Revisiting the Contact Hypothesis: The Case of Public Exposure to Homelessness

by Barrett A. Lee, Chad R. Farrell, Bruce G. Link
Revisiting the Contact Hypothesis: The Case of Public Exposure to Homelessness
Barrett A. Lee, Chad R. Farrell, Bruce G. Link
American Sociological Review
Start Page: 
End Page: 
Select license: 
Select License

Revisiting the Contact Hypothesis:
The Case of Public Exposure to Homelessness

Barrett A. Lee Chad R. Farrell

The Pennsylvania State University The Pennsylvania State University

Bruce G. Link

Columbia UniversiQ

Using data from a national survey ofpublic attitudes toward homeless people, this paper

evaluates the applicability of the contact hypothesis to in-group/out-group relations that fail to meet the optimal conditions speczfied in the contact literature. Past efforts are

extended by (I) moving beyond face-to-face encounters to consider multiple types of in- group exposure to a highly stigmatized out-group, (2) examining a variety of attitudinal

outcomes, and (3) incorporating community context as a possible antecedent of such

outcomes. Even after taking selection and social desirability processes into account, all

types of exposure are found to affect public attitudes in the predicted Cfavorable)

direction. Moreover, the size of the local homeless population-our primary measure of

context-shapes opportunities for most forms of exposure and thus influences attitudes

indirectly. These findings suggest that the scope of the contact hypothesis needs to be

widened rather than narrowed.

ore than a half-century after its initial appearance in an agenda-setting report by Williams (1947), the contact hypothesis con- tinues to stimulate interest across the social sci- ences (for reviews, see Jackson 1993; Pettigrew 1986, 1998; Stephan 1987). Part of the appeal of the hypothesis lies in its simplicity: contact between members of an in-group and an out- group is expected to improve the attitudes of the former toward the latter by replacing in-group ignorance with first-hand knowledge that dis-

Direct correspondence to Barrett A. Lee, Department of Sociology, The Pennsylvania State University, 2 1 1 Oswald Tower, University Park, PA 16802 ( This research has been sup- ported by National Institute of Mental Health Grant MH-46 10 1 (on which the third author served as prin- cipal investigator) and by the Population Research Institute of The Pennsylvania State University, which receives core funding from National Institute of Child Health and Human Development Grant 1-R24- HD4 1025. The authors thank Leon Anderson, Glenn Firebaugh, Thomas Pettigrew, Marylee Taylor, Susan Welch, and the anonymous ASR referees for their helpful comments.

confirms stereotypes. Yet few contemporary scholars accept so simple a version of the hypothesis. In his landmark volume The Nature of Prejudice, Allport (1954) identified several conditions-status equality, common goals, institutional support-believed to enhance the beneficial effects of contact. Since then, addi- tional caveats and qualifications have so nar- rowed the contact hypothesis that it is thought by some to be valid only under optimal cir- cumstances (Amir 1969, 1976; Cook 1970; Stephan and Stephan 1996). This restricted scope, coupled with the shortcomings men- tioned below, has prompted calls for a fresh start. Indeed, a number of models or theories of intergroup relations, posed as alternatives to the hypothesis, are now available (Blumer 1958; Brewer and Miller 1984; Brown and Turner 198 1; Forbes 1997; Hewstone and Brown 1986).

Despite reservations about the contact hypothesis, the empirical evidence supporting it is impressive. Historically, racial and ethnic minorities have been the out-group of choice in research that affirms the hypothesis (Welch et al. 200 1 ;Williams 1964), but recent work doc- uments the positive effects of contact on in- group attitudes toward homosexuals (Herek and Capitanio 1996), older people (Caspi 1984), the mentally ill (Desforges et al. 1991; Link and Cullen 1986), persons with acquired immune deficiency syndrome (AIDS) (Werth and Lord 1992), and persons with physical disabilities (Makas 1993; Yuker 1988). Meta- analyses by Pettigrew and Tropp (2000,2002) show that the effects of contact are robust, occur in the absence of optimal conditions, and achieve greatest magnitude in the most rigorous studies. The meta-analyses also find support for the hypothesis irrespective of mode of data collection (survey, quasi-exper- iment, observation, etc.). These results indi- cate that the contact hypothesis has sufficient merit to warrant reevaluation rather than pre- mature dismissal.

Our research addresses three areas in which the literature on the contact hypothesis is deficient. The first deficiency concerns the types of contact possible between in-group and out-group. Although past research has heavily emphasized face-to-face interaction, contact can take other forms, such as obser- vation of the out-group in public settings. It may also occur vicariously, through media coverage of the out-group or conversations with friends and relatives. A broader con- ception of contact should be complemented by attention to a second problem: the limited range of independent variables considered. Most investigations proceed at the individual level only, examining the influence of in- group members' characteristics and contact experiences on their attitudes toward the out- group. This approach overlooks the possibil- ity that features of the larger context enhance or constrain opportunities for contact and hence, may shape attitudes indirectly. For example, the number of out-group members in a city-or practices that confine them to certain sections of the city (e.g., housing dis- crimination)-could determine the likelihood of residents' encounters with the out-group. Finally, the consequences of contact need fuller exploration. The traditional focus on race relations in the relevant literature means that prejudice and social distance measures are overrepresented to the neglect of other kinds of attitudes. Whether all attitudes are affected similarly by contact, and whether they generalize to all members of the out- group (as opposed to just those members with


whom one has interacted), remain open ques- tions.

We choose the homeless population as a strategic out-group with which to revisit the contact hypothesis. Like many out-groups, homeless people are more visible than they once were due in part to a growth in numbers and in part to extensive media and policy treat- ments of their plight. Thus most Americans are now at least superficially familiar with the homelessness issue. Do the sources of this familiarity-either direct or vicarious-influ- ence domiciled people's opinions of home- less individuals? Null or negative findings would not be surprising since public encoun- ters with the homeless rarely satisfy the opti- mal criteria set forth in the restricted version of the contact hypothesis. Conversely, any positive effects detected would support a sim- pler, less conditional version of the hypothe- sis, implying that its scope needs widening rather than narrowing. Mixed outcomes may result if particular kinds of contact register stronger impacts than others or if attitudes vary in their responsiveness to contact. Whatever the final pattern of results, the con- clusions we draw from them should have sig- nificant implications for the value of the contact hypothesis in general, beyond its appli- cability to attitudes toward homelessness.

The research reported here uses data from a 1990 national survey to examine how con- tact between domiciled persons and homeless persons affects the former's views of the lat- ter. The background section of the paper, which discusses the out-group status of home- less people and the various reactions that this status elicits, provides the rationale for our study. We then argue for reconceptualizing contact in terms of exposure, a move intend- ed to overcome some of the limitations asso- ciated with the older construct. We also propose that an expanded menu of antecedents (including features of community context) and outcomes (different types of attitudes) be considered. Subsequent sections describe the dataset used and present results from a mul- tivariate analysis that tests the contact hypoth- esis with increasingly refined measures of exposure. Finally, our study investigates the possibility of selection-that attitudes influ- ence exposure rather than vice versa-and the role that contextual factors play.



What homeless people share with the out-groups represented in the contact literature is that all have been labeled or marked based on one or more attributes judged undesirable by in-group members. When the label is strong enough, it conforms to Gofian's (1963) notion of stigma (see Jones et al. 1984; Link and Phelan 200 1). Persons belonging to stigmatized groups suffer from a spoiled identity that encourages their devaluation and rejection by "normal" others. They are also more likely to be stereotyped, to be perceived in categorical fashion rather than as individuals. On this point, social psycholog- ical studies suggest that the attention attracted by people who are eye-catching or unusual tends to accumulate over time, fostering the devel- opment of stereotypes about their group (Tajfel 1982).

The homeless population clearly has the potential to be highly stigmatized. National and local surveys of homeless samples yield a dis- proportionate prevalence of poor physical and mental health, substance abuse, criminal involvement (as both victim and perpetrator), weak or absent family ties, and extreme pover- ty (Burt, Aron, and Lee 2001 ;Momeni 1990; Rossi 1989). Homeless persons are often marked in multiple ways, exhibiting ADM (alco- hol, drugs, mental illness) difficulties or other constellations of problems. The fact that some of these problems take a visible form (via appearance or behavior) could further intensi- fy the stigma associated with homelessness. Because personal difficulties are hard to conceal while living on the street, they serve as constant reminders of the out-group status of homeless people and thus reinforce stereotypes. Put dif- ferently, the very conspicuousness of the out- group may lead to oversimplified views that resist contrary evidence (Allport 1954; Rothbart and John 1993)

Several types of findings shed light on the extent to which the public stigmatizes and stereotypes homeless individuals. Ethnographic investigations document the degradation rituals endured by such individuals, who are routine- ly avoided or treated as non-persons by domi- ciled passersby (Anderson, Snow, and Cress 1994; Lankenau 1999a, 1999b). Similarly, the attitudes of domiciled subjects toward hypo- thetical cases in a vignette experiment have led Phelan et al. (1997:334) to conclude that "home- lessness is stigmatized more severely than pover- ty and, generally, more severely than mental illness." The substantial percentages of survey respondents blaming homeless people for being homeless and attributing deviant properties (substance abuse, mental illness, dangerous- ness, etc.) to them would seem to confirm the public's negative view of the homeless (Lee, Hinze Jones, and Lewis 1990; Lee, Lewis, and Hinze Jones 1992; Lee, Link, and Toro 1991 ; Link et al. 1995; Toro and McDonnell 1992).

We should stress, however, that the empirical record is not entirely one-sided. In the same surveys cited above, structural beliefs about the causes of homelessness are more common than individualistic ones. Many respondents remain emotionally engaged, expressing sadness or anger regarding the persistence of the problem and backing government efforts to solve it. In short, public attitudes can best be described as heterogeneous: some are quite unfavorable, con- sistent with the stigmatization of homelessness, but others appear more sympathetic.

Our mission is to determine how contact with homeless people affects the attitudes of in-group (domiciled) members. The very definition of contact complicates this task; sometimes con- tact is taken to mean little more than proximi- ty to an out-group or the potential for acquaintance implied by group proportions in a population (Amir 1969; Forbes 1997; Makas

1993). Because of such conceptual ambiguity, the contact literature conveys contradictory messages concerning the types of contact that have the greatest influence on attitudes. According to the simple version of the contact hypothesis, virtually any kind of interaction seems to suffice. However, proponents of the restricted version argue that positive impacts are most likely when the interaction is face to face, voluntary, sustained, intimate, cooperative, and mutually rewarding (Allport 1954; Cook 1978; Stephan 1987). The actors should also enjoy equal status in the situation and their contact should be institutionally supported.

Few of these conditions apply to interactions between domiciled and homeless individuals. Consider the panhandling encounter, a com- mon form of contact in large cities (Lee and Farrell2003). Asking for money or food high- lights the status differences rather than simi- larities between solicitor and donor. The act itself tends to be brief, anonymous, unidirec- tional (passersby rarely respond), and anxiety provoking for both parties (Lankenau 1999a, 1999b; Snow and Anderson 1993). From the perspective of an in-group member, panhan- dling is often a nuisance at best and a threat to one's safety at worst.' Institutional authorities (municipal government officials, business lead- ers, and the like) take a dim view of the behav- ior as well, formulating ordinances that seek to limit or ban panhandling (National Law Center on Homelessness and Poverty 1996; Simon 1996). In short, this kind of public-homeless contact may make in-group attitudes less sym- pathetic, not more so.

Being panhandled is far from universal and significantly depends on how often a domiciled person enters the urban public spaces where panhandlers congregate. When members of an out-group are unavailable for direct contact, other sources can shape in-group views of the out-group. Recognition of such sources has led some scholars to suggest that the concept of con- tact should be expanded (Pettigrew 1986; Stephan 1987). The term "exposure" strikes us as particularly useful for this purpose. Exposure overcomes the face-to-face implication of con- tact, encompassing a number of vicarious pos- sibilities. It may also refer, at the opposite extreme, to "insider" knowledge of the out-group. We switch the focus from contact to exposure to promote a more differentiated per- spective on how an in-group can learn about an out-group.

At least four categories of exposure are worth examining: information, observation, interac- tion, and membership. First, information from third-party (non-out-group) sources can influ- ence impressions of an out-group. Examples of information exposure to the out-group include media coverage, lectures or presenta-

We recognize that some urban residents have regular and pleasant exchanges with panhandlers whom they "adopt." Such patron relationships can be mutually beneficial, offering solicitors enhanced sta- tus and a steady income and providing donors with psychic gratification (Goldstein 1993; Lankenau 1999b). They are, however, the exception rather than the rule; problematic panhandling encounters far outnumber rewarding ones.


tions, and informal conversations. Second, the in-group's direct observation of the out-group in the course of everyday life can provide famil- iarity with the out-group. The frequency of see- ing out-group members, the setting in which such members are seen, and their behavior all constitute relevant dimensions of observation. The third category of exposure, interaction, approximates the conventional face-to-face meaning of contact. Finally, membership in the out-group carries exposure to its logical extreme. Has an individual ever belonged to the out-group or had a family member or friend who did? As a set, these four types of exposure reflect gradations of increasing depth and inten- sity. Consequently, it seems reasonable to expect the impact of exposure on in-group attitudes to become stronger toward the membership end of the scale.

Existing research tells us little about the prevalence of different types of exposure to homelessness. A survey of Buffalo, New York residents conducted by Toro and McDonnell (1992) reveals mixed levels of exposure through interaction and membership. More than 80 per- cent of the Buffalo respondents had few or no direct contacts with homeless people, but 12 per- cent said they knew a homeless person and 4 percent reported ever being homeless them- selves. Results from a Nashville, Tennessee sur- vey suggest substantial public exposure to homelessness in that community, although the measures used were different than in Buffalo (Lee et al. 1990). A majority of the respondents (76 percent) in the Nashville survey had dis- cussed the issue with friends or family during the past year (the information dimension), and significant proportions had interacted directly with homeless individuals, either in a panhan- dling encounter (68 percent) or in a conversation about something other than money (25 percent). Observation of homeless people in downtown public spaces (outdoors, in stores, restaurants, office buildings, parks, etc.) was also common.

Exposure is not the sole influence on in-group attitudes toward an out-group. Tests of the con- tact hypothesis regularly examine the effects of in-group members' social and demographic characteristics, with results contingent on the


out-group in question. In the case of attitudes about homelessness, national and local surveys indicate that women, younger people, the well educated, and political liberals tend to hold more favorable or sympathetic views of home- less persons than do their male, older, less edu- cated, and conservative counterparts (Lee et al. 1990, 199 1, 1992; Phelan et al. 1995; Toro and McDonnell 1992). The direct effects of these characteristics are stressed, though indirect impacts seem plausible. For example, while lib- eralism may sensitize one to the circumstances of homeless persons, it could also encourage volunteering in a shelter or seeking other forms of exposure that increase opportunities for observation and interaction.

Where both the contact and homelessness literatures fall short is in explaining how con- text shapes such opportunities. Proponents and critics of the contact hypothesis alike emphasize the importance of taking macrostructural fea- tures of a society (e.g., norms, institutional arrangements) into account (Hewstone and Brown 1986; Pettigrew 1986, 1998). Thus far, their call has gone unheeded. Yet it seems obvi- ous that structural factors-especially as man- ifested in the local arena-have the potential to affect in-grouplout-group relations. Perhaps the best illustration of the multifaceted influence of context comes from the body of research on contact between whites and African Americans. From pioneering studies of integrated housing (Deutsch and Collins 195 1 ; Wilner, Walkley, and Cook 1955) and comparative surveys (Williams 1964) to more recent work (Jackman and Crane 1986; Sigelman and Welch 1993; Welch et al. 2001), the guiding principle has been that society-wide patterns of racism and discrimination are key sources of residential propinquity (or lack thereof), which in turn determines the likelihood of contact between whites and Afncan Americans.

Our approach to context acknowledges the national forces producing homelessness but assumes that the centrality of homelessness as a public concern differs by community. On aver- age, the attention paid to the issue-both ameliorative and hostile-is probably greater in large cities than in smaller areas. The importance of community size is reflected in levels of media coverage, availability of services, and involve- ment of municipal government (Berman and West 1997; Burt et al. 2001; Lee et al. 1991), which all have implications for the exposure of domiciled individuals to homeless people. For some types of exposure, demographic aspects of the homelessness problem may be especial- ly relevant. The bigger the local homeless pop- ulation, for example, or the higher the rate of homelessness, the more chances should exist for in-group observation of or interaction with members of the out-group. The magnitude of the problem could also inspire exposure-related policies (e.g., anti-panhandling ordinances, the creation of safe zones) designed to minimize the public's contact with the homeless population.

A satisfactory analysis of attitudes toward homeless people-or toward any out-group- requires that context be taken into account explicitly. In the case of homelessness, we believe that the local context should affect in- group attitudes indirectly, by increasing or decreasing opportunities for exposure. Direct impacts would be more difficult to interpret; the size of the general or homeless population in a community would somehow have to predispose in-group members to see the homeless out- group in a positive or negative light, with no intervening processes operating. Of course, dif- ferent persons may react differently to the same context; that is, their attitudes could be jointly influenced by their social and demographic characteristicsandby features of the local envi- ronment where they reside. Our research con- siders the possibility of such interaction effects as well as the alternatives (direct and indirect effects) mentioned above.

Because many contact studies have targeted specific instances of intergroup tension or con- flict, particular members of the out-group are often identified as the attitude objects. This approach leaves unresolved the question of gen- eralization: whether contact with such a mem- ber changes one's perspective on the group as a whole. Reviews of the contact literature sug- gest that attitude generalization requires repeat- ed exposure to various persons who are judged typical of the out-group but who challenge pre- conceptions of what out-group members are like (Hewstone and Brown 1986; Patchen 1995; Pettigrew 1998; Stephan 1987).* Given the lim-

Meta-analytic evidence reported by Pettigrew and Tropp (2002) raises doubts about the necessity

ited contacts that most domiciled individuals have with homeless people, the chances of sat- isfying the paradoxical criteria for generaliza- tion seem slim. Nevertheless, we take advantage of the abstract referents (to homeless people and homelessness in general) built into our sur- vey items to shed empirical light on this issue.

The survey data also allow us to avoid pre- occupation with a single type of attitudinal out- come. Even prejudice falls short of being a unitary concept despite its dominant status in contact hypothesis research: Allport (1954) dis- cerned early on that prejudice contains favorldis- favor and belief dimensions. Similarly, Wilner et al. (1955) recognized distinctions among beliefs about an out-group, feelings toward that out-group, and relevant policy orientations. Recent scholarship makes the case more broad- ly, suggesting that contact can influence cogni- tive, affective, and behavioral aspects of a person's orientation (Jackson 1993; Pettigrew and Tropp 2002; Stephan 1987). The exact nature of that influence remains unclear, how- ever. Contact may alter the direction of certain attitudes while increasing the intensity of oth- ers (Amir 1969). Moreover, its impact may vary in duration or magnitude, depending on the type of attitude involved. The latter point is underscored in a study reporting the racial atti- tudes held by whites, which finds that beliefs about blacks and associated policy preferences are less sensitive to contact than are measures of warmth or affect (Jackman and Crane 1986). Anticipating such variation, our own analysis will use attitudinal measures that span the cog- nitivelrational (e.g., causal beliefs about and perceived characteristics of the out-group), affective (positive and negative feelings toward the out-group), and policy dimensions.

Existing investigations of the attitudinal con- sequences of exposure to homelessness, though sparse, confirm the value of considering a range of dependent and independent variables. In the Nashville survey discussed above (Lee et al. 1990), the likelihood of attributing a person's homeless status to personal choice-a type of causal belief-is greater among domiciled indi- viduals who have been panhandled than among

of repeated contact, showing attitude generalization to the target out-groupand to other out-groups not involved in the contact-to be more common than expected based on conventional contact theory.


those who have not. Engaging in conversations about homelessness with friends or relatives increases the belief in choice as well. The only germane results based on a nationwide sample come from a previous analysis of the dataset we use here. In a test of the so-called "compassion fatigue" hypothesis, Link et al. (1995) report that the number of homeless persons observed dur- ing an average week has a positive rather than negative impact on domiciled people's willing- ness to do something about homelessness and on their affective responses to the problem. Fewer significant outcomes follow from being panhandled, although they too refute the fatigue hypothesis; if anything, compassion is boosted by exposure.

A more fundamental concern is whether com- passion and other attitudes really constitute out- comes. Given the cross-sectional design of most contact research, it is difficult to dismiss the pos- sibility that initially favorable or unfavorable attitudes prompt in-group members to seek or avoid contact with the out-group (Amir 1976; Hewstone and Brown 1986). So is the possibility that persons sympathetic to the out-group might inflate the extent of their exposure. In both instances, selection processes cast doubt on the causal direction of the contact-attitude rela- tionship. The temptation is to treat the rela- tionship as reciprocal. However, studies designed to cope with selection-for example, by relying on experimental or longitudinal designs or by examining situations in which the ability to choose contact is diminished-find substantial support for the direction of causation explicit in the contact hypothesis (Cook 1984; Link and Cullen 1986; Powers and Ellison 1995; also see Pettigrew and Tropp 2000,2002). To the degree that the cross-sectional survey at our disposal permits, we intend to see if the same causal conclusion holds true with respect to the public's exposure to and attitudes toward home- less people.


Despite the substantive focus on homelessness, our research pursues several broad objectives with respect to the contact hypothesis. We seek to determine (1) whether a simple version of the hypothesis has any merit when applied to in- grouplout-group relations that rarely satisfy the optimal conditions specified in the contact lit- erature; (2) how, if at all, the types of exposure


beyond face-to-face encounters influence in- group attitudes toward an out-group; (3) what types of in-group attitudes are affected and in what ways; and (4) how the larger community context influences such attitudes: directly, indi- rectly (by molding opportunities for exposure), or via interaction with inhvidual characteristics. We also evaluate the appropriateness of the causal logic (exposure +attitudes) underlying these four objectives, given the challenges posed by selection. At stake here, then, is the scope and validity of a version of the contact hypothesis that replaces contact with exposure as the cen- tral concept and that incorporates a wide range of antecedents and outcomes.


To test the hypothesis, we utilize data from a national telephone survey fielded in 1990 by Columbia University researchers (for full details, see Link et al. 1994, 1995; Phelan et al. 1995).3 Multistage probability sampling was used to select (1) more than 200 metropolitan and nonmetropolitan phone exchanges in the continental United States, (2) occupied house- holds within those exchanges, and (3) one adult per sampled household. Each adult was offered a modest monetary incentive ($10) to participate in an interview that averaged 40 minutes in length. The investigators completed 1,507 inter- views, yielding a response rate of 65 percent for English-speaking sample members and 63 per- cent overall (with non-English speakers includ- ed in the denominator). The response rate reflects extensive efforts by the survey team to

Admittedly, survey data are less than ideal for iso- lating the causal influence of exposure on public attitudes. We can imagine a variety of experimental manipulations that would complement the approach taken here by assigning subjects to treatment groups that (1) serve meals to or otherwise interact with homeless clients in a service facility, (2) watch videos of homeless people, or (3) read newspaper or mag- azine articles about homelessness. When experi- mental methods are used to test the contact hypothesis for other, non-homeless out-groups, they generally produce stronger support for the hypothesis than do survey-based studies (Pettigrew and Tropp 2000, 2002). Thus our own results probably should be con- sidered conservative.

reach households (an average of nine calls per interview) and to convert initial refusers to respondents.

When appropriately weighted the survey par- ticipants constitute a reasonably representative cross-section of persons 18 years of age or older in households with phones as of 1990. Census results for that year suggest that the sample underrepresents Latinos and overrepresents women, the well educated, married adults, and people in the 25-54 year age group, but none of the discrepancies are large. A second type of comparison-between initial rehsers (who pre- sumably resemble nonrespondents) and will- ing respondents-also provides little evidence of bias (Phelan et al. 1995). From the full sam- ple of 1,507 adults, we identify a working sam- ple of 1,388. These are the respondents with complete data on all variables incorporated in our analysis.

Much of the survey interview was devoted to collecting information on respondents' attitudes toward homelessness. We focus on nine specif- ic multi-item scales (additive indexes) and an overall attitudes scale, treating them as depend- ent variables. For four of the nine specific scales, the choice of items has been guided by loadings from principal-components factor analyses. The remaining five rest on inter-item correlations and satisfy face validity criteria. Some of the scales are positively framed and others nega- tively, most achieve levels of reliability in the modest to respectable range (as evaluated by Cronbach's alpha), and they tap different types or dimensions of attitudes. The average corre- lation among the nine scales (mean r = .28) reflects their minimal overlap, as does the fact that no two scales share as much as 25 percent of the variance with each other (maximum r = .49). Our ultimate interest is in determining how respondents' scale scores are influenced by variations in exposure.

The first four scales assess what we term the cognitive-rational dimension, probing the per- ceived causes of homelessness and perceived attributes of homeless people. Belief in structural causes (alpha =.62) is measured with four items that trace the roots of homelessness to a shortage of affordable housing and to defi- ciencies in schools, the economic system, and government aid. The three-item individual caus- es scale (alpha = .60) places the blame for home- lessness on laziness, substance abuse, and irre- sponsible behavior. Respondents with high scores on the three-item role competence scale (alpha = .54) think that most homeless people have good job skills, could take care of a home, and would respect their neighbors' property. Finally, perceived dangerousness (alpha = .66) is measured with six items that tap homeless persons' trustworthiness, propensity for vio- lence, ability to inspire fear, dangerousness (in general and compared to other people), and threat to safety when allowed to gather in pub- lic spaces.

The next three attitude scales are primarily affective in nature, describing respondents' feel- ings about homelessness. Positive emotions (alpha = .57), such as expressing sadness or compassion for homeless individuals or becom- ing angry over their number in our affluent soci- ety, are summarized in a three-item scale. On a more negative note, respondents with high scores on the four-item lack of empathy scale (alpha = .65) do not understand how anyone could become homeless and believe that the condition reduces everyday worries, provides an abundance of free time, and can be identified on the basis of appearance alone. The third affective scale, community burden (alpha = .56), consists of five items measuring the perceived harmful impacts of homelessness on taxpay- ers, local businesses, neighborhoods, parks, and quality of city life.

Two other scales measure attitudes as revealed in respondents' policy views. The conviction that homeless persons should be able to beg, set up a tent, or sleep overnight in public places yields the maximum score on a three-item support of rights scale (alpha = .69). And respon- dents' willingness to sacrzfice (alpha = .75) is indicated by their agreement to pay more taxes, do volunteer work, and have a shelter or other emergency housing located in their neighbor- hood if such steps would help address the home- lessness problem.

Our last dependent variable is an overall measure of attitudes (alpha = .78), created by converting the values on the nine specific scales to Z scores and then summing them.4 The cod-

When the 35 interview items that make up the nine scales are standardized and summed (rather than the nine scales themselves), the reliability of the overall attitudes scale climbs substantially (alpha = 36).


ing of the negatively framed scales (individual causes, dangerousness, lack of empathy, and community burden) has been reversed so that the high end of the overall attitudes scale indi- cates a more favorable or sympathetic orienta- tion toward homeless people. In addition to its economy, the overall scale is useful for manag- ing the limited reliabilities of some of the spe- cific scales, which can reduce correlations and explained variance^.^ We take other steps to address this issue and note them during the course of the analysis. The potentially attenuated nature of our findings should be kept in mind throughout. Whenever it occurs, the likely con- sequence of any unreliability in our dependent variables is to reduce the chance of obtaining significant results.

Different types of exposure constitute the prin- cipal independent variables. From the survey interview we have chosen the 14 exposure meas- ures described in Table 1.6 When possible, forms of exposure likely to be judged neutral or neg- ative by domiciled individuals (number of homeless people seen during an average week, observing homeless people acting strange or looking through garbage, frequency of being panhandled) have been chosen to incline our test of the contact hypothesis in a conservative direction. The percentages in the right-hand columns of Table 1 document considerable

Three of the scales described in preceding para- graphs (role competence, positive emotions, and community burden) exhibit alphas that are only mar- ginally acceptable, falling below .6 in magnitude. We have decided to retain these scales because their component interview items-like those of our other scales-are worded in a manner that squarely taps the attitude dimensions of interest.

Despite the explicit nature of the survey items underlying these measures, many ultimately rest on a subjective definition of homelessness. That is, respondents are asked to make a judgment about the housing status (homeless or domiciled) of people whom they have observed or interacted with in pub- lic settings. The accuracy of such judgments cannot be determined. We take comfort, however, from the results in Table 4, which show positive effects of local homeless population size (a measure derived from an independent data source) on different types of exposure.

Table 1. Components of Exposure

Dimensiodvariable Information Television shows Articles Conversation Observation Neighborhood Week Positive behavior Negative behavioP

Interaction Shelter Panhandling Positive contact


Membership Ever homeless Homeless kin Homeless friend

N = 1,388.


Number of television shows watched about homeless Number of newspaper, magazine articles read about homeless Frequency of discussing homelessness with family, friends

Frequency of seeing homeless in own neighborhood Number of homeless seen during average week Saw homeless do something nice for someone in past year Saw homeless sleep on public benches, act strange, look through garbage,

be drunk or high, hit someone, threaten someone in past year

Ever been inside or worked/volunteered in homeless shelter Number of times panhandled by homeless in past year Homeless did something nice for respondent in past year Number of homeless known personally

Ever had time in life when considered self homeless Member of family ever homeless Close friend ever homeless

a One added to index score for each of six types of negative behavior observed. Analysis employs logarithmically transformed version of this variable.

% in Lowest Category


6.3 (none)

11.7 (none)

5I. 1 (never)

37.9 (none)

65.5 (no)

19.5 (none)

56.0 (no)

37.8 (never)

  1. (no)
  2. (none)

86.2 (no)

89.2 (no)

79.4 (no)

% in Highest Category

20.0 (5+)

28.0 (a lot)

10.5 (often)

12.9 (often)

3.2 (50+)

34.5 (yes)

4.1 (6)

44.0 (yes)

14.8 (lo+)

36.5 (yes)

1.2 (98+)

13.8 (yes)

10.8 (yes)

20.6 (yes)

variation in all of the measures, although rela- tively few respondents report ever being home- less themselves or having a family member or friend who was homeless. The 14 individual measures have been combined into multi-item scales representing the four dimensions of expo- sure. By converting the items listed in Table 1 under each dimension (e.g., number of televi- sion shows and articles and frequency of dis- cussions under information) into Z scores and summing them, we have created scales for information (alpha = .59),observation (alpha = .73), and interaction (alpha = .65). In the case of membership, the three dichotomous measures have been added together first, then their sum logarithmically transformed to adjust for mald- istribution.An overall exposure scale (alpha = .74) comprises the sum of the standard scores from the four dimension-specific scales.

To estimate the independent impact of expo- sure, our analysis incorporates a range of respondent characteristics as controls. Sex, age, and education (six categories) are included based on their importance in the research cited above. We examine race (African American and Hispanic dummy variables), employment status, and income (six categories), because each may structure opportunities for exposure to the homeless as well as exert a direct effect on domiciled individuals' attitudes toward that out- group. Ideological influences on such attitudes are measured with a five-point political liber- alism/conservatism scale, ranging from "very liberal" (high) to "very conservative" (low), and with a measure that we label "religious rel- evance." During the interview, respondents were asked how important their religious beliefs were in shaping their views about homelessness. Answers range from "very important" (high) to "not at all important" (low).

The survey data are complemented with three contextual variables drawn from the 1990 cen- sus. We use the log of community population size (for the census-demarcated metropolitan area or nonmetropolitan county of residence for each respondent) as a proxy for the salience of homelessness on the local agenda. But we also tap problem salience more directly. In 1990 the Census Bureau conducted an S-night (street and shelter) enumeration of the U.S. homeless population. Working from master lists of sites based on administrative records and input from knowledgeable informants, census staff attempt- ed to count all individuals staying in emergency


shelters during the evening of March 20, then moved outside early the next morning to search for homeless persons in parks, alleys, and other public spaces (for full details, see Taeuber and Siege1 1991).7 Two contextual measures have been developed from the S-night data and appended to respondents' records: (1) the size of the local (metropolitan area or nonmetro- politan county) homeless population (logged), and (2) the local homelessness rate (logged), defined as the number of homeless people per 10,000 residents. Both of these measures are expected to positively affect exposure.

The primary selection issue to be confronted in our analysis involves causal direction. Contrary to the logic of the contact hypothesis, preexist- ing negative attitudes toward homeless people could prompt domiciled individuals to limit their exposure by avoiding that out-group. One strategy we use to address such a possibility is to pay special attention to less voluntary forms of exposure (e.g., personally experiencing homelessness or having a relative or fnend who has), in which the role of choice should be min- imized. Another is to measure avoidance direct- ly rather than to speculate about it. We have combined two survey items into a dichotomous variable that indicates whether respondents have changed where they go for shopping or enter- tainment or how they use public transportation due to the presence of homeless people. If a selection process is operating-that is, if neg- ative attitudes steer domiciled people away from their homeless counterparts and if such avoid- ance reduces the exposure of the former group to the latter-an inverse relationship should be

'Although the S-night results have been criticized for underestimating the number of homeless (Martin 1992; Wright and Devine 1995), they are compara- ble to counts from rigorous studies in Chicago (Rossi 1989), Los Angeles (Koegel, Burnam, and Morton 1996), Nashville (Lee 1991), Washington, D.C. (Robinson 1985), and elsewhere, and they deviate on the high side, not just the low (also see Garfinkel and Piliavin 1994). Moreover, they appear to capture adequately the "literal" homeless, a subgroup ger- mane to our investigation because of the visibility of its members.


observed between the avoidance variable and most kinds of exposure.

Selection can manifest itself in another way, as social desirability bias. Being favorably pre- disposed toward homeless persons may encour- age the overreporting of exposure, either intentionally or otherwise, and might artifi- cially inflate attitudes. To correct for this ten- dency, we enter a Crowne-Marlowe-style scale in all of our equations. The six-item scale (alpha = .67) consists of true-false statements about always going out of the way to help a homeless person, never having an unkind thought or laughing at a joke about the homeless, and only feeling compassion toward them, among other items (Link et al. 1995). To the extent that the scale works as it should, any effects of exposure detected are net of social desirability. Moreover, such effects are potentially conservative since the scale items used here-unlike those in Crowne and Marlowe's (1964) original instru- ment-are domain specific and may measure truly favorable attitudes about homelessness as well as respondent personality traits. Our social desirability scale could thus inadvertently be diminishing some of the impact of prime theo- retical interest.


We begin by evaluating the simple version of the contact hypothesis: that overall exposure posi- tively influences attitudes toward homeless peo- ple regardless of whether optimal conditions are met. The overall attitudes scale and the nine specific attitude scales are each regressed on respondent characteristics and the social desir- ability scale (which together constitute the base- line model), then the overall exposure measure is entered to complete the equation. Table 2 presents results for the full equations only. Our interest centers on the unstandardized and stan- dardized regression coefficients for exposure and on the increment in R2 added by exposure after the other predictors have explained as much of the variance in attitudes as they can. These statistics, which capture the impact of exposure, are shown near the bottom ofTable 2.

Both the regression coefficients and the R2 increments provide strong support for the con- tact hypothesis. Despite the stigma associated with homelessness and the unfavorable condi- tions normally surrounding public-homeless encounters, the overall exposure measure achieves statistical significance in 9 of the 10 equations, including the equation with overall attitudes as the dependent variable (first column of Table 2). The effects of exposure are always in a more sympathetic direction8 Specifically, respondents experiencing greater exposure are more likely to attribute homelessness to struc- tural causes and less likely to believe in indi- vidual causes. These respondents tend to perceive homeless persons as competent and nonthreatening. With respect to the affective dimension, exposure encourages the develop- ment of positive emotions and empathy. Exposure also influences policy attitudes in the expected (positive) direction. Respondents exposed to homelessness are more willing to

This conclusion holds irrespective of how atti- tudes and exposure are operationalized. Although detailed results are not reported here, we have used principal components analysis to define two attitude factors or components, one on which the four nega- tively framed attitude scales (individual causes, dan- gerousness, lack of empathy, and community burden) load heavily and the other dominated by the positively framed scales. The ordinary least squares (OLS) regressions employing the two sets of factor scores and the predictors from Table 2 show that positive atti- tudes toward homeless people are significantly enhanced by overall exposure while negative atti- tudes are significantly eroded. Similarly, a principal components analysis of the 14 exposure items in Table 1 yields three new measures that significantly influence attitudes in the expected direction in 2 1 of 30 equations (10 attitude scales x 3 exposure com- ponents). As a final sensitivity check, we have treat- ed attitudes and exposure as latent variables in a structural equation model (Maruyama 1998), with the former variable represented by nine indicators (the specific attitude scales) and the latter by four (the information, observation, interaction, and member- ship scales). The modeling exercise, which relies on maximum likelihood methods available in the Amos software package (Arbuckle and Wothke 1999), doc- uments a highly significant effect of exposure on attitudes. Furthermore, a comparison of the stan- dardized coefficients for this effect obtained from equations incorporating exposure and attitudes as latent variables (B = .26) and as overall scales (B = .21) gives some sense of what is gained when our results are corrected for the attenuation associated with less than perfect measurement (see note 10 for more evidence on this point).

support the rights of homeless people and to make sacrifices to help them.

These effects exhibit consistency not only in direction but in magnitude. The standardized regression coefficient or beta (symbolized by B in the table column headings) for the overall exposure measure ranks among the five strongest in most equations, and it is larger than that of any other predictor in the last equation (with willingness to sacrifice as the dependent variable). As seen in the bottom row ofTable 2, exposure explains at least 1 percent of unique variance beyond that accounted for by the base- line model in six of the equations.

Additional determinants of domiciled respon- dents' attitudes toward homelessness are appar- ent in the remainder of Table 2. Age, education, and political liberalism register significant effects on most attitudes. In general, older age corresponds to decreased sympathy for home- less persons, while education and liberalism correspond to increased sympathy. Minority status is another characteristic that frequently has significant effects. African Americans and Hispanics are more likely to lack empathy for homeless persons and to consider them dan- gerous and a burden on the community. The direction of influence here seems reasonable because members of minority groups may be less able than whites to insulate themselves from the more visible and problematic aspects of homelessness on a daily bask9 Hispanics are also less supportive of homeless people's rights and less willing to help them. These negative

Black-white and Hispanic-white t-tests yield a series of significant differences, with members of both minority groups seeing homeless people more often in their neighborhoods, seeing more homeless people in an average week, and being panhandled more frequently than their white counterparts. These results, in conjunction with the African American and Hispanic coefficients in Table 2, suggest that the favorable influence of exposure may weaken when it surpasses some threshold level and home- lessness becomes impossible to ignore. Even under such conditions, however, exposure still has certain effects consistent with the contact hypothesis. A replication of the Table 2 analysis limited only to African American respondents-though hampered by small sample size (N = 142)-finds a significant positive impact of exposure on willingness to sacri- fice and nearly significant positive impacts on belief in structural causes and support for the rights of


policy attitudes, which align with the ideology stressing individual effort and social mobility to which some Hispanic groups subscribe, could be reinforced by the apparent success group members have had in staying off the streets themselves through doubling up and other net- work-based residential strategies (Gonzalez Baker 1996; Lee et al. 1992).

Finally, the substantial coefficients displayed across the board by the social desirability meas- ure underscore the need to control for this kind of bias when homeless people are the attitude object. Such a conclusion assumes, of course, that the measure is not tapping very positive atti- tudes.

Though useful, our overall exposure scale could obscure variation in the attitudinal con- sequences of different dimensions of expo- sure. For example, learning about homelessness vicariously (through media reports or conversation with friends) might have an effect opposite that of being panhan- dled or of interacting with homeless individ- uals in other face-to-face situations. Or perhaps more intense types of exposure exert a stronger impact on attitudes. To address these possi- bilities, we retest the contact hypothesis with increasingly specific exposure measures.

Table 3 summarizes key results from regres- sions using the four exposure scales described earlier. Model specification is the same as in Table 2 but with the information, observation, interaction, or membership scale entered instead of the overall exposure variable. What the results tell us is that the impact of overall exposure is not due to the disproportionate influence exer- cised by a particular exposure dimension. In Table 3, the information dimension has a sig- nificant effect on nine of the attitude scales, membership on eight, and interaction on seven; in every instance, at least one scale within every category of attitudes (cognitive, affective, pol- icy) is affected. The observation dimension, though less often important, still reaches significance in 4 of 10equations.lo The effects

homeless persons. (Too few Hispanics are available to conduct a similar analysis of the exposure dynam- ic within that group.)

lo These results change little when structural equa- tion modeling is used to correct for attenuation. Each

Table 2. OLS Regression of Attitudes on Respondent Characteristics and Overall Exposure

Predictor Female Agea lack^ is panic^ Educationc Employedd Incomee Political liberalf Religious relevances Social desirabilityh Exposureh

Intercept Total R2 R2 added by exposurei

Predictor Female Age" lack^

Attitudes b (SE) B

Positive Emotions

b (SE) B

.176*** ,153 (.030) .003*** ,085 (.001) .075 .040 (.047)

Structural Causes b (SE) B

Lack of Empathy

b (SE) B

,019 -.013 (.037) .006*** ,109 (.001) .139* ,059 (.059)

Individual Causes b (SE) B

Community Burden

b (SE) B

.005 -.005 (.027) .005*** ,166 (.001) .085* ,053 (.043)

Role Competence b (SE) B

Support of Rights

b (SE) B

-.060 .039 (.041)

-.004*** -.080 (.001)

-.050 ,020 (.065)

Perceived Danger b (SE) B

Would Sacrifice

b (SE) B

.053* ,054 (.024) -.003*** -.I10 (.001)

,049 .03 1 (.038)

(continued on next page)

Table 2. (Continued)  
    Positive Emotions Lack of Empathy Community Burden Support of Rights Would Sacrifice
Predictor b (SE) B b (SE) B b (SE) B b (SE) B b (SE) B
Hispanicb ,126 -.044 .318*** .088 .218*** .089 -.206* -.054 -. 134* -.054
  (.07 1)   (.089)   (.064)   (.098)   (.057)  
EducationC .030** ,086 .119*** -.277 -.028*** .095 .037** ,082 .041*** ,141
  (.O 10)   (.012)   (.009)   (.O 13)   (.008)  
Employedd -.009 -.007 ,010 ,007 ,017 -.017 .056 ,036 -.002 ,002
  (.03 1)   (.039)   (.028)   (.043)   (.025)  
Incomee .023* .065 -.035** .081 -.003 .010 -.010 -.022 .001 .003
  (.010)   (.012)   (.009)   (.013)   (.008)  
Political liberalf .085*** .I59 -.087*** ,131 .033** -.074 .115*** ,162 .071*** ,155
  (.013)   (.017)   (.012)   (.019)   (.011)  
Religious relevanceg .089*** .I70 ,004 ,006 -.O 16 -.036 .039* ,056 .054*** ,121
  (.014)   (.017)   (.012)   (.O 1 9)   (.011)  
Social desirabilityh .504*** ,246 ,020 -.008 -.428*** -.246 .581*** .214 .439*** .250

(.056) (.071) (.05 1) (.078) (.045) Exposureh .026*** .I36 -.033*** -.I38 .006 -.037 .038*** ,153 .053*** .329 8(.005) (.006) (.004) (.007) (.ow) 2


Intercept 1.900*** -1.988*** 3.363*** 1.095*** .917***

Total R2 .190*** .190*** .088*** .120*** .286***


R2 added by exposurei .017*** .017*** ,001 .021*** .098*** N = 1,388. m


a In years, from 18 through 90.


Whitelother is reference category. 9


Six categories, from eighth grade or less through graduate work. U


Full-time outside home.

0Family total, before taxes; six categories, from less than $20,000 to $75,000 or more. Five categories, from very conservative to very liberal.



g Importance of religious beliefs in shaping views about homeless; four categories, from not at all important to very important. wl

See text for description. wl5 'Additional increment of variance explained by exposure when entered in equation. wl *p < .05, **p < .01, ***p< .001.




of each exposure dimension are in the direction of greater sympathy toward homeless people, and-as judged by their standardized coeffi- cients-they tend to be sizeable in magnitude when compared with the effects of respondents' characteristics.

The next stage in our analysis is to check the coefficients of the individual exposure meas- ures that make up the overall and dimension- specific scales already examined. Each variable from Table 1has been entered separately in the 10 attitude equations, along with the other pre- dictors from Table 2 (not shown). Even at this level of disaggregation, support for the contact hypothesis remains abundant. The three infor- mation measures-number of television shows watched about the homeless, number of arti- cles read about them, and frequency of dis- cussing homelessness with friends and relatives-are significant in 25 of 30 equations (10 dependent variables x 3 exposure measures). We obtain similar proportions of significant effects for the four interaction and three membership measures (26140and 20130, respectively). The four observation measures achieve significance in half of their equations


Though the vast majority of these effects are in the expected direction, with greater expo- sure increasing sympathy, several intriguing complexities emerge. Among the information variables, for example, acquiring information about homelessness through television is less likely to shape a domiciled person's attitudes than is reading about or discussing the topic. Because television viewing is a relatively pas- sive activity, it may require less critical think- ing or articulation of one's own position. In fact, media consumption of any type, including reading relevant newspaper or magazine articles, does not improve attitudes when it is the only

of the 12 equations in Table 3 in which an exposure dimension does not have a significant attitudinal impact has been rerun, but with the specific types of exposure and attitudes operationalized as latent vari- ables and with individual interview items (described in the methods section) serving as indicators. The effect of interaction on role competence-already nearly significant in the table-becomes conventionally significant. However, none of the other insignificant coefficients are boosted into signifi- cance by this alternative estimation procedure.

manner of exposure." At the other extreme, the most intimate type of exposure-experi- encing homelessness during one's life-has a limited impact as well. We suspect that for- merly homeless individuals are more detached in their assessments of why people wind up on the streets. Because they have escaped the con- dition, they believe that others should be able to follow suit. By contrast, second-hand member- ship (e.g., having a homeless family member or fnend) tends to elicit a sympathetic reaction.

The disaggregated exposure measures are also helpful in addressing the issue of selection. Significant effects could potentially lead to incorrect causal inferences for those types of exposure (such as acquiring information from the media or being an employee or volunteer in a shelter for the homeless) that might be due to the choices of persons already favorably dis- posed toward homeless people. To return to one of the complexities just mentioned, perhaps some initial degree of sympathy is a prerequi- site for bothering to read a newspaper or mag- azine article about homelessness. But other kinds of exposure appear less voluntary; exam- ples include having a homeless family member or friend and seeing homeless people in one's own neighborhood or during the average week. In these instances, more exposure is once again statistically associated with more positive atti-

l1 In a separate analysis, we make use of a dummy variable that identifies all respondents who report any media exposure to homelessness-via television shows, newspapers, or magazines-but who have not observed or interacted with homeless persons or experienced homelessness more directly. Including this media-only variable as a predictor in our 10 full models, we find that media contact in the absence of other kinds of exposure can lead to a less sympathetic orientation. Media-only respondents score lower on the overall attitudes scale, and they are more likely to view homeless people as dangerous and less like- ly to support homeless rights to public space or to express a personal willingness to sacrifice. A demographic profile of these respondents shows them to be disproportionately female, older, rural, and not employed full-time when compared to the remainder of the survey sample. In other words, they reside in areas isolated from the most visible manifestations of homelessness and may have fairly circumscribed spatial routines. We should stress that they constitute a small proportion of all respondents (1 17 of 1,388; or 8.4 percent).

Table 3. Summary Results from OLS Regression of Attitudes on Respondent Characteristics and Types of Exposurea

Type of Structural Individual Role Perceived Positive Lack of Community Support of Would Exposure Attitudes Causes Causes Competence Danger Emotions Empathy Burden Rights Sacrifice

Information b .534*** .043*** -.023*** .027*** -.016** .044*** -.040*** -.011 .035*** .064*** SE ,057 ,007 ,007 ,006 .006 .006 ,008 ,006 ,009 ,005 B (rank)b ,219 (4) ,156 (4) -.085 (6) ,128 (3) ,073 (6) ,174 (2) -. 126 (4) ,053 (6) .lo3 (3) ,292 (1)

Total R2 .308*** .180*** .106*** .108*** .155*** .201*** .188*** .090*** .109*** .266*** R2 addedC .044*** .022*** .007*** .015*** .005** .028*** .015*** ,003 .010*** .078*** Observation

b .187*** .021*** -.003 ,005 .OO 1 ,006 -.010 -.002 .025*** .036*** SE ,045 .005 ,005 ,004 ,004 ,005 ,006 .OM .007 ,004 B (rank)b ,102 (6) .lo1 (4) -.015 (10) ,030 (6) -.008 (10) ,031 (10) -.041 (7) -.012 (9) .lo0 (3) ,217 (2) Total R2 .273*** .166*** .loo*** .093*** .150*** .174*** .175*** .087*** .log*** .230*** R2 addedc .009*** .009*** .OOO .OO 1 ,000 .OO 1 .002 .OOO .009*** .041***

Interaction b .361*** .024*** -.010 ,008 -.016*** .026*** -.034*** ,004 .040*** .048*** SE ,046 ,006 .006 ,005 ,005 .005 ,007 .005 ,007 .004


B (rank)b .I85 (4) .l08(4) .046(10) .050(4) -.093(6) ,125 (5) -. 135 (2) -.024 (8) .I49 (3) .275 (1) Total R2 .295*** .168*** .101*** .095*** .158*** .187*** .189*** .087*** .119*** .257***


R2 addedC .031*** .01 I*** .002 ,002 .008*** .014*** .016*** ,001 .020*** .068*** ' Membership $ b 1.949*** .122*** .077* .045 -.143*** .103** -.187*** -.023 .194*** .235***


SE .3 1 1 ,038 .038 .03 1 ,030 .035 ,044 .03 1 ,048 ,029 m



B (rank)b .I49 (5) ,082 (4) -.054 (8) .039(5) -.121(4) ,075 (7) -. 109 (4) ,020 (8) ,107 (3) .200 (2) Total R2 .284*** .164*** .102*** .094*** .163*** .178*** .184*** .088*** .109*** .225*** %


R2 addedC .020*** .006*** .003* ,001 .013*** .005** .011*** .OOO .010*** .037*** x

0 N = 1,388.


a Each equation summarized involves regression of attitude on 10 respondent characteristics in T. 2 and one type of exposure; only exposure regression coefficients (b, B, SE) are


shown. m Number in parentheses indicates how exposure B ranks (in terms of magnitude) compared to Bs for all other variables in equation. 5m


Additional increment of variance explained by type of exposure when entered last in equation. *p< .05, **p < .01, ***p < ,001. VI



tudes, as predicted by the contact hypothesis. Even being panhandled and observing the homeless engage in negative behavior signifi- cantly increase respondents' willingness to sac- rifice, although such encounters are related to the perception of homelessness as a burden on the community.

At the heart of the selection dynamic is whether respondents' views on homelessness prompt them to seek out or avoid situations where contact can be anticipated to occur, ulti- mately affecting their level of exposure. The avoidance measure introduced earlier allows us to shed light on this possibility. Approximately 1 in 10 members of our sample (10.6 percent), including a disproportionate share of racial or ethnic minority respondents (1 8.3 percent), report altering where they go for shopping or entertainment or how they use public trans- portation because of the presence of homeless people. As predicted by the selection hypothe- sis, their attitudes about homelessness are less sympathetic, presumably motivating their avoid- ance behavior. (All 10 of the attitude scales exhibit significant zero-order correlations with avoidance.) Contrary to the hypothesis, however, the avoiders' exposure levels are higher. Avoidance is positively correlated with the fre- quency of sighting homeless people in one's neighborhood and during the average week, witnessing undesirable behaviors by homeless people (acting strange, looking through garbage, getting drunk, or getting high on drugs, etc.), and having first-hand experience with such behaviors (e.g., being panhandled). Do those members of the public most bothered by home- lessness really pursue opportunities for more exposure? The logic of selection seems coun- terintuitive here. Instead, the contact hypothe- sis offers the more credible causal interpretation: when extreme, exposure can reduce sympathy and promote avoidance among a handful of domiciled individuals. l2

l2 We have also attempted to address the selec- tionicausation issue statistically, by estimating the rel- ative effects of the exposure and attitude measures on each other in a series of nonrecursive structural equa- tion models. When all of the predictors from Table 2 plus the size of the local homeless population (introduced in the next note) are included in the


Thus far our research has focused on respon- dents' characteristics and exposure as antecedents of their attitudes. To assess the influence of contextual factors, we conduct several multilevel analyses (Bryk and Raudenbush 1992).13 In the first of these, the

model, the path from each exposure dimension to our overall attitudes scale exhibits a larger standardized coefficient than does the path running in the oppo- site direction, consistent with the contact hypothesis. However, the sensitivity of these results to minor adjustments in model specification, the inability of the Amos software package to handle weighted data, and other technical ambiguities lead us to de-empha- size the structural equations approach. We have greater confidence in causal inferences based on analysis of the avoidance variable and less voluntary forms of exposure.

I3 The hierarchical linear model underlying these analyses estimates both individual and contextual contributions to the attitudinal outcomes of interest while taking statistical dependence into account. This nested-observation framework can be expressed in terms of intracornmunity, individual-level (level l), and intercommunity (level 2) models. The simplest form of our level 1 model is

where x,= attitude score for respondent i in community j Po, = mean attitude score in a community after con- trolling for exposure PI, = effect of exposure on the attitude score and rij=residual variance after controlling for respon- dent's exposure.

The level-2 model takes the form

Pol = yoo+ yol(Homeless Population), + uoj

where yoo= mean attitude score across respondents yo, = effect of community homeless population size on the adjusted attitude score and uoJ = error term for community-level random effects.

For our full models, all individual-level variables from Table 3 appear in the level 1 equations. To determine the need for fixed or random coefficients, we conducted preliminary tests of the level 1 variance components. In the final analysis, however, we decid- ed to constrain all regression coefficients (except for the intercept) to be constant within communities due to small sample sizes per community. (The aver- age sample size comprises fewer than eight respon-


10 attitude scales are regressed on the same predictors as in Table 2, plus one of the three contextual measures described earlier: com- munity population size, the local homeless- ness rate, or homeless population size. Across the resulting 30 equations (not shown), over- all exposure has a statistically significant effect 27 times. By contrast, only three of the con- textual coefficients achieve significance, all in relation to the same dependent variable: the domiciled persons less willing to make sacri- fices to help homeless people are those who live in communities with higher rates of home- lessness and with larger homeless and gener- al populations. Aside from this handful of effects, no evidence exists that context directly

influences attitudes.

A second possibility-that contextual and individual characteristics interact, jointly shap- ing the public's views on homelessness-also receives little support. We search for cross-level interactions by estimating the impact of our con- textual measures on the slope coefficients of the individual-level predictors in the equations just discussed (not shown). Due to the high correla- tions among the three contextual variables, the findings from the equations involving the size of the local homeless population are typical, with only two of 110 interaction terms [l contextual variable x 1 1 respondent characteristics (includ- ing overall exposure) x 10 attitude scales] attain- ing significant magnitude. Interactions also have been checked via subdivision of the survey sam- ple into persons in communities containing large (500+) and small (< 500) homeless populations. In the vast majority of cases, respondent char- acteristics and the overall exposure measure influence attitudes consistently irrespective of homeless population size. In fact, there is not a single case where a predictor registers significant effects in opposite directions in the two context- specific subsamples.

Context does, however, play a pivotal indirect role. As anticipated, opportunities for exposure to homelessness depend to a nontrivial extent on local circumstances. Table 4 documents this point, displaying equations from the hierarchi- cal regression of overall exposure and the four exposure dimensions on individual character-

dents.) The results are virtually identical whether we use random or fixed coefficients; the latter are report- ed in the text and Table 4.

istics and homeless population size.14 The key result in the table is that as the homeless popu- lation increases, so does overall exposure (Table 4, bottom of left column). The homeless popu- lation variable registers a significant impact on three of the four types of exposure as well; the effects are strongest for observation and inter- action. Thus it is through exposure, especially at the street level, that aspects of community context make a difference in the public's atti- tudes.

Several other results in Table 4 are worth noting. Those in the top portion demonstrate the spatially stratified nature of daily activities for certain demographic groups. By way of illus- tration, women and the elderly are less likely to see or interact with homeless people while African Americans are more likely to have such encounters. The significant effects of political liberalism on all dimensions of exposure, and of religious relevance on most, hint at an ideo- logically inspired sensitivity to homelessness and, conceivably, a tendency toward overre- porting. The plausibility of this scenario is rein- forced by the sizeable coefficients for the social desirability measure. Given that ideological pre- dispositions are controlled in the exposure equa- tions, the persistent impact of homeless population size becomes all the more impres- sive.


Our analysis yields consistent support for the contact hypothesis. The effects on in-group atti- tudes of exposure to a highly stigmatized out- group are not only in the predicted direction; they are quite robust. Most of the exposure coefficients achieve substantial magnitude with

l4 Initial estimates of intraclass correlation coef- ficients justify our hierarchical framework. The intra- class correlation is calculated using the formula

For our purposes, this indicates how much of the total variation in individual levels of exposure (7+ 02)is due to between-community variation (7).For example, 16.4 percent of the unconditional varia- tion in overall exposure to the homeless is between communities. This between-community component ranges from 5.4 percent (information) to 28.9 percent (observation) for our four different types of exposure.


Table 4. Multi-Level Regression of Overall Exposure and Types of Exposure on Respondent Characteristics and Homeless Population




Age Black Hispanic Education Employed Income Political liberalism Religious relevance Social desirability Intercept


Homeless populationb

N = 1,388.

Exposure Information Observation Interaction Membership b (SE) b (SE) b (SE) b (SE) b (SE)

a See Table 2 (notes a-h) for coding of respondent characteristics. Size of local homeless population, logarithmically transformed.

*p< .05, **p< .01, ***p< ,001.

multiple controls in place and with selection and social desirability processes taken into account. These coefficients hold up despite the fact that the optimal conditions set forth by the restrict- ed version of the contact hypothesis rarely apply to homeless-public encounters. Indeed, the homeless people most available for such encounters-those on the streets-are likely to exhibit stereotype-confirming attributes and behaviors. That their visible stigma fails to dampen or reverse the beneficial influence of contact highlights the impressive durability of the hypothesis, even in its simple form. Pettigrew and Tropp (2002) reach a similar con- clusion about the hypothesis based on their sys- tematic review of more than 500 studies.

The support described here is robust in anoth- er sense, extending beyond the conventional definition of contact. In particular, our results attest to the promise of exposure as a concep- tual alternative. Regardless of whether exposure is via information from third-party sources, observation in everyday settings, face-to-face interaction, or out-group membership, it seems to make a difference in how the public views homelessness. Such dimensions of exposure should be explored in research on attitudes toward other out-groups. Of course, investigat- ing the most direct form of exposure-actual- ly belonging to the out-group in question- requires the characteristic that defines out-group status to be mutable. Some mental and physi- cal illnesses satisfy this criterion, but persons rarely "recover" (i.e., join or rejoin the in-group) when the basis of their stigma lies in race, sex- ual orientation, or old age. We are intrigued by the finding that experiencing homelessness first-hand has few attitudinal consequences. As suggested earlier, some formerly homeless indi- viduals may be hardened by their own success, making them less sympathetic than their never- homeless counterparts.15 At the other extreme, indirect exposure to information registers numerous impacts.

The weak association between exposure intensity and attitudinal outcomes is one of sev- eral issues that cannot be addressed with the data available in this study. The second issue is how exposure works. We assume that its various dimensions operate through the same mecha- nisms believed responsible for the effects of contact. According to the contact literature, the most obvious mechanism is simple learning: exposure provides in-group members with new information about the out-group, reducing the former's ignorance and thus their need to resort to stereotypes (Pettigrew 1998; Stephan and Stephan 1984). Similarly, anxiety, discomfort, and other negative emotional responses to the out-group are lessened by habituation, which entails a growing familiarity born of repeated or prolonged exposure to unsettling stimuli (Makas 1993; Patchen 1995). Positive emotions may emerge from such exposure as well, when interaction fosters relationships with out-group members or when exposure is to friends and rel- atives who belong to the out-group. Exposure (direct or indirect) can change in-group mem- bers' actions before it changes their attitudes; the resulting dissonance is resolved by adopting more favorable views of the out-group (Patchen 1995; Pettigrew 1998). What seems clear is that no single mechanism explains the influence of exposure on attitudes. Rather, several distinct social psychological processes are involved that need to be better understood.

Much also needs to be learned about the larg- er context within which exposure occurs. Responding to past critiques, our test of the contact hypothesis incorporates community contextual features. We find that the size of the

lS This hardening process may begin while they are still homeless and on the streets. Snow and Anderson (1987) analyze how some homeless people construct personal identities that allow them to distance them- selves from other homeless people. Critical for our purposes is the finding by Snow and Anderson that recently dislocated individuals and those on the verge of exiting homelessness are especially likely to employ distancing strategies. Once domiciled, they could "graduate" to the generally unsympathetic atti- tudes documented here.


local homeless population registers few signifi- cant direct or joint (interactive) effects on pub- lic attitudes toward homelessness. It does, however, determine opportunities for most types of exposure, which in turn shape attitudes. Future work should strive for a more thorough specification of context. For example, our indi- vidual-level results suggest that general aspects of the community environment such as an "enlightened" climate-reflected in aggregate measures of educational attainment and politi- cal liberalism-may promote awareness of out- group circumstances and ultimately more positive attitudes. Aspects of context relevant to the out-group in question should be considered as well. In the case of homelessness, some cities have enacted anti-panhandling ordinances, con- ducted periodic sweeps of downtown areas, and encouraged the spatial dispersal of services (National Law Center on Homelessness and Poverty 1996; Simon 1996). Each of these steps can affect opportunities for public exposure. The volume and content of local media cover- age devoted to out-group members can also affect such opportunities.

Like its contextual antecedents, the conse- quences of exposure need further attention. Ideally, one would prefer preexposure and post- exposure data for a direct assessment of attitude change. We make do with cross-sectional sur- vey measures, showing that cognitive, affec- tive, and policy-oriented attitudes toward homelessness are influenced by exposure, usu- ally in sympathy-heightening fashion. The word- ing of the survey questions hints that these attitudes apply to homeless people as a whole, not just to individuals with whom respondents have had face-to-face encounters. Why exposure registers such across-the-board impacts remains a bit puzzling. One explanation is offered by Pettigrew (1998), who contends that in-group attitudes are most likely to change as contact sit- uations accumulate over time. When an out- group grows or becomes more visible, as the homeless population has during recent decades, direct and indirect forms of exposure can be expected to increase in frequency and duration.

Normally the attitudinal outcomes associat- ed with this escalation are positive, but not always. The results from our selection analysis indicate that heavy exposure to homelessness- especially through observation or interaction- may push a small segment of the public past the "tipping point," eroding sympathy and promot-


ing avoidance. The susceptibility of minority group members to the tipping dynamic has already been noted, as has the caveat that expo- sure to homelessness makes some of the views held by African Americans more favorable (see note 9). At the opposite end of the continuum, media consumption without any concomitant close-range exposure appears to have a negative impact on the attitudes of those few persons who remain far removed from the problem (see note 11). Whether similar floors and ceilings can be detected in the exposure-attitude relationship for other out-groups is worth additional examination.

Lastly, existing research sheds only limited light on the behavioral implications of exposure. In line with the logic of the contact hypothesis, attitudes toward homeless people, molded by exposure, should translate into predictable actions on the part of the in-group. A domiciled person who attributes homelessness to structural causes, for instance, should be more inclined to rent housing to or hire the homeless than would someone who believes in individual causes such as laziness or an aversion to work. The plausi- bility of such behavioral outcomes is revealed by a single, fortuitous variable in our survey

Attitudes ::

Positive Emotions



Lack of Empathy 197

Would Sacrifice

dataset: whether, upon completion of the inter- view, respondents spontaneously donated the $10 payment they had been offered for survey participation. The one-fourth (23.3 percent) who did so exhibit a distinctive attitudinal pro- file, expressing positive emotions toward home- less people, perceiving them as less dangerous, and holding favorable policy views (supporting homeless persons' rights, being willing to sac- rifice to help them). Each of these attitudes is positively and significantly related to the like- lihood of making a donation. By contrast, lack- ing empathy for homeless people and seeing them as a community burden significantly reduces the chances of donating one's payment. Figure 1 visually demonstrates the beneficial effects of favorable scores on the six attitude scales just mentioned and the overall attitudes scale. Donation rates are typically 10-15 per- centage points higher for respondents a standard deviation above the scale mean (i.e., more sym- pathetic) when compared to respondents a stan- dard deviation below.

Though encouraging in direction, the rela- tionships summarized in Figure 1 do not preclude instances of attitude-behavior inconsistency. As


i 1

! I

i 1

1 Favorable

1 1 /


Percentage Donating

Figure 1. Differences in Donation of Payments by Favorability of Attitudes

Note: Favorable and unfavorable categories include those respondents falling one standard deviation above or below sample mean of attitude scale. The total number (N) of respondents is shown at the beginning of each bar.

Dear and Gleeson (1991) argue, favorable views toward a stigmatized out-group can quickly "mutate" into antagonistic opposition when the out-group becomes threatening in a concrete or proximate way. Illustrations of this phenom- enon include residents' resistance to the location of a shelter in their neighborhood and mer- chants' attempts to get rid of street people thought to be scaring away customers. In the future, investigators should seek to identify the conditions under which exposure to an out- group produces different types of behavioral outcomes. Ironically, it is these very outcomes, though under-studied, that provide the primary justification for interest in the contact hypoth- esis.

Barrett A. Lee is Professor of Sociology and Demography and a Faculty Associate of the Population Research Institute at The Pennsylvania State University. In addition to continuing interests in urban homelessness, his work focuses on neigh- borhood change, residential mobility, community attachment, and sociospatial aspects of racial and ethnic diversity.

Chad R. Farrell is a Ph.D. candidate in sociology at The Pennsylvania State University. His general inter- ests include urban inequality, racial and ethnic seg- regation, and neighborhood change. He is currently conducting research on the stratification of public space, multiracial diversity and segregation in hous- ing, and immigrant enclaves.

Bruce G. Link is Professor of Epidemiology and Sociomedical Sciences at the Mailman School of Public Health of Columbia University and a Research Scientist at New York State Psychiatric Institute. His interests include the nature and consequences of stigma for people with mental illnesses, the social epi- demiology of homelessness, the connection between mental illnesses and violent behaviors, and expla- nations for associations between social conditions and morbidity and mortality.


Allport, Gordon. 1954. The Nature of Prejudice. Cambridge, MA: Addison-Wesley. Amir, Yehuda. 1969. "Contact Hypothesis in Ethnic Relations." Psychological Bulletin 7 1 :3 1942.

. 1976. "The Role of Intergroup Contact in

Change of Prejudice and Ethnic Relations." Pp.

245-308 in Towards the Elimination of Racism,

edited by P. Katz. New York: Pergamon. Anderson, Leon, David A. Snow, and Daniel Cress. 1994. "Negotiating the Public Realm: Stigma Management and Collective Action Among the Homeless." Pp. 12143 in Research in Community


Sociology: The Community of the Streets

(Supplement l), edited by S.E. Cahill and L.H. Lofland. Greenwich, CT: JAI Press. Arbuckle, James L., and Werner Wothke. 1999. Amos

4.0 User's Guide. Chicago: Smallwaters Corporation.

Berman, Evan M., and Jonathan P. West. 1997. "Municipal Responses to Homelessness: A National Survey of 'Preparedness."' Journal of Urban Affairs 19:303-18.

Blumer, Herbert. 1958. "Race Prejudice as a Sense of Group Position." Pacific Sociological Review 1 :3-7.

Brewer, Marilynn B., and Norman Miller. 1984. "Beyond the Contact Hypothesis: Theoretical Perspectives on Desegregation." Pp. 281-302 in

Groups in Contact: The Psychology of Desegregation, edited by N. Miller and M.B. Brewer. Orlando: Academic.

Brown, Rupert J., and John C. Turner. 1981. "Interpersonal and Intergroup Behavior." Pp. 33-65 in Intergroup Behavior, edited by J.C. Turner and H. Giles. Chicago: University of Chicago Press.

Bryk, Anthony S., and Stephen W. Raudenbush. 1992.Hierarchical Linear Models. Newbury Park, CA: Sage.

Burt, Martha R., Laudan Y. Aron, and Edgar Lee. 2001. Helping America's Homeless: Emergency Shelter or Affordable Housing? Washington, DC: Urban Institute Press.

Caspi, Avshalom. 1984. "Contact Hypothesis and Inter-Age Attitudes: A Field Study of Cross-Age Contact." Social Psychology Quarterly 47:74-80.

Cook, Stuart W. 1970. "Motives in a Conceptual Analysis of Attitude-Related Behavior." Pp. 179-23 1 in Nebraska Symposium on Motivation, 1969, edited by W.J. Arnold and D. Levine. Lincoln, NE: University of Nebraska Press.

. 1978. "Interpersonal and Attitudinal Outcomes in Cooperating Interracial Groups."

Journal of Research and Development in Education 12:97-113.

. 1984. "Cooperative Interaction in Multiethnic Contexts." Pp. 155-85 in Groups in Contact: The Psychology of Desegregation, edited by N. Miller and M.B. Brewer. Orlando: Academic.

Crowne, Douglas P., and David Marlowe. 1964. The Approval Motive: Studies in Evaluative Dependence. Westport, CT: Greenwood.

Dear, Michael, and Brendan Gleeson. 1991. "Community Attitudes Toward the Homeless." Urban Geography 12: 155-76.

Desforges, Donna M., Charles G. Lord, Shawna L. Ramsey, Julie A. Mason, Marilyn D. Van Leeuwen, and Sylvia C. West. 1991. "Effects of Structured Cooperative Contact on Changing Negative Attitudes Toward Stigmatized Social Groups."


Journal of Personality and Social Psychology

60:53 144. Deutsch, Morton, and Mary Evans Collins. 195 1.

Interracial Housing: A Psychological Evaluation of a Social Experiment. Minneapolis: University of Minnesota Press.

Forbes, H.D. 1997. Ethnic Conflict: Commerce, Culture, and the Contact Hypothesis. New Haven, CT: Yale University Press.

Garfinkel, Irwin, and Irving Piliavin. 1994. "Trends in the Size of the Nation's Homeless Population During the 1980s: A Surprising Result." Institute for Research on Poverty Discussion Paper, No. 1034-94. Madison, WI: University of Wisconsin. Goffman, Erving. 1963. Stigma: Notes on the Management of Spoiled Identity. Englewood Cliffs, NJ: Prentice-Hall. Goldstein, Brandt J. 1993. "Panhandlers at Yale: A Case Study in the Limits of Law." Indiana Law Review 27:295-359. Gonzalez Baker, Susan. 1996. "Homelessness and the Latino Paradox." Pp. 13240 in Homelessness in America, edited by J. Baumohl. Phoenix, AZ: Oryx Press. Herek, Gregory M., and John P. Capitanio. 1996. "'Some of My Best Friends': Intergroup Contact, Concealable Stigma, and Heterosexuals' Attitudes Toward Gay Men and Lesbians." Personality and Social Psychology Bulletin 22:4 12-24. Hewstone, Miles, and Rupert Brown. 1986. "Contact is Not Enough: An Intergroup Perspective on the 'Contact Hypothesis."' Pp. 144 in Contact and Conflict in Intergroup Encounters, edited by M. Hewstone and R. Brown. Oxford, UK: Basil Blackwell. Jackman, Mary R., and Marie Crane. 1986. "'Some of My Best Friends are Black. . .': Interracial Friendship and Whites' Racial Attitudes." Public Opinion Quarterly 50: 459-86. Jackson, Jay W. 1993. "Contact Theory of Intergroup Hostility: A Review and Evaluation of the Theoretical and Empirical Literature." International Journal of Group Tensions 23:4345. Jones, Edward E., Amerigo Farina, Albert H. Hastorf, Hazel Markus, Dale T. Miller, and Robert A. Scott. 1984. Social Stigma: The Psychology of Marked Relationships. New York: W.H. Freeman. Koegel, Paul, M. Audrey Burnam, and Sally C. Morton. 1996. "Enumerating Homeless People: Alternative Strategies and Their Consequences." Evaluation Review 20: 378403. Lankenau, Stephen E. 1999a. "Panhandling Repertoires and Routines for Overcoming the Nonperson Treatment." Deviant Behavior

20: 183-206.

. 1999b. "Stronger than Dirt: Public Humiliation and Status Enhancement Among Panhandlers." Journal of Contemporary Ethnography 28:288-3 18.

Lee, Barrett A. 1991. "The Nashville Method:

Unobtrusive Observation." Pp. 139-46 in

Enumerating Homeless People: Methods and Data

Needs, edited by C.M. Taeuber. Washington, DC:

U.S. Bureau of the Census.

,and Chad R. Farrell. 2003. "Buddy, Can You Spare a Dime? Homelessness, Panhandling, and the Public." Urban AfSairs Review 38:299-324.

,Sue Hinze Jones, and David W. Lewis. 1990. "Public Beliefs About the Causes of Homelessness." Social Forces 69:253-65.

,David W. Lewis, and Sue Hinze Jones. 1992. "Are the Homeless to Blame? A Test of Two Theories." Sociological Quarterly 33535-52.

, Bruce G. Link, and Paul A. Toro. 1991. "Images of the Homeless: Public Views and Media Messages." Housing Policy Debate 2:649-82.

Link, Bruce G., and Francis T. Cullen. 1986. "Contact with the Mentally I11 and Perceptions of How Dangerous They Are." Journal of Health and Social Behavior 27:289-303.

Link, Bruce G., and Jo C. Phelan. 2001. "Conceptualizing Stigma." Annual Review of Sociology 27:363-85.

Link, Bruce G., Sharon Schwartz, Robert Moore, Jo Phelan, Elmer Struening, Ann Stueve, and Mary Ellen Colten. 1995. "Public Knowledge, Attitudes, and Beliefs About Homeless People: Evidence for Compassion Fatigue?" American Journal of Community Psychology 23533-55.

Link, Bruce G., Ezra Susser, Ann Stueve, Jo Phelan, Robert E. Moore, and Elmer Struening. 1994. "Lifetime and Five-Year Prevalence of Homelessness in the United States." American Journal ofpublic Health 84: 1907-12.

Makas, Elaine. 1993. "Getting in Touch: The Relationship Between Contact with and Attitudes Toward People with Disabilities." Pp. 121-36 in Perspectives on Disability, edited by M. Nagler. Palo Alto, CA: Health Market Research.

Martin, Elizabeth. 1992. "Assessment of S-Night Street Enumeration in the 1990 Census." Evaluation Review 16:4 18-38.

Maruyama, Geoffrey M. 1998. Basics of Structural
Equation Modeling. Thousand Oaks, CA: Sage.
Momeni, Jamshid A., ed. 1990. Homelessness in the
United States: State Surveys. New York: Praeger.

National Law Center on Homelessness and Poverty. 1996. Mean Sweeps: A Report on Anti-Homeless Laws, Litigation, and Alternatives in 50 United States Cities. Washington, DC: National Law Center on Homelessness and Poverty.

Patchen, Martin. 1995. "Contact Between Ethnic Groups: When and How Does It Lead to More Positive Relations?" International Journal of Group Tensions 25:271-87.

Pettigrew, Thomas F. 1986. "The Intergroup Contact Hypothesis Reconsidered." Pp. 169-95 in Contact and Conflict in Intergroup Encounters, edited by

M. Hewstone and R. Brown. Oxford, UK: Basil Blackwell.

. 1998. "Intergroup Contact Theory." Annual Review of Psychology 49:65-85.

,and Linda R. Tropp. 2000. "Does Intergroup Contact Reduce Prejudice? Recent Meta-Analytic Findings." Pp. 93-1 14 in Reducing Prejudice and Discrimination, edited by Stuart Oskamp. Mahwah, NJ: Lawrence Erlbaum Associates.

,and Linda Tropp. 2002. "A Meta-Analytic Test and Reformulation of Intergroup Contact Theory." Department of Psychology, University of California, Santa Cruz, CA. Unpublished manu- script.

Phelan, Jo, Bruce G. Link, Robert E. Moore, and Ann Stueve. 1997. "The Stigma of Homelessness: The Impact of the Label 'Homeless' on Attitudes Toward Poor Persons." Social Psychology Quarterly 60:323-37.

Phelan, Jo, Bruce G. Link, Ann Stueve, and Robert

E. Moore. 1995. "Education, Social Liberalism, and Economic Conservatism: Attitudes Toward Homeless People." American Sociological Review 60:12-0.

Powers, Daniel A,, and Christopher G. Ellison. 1995. "Interracial Contact and Black Racial Attitudes: The Contact Hypothesis and Selectivity Bias." Social Forces 74:205-26.

Robinson, Frederic G. 1985. "Homeless People in the Nation's Capital." Center for Applied Research and Policy, University of the District of Columbia, Washington, DC. Unpublished manuscript.

Rossi, Peter H. 1989. Down and Out in America: The Origins of Homelessness. Chicago: University of Chicago Press.

Rothbart, Myron, and Oliver P. John. 1993. "Intergroup Relations and Stereotype Change: A Social-Cognitive Analysis and Some Longitudinal Findings." Pp. 32-59 in Prejudice, Politics, and the American Dilemma, edited by P.M. Sniderman, PE. Tetlock, and E.G. Carmines. Stanford, CA: Stanford University Press. Sigelman, Lee, and Susan Welch. 1993. "The Contact Hypothesis Revisited: Black-White Interaction and Positive Racial Attitudes." Social Forces 71:781-95. Simon, Harry. 1996. "Municipal Regulation of the Homeless in Public Spaces." Pp. 149-59 in Homelessness in America, edited by J. Baumohl. Phoenix: Oryx Press.

Snow, David A,, and Leon Anderson. 1987. "Identity Work Among the Homeless: The Verbal Construction and Avowal of Personal Identity." American Journal of Sociology 92: 1336-7 1.

. 1993. Down on Their Luck: A Study of Homeless Street People. Berkeley: University of California Press.


Stephan, Walter G. 1987. "The Contact Hypothesis in Intergroup Relations." Pp. 1340 in Group Processes and Intergroup Relations, edited by C. Hendrick. Review of Personality and Social Psychology, Vol. 9. Newbury Park, CA: Sage.

,and Cookie White Stephan. 1984. "The Role of Ignorance in Intergroup Relations." Pp. 229-57 in Groups in Contact: The Psychology of Desegregation, edited by N. Miller and M.B. Brewer. Orlando: Academic.

,and Cookie White Stephan. 1996. Intergroup Relations. Boulder, CO: Westview.

Taeuber, Cynthia M., and Paul M. Siegel. 1991. "Counting the Nation's Homeless Population in the 1990 Census." Pp. 92-122 in Enumerating Homeless People: Methods and Data Needs, edited by C.M. Taeuber. Washington, DC: U.S. Bureau of the Census.

Tajfel, Henri. 1982. "Social Psychology of Intergroup Relations." Annual Review ofPsychology 33:l-39.

Toro, Paul A,, and Dennis M. McDonell. 1992. "Beliefs, Attitudes, and Knowledge About Homelessness: A Survey of the General Public."

American Journal of Community Psychology


Welch, Susan, Lee Sigelman, Timothy Bledsoe, and Michael Combs. 2001. Race and Place: Race Relations in an American City. Cambridge, UK: Cambridge University Press.

Werth, James L., and Charles G. Lord. 1992. "Previous Conceptions of the Typical Group Member and the Contact Hypothesis." Basic and Applied Social Psychology 13:351-69.

Williams, Robin M., Jr. 1947. The Reduction of Intergroup Tensions: A Survey of Research on Problems of Ethnic, Racial, and Religious Group Relations. New York: Social Science Research Council.

. 1964. Strangers Next Door: Ethnic Relations in American Communities. Englewood Cliffs, NJ: Prentice-Hall.

Wilner, Daniel M., Rosabelle Price Walkley, and Stuart W. Cook. 1955. Human Relations in Interracial Housing: A Study of the Contact Hypothesis.Minneapolis: University of Minnesota Press.

Wright, James D., and Joel A. Devine. 1995. "Housing Dynamics of the Homeless: Implications for a Count." American Journal of Orthopsychiatry 65:320-29.

Yuker, H.E. 1988. "The Effects of Contact on Attitudes Toward Disabled Persons: Some Empirical Generalizations." Pp. 262-74 in Attitudes Toward Persons Kth Disabilities, edited by H.E. Yuker. New York: Springer.

  • Recommend Us