Journal of Social Issues, Vol. 58, No. I, 2002, pp. 177-193 Studying Hate Crime with the Internet: What Makes Racists Advocate Racial Violence? Jack Glaser* University of California, Berkeley Jay Dixit New York City Donald P. Green Yale University We conducted semistructured intervie\vs with 38 participants in White racist In- ternet chat rooms, examining the extent to which people would, in this unique environment, advocate interracial violence in response to purported economic and cultural threats. Capitalizing on the anonymity and candor of chat room in- teractions, this study provides an unusual perspective on extremist attitudes. We experimentally manipulated the nature and proximity of the threats. Qualitative and quantitative analyses indicate that the respondents were most threatened by interracial marriage and, to a lesser extent, Blacks moving into White neighbor- hoods. In contrast, job competition posed by Blacks evoked very little advocacy of violence. The study affords an assessment of the advantages and limitations of Internet-based research with clandestine populations. Under what conditions do people advocate racial violence? In particular, are people who openly embrace ideologies of racial hierarchy in a constant state of readiness to respond to racial threats, or are certain types of threats particularly evocative? Since Myrdal (1944), scholars have observed that threats of miscegena- tion tend to evoke the strongest emotional reactions from avowed racists, yet at *Correspondence concerning this article should be addressed to Jack Glaser, Goldman School of Public Policy, 2607 Hearst Ave., University of California, Berkeley, CA 94720-7320 [e-mail: glaserj@socrates.berkeley.edu]. We wish to thank Jon Drummond, Rob MacCoun, and seven anony- mous reviewers for their very helpful comments on earlier drafts. 177 © 2002 "The Society for the Psychological Study of Social Issues 178 Glaser, Dixit, and Green the same time, there exists an extensive literature that explains racial animus by reference to economic competition and other "realistic" sources of conflict (e.g., Olzak, 1990; Tolnay & Beck, 1995). This topic, like so many in the domain of intergroup conflict, presents the researcher with a wide array of measurement problems. In addition to problems of dissembling and self-presentation (Crosby, Bromley, & Saxe, 1980), special ideo- logical populations, such as those belonging to White separatist movements, are often deeply suspicious of authorities and outsiders. Although interviews with avowed racists have been conducted in past research (Ezekiel, 1995), it remains unclear whether the views expressed have been tailored to the interview setting. In order to observe this population as unobtrusively as possible, we used the Internet, applying a randomized survey procedure to examine the conditions under which individuals who participate in White racist chat rooms would be willing to advocate violence in response to a racial threat. This approach benefits from a combination of features unique to the Internet: the guarantees of anonymity in communication, and the abundance of hate group communication forums there. It should be noted at the outset, however, that our study does not measure actual illegal conduct. Rather, our approach is to study the advocacy of racially motivated crime, under the assumption that the causes of advocacy are related to the causes of illegal conduct itself. Motifs of Ethnic Violence It has long been observed that different types of ethnic groups are targets of different types of prejudice, discrimination, and violence (Horowitz, 1985). Research on hate crime, unlawful conduct motivated by prejudice against a social group, emphasizes this theme by demonstrating quantitatively how different groups are victimized. For example, Jews, often stereotyped as affluent and greedy, are more frequently the victims of vandalism than assault. Gays and lesbians, who are perceived to be a moral and sexual threat, are more often the victims of physical assault. Blacks are attacked in a variety of ways, including acts of vandalism and intimidation, apparently in an attempt to confine African Americans to a subordinate status and keep them physically separate from Whites. The present analysis focuses on hate crime against African Americans and will compare the factors that have most prominently been thought to precipitate such acts: economic competition, turf violation, and interracial dating, marriage, and/or sex. We focus on anti-Black hate crime for several reasons. First, because the nature of hate crime appears to vary as a function of the target group, it is prudent to focus on one group in order to isolate predictor variables. Given that, we chose to focus on anti-Black hate crime in part because Blacks are by far the most frequently victimized group according to federal statistics (e.g., U.S. Department of Justice, 1997). Additionally, our previous research (e.g., Green, Glaser, & Rich, 1998) on Hate on the Internet 179 causes of hate crime also focused on Blacks, and the present study allows us to clarify and extend that analysis. Societal factors and motivation. What kinds of threats trigger the most vig- orous racist reaction? Historical analyses have attributed Hitler's rise to poor macroeconomic conditions (e.g., Eberhard, 1998; Turner, 1996). Similarly, dom- inant social science approaches to understanding hate crime assume instrumental antecedents like competition for material resources. This dates back at least to Hovland and Sears's (1940) demonstration that the price of cotton (and other eco- nomic indicators) in the post-Civil War Deep South correlated negatively with the number of lynchings of Blacks. Although Hovland and Sears conducted their analyses to find support for the frustration-aggression hypothesis (Bollard, Doob, Miller, Mowrer, & Sears, 1939), their finding quickly became a cornerstone of so- ciological studies of economic threat and intergroup violence (e.g., Olzak, 1990; Pinderhughes, 1993; Tolnay & Beck, 1995). Theories of realistic group conflict (e.g., Bobo, 1988), wherein competition for material resources underlies intergroup strife, are also bolstered by such findings. However, recent analyses of historical and contemporary data indicate that economic variables are not reliable predictors of hate crime (Green, Abelson, & Garnett, 1999; Green, Glaser, & Rich, 1998; Green, Strolovitch, & Wong, 1998). Green, Glaser, and Rich (1998) found that when time series data involved in earlier analyses were reanalyzed and extended in time, economic conditions bore a weak and inconsistent relationship to lynching. Furthermore, contempo- rary data revealed no reliable correlations between unemployment, an important economic factor, and hate crime against Blacks, Asians, Jews, or gays and les- bians (Green, Glaser, & Rich, 1998). Rather than flowing directly from economic downturns, hate crime seems to coincide with hard times when political leaders seize on economic conditions in order to marshal resentment against minority groups. What little is known about the motives of those who belong to racist groups or engage in racially motivated crime also suggests the importance of noneco- nomic factors. Green et al. (1999) conducted a telephone survey of known White supremacists and hate crime perpetrators in North Carolina, having identified such people based on media reports. The respondents were not aware of the selection criteria. Green et al. (1999) found that, in comparison to a general-public control sample, supremacists and hate crime perpetrators were not more economically frustrated or pessimistic about future finances. However, they were considerably more opposed to interracial marriage and migration of minorities into tradition- ally White communities. Supremacists in particular also appeared more concerned with threats to cultural identity (e.g., banning of the Confederate flag). With respect to anti-Black violence in particular, historical analyses reveal that lynchings often resulted from the perception and accusation that a Black 180 Glaser, Dixit, and Green man had committed sexual assault against, had a consensual relationship with, or even simply looked at a White woman (Hodes, 1997; Tolnay & Beck, 1995; Wright, 1990). More contemporary anti-Black hate crimes (e.g., cross burnings) are often precipitated by territorial incursion by Blacks into predominantly White communities (Green, Strolovitch, & Wong, 1998). Interracial dating and marriage is still taboo, at least implicitly, and in some circles explicitly, with Alabama taking until the year 2000 to expunge a law (albeit one rendered moot by federal legislation) prohibiting interracial marriage, and this by only a 60% to 40% popular vote. Thus, we theorize that hate crime against African Americans typically results not so much from economic concerns or frustrations, or competition for material resources, but more often from the perceived threat to the integrity, separateness, and hegemony of the ingroup. We set out to test this thesis by examining the differential effects of economic, geographic, and genetic threat on White racists' advocacy of hate crime. All three threats are interrelated, but our hypothesis is that the threat of genetic incursion is the most evocative. This conceptualization is in some ways consistent with Ezekiel's (1995) the- ory of the racist mind. Ezekiel argues that White racists often fear for their own survival and that belonging to supremacist groups gives them comfort and re- assurance. Support for the importance of cultural threat, especially with regard to interracial sex, also comes from Hodes (1997), whose historical analysis sug- gests that miscegenation in the 19th-century Deep South came to be perceived by Whites as a threat and a justification for lynching primarily after emancipation, when subordination of Blacks was no longer facilitated through the institution of slavery. The purpose of the present research is to employ and test an Internet-based, unobtrusive survey method to determine the social factors that are most likely to incite bias-motivated violence. The question is, what form of threat by minorities, in this case African Americans, is most likely to yield an inclination by racists to commit violence against them? Accordingly, we posed scenarios that might be perceived as threatening, regarding interracial marriage, minority in-migration (i.e., Blacks moving into one's neighborhood), and job competition (i.e., compet- ing with a Black person for a job). These were selected to reflect genetic (cultural), geographic (cultural), and economic threats, respectively, each representing funda- mental and prototypical examples of these types of threats in the interracial realm. In addition to investigating the effects of different types of threats, we assess the impact of threat proximity, that is, whether the threat occurs at the personal, lo- cal, or national level. This factor is potentially important because White separatist rhetoric often dovetails with nationalistic concerns (i.e., many separatists advo- cate deporting racial and ethnic minorities to other countries or continents), but for practical reasons it seems likely that action (e.g., violence) will be taken or advocated when threat is immediate or proximal. Hate on the Internet 181 Method The Utility of the Internet In order to test hypotheses about the likelihood of committing a hate crime, we required an unusual sample, one that would have sufficient variability in this regard (i.e., contain a relatively large portion who would be inclined to condone hate crimes), and one that would be willing to express such dispositions. Similarly, we needed to conduct this study in a milieu that would enable the expression of hate crime advocacy, with sufficient anonymity and confidentiality for our participants and ourselves. The Green et al. (1999) survey of supremacists and hate crime perpetrators made significant inroads in this regard, but the sample size was very limited and the telephone public opinion poll format restricted the survey to relatively formal and tame questions. Fortunately, with the advent of the Internet, we were afforded the opportunity to gain entree to the world of White racist hate groups, surveying them in an environment that encourages open expression. Over 100 million Americans used the Internet in 1999 (Cole et al., 2000). White pride Web sites appear to be particularly active on the Internet (Anti- Defamation League, 1999; Whine, 1999). Organized racist presence on the Internet began no later than 1985 (Anti-Defamation League, 1999) and has increased to include hundreds of Web sites, file archives, chat rooms, mailing lists, newsgroups, etc. (Franklin, 2000; Klanwatch, 1998). Racists on the Internet tend to express their views rather freely, at least when they are interacting with those they perceive to be like-minded. For this reason, and because of their prevalence there, we decided to use the Internet Relay Chat (IRC) "rooms" affiliated with racist organizations as our venue, and the partici- pants therein as our sample. In this manner we can assess attitudes and behavioral inclinations that racists might otherwise be reluctant to reveal. Our goal was to compare factors that are likely to inspire hate crime, specif- ically those discussed above: economic threat (i.e., job competition), territorial threat (i.e., minority in-migration to neighborhoods), and genetic threat (interra- cial marriage). In order to accomplish this, we visited various IRC chat rooms sponsored by White supremacist groups and conducted randomized interviews. Posing as a new visitor to the chat rooms, our interviewer presented scenarios of different kinds of threats and recorded the responses. These responses were then coded for their advocacy of violence so that we could compare the extent to which different types of threat differentially inspire advocacy of hate crime. Design We employed a 3 (threat type: interracial marriage, minority in-migration, job competition) by 3 (threat level: personal, local, national) design. One important 182 Glaser, Dixit, and Green feature of the design is that it was neither fully within- or between-participants, nor was it a mixed-factorial design in the typical, orthogonal sense. We adopted an unusual, pseudo-Latin square design to maximize the number of respondents in each condition of the experiment. Although there are a large number of racist Web sites, the number of chat rooms is more limited, and our preliminary monitoring of these rooms indicated that many of the participants were the same people, jumping around from room to room. Consequently, there appeared to be a universe of only several hundred at the most. Because we anticipated that only a subset of this population would be willing to converse one-on-one with a chat room neophyte (the interviewer) with whom they were not familiar, we attempted to increase the number of respondents in each condition of the experiment by utilizing a partial within-participants design. It most likely would be awkward and suspicious for the respondents if all nine scenarios were posed to each of them, or if they were posed a series of similar questions with only one parameter changed. It was decided that we could pose three scenarios each, if only one was drawn from each level of each variable. Specifically, one group of respondents got all the scenarios in the diagonal of Table 1 (personal interracial marriage, local in-migration, and national job competition). Similarly, another group heard about personal in-migration, local job competition, and national interracial marriage. Finally, another group was asked about personal job competition, local interracial marriage, and national in- migration. In this manner, all nine conditions were filled with a limited sized sample without raising suspicion. Ethics In order for us to gather candid responses without raising suspicion, we could not obtain informed consent. The Yale human participants committee agreed that respondents would have been very unlikely to participate, that those who did would not have been representative, and that responses would have been sig- nificantly biased. We believe, and the Yale committee concurred, that the lack of informed consent was acceptable because respondents participated without Table 1. Scenarios Comprising 3 x 3 Design of the Quasi-Experimental Survey Marriage In-migration Job competition Personal My sister is talking about I found out this Black J found out I'm competing getting manned to this couple is moving in with a Black man for my Black man. next door to me. promotion at work. Local Lots of White women in my Lots of Blacks are At my work, White people neighborhood are getting moving into my have to compete with Blacks married to Black men. neighborhood. for promotions. National All over the country. Black All over the country. All over the country, Blacks men are getting married Blacks are moving into are taking White people's to White women. White neighborhoods. jobs. Hate on the Internet 183 coercion, in a public forum, discussing topics that were common subjects of conversation there. Finally, respondents' identities were protected, through the use of their own pseudonyms and our careful separation in the data set of these pseudonyms from the responses they provided. Procedure The interviewer entered various chat rooms, posing as a curious neophyte. In an IRC chat room one can engage in real-time conversations with any or all of the visitors there. Chat rooms are ideal for this study because their participants are especially likely to express otherwise socially taboo sentiments and proclivities (McKenna & Bargh, 1998, 2000). In each visit, the interviewer made small talk until he was able to engage an individual in a dialogue. At this point the interviewer randomly selected a sequence of experimental conditions (e.g., personal-marriage, local-in-migration, and then national-job competition). There were 18 possible orderings to ensure that scenario order would be a random variable and that the interviewer would not be able to anticipate the next scenario he was to give. The interviewer posed each scenario after the respondent appeared to have played out his or her response to the previous scenario. The interviewer typically carried out the discussion "in view of" the chat room but did not engage other participants in the same room on the same day. We employed a semistructured interview approach (Smith, Harre, & Von Langenhove, 1995) wherein the interviewer engages in a relatively free-flowing discussion, making sure to cover certain topics. This approach allowed for a real- istic discussion with survey questions embedded naturalistically, thereby increas- ing the likelihood of candid and representative responses. The danger that the interviewer's own hypotheses might bias his interactions with respondents was mitigated by the random selection of the conditions just prior to posing each scenario. Thirty-eight people were interviewed in all. They were all voluntary partici- pants in the chat room in which they were surveyed. No demographic data were obtained on the respondents because this might have served to undermine their anonymity and to raise suspicion about the interviewer. As described above, three of the nine scenarios were posed to each respondent. A few respondents left the chat room after responding to only one or two of the three scenarios, thus reducing the total number of responses from the expected 114 to 107. Responses were recorded by logging the text of the discussions and downloading it for later coding. The responses were coded for the extent to which the respondent advocated violence in some form during the course of his or her response. Blind to the condition of the experiment, the interviewer and another author rated each response on a 6-point scale from 0 (no advocacy of violence) to 5 (advocacy of extreme violence). It is plausible that the interviewer could have T84 Glaser, Dixit, and Green recognized the responses and recalled the condition, but this was impossible for the other author, who had not seen the data prior to coding. Nevertheless, the interrater correlation was reasonably high (r — .82). Although the author who had not participated in data collection was not able to recognize the specific responses, he nevertheless was aware of the design of the experiment, may have been able to infer the scenarios to which some of the responses were made, and may have had hypotheses similar to those of the experimenter, which may have served to bias his ratings. To address this concern, we had a third, doubly blind (i.e., unaware of design or conditions, or for that matter the nature of the study in general) person rate the responses on the same scale for their advocacy of violence. The correlation between this rater's ratings and the average ratings of the first two raters (used in the data analyses) was .90, indicating strong correspondence. Results Narrative Analysis Perhaps the greatest power of this study lies in the qualitative analysis of the types of responses the different scenarios evoked. The specific statements made by our respondents were often very strongly worded, revealing clear opinions that differed substantially as a function of threat type. They offer a rich source for analysis of racists' attitudes toward minorities and committing violence against them. It should be noted that the style of communication in chat rooms is very informal and, even more so than with e-mail, typing is often sloppy. We have not corrected errors in the following quotes. Interracial marriage. The data clearly reveal that responses with regard to interracial marriage, especially at the personal level, were the most volatile. For example, one respondent stated, "better kill her. kill him and her. pull a o j . . . im not kidding, i would do it if it was my sister, i would gladly go to prison then live a free life knowing some mud babies were calling me uncle whitey." Another respondent advised, in a manner reminiscent of the days of lynchings, "Hang his black ass." One person cautiously suggested the following: "Im not saying how I would stop it, nor am I encouraging you to do this, but, there are many murders in America today, some for the better of our race, and it is fairly easy to get away with one. Not that I am telling you to go out and murder someone though!" It is not certain, but this respondent appears to be advocating violence while being careful not to do so directly. There is certainly an aspect of encouragement (i.e., "it is fairly easy to get away with one") in his comments. A few people responded to the personal-interracial marriage scenario without advocating violence (e.g., "well I wouldn't have anything to do with her if I were you."), but the clear majority (9 out of 13) expressed some advocacy of violence, Hate on the Internet 185 mostly extreme violence. Further evidence for the evocative nature of interracial marriage is available in spontaneous responses to other threats. For example, one respondent stated with regard to in-migration, "I don't let it get to me, they stay with their space and I stay in mine.... The thing that upsets me most is seeing so many white girls dating black guys." Responses to interracial marriage at the local level were, in contrast, notably brief and bland. Respondents expressed some concern (e.g., "Thats sad. Yuck that makes me sick," "man that to bad," "then move"), but were generally more passive and helpless (e.g., "sorry man, I don't know what to say"). Only 1 respondent out of 14 in this condition indicated a disposition toward violence, recounting his own assault of a Black man who married his cousin, for which he claimed to have been prosecuted. At the national level, the responses to interracial marriage were somewhere between the personal and local in terms of advocacy of violence, with most being tepid, but a few exhibiting support for extreme violence (e.g., "They should all be shot."). Minority in-migration. At least at the personal level, in-migration evoked a range of responses. Many were mild and even restrained, such as, "I don't like it but we canf't] do anything." Others were more extreme: "I would run the niggers and all non-whites oil of my city... kill some nigger ass." Others recommended specific action that would qualify as hate crime but was not necessarily violent (e.g., "Spraypaint 'niggers beware' on the door before they even move in. If they catch wind of it, I doubt they will even finalize the buy"). At the local and national levels, in-migration moved only a few respondents to advocate some form of action (e.g., "make his ass move out of there") but for the most part responses were passive and even resigned (e.g., "move." "Yep.... It's happening everywhere"). Job competition. The threat of job competition elicited a more consistently mild reaction. Although one respondent did advocate extreme violence ("kill him") with regard to a personal-level scenario, most responses were tepid and reasoned, such as, "all you can do is try your best." In response to the personal-level threat, several inquired about the qualifications of the Black competitor, one suggesting filing suit for reverse discrimination. At the local level, responses were similar, with the exception of one person who advocated framing the Black competitor: "I say set em up for a bust get em fired and away from our women." Notably, in this case, spontaneous reference is again made to interracial dating/marriage, suggesting that it is a chronically salient threat. At the national level as well, responses tended to be political, focusing on issues of affirmative action, rather than violent. One person volunteered, "asian are taking white jobs too!" suggesting that whereas job competition from Blacks may not be much of a threat, competition from other groups may be. 186 Glaser, Dixit, and Green The clear differences in the types of reactions to different scenarios reveals to some extent that the threat to the integrity of the group, be it cultural or genetic, is a relatively potent predictor of violent tendencies toward outgroup members. The genetic aspect of this is evidenced by the specific references to the outcome of inter- racial marriage: "mud babies." This conclusion is bolstered by data from additional inquiries we made. On one occasion, after completing the relatively structured part of the interview, the experimenter asked, more comparatively, "Of blacks marry- ing white women, blacks moving into white neighborhoods, and blacks taking white jobs, which of these do you think is the most threatening?" The response was "integration of races, obviously!" When the interviewer pressed further the respondent indicated that he meant integration through intermarriage, restating, "obviously." Quantitative Analysis Although the analytic strength of this research appears to lie in the com- pelling narrative of the responses, the data can also be analyzed quantitatively. Accordingly, Table 2 reports the mean degree of advocacy of violence in each condition of the experiment, based on the 6-point (0-5) scale ratings. These results reinforce our qualitative assessment. The average level of advocated violence was greatest when respondents were presented with the issue of interracial marriage as compared to job competition or in-migration. Averages were also higher when these threats were framed in personal terms. Local threats elicited low levels of advocated violence, with national threats falling in between. These results can be tested more rigorously using regression analysis, mak- ing allowances for the unusual design of this experiment. The quasi-Latin square Table 2. Advocacy of Violence as a Function of Threat Type and Level Threat level Threat type Interracial marriage In-migration Job competition Column mean Personal 2.46 (2.21) n = 14 1.5 (2.23) n = I l 0.54 (1.47) ( 7 = 1 2 1.54 (2.11) n = 37 Local 0.18 (0.67) n = 14 0.0 (0.0) « = 16 0.0 (0.0) n = l 0.07 (0.41) « = 37 National 1.43 (2.44) n = 7 0.0 (0.0) » = 9 0.29 (1.21) n = 17 0.46 (1.46) « = 3 3 Row mean 1.34 (2.04) « = 35 0.46 (1.38) n = 36 0.32 (1.18) « = 36 Note. Scores represent mean responses from ratings on a 6 -point scale from 0 (no advocacy of violence) to 5 (advocacy of extreme violence). Standard deviations are in parentheses. Hate on the Internet 187 design of the experiment, in which each respondent was presented with multi- ple scenarios (but not all scenarios), increased the number of participants in each condition, but the observations are not statistically independent. The disturbances associated with one observation are likely to be correlated with others. In order to address this problem, we report robust standard errors derived from bootstrap- ping. That is, 150 samples of size 107 (the size of the total sample) were drawn (with replacement, necessarily) from the original data set. The empirical sampling distribution of these regression estimates is used to calculate the standard errors. The second problem also arises from the limited size of each treatment condi- tion. In randomizing the placement of respondents into conditions, and due in large part to false starts (e.g., refusals to engage in discourse), our group sample sizes became lopsided. Before we could fully rectify this, it became evident that we had tapped out much of the available sample and suspicion was beginning to mount about our presence and repeated statements and questions. As a result, we were left with condition cells of varying sizes (ranging from 7 to 17). This imbalance reduces the statistical power of our test but does not lead to biased inference. The regression includes four predictors of advocated violence. The first two are dummy variables associated with the type of threat: one dummy marks inter- racial marriage and the other in-migration (with job competition being the omitted category). The second set of dummy variables marks the level of threat: one dummy for personal threat and the other for local (with national threats being the omitted category). The regression reported in Table 3 indicates that interracial marriage significantly increases advocated violence by an average of 1.1 scale points rela- tive to job competition (t = 2.7, p < .01, two-sided test). Similarly, personal threats significantly increase advocated violence (/ = 2.22, p <.05, two-sided), relative to national threats, whereas local threats significantly decrease them (£ = 2.03, Table 3. Regression Analysis of Effect of Threat Type and Threat Level on Advocacy of Violence Estimates Robust standard errors Threat type Interracial marriage 1.09** .41 In-migration .32 .29 Threat level Personal .91* .41 Local -.61* .30 Constant .14 .25 Note. Threat types include interracial marriage, minority in-migration, and job competition. Threat levels include personal, local, and national. Job competition and the national threat level serve as omitted categories and can be considered the neutral (zero) reference point against which to compare the other parameter estimates. One hundred fifty random replications were conducted to generate robust standard errors in order to allow for the nonindependence of the conditions (i.e., that each respondent was in conditions at multiple, but not all, levels of both variables) and uneven sample sizes across conditions. Adjusted R1 = .20, N = 107. Asterisks indicate that the estimate differs significantly from the neutral (zero) reference point for that variable (i.e., job competition for threat type, national for threat level) as follows: * p <.05, two-tailed; **p<.0l, two-tailed. 188 Glaser, Dixit, and Green p < .05, two-sided). No significant interactions between the level and type of threat emerge, using two-sided tests. Discussion "The greatest existing cause of lynching is the perpetration, especially by black men, of the heinous crime of rape..." —Theodore Roosevelt (cf. Wright, 1990, p. 77) The thousands of lynchings of Southern Blacks, often on trumped up charges, during the post-Reconstruction period represent an early form of what we now might call "hate crime." President Roosevelt's take on the cause of lynching about a century ago must be considered within the historical context of the time. Accounts of lynchings often involved the accusation of the rape of a White woman by a Black man. Although Roosevelt's conclusion, influenced by the propaganda of the time, was almost certainly misguided, it nevertheless may offer a valuable insight into the mind of the hate crime perpetrator: that violence against members of other races may be particularly linked to concerns over interracial mixing. Indeed, it is clear from the present study that interracial marriage is the idea that most upsets racists on the Internet and is likely to drive them to advocate anti-Black hate crime. Consistent with Green, Strolovitch, and Wong (1998) and Green et al. (1999), there is also some response to territorial incursion, but only at the personal level. In keeping with the findings of Green, Glaser, and Rich (1998) as well as Green et al. (1999), but inconsistent with Hovland and Sears (1940) and studies that followed, job competition, an economic variable, inspired very little advocacy of violence against Blacks at any level. The propensity for interracial marriage and minority in-migration to evoke extreme responses among White racists on the Internet is perhaps exemplified in the frequent, spontaneous invocations of the expression "14 words." This is White separatist code for, "We must secure the existence of our people and a future for White children," and it serves as a rallying cry of sorts, but clearly reflects concerns about race mixing. The effect of threat level, although of less theoretical interest than threat type, was fairly clear. Scenarios posed at the personal level were by far the most evocative. This may be the case because scenarios of this sort are the most concrete and conducive to giving advice. Specifically, it seems more likely that one would advocate some form of action to someone who faces a "problem" personally and so, in this respect, this variable is not very informative or interesting. However, this was not equally true across types of threat. Job competition had the same small effect at the national and personal levels, where responses often indicated that such threats either were beyond one's control or would work themselves out, this based on the belief that Blacks are inferior and will not be able to retain such jobs. Hate on the Internet 189 We must also consider why the national level evoked some advocacy of vio- lence but the local level had no effect whatsoever. The most plausible explanation for this is probably that threats posed at the national level triggered thoughts tied to the rhetoric of hate groups who tend to speak nationalistically and lament the declining state of the country, which they often attribute to "lazy" and criminal minorities and immigrants. The local threat level, on the other hand, may have been too abstract, tapping neither the empathic response of the personal level nor nationalist dogma. It is nevertheless surprising that this level evoked virtually no advocacy of violence, especially with regard to minority in-migration, which is often a community concern, or at least perceived to be. Impact of the Internet on Race Relations The utility of the Internet as a venue for studying racist extremists is a derivative of a potentially troubling reality: that such groups are prevalent in cyberspace. Although their prevalence is undisputed, there is some debate over whether or not the Internet has been a boon for racist groups. In fact, a Web site called Hatewatch that had functioned to list and monitor Web-based hate groups since 1995 recently shut itself down, offering the explanation that it had completed its mission and concluding that the Internet has had a negative impact on hate (Dixit, 2001). Although White racist groups have proliferated on the Internet in recent years, there appears to have been no corresponding increase in membership in these groups or in hate crime rates. In fact, one might argue that the prevalence of racist groups on the Internet works to reduce hate crime, perhaps by providing less physical, more rhetorical outlets for hate. Furthermore, the presence of hate groups on the Internet has in many ways made them more transparent to the public, which in turn facilitates monitoring by watchdog groups, government and private alike, not to mention social scientists. Nevertheless, perhaps in part because of the inherently underground nature of White separatist groups, extending back to the days of the hooded Ku Klux Klan, there is no direct evidence available that the Internet has not helped to proliferate hate groups. Furthermore, the potential of the Internet as a tool for communication, organization, and information dissemination among extremist groups is undeniable, and although it may have its incidental benefits (e.g.. transparency), it clearly warrants scrutiny. Methodological Considerations In addition to providing insight into the mentality of White separatists and perhaps the antecedents of hate crime, the present study offers a methodological innovation that should generalize to other research questions and prove useful in the future. Specifically, by unobtrusively surveying people on the highly anony- mous, yet public, forum of the Internet, and specifically in chat rooms, we are 190 Glaser, Dixit, and Green able to open up avenues of research not previously available to most researchers. There are numerous groups that are difficult to gain access to, either because of their marginality or because of illicit aspects of their behaviors that make self- disclosure potentially costly. McKenna and Bargh (1998, 2000) have also suc- cessfully employed Internet-based research to reach populations that would be difficult to access. The present study adopted a more surreptitious approach, per- haps more appropriate when measuring variables relating to illegal behavior such as hate crime and when dealing with populations that are suspicious of researchers and other outsiders. This approach could be adopted for other populations, such as terrorist or militia groups, child pornographers, and illegal weapons traders, whose members would also be unlikely to respond candidly to explicit survey questions, and the potential is clear for public good arising from greater understanding of such groups. In order to carry out such research, however, we must consider the thorny issues of deceit and informed consent. As noted above, deceit with regard to the identity of the interviewer (and, in fact, that he was an interviewer at all) was essential for the success of the study. Otherwise, respondents would have been very unlikely to participate, or to respond candidly if they did. For the same reasons, obtaining informed consent was impossible. Fortunately, the use of the Internet had the added advantage of ensuring the anonymity of our respondents, all of whom use pseudonyms in the chat rooms. We further promoted their anonymity and confidentiality by separating even their pseudonyms from their data, assigning random numerical codes. Additionally, the public forum nature of the chat rooms mitigates the need for informed consent. Nevertheless, such research should always be carried out with the utmost regard for the confidentiality and safety of the sample. Limitations Despite the seeming clarity of the results, there are several limitations of the study that should be acknowledged. First, this study was conducted at one moment in time, and only through replication can one ascertain whether, for example, the motive power of economic threat was undermined by the generally favorable economic conditions that prevailed at the time of our study. This investigation is also limited by the relatively small size of the sample, which was compromised in part by rising suspicion among potential respondents. McKenna and Bargh (1998) faced similar problems in their e-mail survey of posters on White supremacist (and other ideologically extreme) newsgroups, where some posters were warning others to not respond to the survey and generating rumors about its being conducted by government agencies like the FBI. This may prove to be a chronic problem with research on this population, which appears to have limited numbers participating in chat rooms at any given time. However, incremental replications over time, as new Hate on the Internet 191 cohorts join and grow tired of the chat rooms, could ultimately yield a substantial database of knowledge. Furthermore, we would recommend that future studies of this sort employ a purely between-participants design to minimize the appearance of inquisitiveness on the part of the ostensive neophyte (i.e., the interviewer). We are somewhat more sanguine about other aspects of the study, in particular the sincerity with which respondents expressed their views. One respondent went so far as to reveal the types of code words that are used over ham radios to form posses and plan hate crimes. Of greater concern than the sincerity of the responses is the possibility that, because of the anonymity of the Internet and the culture of the chat rooms, responses reflected a greater level of endorsement of violence than respondents actually felt. Nevertheless, as far as we can tell, there is little cause for concern that respondents were not genuine. We must also consider the possibility that the very presence of the interviewer, as a curious neophyte, influenced the responses, perhaps invoking more bravado. Although we made every effort to be nondescript, this possibility is still real, but it should not represent a confound in the design and results with regard to relative differences in advocacy of violence as a function of threat type and level. Perhaps more complex is the issue of how advocacy of violence relates to actual illegal conduct. It is telling that several respondents, but by no means a majority, indicated that they had, indeed, themselves committed violent hate crimes in the past and provided some detail of them. Needless to say, we cannot be sure that such statements are true. They could simply reflect a certain form of false bravado, even under conditions of anonymity. The possibility remains, however, that some overlap exists between racist ideology and racist action (Green et al., 1999). Consequently, variables that are diagnostic for this group may be important in predicting hate crime. Conclusion In sum, because of the nature of Internet-based communication, particularly chat rooms, we were able to observe a group and form of behavior that would otherwise be difficult for scientists to study. By assessing their responses to vari- ous types of threat, we develop a better understanding of the motives and beliefs that animate those who advocate and perhaps commit hate crimes. These results are consistent with our past findings that economic threat, whether in the form of declining cotton prices, increases in unemployment, or heightened job com- petition, does not in itself trigger violent ideation. Rather, perceived threats to White hegemony and separateness, via in-migration and especially interracial sex and marriage, generate a visceral reaction against outgroups. These findings are important insofar as they shed light on the parlance used to incite fear by racist ideologues, both past and present. 192 Glaser, Dixit, and Green Although further research is clearly warranted, the present study has some important policy implications. First, it offers a methodological approach that policy analysts can employ to survey otherwise inaccessible groups (e.g., separatists, child pornographers, weapons traders) to better understand their motivations and behaviors. Second, as noted above, the findings, in conjunction with past research, challenge the presumption that economic conditions are direct determinants of intergroup violence. Those concerned with reducing such socially destabilizing phenomena should look to other factors, including extremist group belief systems, which may prove more worthwhile targets of change. Similarly, it is questionable as to whether the Internet itself serves to promote racial violence, but clearly it can be employed as a means to understanding, and perhaps reducing, such bigotry and conflict. References Anti-Defamation League. (1999). Poisoning the Web: Hatred online. Washington, DC: Author. Bobo, L. (1988). Group conflict, prejudice, and the paradox of contemporary racial attitudes. In P. A. Kate & D. A. Taylor (Eds.), Eliminating racism (pp. 85-114). New York: Plenum. Cole, J. I., Suman, M., Schramm, P, van Bel, D., Lunn, B., Maguire, P., Hanson, K., Singh, R., Aquino, J., & Lebo, H. (2000). The UCLA Internet report: Surveying the digital future. Los Angeles: University of California at Los Angeles, Center for Communication Policy. Crosby, F., Bromley, S., & Saxe, L. (1980). Recent unobtrusive studies of Black and White discrimi- nation and prejudice: A literature review. Psychological Bulletin, 87, 546-563. Dixit, J. (2001, May 9). A banner day for neo-Nazis [On-line]. Available: www.salon.com/tech/ feature/2001/05/09/hatewatch/index.html Dollard, J., Doob, L. W., Miller, N. E., Mowrer, O. H., & Sears, R. R. (1939). Frustration and aggression. New Haven, CT: Yale University Press. Eberhard, K. (1988). The Weimar Republic. Boston, MA: Unwin Hyman. Ezekiel, R. S. (1995). The racist mind: Portraits of American neo-Nazis and Klansmen. New York: Viking. Franklin, R. A. (2000). The hate directory [Internet Website]. Available: www.hatedirectory.com Green, D. P., Abelson, R. P., & Garnett, M. (1999). The distinctive political views of hate-crime perpetrators and White supremacists. In D. A. Prentice & D. T. Miller (Eds.), Cultural divides: Understanding and overcoming group conflict (pp. 429-464). New York: Russell Sage Foundation. Green, D. P., Glaser, J., & Rich, A. O. (1998). From lynching to gay-bashing: The elusive connection between economic conditions and hate crime. Journal of Personality and Social Psychology, 75, 82-92. Green, D. P., Strolovitch, D.. & Wong, J. (1998). Defended neighborhoods, integration, and racially motivated crime. American Journal of Sociology, 104, 372^03. Modes, M . E . (1997). White women. Black men: Illicit sex in the nineteenth-century South. New Haven, CT: Yale University Press. Horowitz, D. (1985). Ethnic groups in conflict. Berkeley and Los Angeles: University of California Press. Hovland, C. I., & Sears, R. R. (1940). Minor studies of aggression: VI. Con-elation of lynchings with economic indices. Journal of Psychology. 9, 301-310. Klanwatch. (1998). 474 hate groups blanket America: God, rock 'n' roll and the Net fuel the rage. Intelligence Report Special Issue: 1997, the Year in Hate, Winter, 1998, 89. McKenna, K. Y. A., & Bargh, J. A. (1998). Coming out in the age of the Internet: Identity "demarginal- ization" through virtual group participation. Journal of Personality and Social Psychology, 75, 681-694. Hate on the Internet 193 McKenna, K. Y. A., & Bargh, J. A. (2000). Plan 9 from cyberspace: The implications of the Internet for personality and social psychology. Personality and Social Psychology Review, 4, 57-75. Myrdal, G. (1944). An American dilemma: The Negro problem and modern democracy. New York: Harper. Olzak, S. (1990). The political context of competition: Lynching and urban racial violence, 1882-1914. Social Forces, 69, 395-421. Pinderhughes, H. (1993). The anatomy of racially motivated violence in New York City: A case study of youth in Southern Brooklyn. Social Problems. 40, 478^192. Smith, J. A., Harre, R., & Von Langenhove, L. (1995). Rethinking methods in psychology. New York: Sage. Tolnay, S. E.. & Beck, E. M. (1995). A festival of violence: An analysis of Southern lynchings, 1882- 1930. Urbana, IL: University of Illinois Press. Turner, H. A. (1996). Hitler's thirty days to power: January 1933. Reading, PA: Addison-Wesley. U.S. Department of Justice. (1997). Hate crime statistics 1997. Washington, D.C.: U.S. Department of Justice, FBI Criminal Justice Information Services Division. Whine, M. (1999). The use of the Internet by far right extremists [On-line]. Available: http:// www.ict.org.il/articles/right-wing-net.htm Wright, G. C. (1990). Racial violence in Kentucky, 1865-1940: Lynchings, mob rule, and "legal lynchings." Baton Rouge, LA: Louisiana State University Press. JACK GLASER is an Assistant Professor in the Goldman School of Public Policy at the University of California, Berkeley. After earning his PhD in psychology from Yale University in 1999 he received a National Institute of Mental Health National Research Service Award to serve as a postdoctoral fellow at UC Berkeley's Institute of Personality and Social Research. Glaser studies intergroup bias at multiple levels of analysis and manifestations, including implicit stereotyping and prejudice, racial profiling, and hate crime, and has interests in the psychology of electoral politics and political ideology. JAY DLXIT is a prize-winning New York City-based freelance writer. His work has appeared in Rolling Stone, The Village Voice, Salon.com, and Psychology Today. He received his bachelor's degree, cum laude, in psychology, with distinction in the major, from Yale University in 1998, where he wrote for The New Journal, including an article about White supremacists on the Internet. DONALD P. GREEN is A. Whitney Griswold Professor of Political Science at Yale University, where he has taught since receiving his PhD from UC Berkeley in 1988. Since 1996, he has served as Director of the Institution for Social and Policy Studies, an interdisciplinary research unit at Yale. His scholarly interests include hate crime, political behavior, public opinion, campaign finance, and research methodology.