reform. Illustrating statistical trends with "real-life" examples of the dilemmas and success stories of former welfare recipients is of great value in this regard. Researchers should not cede to journalists the entire responsibility for telling the story of welfare reform "with a human face" because reporters rarely select their informants systematically and there is no guarantee that their accounts will be anything more than anecdotal.
Qualitative panel studies can be developed with an original sampling strategy that picks up a representative population based on neighborhood residence or participants and matched controls who participate in a social service program. However, they are probably most valuable when they are embedded in panel studies using a survey design and are therefore subsets of the much larger population of survey respondents that can serve as a better basis for statistical analysis.
This embedding strategy has one disadvantage: If the underlying survey is part of a longitudinal panel study, the selection of a subsample that will be accorded more attention may bias the responses of this group to succeeding waves of the survey. Researchers need to evaluate this possibility, though it need not be a serious flaw. Most surveys seeking to track the consequences of welfare reform are going to focus on "objective" and measurable outcomes: hours worked, income earned, jobs acquired, jobs lost, health insurance enrollment, and so forth. Qualitative studies may yield additional information on how these states of being were reached (job search strategies, barriers to insurance enrollment), but in most instances will not compromise the underlying information in a negative (concealing) direction. The experience of providing more information through open-ended interviews may, in fact, encourage greater revelation among the participants in the qualitative study. Researchers will want to check for any systematic biases that may be emerging and, for some purposes, exclude the subsample from statistical analyses of the survey population.
However, I would argue that the advantages of selecting the qualitative sample from an original panel population far outweigh the disadvantages. "Soft" studies of this kind are often suspect on grounds of representativeness and the value of their contribution dismissed as a result. Although one could, in theory, recruit participants in a qualitative study who are similar to those in the survey population, it is always possible that these "add ons" differ enough from the participants to raise doubt. Hence, in my view, it is a safer bet to draw the qualitative sample from the original research universe and risk the chances that their involvement may adversely alter their responses to a longitudinal study. (Obviously this is not a problem if the underlying survey is cross-sectional.)
"01.pdf" (pdf, 472.92Kb)
"02.pdf" (pdf, 395.41Kb)
"03.pdf" (pdf, 379.04Kb)
"04.pdf" (pdf, 381.73Kb)
"05.pdf" (pdf, 393.7Kb)
"06.pdf" (pdf, 415.3Kb)
"07.pdf" (pdf, 375.49Kb)
"08.pdf" (pdf, 475.21Kb)
"09.pdf" (pdf, 425.17Kb)
"10.pdf" (pdf, 424.33Kb)
"11.pdf" (pdf, 392.39Kb)
"12.pdf" (pdf, 386.39Kb)
"13.pdf" (pdf, 449.86Kb)