The data derived from interview and participant observation projects can be used in at least three ways: (1) to generate hypotheses that might be turned into survey research questions; (2) to complement research based on large-sample statistical analyses; or (3) as an end in and of themselves. These three aims are not mutually exclusive. The difficulty, of course, with the complementary research and "end in itself" approach is that questions of representativeness are always vexing with very small samples and for most research in this genre, small samples are the only affordable possibility.
My own approach has involved embedding the selection of informants within a larger survey design in order to respond to this concern. In 1995-96, we undertook a survey of 900 middle-aged African Americans, Dominicans and Puerto Ricans in New York City. They were chosen to be representative of ethnically diverse and ethnically segregated neighborhoods, with both high and low levels of household income. From this population, a random subsample of 100 respondents was chosen for in-depth interviews at 3-year intervals (1998 and again in 2001). Finally, 12 individuals--4 from each of the ethnic groups of central concern--living in the three neighborhoods described in the previous section were selected from this qualitative subsample. The choice of these particular 12 people was guided mainly by their employment status and family type, with a mix of single parents and intact couples. This nested design has enabled us to generalize from the families we have come to know best to the population as a whole with which we began.
A similar approach has been pursued by the Manpower Demonstration Research Corporation's "Urban Change" project, a study of the impact of devolution and the time limits of the TANF system on poor families in four cities: Philadelphia, Cleveland, Miami, and Los Angeles. A multidisciplinary team of social scientists are drawing on "administrative records; cross-sectional surveys of food stamp recipients; census tract-level neighborhood indicators; repeated interviews with Executive Directors of community-based social service organizations; repeated ethnographic interviews with welfare-reliant women in selected neighborhoods; and repeated interviews with and observations of welfare officials and line staff" (Edin and Lein, 1999:6).(5)
The qualitative interview part of the Urban Change project has been following 80 families from high- and medium-poverty neighborhoods in Cleveland and Philadelphia. Under the direction of Edin at the University of Pennsylvania, this project has thus far collected a large amount of baseline information on a series of topics including:
Aspirations for [women's lives] and their children; experiences with case workers and the welfare system; knowledge about and attitudes toward welfare reform; income and expenditure patterns; educational and work experiences; family life; attitudes toward marriage and future childbearing; health and caregiving; social support; material hardship; use of social service agencies; and perceptions of the quality of their neighborhoods (Edin and Lein, 1999:6).
Families were chosen for this part of the study by selecting three neighborhoods 6 in each city with moderate to high concentrations of poverty (more than 30 percent living below the poverty line) and welfare receipt (20 percent or more of families receiving welfare). Ten to 15 families were recruited in each neighborhood by posting notices in the target neighborhoods, knocking on doors, and requesting referrals from community leaders and local institutions. They attempted to guard against the overrepresentation of any given social network by utilizing no more than two recruits through any of these sources. This strategy avoided the liabilities of drawing from lists provided by TANF offices (which would necessarily skew the research toward welfare recipients alone). The strategy also allowed the researchers to present a truly independent face to their informants, untainted by connection to enforcement agencies that could affect their cash benefits.
A strategy of this kind probably overrepresents people who are higher on social capital than some of their more isolated counterparts. They have connections. A strict sampling design from an established list may pick up people who are less "hooked in" to institutional resources or private safety nets and will therefore tell us something about people who confront welfare reform from a socially isolated vantage point as well as those who are more connected. However, the liabilities of this approach are considerable, for it is much harder to disassociate from official agencies when pursuing a sample generated randomly from, for example, a TANF office caseload.
The neighborhood strategy employed by the Urban Change project ensures that the qualitative study includes white, black, and Latino families who are particularly disadvantaged. As Edin and Lein (1999:7) have explained, the design will not pick up welfare recipients who live in mixed-income or more affluent neighborhoods. It is possible that this strategy yields a slightly more pessimistic perspective on the consequences of welfare reform as compared with what we would have seen had the study included the entire range of long term-recipients, many of whom moved off of the rolls with apparent ease as unemployment declined. These are the people whose human capital, including prior work experience, made them relatively easy to place. The Urban Change project will tell us how this transition affected those with less going for them, because their neighborhoods (and the contacts they derive from them) are less likely to provide useful information for job hunting. The communities selected as the focus neighborhoods undoubtedly present safety concerns that mothers will have to consider as they scramble to figure out how to care for their children. In the end, these are the more pressing questions in need of answers, hence the wisdom of the Urban Change project's approach.
Urban Change is not an ethnographic project in the strict sense of the term. Contact is maintained intermittently with the target families, often utilizing telephone interviews in place of face-to-face contact. Intervals of contact are approximately 6 weeks, though this varies by the informants' situation. Nonetheless, it will provide a very rich database, spanning the before and after of the imposition of time limits, that will tell us an enormous amount about the challenges women and their families have faced in transitioning from public assistance to the world of work. The size and ethnic diversity of the sample (including poor whites, often overlooked in studies of the poor), the multicity approach, and the fusion of administrative records, expert perspectives, and the inclusion of welfare-reliant families in communities with varying levels of poverty will help to address many of the more important theoretical questions before us, especially the consequences of race and ethnic differences, neighborhood effects, and human capital differences in the unfolding of welfare reform.
Angel, Burton, Chase-Landsdale, Cherlin, Moffitt and Wilson are in the midst of a similar study of welfare reform and its consequences, the Three-City Study. This project involves a survey, which began in 1999, of 2,800 households from poor and moderate income. The sample is divided between TANF recipients and those who do not receive these benefits. It is restricted to households with young children (younger than age 4) and those with children between 4 and 14. A developmental study of 800 of these families who have children ages 2-4 will be embedded in this larger design. This embedded study will include interviews with caretakers and the fathers of these children.
The Three-City Study also has an ethnographic component directed by Burton. The study will follow 170 families to track how welfare policies affect the daily lives and neighborhood resources of poor families. In-depth interviews will be conducted over the course of 2 years and will cover topics such as the respondent's life history and daily routines. This component also includes diary studies and observations of the participant when she goes to social service offices for assistance. (Winston et al., 1999). The great advantage of the three-city study is the way in which the ethnographic sample is nested inside a larger, more representative survey sample and contextual data set that can analyze neighborhood variables, state- and local-level employment data, and the repeated interviews and family assessments in the child development portion of the project.
This project has an enormous budget and is therefore the "Cadillac" model that few other studies of welfare reform will be able to match. Nonetheless, it is theoretically possible to use a rich fieldwork approach as long as the resources for this labor-intensive form of data gathering are available. Few social scientists would disagree that moving from macrolevel findings based on surveys to the most microlevel data drawn from fieldwork, with mid-range interviews and focus groups in between, is the best possible approach for preserving representativeness but building in the richness of qualitative research.
Few research projects will be able to match the scale of the Urban Change and the Three-city projects. Indeed, even my own more modest study of 100 families in one city required a substantial research budget and a rotating team of fieldworkers willing to commit a total of more than 6 years to the enterprise. Of course, not all studies of welfare reform need to be as long in duration as the ones described here. For state and local officials whose aim is less to explore the theoretical questions that motivated these studies and more to learn in depth about the family management problems of their caseloads, it may be possible to arrange with local universities to organize neighborhood-based research projects that will provide "snapshot" versions of the same kinds of questions.
Another sampling strategy involves the use of "snowball" samples that attempt to capture respondents who share particular characteristics (e.g., low-wage workers or welfare-reliant household heads) by asking those who meet the eligibility criteria to suggest friends or neighbors who do as well. Some classic studies in the annals of poverty research have used snowball samples to great effect (e.g., Lillian Rubin's Worlds of Pain , Elliot Liebow's Tally's Corner ). More recently, Edin and Lein's Making Ends Meet relies on referrals from a variety of sources, including the personal contacts of individuals already in their study population, to build a sample in four cities. The defining feature of a snowball sample is that it gathers individuals into a sample that have some acquaintance with those who are already involved. Multiple snowball techniques seek to maximize the heterogeneity of the sample, while single snowballs maximize the homogeneity of the sample. Neither approach results in a sample that is genuinely random, though the former seeks diversity while the latter explicitly seeks purposive groups.
Snowballs can be bound tightly to a particular network, as was the case in Tally's Corner, or can guard against the possibility that membership will not represent truly independent cases. When the object of study is densely connected webs of friends and relatives, it is important to capture naturally occurring social networks. In this case, the initial selection of the key informant needs to pay attention to representativeness. Thereafter, however, there will be nothing random about the study participants: They will be selected members of the original informant's trusted associates.
For example, in my recent study of the working poor in central Harlem (Newman, 1999), a representative sample of workers in fast food restaurants formed the core of the research, but a selected subsample was central to a final phase of intensive participant observation that focused on the survival strategies of 10 households and the social networks attached to them. The ten key informants were selected to represent the racial and gender diversity of the universe of workers. Branching out from there, in concentric circles around the 10 key informants, we took in the friends, neighbors, schoolmates, teachers, preachers, distant relatives, and street contacts of these individuals. Hence, although the original subsample was representative, the snowballs grew around them because the purpose of the study was to learn about how these households managed the many challenges of low-wage work in naturally occuring contexts (school, home, church, extended family, etc.). Ultimately, perhaps as many as 500 additional people were included in this phase of the research, though they were hardly a random sample.
Others have used snowballs to generate the "master sample." However in this situation it is important to guard against the possibility that network membership is biasing the independence of each case. Some snowball samples are assembled by using no more than one or two referrals from any given source, for example. Edin and Lein's (1997), Making Ends Meet is a good example of a partial snowball strategy that has made independence of cases a high priority. Initially, they turned to neighborhood block groups, housing authority residents' councils, churches, community organizations and local charities to find mothers who were welfare reliant or working in the low-wage labor markets in Boston, Chicago, Charleston, and San Antonio. Concerned that they might miss people who were disconnected from organizations like those who served as their initial sources, Edin and Lein turned to their informants and tried to diversify:
To guard against interviewing only those mothers who were well connected to community leaders, organizations and charities, we asked the mothers we interviewed to refer us to one or two friends whom they thought we would not be able to contact through other channels. In this way, we were able to get less-connected mothers. All in all we were able to tap into over fifty independent networks in each of the four cities (1997:12).
Using this approach, Edin and Lein put together a heterogeneous set of prospective respondents who were highly cooperative. Given how difficult it can be to persuade poor people who are often suspicious of researchers' motives (all the more so if they are perceived as working for enforcement agencies), working through social networks often can be the only way to gain access to a sample at all. Edin and Lein report a 90 percent response rate using this kind of snowball technique. Because this rate is higher than one usually expects, there may be less independence among the cases than would be ideal under random sample conditions, but this approach is far preferable to one that is more random but with very low response rates.
Sample retention is important for all panel studies, perhaps even more so for qualitative studies that begin with modest numbers. Experience suggests that studies that couple intensive interviews with participant observation tend to have the greatest success with retention because the ethnographers are "on the scene," and therefore have greater credibility in the neighborhoods from which the interview samples may be drawn. Their frequent presence encourages a sense of affiliation and participatory spirit into studies that otherwise might become a burden. However, my experience has shown that honoraria make a huge difference in sample retention when the subjects are poor families. I have typically offered honoraria of $25)$100, depending on the amount of time these interviewers require. Amounts of this kind would be prohibitive for studies involving thousands of respondents, but have proven manageable in studies of 100, tracked over time. The honoraria demonstrate respect for the time respondents give to the study.
Though design features make a difference, retention is a problem in all studies that focus on the poor, particularly those that aim at poor youth. The age range 16-25 is particularly complex because residential patterns are often unstable and connections between young adults and their parents often fray or become less intense. Maintaining contact with parents, guardians, or older relatives in any study dealing with poor youth is important because these are the people who are most likely to "stay put" and who have the best chance of remaining effective intermediaries with the targets of these longitudinal studies. Retention problems are exacerbated in all studies of the poor because of geographic mobility. One can expect to lose a good 25-40 percent of the respondents in studies that extend over a 5-year period. This may compromise the validity of the results, though it has been my experience that the losses are across the board where measurable characteristics are concerned. Hence one can make a reasonable claim to continued representativeness. Such claims will be disputed by those who think unmeasured characteristics are important and that a response rate of 60-75 percent is too low to use.