racial rioting in the 1960's: an event history analysis of local conditions

7/30/2019 Racial Rioting in the 1960's: An Event History Analysis of Local Conditions

http://slidepdf.com/reader/full/racial-rioting-in-the-1960s-an-event-history-analysis-of-local-conditions 1/20

Racial Rioting in the 1960S: An Event History Analysis of Local ConditionsAuthor(s): Daniel J. MyersReviewed work(s):Source: American Sociological Review, Vol. 62, No. 1 (Feb., 1997), pp. 94-112Published by: American Sociological Association

Stable URL: http://www.jstor.org/stable/2657454 .

Accessed: 21/01/2013 18:26

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

.

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to

American Sociological Review.

http://www.jstor.org

This content downloaded on Mon, 21 Jan 2013 18:26:05 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=asa

http://www.jstor.org/stable/2657454?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp





http://www.jstor.org/stable/2657454?origin=JSTOR-pdf

http://www.jstor.org/action/showPublisher?publisherCode=asa



RACIAL RIOTING IN THE 1960S:

AN EVENT HISTORY ANALYSIS OF LOCAL CONDITIONS*

Daniel J. MyersUniversity of Wisconsin, Madison

Using Spilerman's (1970b) data on the timing and locations of race riots

from 1961 to 1968, I use event history analysis to investigate the effects of

local conditions on city-level hazard rates of rioting. First, several struc-

tural strain and social-psychological arguments, which were the focus of

Spilerman's original studies, are reexamined in light of event history analy-

sis. Second, I reinterpret ethnic competition arguments recently used to ex-

plain a wide variety of collective violence, and apply these ideas to the

1960s' riots. Third, I test two models of the diffusion of rioting. Like previous

analyses, my findings fail to support structural strain theories. Contrary toprevious analyses however, the size of the non-White population is not the

singular predictor found for the 1960s' riot locations. Instead, my results

support both competition and diffusion arguments.

uring the 1970s, Spilerman(1970a,1971, 1972, 1976) published an influ-

ential set of studies examining racial riotingin the 1960s. For many, this research became

the definitive word on the 1960s riots, and itremains influential today (McPhail 1994).

Spilerman examined a numberof theoretical

explanations for the occurrence and severity

of riots, including structural trainarguments(Smelser 1962), absolute deprivation (Dow-

nes 1968), relative deprivation (Gurr 1968,1970), unresponsivepolitical structures Lie-berson and Silverman 1965), and geographi-cal contagion (National Advisory Commis-

sion on Civil Disorders 1968). In summary,

he concluded that differences among citieswere unrelatedto the frequency and severity

of rioting. In fact, only two variables, thesize of the non-White population and a

dummy variable for region in the United

States, were better predictors of riots andtheir severity than the 16 theoretical indica-tors combined.

Despite the apparentclarity of Spilerman's

findings, scholars did not abandon the theo-retical constructs he tested. Much subsequentresearch has attempted to challenge his find-

ings and revive older explanations of the1960s riots (for reviews, see Bryan 1979;McPhail 1994; McPhail and Wohlstein 1983;

Schneider 1992). Despite slight deviations,however, the results of these studies gener-ally support Spilerman's findings: The sizeof the non-White population accounted forthe majority of the variance of riot frequency

explained, and other theoretical variablesex-plained only small increments beyond it.

Ultimately, however, there is still a consid-erable lack of agreement about what factors

should remain under consideration as poten-tial causes of rioting. Some scholars con-clude that structural strain explanations andother social-psychological explanationshavebeen thoroughly trounced and that new ap-proaches must be developed to identify the

underlyingdeterminantsof rioting and other

collective behaviors (McPhail 1994). Othersmaintain the worthiness of grievance and

deprivation explanations and continue to in-

voke slightly recast versions of these theo-

ries (Heskin 1985; Koomen and Frankel

*Direct correspondence to Daniel J. Myers,

Department of Sociology, 1180 Observatory

Drive, Madison, WI 53706 ([email protected].

edu). I thank Pamela Oliver, Gerald Marwell,

Charles Halaby, Seymour Spilerman, Robert

Mare, Alberto Palloni, Ronald Wohlstein, RalphTurner,Alexandra Corning, and the ASR review-

ers for comments on earlier drafts. Parts of this

paper were presentedat the 1995 annual meetingof the American Sociological Association inWashington, D.C.

94 American Sociological Review, 1997, Vol. 62 (February:94- 12)







RACIAL RIOTING IN THE 1960S 95

1992; Polletta 1992). These ideas may con-tinue to be popular in part because of theambiguity resulting from different samples

and methods, or simply because "it is very

difficult to abandon old explanations" (Mc-Phail 1994:5). However, the continued at-

tractiveness of these previous explanationsmay hint that the empirical examinationsthattested them were inadequate analyses of theprocesses underlying rioting. This question

can be answered by critically examining pre-vious empirical work using new analytical

procedures and recent theoretical develop-ments that are bettersuited to the problem.

Thus, I reexamine Spilerman's data on ra-

cial rioting in the 1960s using event historyanalysis. First, I closely follow his analysis,re-examining his conclusions regarding the

structural strain and deprivation arguments.Then, I investigate two additional explana-tions for rioting-one based on notions of

social diffusion and the other based onOlzak's (1987, 1992) recent work on ethniccompetition.

BACKGROUND

Structural Strain and Deprivation

Spilerman (1970a, 1971, 1976) conductedhis

analysis of the outbreakand severity of riots

by groupingstructural trainand other relatedstructuralexplanations into four basic clus-ters and adopting indices for each cluster of

arguments (Spilerman's variable clusters ap-pear in the stub of Table 2 on page 103). His

first variable cluster tested the "social disor-ganizationthesis"-that poorly integrated n-

dividuals are outside the general control of

communitynormsandhaveless access to tra-

ditional mechanisms for addressing socialgrievances (Downes 1968). Spilerman'ssec-

ond and third clusters of variables examinedthe "absolute"and "relative"versionsof dep-rivationarguments.The absolute deprivation

argument posits that the most severely dis-

possessed in society will be the ones who

engage in collective violence (Downes 1968).The relative deprivation argument (Gurr

1968, 1970) statesthata disadvantagedgroupdetermines its own level of social and eco-

nomic deprivation by comparing itself to

some reference group; in the case of

Spilerman's studies, Blacks would develop a

sense of relative deprivation by comparingtheir situationto that of proximalWhites.

Spilerman'sthird variableclustertested the"expectations"argument Berkowitz 1968) in

which expectations for improved social andeconomic conditions increase as the disad-vantagedgroup's position improves, therebyleading to heightenedfrustration f the risingexpectations are not met. In this argument,the opposite relationshipbetween deprivationvariables and racial rioting is hypothesized:As conditions improvefor Blacks, an increasein expectations, frustration, and ultimatelythe number of riot events occurs.

Finally, Spilermantested the idea that the

lack of access to political representation in-creases rioting (Lieberson and Silverman1965): With no way to addressgrievances orto have group interests represented in gov-ernmental decision-making bodies, Blacksmay turn to violence as a method of express-ing their demands. As noted above, Spiler-man (1970a) found little or no support forany of these arguments n his initial work, inwhich the frequency of rioting was the de-pendent variable.When riot severity was thedependent variable and also failed to showsupport (Spilerman 1976), these ideas werefurtherdiscredited.

In addition to the original four arguments,Spilerman (1970a) also noted several other

potential sources of the tendency to riot, buthe providedonly partialtests. Of these, I ex-amine competitionand diffusion.

CompetitionModels

Competitionfor scarce resources is often as-sumed to be a key process underlying mostconflicts. When two or more groupscompetefor a common pool of resources, they inevi-tably come into conflict, and as the demandfor a good increases and the supply de-creases, the conflict will increase. Park(1950) applied such an economic model toethnic andracial conflict. His faith, however,that assimilation would in time relieve eth-

nic conflict and lead to peaceful accommo-dation has not been substantiated;thus, the

continued application of economic competi-tion models to ethnic conflict in general andethnic collective violence in particular(Lieb-erson and Silverman 1965; Nielsen 1985;Olzak 1987, 1992; Olzak and Nagel 1986;







96 AMERICAN SOCIOLOGICAL REVIEW

Shorter and Tilly 1974; Spilerman 1970a,1971, 1976; Tilly, Tilly, and Tilly 1975;Tolnay and Beck 1995).

Olzak's (1992) recent work offers a prom-

ising model that connects economic pro-cesses with collective action using ethniccompetition as its main mechanism. Herstudy of ethnic conflict and protest supportsthree main competition mechanisms: (1) Thebreakdown of labor market segmentationleads to increased competition between eth-nic groups, thereby leading to increased con-

flict and collective action; (2) immigrationincreases competition directly (Bonacich1972; Lieberson 1980) and also through its

effects on labor marketsegmentation;and (3)economic contraction exacerbates competi-tion and conflict as it increases competitionfor employment.

While Olzak's analysis is compelling, thecollective violence she and others attempttoexplain is predominantly that of Whitesagainst ethnic and racial minorities. Riotingin the 1960s, however, was an exception tothat pattern. Therefore,if ethnic competitionarguments are to be used to explain theBlack-initiated property riots of the 1960s,the arguments must be congruent with mi-

nority-initiated as well as majority-initiatedcollective violence.

I propose that while changes in competi-tion may in fact change the levels of collec-tive violence, predictingwho will initiate the

violence and who will be the target of vio-lence requires knowledge of the outcomes ofincreased competition. In the cases examined

by Olzak, White-initiated violence in re-sponse to labormarketde-segregationmakesintuitive sense because White advantage is

threatenedby desegregation. However,apply-ing the labor market desegregation argumentto Black riots is problematicbecause it is notclear that Blacks would respondto improvedmarket positions with violence againstWhites.Yet it is possible thatdecreased labormarketsegregationcould lead to violence byBlacks if desegregation led to worse market

positions for Blacks. Consider the possibilitythat desegregationresulted from White inva-

sion of labor niches traditionally dominated

by Blacks; the result for some Blacks thus

would be unemployment.Ratherthan an im-

proved labor marketposition for Blacks, de-segregation would have the opposite effect.

Black-initiated violence is more easily under-stood undersuch circumstances.

Using the above insight, I treatnon-Whiteunemploymentas an indicatorof competitive

labor market outcomes that disadvantageBlacks. I also incorporatevariables indicat-ing increasedcompetition in the traditionallyBlack labor market stemming from non-White migration to a city and all immigra-

tion to a city from outside the United States.

Both of these variables are particularly rel-evant to the 1960s because of high levels ofinterstatemigration of Blacks and increasingimmigration throughout the 1950s and1960s. And because competition theory pre-

dicts that economic downturns exacerbateany competition-driven ethnic violence, I in-clude indicators of general economic con-

tractions in the analysis. Finally, I examineinteractions between immigration and eco-

nomic conditions implied by competition ar-guments. Intuitively,the effect of any in-mi-gration should be greater when the economicsituation is poor in a particularcity (Olzak1989b). Conversely, an influx of foreign im-migrants or American non-Whites shouldhave a smallereffect on competitionand sub-sequent violence when the economy is far-ing well.'

Diffusion of Rioting

Diffusion refers to a process in which the oc-currence of an event to one member of apopulationchanges the likelihood of such anevent occurring to other members of the

population (Strang 1991). So when a riot oc-curs in one city in the United States, the ratesof future rioting in other cities are hypoth-esized to increase as a result. Many analystshave recognized that collective protests andcollective violence tend to cluster both tem-porally and geographically (McAdam 1982;Pitcher, Hamblin, and Miller 1978; Olzak1987; Koopmans1993;Lichbach 1985;Diek-mann1979). This is not imitationor mindlesscopying, (as was the view in early contagion

theories (see McPhail 1991 for a review). In-stead, it is viewed as a social learning processin which the favorable outcomes of others'

I For a complementary analysis of racial riot-ing that also supports competition theory, seeOlzak and Shanahan(1996).








collective behaviors provides a model forsubsequentprotests (Pitcher et al. 1978).

The process of social diffusion is driven bya variety of communication processes among

established social networks (Hamblin,Jacob-sen, and Miller 1973; McPhail and Wohlstein1983; Paige 1971). These networks may bedefined by friendship and family connec-tions, or they may be more dispersed, aswhen individuals are connected through themass media (Oliver 1989). For example,Rude (1964, 1972) documented that rebel-lion in England and France during the eigh-teenth and nineteenth centuries diffusedalong majortransportation outes suggesting

that information about collective actions wasbeing communicatedby travelersalong traderoutes. More recently, the mass media andthe telephone have been considered to be themost prominentcommunicationmodes driv-ing the diffusion of collective behavior(Spilerman 1970a; Morris 1984). Clearly,thetelephone is more apt to be used by networksof acquaintances, while the mass media hasa more widely dispersed influence governedin scope by its audience (i.e., a national orlocal audience) (Oliver 1989).

I assume that these communication net-

works are geographically concentrated.Friends and family (and thus their face-to-face and telephone contacts) are likely to be

geographically concentrated.And, althoughsome riots receive national media attention,media coverage of riots and civil violence is

generally concentrated in the regions wherethey occurred. Thus, while a largeriot in De-

troit may receive national newspaper andtelevision coverage, news broadcasters andnewspaper editors in and around Michiganwould devote higherproportionsof their cov-

erage to the riot than would the media inTexas or Florida. Small riots that receive re-gional media attention may not be coverednationally and therefore may be completelyunknownto people in distantregions.

Any adequatemodel of a diffusion processmust identify the underlyingfactors that ac-

count for variation n the "contagiousness"ofa social phenomenon. The two most impor-tant of these factors are temporal proximityand spatial proximity (Hamblin et al. 1973;Rogers 1983; Brown 1981; Mahajan andPetersen 1985; Strang and Tuma 1993). Inthe case of riots and other collective events,

spatialproximityis usually indicated by geo-graphicdistance (Hedstrom 1994; Spilerman1970a), and temporalproximity is indicatedby the amount of time that has passed since

a collective event (Olzak 1987, 1992). I usespatial and temporal proximity to createthree diffusion indicators.

The first diffusion variable is a decliningfunction of geographical distance between agiven city and each city that experienced ariot at time t - 1. The amount of contagion,

c,, experienced by a given city at time t isthus given by:

ct =Xd7mti , (1)j

where di1 is the distance in miles betweentwo cities, i andj, and m is a dummy vari-able indicating whether or not a riot occurredin city j at time t - 1. This contagion variabletests the hypothesis that riots in one city in-crease the likelihood of rioting in other cit-ies and that otherriots are more likely in cit-ies closest to where the original rioting oc-curred.

Given thatdiffusion is driven by commu-nication processes, the media in particular,

an alternativepattern of diffusion may haveoperatedin the 1960's rioting. Given the sub-stantial nationalmedia coverage of the riots,particularlyof large riots, rioting may havediffused nationally rather than regionally. Inotherwords, the proximityof cities may havebeen irrelevant if all cities nationally re-ceived equivalent information about the riotsas they occurred.To test for this possibility,

I examined two additional diffusion vari-ables. The first variable measures the num-ber of riots in the United States at time t - 1

(M,_1where Mt-1 = Ymj1t,). This variable as-

sumes that a riot in any city has an equal ef-fect on other cities, regardless of relative lo-cations. The effect of the number of riots attime t - I may not be linear however. Thefirst riot at t - 1 may contributesubstantiallyto the occurrence of additional riots, but thetenth riot may contribute little more beyond

thatcontributedby the ninth riot or may evendecrease the likelihood of rioting. To reflectthis possibility, I incorporatethe number ofriots squared,(M,_1)2, in the analysis.

Unlike other social diffusion processes,such as the adoption of a new technology(Hamblin et al. 1973; Rogers 1983), disrup-








tive protest activities do not continue to dif-fuse unabated. Rather, protest activity ini-tially accelerates, then peaks, and then be-gins to decline. In the case of rioting, each

riot subsequent to the apex slows the rate ofrioting, almost as if the process were becom-ing distasteful to its participants. This ex-haustion effect has been hypothesized to re-sult from either the tiring and calming of theparticipants or the increase in repressivemechanisms by agents of social control (Kel-ly and Isaac 1984; Koopmans 1993; Ober-

schall 1978). Several analyses have sug-gested that the diffusion and exhaustion pro-cesses for collective violence require rela-

tively brief periods. Olzak (1987) found thediffusion-exhaustion cycle to be completedwithin a period of 30 days. With regard tothe 1960s rioting, the Kerner Commissionreport (National Advisory Commission onCivil Disorders 1968) suggested that eventsclustered in two-week periods. Given theseobservations and the nature of the data on1960s riots, I estimate four sets of modelsusing rioting in the past one week, the pasttwo

weeks, andthe

past fourweeks.

Modelsusing the one-week period were clearly su-perior; the addition of longer time periodsdid not improve prediction over the one-week model. Therefore, here I present onlymodels using the one-week measures.

What Can and Cannot Be Explained

Every analysis is boundedby the scope of thedata examined, and the current study is no

exception. As were previous studies of the1960s riots, this analysis is also limited bytwo key aspects of the data: their temporalscope and their units of analysis. Because thedata I examine are limited to the 1960s andcompare city-level characteristics, they can-not explain the rise of the unprecedentedwave of Black-initiated rioting thatoccurredin the 1960s. A different type of data set,which provides informationon long-termna-

tional conditions, is necessary to explain the

riot wave (see McAdam 1982; Jenkins andEckert 1986; and Kelly and Isaac 1984 for

empirical investigations of these long-termtrends and contexts). In fact, it is exactly this

limitation that led Spilerman(1970a) to con-

jecture about the causes of Black-initiatedri-oting in the 1960s. Given that Spilermandid

not find any city-level differences, he con-cluded that rioting must be the product of anational phenomenon. Drawing on Tom-linson (1968), Spilerman suggested that the

process was actually driven by nationwideconditions thataffected all Blacks and by theacceptance of a "riot ideology" among allBlacks; therefore riots would break out ran-domly, and when and where were governedonly by the number of available rioters.

Thus, the current data cannot provide forcomprehensive tests of theories about thecauses of racial rioting; but they can bemore appropriatelyapplied to the questionof which cities were more likely to experi-ence rioting given that the riot wave was oc-

curring. Conclusions drawn from this datamust take into account these limitations.Thus, my aim is to examine differences inlocal conditions and to determine if thesedifferences contributed in any systematicway to the pattern of rioting observed dur-ing the 1960s.2

DATA

The riot data were collected by Spilerman(1970b), who cataloged urban riots in theUnited States for the years 1961 through1968. He used these in three importantstud-ies that tested alternativeexplanations of riotfrequency and riot severity (Spilerman1970a, 1971, 1976). His original data weredrawn from a numberof sources, includingthe CongressionalQuarterly'sCivil DisorderChronology, the New York Times Index, the

Report of the National Advisory Commissionon Civil Disorders, and the Lemberg

Center's Riot Data Review (for details, see

Spilerman 1970a, 1970b). The 673 cities in-

cluded in the data are located in the contigu-ous United States and had populations of atleast 25,000 in 1960. For each city, theweeks in which a riot occurred are recordedfrom the beginning of 1961 throughthe endof 1968.

2

Although the data I analyze do not speak tolong-term cycles of collective behavior, evidenceis mounting indicating that competition processescontributeto long-term patterns of collective vio-lence (Olzak 1992). Olzak's work and the resultsof the present study suggest that competition ar-guments provide strong hypotheses for explain-ing long-term patterns of racial rioting.








150 134 141

z 100

0 5

0 ~~~~~~~~~~~~~36

.0 0

1961 1962 1963 1964 1965 1966 1967 1968

Year

Figure 1. Distribution of Riots by Year in 410 U.S. Cities: 1961 to 1968

For the current study,I furtherreduced theset of cities examined by eliminating thosethat had 1,000 or fewer non-Whites in 1960.My rationale for this procedure was that acritical mass of Blacks was necessary tocause rioting. Obviously, a city with noBlacks cannot experiencea spontaneous riot,and including cities with few or no Blackshas a powerful effect on the statistical rela-tionship between non-White population size

and racial rioting. Withno correction,a largeportion of the relationship between non-White population size and rioting would be

due to the fact that Blacks must be presentfor Black riots to occur. This ratheruninter-

esting finding would confound subsequentinterpretations of the effects of non-Whitepopulation size on rioting. A more interest-ing question asks whether increases in thenon-White population above an essentialminimum increase the likelihood of rioting.

Spilerman (1971) providesevidence suggest-ing that the critical numberof Blacks neededfor rioting in the 1960s was somewhere in

excess of 1,000. Therefore, I chose a non-

White population of at least 1,000 as thecritical criterion for including a city in the

study. I eliminated 263 cities by this proce-dure, and it is not surprising that none ofthese cities experienced a riot during the

study period.Riots included in Spilerman's (1970b) data

were those involving at least 30 people, thatresulted in violence or the destruction of

property, and that were not focused on insti-tutional conflicts (such as those in schoolsand union halls). Spilerman also classifiedriots into five types according to the race ofthe participants and whether the riot was

"spontaneous"or arose from a protest (e.g.,a civil rights demonstration).Because Spiler-man was primarily concerned with randomrather than coordinated rioting and becausehe wished to reduceheterogeneity among thetypes of riots, he limited his analysis to spon-taneousBlack aggression: This accounted for341 separate riot events from 1961 to 1968(see Spilerman 1970b for details). In the in-terest of comparabilitywith Spilerman's and

subsequent studies and because Spilerman'sdocumentation suggests that his tabulationsof the othertypes of riots may be incomplete,I also limit my analysis to spontaneous Blackaggression. Figure 1 presents a tabulation ofspontaneous Black riots by year in the 410U.S. cities in the study.

The Governmental Units Analysis Data,1960 (Alford and Aiken 1970) provides thecity-level economic, structural,and politicalinformation used as independent variables in

both this study and in Spilerman's studies.The sources of these data are documented inAlford and Aiken (1970) and include the1960 Census and The Municipal Yearbook.

ANALYSIS

Past Analysis of Racial Rioting

Since the eruption of racial rioting in the

1960s, research methods for examining riot

patterns have made importantadvances. Ef-forts prior to Spilerman's landmark studyhad employed pairings of similar cities, oneof which hadexperiencedrioting and anotherwhich had not (Lieberson and Silverman

1965). As riots occurred in more and morecities throughout he 1960s, adequatepairsof








riot and non-riot cities matched on regionand populationsize became increasingly dif-ficult to find. Furthermore, hese paired-city

studies allowed only a dichotomous classifi-

cation of cities (riot or non-riot), thereby ig-noring differences among cities in whichmore than one riot had occurred.In response

to these problems, Spilerman (1970a) took alarge step forwardwhen he used multivariate

analysis in which the number of riots in a

city was the dependent variable. Since that

study and his subsequentexamination of riot

severity (Spilerman 1976), most studies have

used some version of multivariatelinear re-gression to examine both riot occurrence and

riot severity (Carter 1986, 1990, 1992; Dot-son 1974; Jiobu 1971; McElroy and Singell1973).

Despite the advances of the multivariate

approach, the techniques used were not ide-ally suited for examining riot data. The firstproblem arises from using frequencies as the

dependent variable. While this approachin-cludes more informationthandoes a dichoto-mous indicator, it nonetheless represents theset of riots as more homogenous than it actu-ally is. In particular,any temporal relation-

ship within the riots is ignored.That is, a dif-

ferent underlyingprocess may operatefor ri-ots that are temporally clustered than for

those that are spread evenly across the time

period studied.A second importantshortcoming of previ-

ous analyses is their inability to introduce

covariates that change over time, a particu-larly difficult problem when attempting to

understanddiffusion processes (Olzak 1992).Faced with such a limitation, Spilerman(1970a) selected one majorriot incident and

attemptedto discern geographicdiffusion ef-

fects. Although his results did not supportgeographic diffusion, the method itself

makes the results unconvincing. In addition

to limiting the analysis to only one incident,Spilermanalso chose a large, nationally pub-licized riot (in Newark, New Jersey) as the

incident hypothesized to produce diffusion.

Because of the large amount of publicity as-sociated with this riot, geographic diffusion

effects would be much less apparentthan in

the case of riots that warranted ess national

publicity. Furthermore, Spilerman's proce-dure ignored the possibility that a riot otherthan a large, nationally publicized one could

have diffusion effects. In short, to draw a le-gitimate conclusion, a comprehensive analy-sis of the data must incorporateexplanatoryvariables that change as functions of both

geographic proximity and time.

Event HistoryAnalysis

In response to some of the shortcomings ofearlierapproachesto rioting, I apply survivalanalysis or event history analysis to the datadescribed above (Tuma and Hannan 1984;

Allison 1984;Yamaguchi 1991). Rather thandifferentiatingcities on the basis of the pres-ence or the number of riots, event historyanalysis focuses on the durationof time be-tween riot events for each individual city. Byapproaching the analysis of events in thisway, researchers can examine processes thatcluster differently in time and use time-vary-ing covariates to examine dynamic processes(see Olzak 1989a and 1992 for a completediscussion of using event history analysis toanalyze collective action events).3 Recently,Olzak (1989b,1990, 1992; Olzak and West1991) has successfully applied event historyanalysis to a number of different types ofcollective events, including laborstrikes,eth-nic violence, lynching and the establishmentof ethnic newspapers.

Formally, an event history analysis thatmodels an event that can occur at any pointin continuous time specifies the instanta-neous rate of transition from one state to an-other; it is defined as

A(t)= limP(t

?T<t+ At

It<T

At--->o At

where T is the time of the event and A(t) isthe hazard rate. In other words, the instanta-neous hazard rate (the hazard rate at a giveninstant) is a function of the probability thatan event will occur between t and t + At,given that it has not yet occurred at time t. Inthis study, this hazard rate is the instanta-neous probability that a riot will occur in a

given city, given that the city is not rioting as

it enters the time point in question.

3 As in many studies, limitations in the currentdata prevent broad application of time-varyingcovariates. Therefore, time-varying variables arelimited to the control variable for prior riotingand the variables critical to representing diffusionprocesses.








I model the hazardof rioting using the pro-

portional hazards model and estimate param-eters using the method of partial-likelihood(Cox 1972).4Parametersare estimated using

an ordinarymaximizationroutine such as theNewton-Raphson algorithmby maximizing

- S.

I exp[ J&Xik (ti )] j

PL Yk ---

IL Xexp[XfPkXk (0)]

where t1 s the time of occurrenceof either an

event or censoring, the Plsare coefficients es-timating the effects of the k hypothesizedex-

planatory variables, and 3i is a dummy vari-able that indicates whether the event associ-ated with case i was censored (8i = 0) or wasa failure (3i = 1). The coefficients generated

are most readily understoodby interpretingeP(the hazardratio) as the amountby whichthe hazard s multipliedforeach unit increasein the related independentvariable.

In event history analysis, the effects of un-observed heterogeneity must be carefullyconsidered (Allison 1984; Yamaguchi1991).Of particularconcern here is the lack of in-

dependence among repeated observations

made on the same city that occurs when all

sources of inter-city dependence are not ac-

counted for in the estimated models. Because

it is impossible to know if all relevant vari-

ables have been included, unobservedhetero-

geneity usually introduces a downward bias

in standarderror estimates. In the data stud-ied here, 76 cities experienced more than one

spontaneous Black-initiated riot (up to amaximum of 11 in Chicago), therefore the

present analysis is highly susceptible to this

problem.

Many methods have been suggested to cor-

rect for bias introduced by unobserved het-

erogeneity in survivalanalysis (Allison 1984;Flinn and Heckman 1982a, 1982b; Heckmanand Singer 1982, 1984; Yamaguchi 1986;Tuma 1985; Trussell and Richards 1985). Of

these, a numberof general methods require

assumptions either about the distribution ofunobserved characteristics or the functionalform of the baseline hazard; none of whichare tenable for the present analysis. There-

fore, I correct for unobserved heterogeneityby incorporatinga control variable:the num-ber of prior riots that have occurred in a city.Including variables that represent the priorhistory of the individual city is a practicalprocedurewhich minimizes the effects due tocorrelationwithin the same city and does notdemand the restrictive assumptions of moregeneral methods (Allison 1984).5

RESULTS

Spilerman'sAnalysis Revisited:Disorganization, Deprivation, andPolitical Structure

Spilerman's (1970a) analysis of riot fre-

quency led him to draw a number of conclu-sions aboutpropensities to riot. In summary,he found that non-White population size andregion (south versus non-south) were by farthe most powerful predictors of riot fre-quency.These two variables subsumed near-ly all the effects of the variables represent-ing social disorganization, deprivation, andpolitical structurearguments. He concludedthat ". . . the racial disturbances of the 1960s

were not responses to conditions in the local

community,"and that non-White populationsize was really the only relevant variable:"[T]he larger the Negro population, thegreater the likelihood of a disorder. Little

else appearsto matter"(p. 645).Tables 1 and 2 present analyses that paral-

lel those conducted by Spilerman (1970a).Table 1 reiterates a portion of Spilerman'sTable 6, reportingthe increments in the vari-ance of riot frequency explained. It also pre-sents an event history analysis analogous to

Spilerman's. I tested the same variables ac-

4 For the basics of survival analysis and par-tial-likelihood estimation see Tuma and Hannan

(1984), Cox (1972), Allison (1984), and Yama-

guchi (1991). Details on the application of eventhistory and partial likelihood to the riot data usedherein are available from the author.

5 False duration dependence due to differentrisk levels among unidentified strata within the

sample also can arise in survival analysis. How-ever, this problem is not relevant here because the

approach to repeated events does not allow themembership of the population atrisk to vary overtime. Details regarding this issue as well as alter-native methods considered for controlling unob-served heterogeneity in the riot data are availablefrom the author.








Table 1. A Comparison of the Results of Spilerman's Analysis with Results from Event History

Analysis

Regression Analysis of Riot CountsSpilerman (1970a) Event History Analysis

(1) (2) (3) (4)Likelihood

Likelihood Ratio TestPercent of Percent of Ratio Test ComparingVariance Variance Comparing Model with

Explained by Explained by Model with Non-WhiteNon-White Variable Cluster Variable Cluster PopulationPopulation when Entered and South to and South to

when Entered after Non- Same Model Same Modelafter Variable White Popu- Adding Non-White Adding

Cluster and lation and Population Variable ClusterVariable Cluster a South South x2 (d.f.) x2 (d.f.)

(A) Social disorganization 43.1 2.3 179.32*** (1) 8.33* (3)

(B) Absolute deprivation 40.6 .1 187.28*** (1) 2.42 (4)

(C) Relative deprivation 27.7 1.3 126.04*** (1) 2.68 (5)

(D) Political structure 19.2 .9 137.58*** (1) 12.95* (4)

All variables except 9.3 4.5 70.04*** (1) 30.86* (16)non-White population

a See the variables listed in Table 2 for indices that Spilerman included in each variable cluster. Numbersin parentheses are degrees of freedom.

* .05 **< .01 ***< .001

cording to their contribution to the hazardrate of rioting; the values reported are incre-ments in chi-squares resulting from a likeli-hood ratio test comparing two models. Al-though minor differences exist, an obviousparallel occurs between these results andSpilerman's regression analysis. The incre-ment gained by addingnon-Whitepopulationto each variable cluster is large (columns 1

and 3) compared with the increment gainedby adding the variable cluster after non-White population size (columns 2 and 4).Furthermore,while two variableclusters pro-duce significant gains when added after thesouth dummy variable and non-Whitepopu-lation, the improvement is marginal.

Despite their seeming clarity, interpreta-tion of these results is problematic becauseof collinearities among the variables. In

Spilerman's analysis, 51.3 percent of the

variance in riot frequency was explained bythe four variable clusters in Table 1, and an

additional 9.3 percent of the variance wasexplained by non-White population size.Thus 42 percent of the variance in riot fre-quency was explained by the four clusters ofvariables. Takentogether with the result that

46.8-percent of the variance was explainedby only non-White population and the southdummy variable, these findings mean thatthere is a large overlap among non-Whitepopulation size, the south dummy, and thevariables in the four clusters. Thismulticollinearityis also apparent n the eventhistory analysis presented in Table 2. Mod-els run with the individual variable clusters

(Models A throughD) show many significantrelationships between the theorized vari-ables. As the models are combined andSpilerman's key variables are added, nearlyall of the theoretical variables become non-significant.

Due to these patterns of collinearities, re-sults relying on variance explained or com-parisons of models using likelihood ratiotests fail to provide convincing evidenceagainst the theories tested or against the no-

tion thatcommunityconditions contribute torioting. Rather than interpreting the differ-ences in communityconditions as negligible,a more reasonable interpretation s that dif-ferences are related to rioting, but that theconditions are related to the size of the non-White population. In fact, the relationships








Table 2. Partial-Likelihood Estimates Showing the Effects of Social Disorganization, Deprivation, and

Political Structure on Hazard Rates for Racial Riots: 410 U.S. Cities, 1961 to 1968

Model Model Model Model Model ModelIndependent Variable A B C D 1 2

South (vs. non-south) a - 1.17 ***

(.304)

Log non-White population .610***

(.074)

(A) Social Disorganization

Percent change in total -.008*** - -.005** -.004*population (.002) (.002) (.002)

Percent change non-White .138 .024 .466population (.699) (.637) (.689)

Percent dilapidated housing -.026* -.047** -.021

(.012) (.018) (.019)

(B) Absolute Deprivation

Percent of non-White males in -.014** -.002 -.002traditionally Black occupations b (.005) (.006) (.007)

Non-White male unemployment .033** - .019 .003rate (.011) (.017) (.019)

Non-White median family income .019* .032* .004(in 100s of dollars) (.008) (.014) (.017)

Non-White median education -.051 -.132 -.119(.056) (.079) (.085)

(C) Relative Deprivation and Socioeconomic Expectations

Percent of males in traditionally -.072* -.024 .050

Black occupations C (.033) (.049) (.056)

Income - 1.74** -1.42 -.030(.639) (1.21) (1.30)

Unemployment C - .067 -.018 -.090(.048) (.075) (.085)

Education - 1.21* - 1.21 -1.02

(.544) (.684) (.770)

Non-White population percentage - .281*** - .268*** --.057

(.038) (.048) (.070)

(D) Political Structure

Population (in 1,000s) per council -- .007*** .006*** -.002

member (.001) (.001) (.001)Percent of council elected at-large - .001 .001 .002

(.002) (.002) (.002)

Presence of non-partisan elections - .507*** .658*** .375*(.129) (.138) (.148)

Presence of mayor-council .693*** .632*** .155government (.137) (.140) (.157)

Previous riots (control) ,414*** ,413*** .374*** .279*** .244*** .180***

(.027) (.027) (.028) (.033) (.033) (.033)

Model chi-square (d.f.) 224.27 214.57 256.58 257.37 354.92 434.82

(d.f. = 4) (d.f. = 5) (d.f. = 6) (d.f. = 5) (d.f. = 17) (d.f. = 19)

Note: Numbers in parentheses under coefficients are standard errors. The four clusters of indices aretaken from Spilerman (1970a).

a "South" indicates Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Caro-

lina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia.

b Traditionally Black occupations are service, household, and laborers.C

Relative deprivation variables are calculated by dividing figure for non-Whites by figure for Whites.

n<.05 "n <.01 "p < .001 (two-tailed tests)








between the non-White population size andthe various city characteristics are them-selves of considerable theoretical interest.

In the end, the comparison reported in

Table 1 does little to either refute or validatethe theories tested. It is the patternof coeffi-cients associated with the theoretical vari-ables in Table 2 that provides evidenceagainst the structural and deprivation theo-ries examined by Spilerman. In each theo-retical cluster, some significant coefficientsare in the predicted direction, but others arenot. Across the four clusters approximatelyone-half of the theoretical variablesproducecoefficients with signs opposite those which

the theories would have predicted. Becausenone of the four theories finds reasonablesupport in the data, alternativeexplanationsmust be pursued.

Competition Models

To test intergroup competition perspectives,I adopted variables to indicate three centralnotions of competition models: economic af-fects of Blacks, general economic factors.and minority group migration. First, I in-clude non-White unemployment rates andraw numbers of non-Whites unemployed asindicators of labor marketcompetition out-comes that affect primarily Blacks. Bothvariables are predicted to have positive rela-tionships with riot risk. Although non-White

unemployment rate is often used in such

analyses, I introducethe absolute number ofunemployed non-Whites because it may have

different effects in cities where the unem-ployment rate is the same. Larger numbersof unemployed non-Whites may make the

unemployed segment of the groupmore vis-ible and thus exacerbatefeelings of competi-tion with other groups.

The general economic factors included arethe median manufacturing worker's wageand the overall unemploymentrate. Highermanufacturingwages signal a growing econ-

omy, which according to competition theory

would reduce competition and thereby re-duce intergroup collective conflict. Unem-

ployment indicators are predicted to have

positive effects on rioting.Minority groupmigrationis representedby

two variables, percent foreign-born and thechange in non-White population from 1950

to 1960. Both variables increase the level ofcompetition for jobs in labor market seg-ments occupied by Blacks and thus are pre-dicted to have positive effects on rioting.

Table 3 presents partial likelihood esti-mates of the effects of competition modelvariables on the hazard rates of racial riot-ing. Models A, B, and C estimate the effectsfor each respective variablecluster.The chi-squareassociated with each of these modelsis highly significant (p < .001), although theeconomic effects on Blacks (Model A) isclearly a morepowerful model thanthe othertwo. Although nearly all significant coeffi-cients are in the direction predicted by theintergroup competition approach, there isone notable exception: In Model B, themanufacturing wage was predicted to de-crease riot risk by alleviating competition,but the data indicate that higher wages actu-ally contributedto an increased risk of racialriots. All of the other significant indicatorstesting competition theory, including botheconomic andmigrationindicators,producedcoefficients that supportthe hypotheses.

Model 1 in Table 3 combines selectedvariables from Models A, B, and C and of-fers a significant improvement over eachsingle-cluster, as shown by the likelihood-ratio test (p < .001 in each case). The coeffi-cients indicate that each variable clusteradds explanatorypower above that achievedby Model A, B, or C alone. Two nonsignifi-cant variables, the non-White unemploy-ment rate and the change in non-Whitepopulation, are droppedfrom the analyses at

this point in the interest of parsimoniouspresentation. Model 2 indicates that drop-ping these two variables does not result in apoorer prediction of rioting rates comparedto Model I (Model I vs. Model 2 in Table 3is nonsignificant).6Therefore I use Model 2as the baseline against which to examine in-teraction effects between economic factorsand migrationfactors.

While most coefficients in Model 2 remainconsistent with Models A, B, and C, the co-

6 I conducted alternative analyses carrying for-ward the variables measuring non-White unem-ployment rate and the 1950-1960 change in non-White population through the remainder of themodels presented in this paper. Neither variableachieved significance any model.








Table 3. Partial-Likelihood Estimates Showing the Effects of Competition Variables on Hazard Ratesfor Racial Riots: 410 U.S. Cities, 1961 to 1968

Model Model Model Model Model ModelIndependent Variable A B C 1 2 3

(A) Economic Effects on Blacks

Non-White unemployment rate -.005 .002

(.012) (.014)

Ln of number of non-White .497*** .504*** .500*** .629***unemployed (in lOGOs) (.037) (.039) (.038) (.059)

(B) General Economic Factors

Median manufacturing wage .029* .028*** 029*** .0268**(in lOs of dollars) (.007) (.008) (.008) (.008)

Unemployment rate - .0838** -.088* -.085* -.090*

(.030) (.042) (.035) (.035)

(C) Minority Group Migration

Change in non-White population, -.574 .3131950-1960 (in 1,OOOs) (.628) (.560)

Percent foreign-born - 048*** .029 * .030** .026*(.010) (.01 1) (.01 1) (.01 1)

Ln non-White unemployed x - - - -.016**percent foreign-born (.0055)

Prior rioting (control) .214*** .430*** .399*** .188*** .188*** .205***

(.031) (.026) (.027) (.032) (.032) (.033)

Model chi-square (d.f.) 359.86 208.25 202.04 387.26 386.96 395.32

(d.f. = 3) (d.f. = 3) (d.f. = 3) (d.f. = 7) (d.f. =5) (d.f. = 6)

Likelihood Ratio Tests Chi-Square d.f.

Model A vs. Model 1 27.40** 4

Model B vs. Model 1 179.01*** 4

Model C vs. Model 1 185.22*** 4

Model 1 vs. Model 2 .30 2

Model 2 vs. Model 3 8.36*** 1

Note: Numbers in parentheses under coefficients are standard errors.

*p < .05 ** < .01 ***p< .001 (two-tailed tests)

efficient for the general unemployment ratebecomes negative in Model 2, indicating thatlower unemploymentleads to increased riot-ing. It is apparent hat the change in sign re-sults from adding the (In) number of non-White unemployed to Model B. Thus, the

coefficient for the general unemploymentrate in Model 2 represents its independenteffect when holding non-White unemploy-ment constant, an effect that must be duelargely to White unemployment.While gen-eral unemployment may produce the rela-tionship predicted by competition theory,

singling out White unemployment reveals acompletely different effect.

I also examined a series of interaction ef-fects suggested by competition theory. I ex-amine interactions between economic con-traction and migration because the effect of

migration on rioting should differ at varyinglevels of economic contraction. Specifically,when the economy is faring poorly, migra-tion into a city should have greater effects oncompetition and violence than when theeconomy is healthy. Thus, I predict a posi-tive coefficient for interactions between the








unemployment variables and the migrationvariables, and a negative coefficient for theinteraction between manufacturing wagesand the migration variables.

In general, the data do not support thetheory in this respect. Of the three possibleinteractions, two (unemploymentrate x per-cent foreign-born andmanufacturingwage xpercent foreign-born) do not achieve statisti-cal significance (results not shown). Further,the one significant interaction term (In num-ber of non-Whites unemployedx percentfor-

eign-born) is in the opposite direction of thatpredicted. The negative coefficient indicatesthat increases in the percentage of immi-

grants has a lower effect on rioting whennon-White unemployment is high and agreater effect when Black unemployment islow. The coefficients for the model contain-ing this significant interaction term is pre-sented in Model 3 in Table 3. Instead of theseeconomic and migration variablescompoun-ding each other, the effect of one appears toattenuate the effect of the other, thereby sup-pressing the main effects of both. In sum,however, the interaction effects I have exam-ined contribute little to the prediction or ex-

planation of racial rioting.

Diffusion of Rioting

Analyses of threediffusion variables arepre-sented in Table 4. First, c,, the spatial diffu-sion variable defined in equation 1, is incor-

porated to model the effects of spatial het-erogeneity among the riots occurring in the

previous week. If the diffusion hypothesis iscorrect, this diffusion variable should pro-duce a positive coefficient. That is, recentri-

oting in cities proximal to a given cityshould increase that city's hazard of experi-encing a riot. Second, the measure for na-

tional-level diffusion effects (Mti) simplycalculates the total number of riots nation-ally in the previous week. Again, if national-level diffusion is occurring, a positive coef-ficient will be observed. In addition, the

square of the national-level diffusion vari-able represents the tapering effect of addi-tional riots over time. A negative coefficientis expected when this term is added after the

unsquared national-level variable, meaningthat a high numberof recent riots would pro-duce a diminishing hazardfor rioting.

Table 4 presents results supporting diffu-sion predictions: All coefficients are in thepredicted directions. In Model 1, the posi-tive and significant spatial diffusion coeffi-

cient indicates that riots in the prior weekincrease the likelihood of riots in proximalcities and that this effect diminishes as thedistance increases. In Model 2, the national-level diffusion variable also has positive andsignificant effects. The number of riots inthe previous week increases the likelihoodof rioting in any given week. Also, the num-ber of riots in the previous week has a di-minishing effect as the number of riots inthe prior week increases, as evidenced by

the significant and negative coefficient asso-ciated with the squared national-level diffu-sion term, (Mt-1)2.

The natureof this particular data set sug-gests an alternative to the diffusion models Ioffer. As is apparent n Figure 1, the vast ma-jority of riot events during the study periodoccurredduring 1967 and 1968. Because ofthe unique character of this period and thespecific events that occurred then (such asthe Vietnam War and the assassinations ofMartin Luther King, Jr. and RobertKennedy),it is possible thatthe observed dif-fusion effect is merely a result of the tightclustering of riot events during a relativelyshort time period. To test for this possibility,I introduce a dummy variable indicatingwhether the time period for each observationfell in the 1967-1968 time period. Given thedistributionof the riotevents across the studyperiod, it is clear that the dummy variable

will be highly significant. Interestingly, itseffects on the diffusion variables are muchless dramatic.As Model 3 in Table 4 shows,when the 1967-1968 dummy is entered, thenational diffusion effect decreases slightlybut remains highly significant. And there isvirtually no effect on the spatial diffusionvariable.Despite the unique character of the1967-1968 period, clearly the diffusion pro-cess was operating.

What remains ambiguous to this point is

the relative importance of the competitionand diffusion variables when compared tothe influence of non-White population size.Thus, to complete the analysis, the size of thenon-White population must be introduced toevaluate its effects in conjunction with thosederived fromcompetition and diffusion argu-








Table 4. Partial-Likelihood Estimates Showing the Effects of Competition and Diffusion Variableson Hazard Rates for Racial Riots: 410 U.S. Cities, 1961 to 1968

Independent Variable Model 1 Model 2 Model 3 Model 4

Ln of number of non-White .617*** .625*** .644*** .482*unemployed (in 1000s) (.059) (.059) (.061) (.203)

Median manufacturing wage .026** .024** .021** .024**(in 100s of dollars) (.008) (.008) (.008) (.008)

Unemployment rate -.081 * -.101** -. 139*** -.113*

(.035) (.036) (.036) (.048)

Percent foreign-born .020 .023* .035** .034**

(.01 1) (.01 1) (.01 1) (.01 1)

Ln non-White unemployed x -.013* -.012* -.004 -.005percent foreign-born (.006) (.006) (.006) (.006)

Spatial diffusion (c,) .034*** .024** .025** .025**(.007) (.008) (.009) (.009)

National-level diffusion (M,-1) .295*** .170*** .170***

(.025) (.029) (.030)

National-level diffusion squared -.005*** -.003*** -.003***(M_ l)2 (.000) (.001) (.001)

Dummy indicating years 2.48*** 2.48***1967-1968 (.260) (.260)

Ln of non-White population .166

(.199)

Prior rioting (control) .202*** .132*** -.0470 -.047

(.033) (.036) (.042) (.042)

Model chi-square (d.f.) 412.49 525.98 629.86 630.5

(d.f. = 7) (d.f. = 9) (d.f. = 10) (d.f. = 11)

Likelihood Ratio Tests Chi-Square d.f.

Model 1 vs. Model 2 113.49 2

Model 2 vs. Model 3 103.88 1

Model 3 vs. Model 4 .69 1

Note: Numbers inparentheses under coefficients are standard errors.

*p < .05 p < .01 *** < .001 (two-tailed tests)

ments. If demographic aggregation of na-tional discontent is in fact operating, effectsof the non-White population size should ex-ist above and beyond any community char-acteristics that are not directly associatedwith non-White population size. In otherwords, if two cities have equal Black unem-

ployment, equal manufacturingwages, equi-valent regional histories, and so forth, butone city has a larger Black population thanthe other, this city should have a higher riskof rioting thanthe city with fewer Blacks.

Model 4 in Table 4 adds the naturalloga-rithm of the non-White population to Model

3. The results for variables, other than non-White population variable itself, are consis-tent with the previous three models. And theaddition of the logged size of the non-Whitepopulation, however, has no effect on thehazard of rioting.7 Therefore, although thevariable may still have effects collinear to the

competition and diffusion variables (as indi-cated by the attenuation of the coefficients

I This finding is not anartifact of the particularmodel. When added to any model beginning withModel 1 of Table 3, the logged size of the non-White population is not significant.








for non-White unemployment and generalunemployment), the demographic aggrega-tion argument is not supported here becauseno additional effect is shown.

SUMMARY AND DISCUSSION

Contraryto the conclusions drawnby Spiler-man (1970a, 1971, 1976) in his classic stud-

ies, I demonstrate that local. conditions did

indeed contributeto the occurrence of racialrioting in the 1960s. The propensity to riot

was a function of far more than simply the

number of Blacks available for rioting in aparticularcity. I find strong supportfor argu-

ments drawn from ethnic competition andconflict literatureand for the existence of riotdiffusion processes. Economic andmigrationpressures in U.S. cities created different lev-

els of intergroupcompetition, which in turncontributed to the chances of experiencing

racial riots. Once a riot broke out, that eventincreased the likelihood that otherriots wouldbreakout elsewhere in the near future-par-ticularly in cities geographically close to the

original riot. These resultsare robustacross avariety of models, including models contain-ing very conservative control variables.

The structural strain argumentsoriginallyexamined by Spilerman (1970a) again failed

to find support. The argumentsdo not fail,

however, because of the lack of significantrelationships between individual variables

and rioting. The variables associated with

structural strain arguments actually do have

importantrelationshipswith collective racial

violence. Although these relationships areoften obscured in multivariateanalysis dueto collinearity among the variables, this is

not the important finding-instead, it is the

lack of a coherent pattern among these vari-

ables relative to any of the tested theories

that stands out. Thus I do not conclude that

community characteristicshave no relation-

ships with rioting, but rather that the rela-

tionships which do exist do not form a co-

herent patternto supportthe specific theories

Spilermantested.

Diffusion

My conclusions regarding the diffusion of

racial rioting are quite straightforward, as

strong evidence emerged for both national

and regional diffusion processes. Althoughthe definitions of diffusion I used here arerobust, many questions remainregarding theexact nature of the diffusion process. First,

given that the process appears to trail offquickly over a two-week period, a more de-tailed approach that examines riot data on adaily ratherthan on a weekly level may pro-vide a fuller picture of this dynamic process.Models incorporatingvariousdecaying func-tions of time may also be useful. Second, al-though the measure for proximity I used herehas proven effective in other studies (Hed-strom 1994), future work should test mea-sures thatrecognize boundariesthat may fil-

ter or limit diffusion processes. Third, het-erogeneity within the diffusion phenomenonsuggests interesting possibilities for model-ing diffusion. What differences among riotscan cause them to contribute more or less tothe diffusion process? Size, severity, amountand type of media attention, and repressiveresponses are all possible riot characteristicsthat could affect the diffusion process. Fi-nally,advances in understanding iot and col-lective action diffusion processes will alsodepend on learning more about the differentcommunicationprocesses and networks thattransmitthe informationdriving the processitself. It may be that information communi-catedthroughacquaintancenetworks has dif-ferent effects at different rates than does in-formation gained through the mass media.Furthermore, he dynamic interaction of allthese possible factors suggests fascinatingcomplexities and hypotheses for futuremod-

els of collective violence diffusion.

Competition

The hypotheses derived from competitiontheory found support here, although the in-teractions suggested by the theory were notimportant. A city9s median manufacturingwage does not appearat first to supportcom-

petition notions. Because manufacturingwage is taken as an indicator of a healthy city

economy, it was predicted to reduce inter-group competition and thus to decrease col-

lective violence. The apparently contradic-

tory finding can be explained easily if it isunderstood how economic conditions mayeffect Blacks and Whites differently. Whilehigh manufacturingwages may indicate that








the economy is generally strong, labor mar-ket segregation causes a strong manufactur-ing sector to benefit Whites more thanBlacks. As Olzak (1992) and Thernstrom

(1973) point out, Whites (both foreign-bornand native-born) dominate skilled occupa-tions and expanding industries. Thus, whenmanufacturing wages are high, Whites dis-proportionately reap the benefits. Further-

more, Olzak's analysis supportsthe idea thatduring periods of expansion, foreign-bornWhites are able to move up andout of segre-

gated occupations, which has the effect ofincreasing the concentration of Blacks inlow-paid and low-prestige jobs. Thus, one

can see how general economic expansion,which might be expected to lower competi-tion, could actually increasethe likelihood ofcollective violence on the partof Blacks.

Second, while greater overall unemploy-ment would seem to indicate greater compe-tition for jobs, my analysis suggests thatBlack unemployment and White unemploy-ment have very different effects on rioting,particularly in segregated labor markets.While White unemploymentmay cause moreWhite aggression against Blacks, and Blackunemployment can be expected to causemore Black aggression against Whites, it ismuch less reasonable to expect that White

unemployment would cause Black aggres-sion. Thus, overall unemployment must bedecomposed by race to make sense of its ef-fects on the 1960s rioting. If unemploymentis viewed as an outcome of losing to corn-petitors in the marketplace, then Black un-

employment, not White unemployment,would be expected to increase Black collec-tive violence.

Non-White Population Size and

Unemployment

In this study,the combination of competitionand diffusion provides a model that most co-

herentlyaccounts for differences in riot risks

among U.S. cities in the 1960s. This model

is superior to models using structural-strainvariables and to a model dependent only on

non-White populationsize. The competition/diffusion model is superiorto the non-White

population size argument because it illumi-

nates theoretical explanations for rioting.The notion of demographic aggregation or

opportunity (via non-White population size)as the sole source of rioting variability is notsupported because increases in populationsize do not increase riot propensity beyond

that of other explanatory variables. Clearly,the argument that 1960s rioting was an es-sentially random process once the demo-graphic aggregation of potential rioters wascontrolled is not adequate.

A detailed stepwise examination of Model4 in Table 4 reveals that one of the main vari-ables superseding the size of the non-Whitepopulation variable is the number of non-Whites unemployed in a given city. In thisanalysis, regardless of the percent unem-

ployed, it is the numberunemployed that hasthe most powerful effect on riot rates. Threetempting interpretations come to mind. Thefirst is that the non-White unemploymentrate is simply a proxy for the non-Whitepopulation size. While this interpretationispartiallytrue andclearly plausible in the caseof cities with similarunemployment rates butdifferent numbers of Blacks (those withhigher populations will experience higherrates of rioting), the interpretation breaksdown when Black unemploymentrates vary.Under such circumstances, cities with largerBlack populations and lower unemploymentrates could have lower raw numbers of un-

employed Blacks thansome cities with smal-ler Black populations and higher Black un-employment rates. Given the results pre-sented here, the cities with larger Black

populations would experience less riotingthan cities with smaller Black populations, a

result which is exactly the opposite of whatwould be predictedfrom the proxy argument.

The position I take, however, is not that thesize of the Black population is unimportantto riot rates. In fact, the models presented inTable4 demonstratethat the number of non-Whites unemployed and the size of the non-White populationarecollinear with regardtotheir ability to predict rioting. Nevertheless,the parameterestimate for the numberof un-

employed non-Whites is significant, indicat-

ing that there are important differences inriot rates that cannot be explained simply bythe overlap between these two variables.

A second interpretationof the relationshipbetween the numbers of non-White unem-

ployed and riot rates is that rioting requiresbiographical availability, and the unem-








ployed, given the lack of demands on their

time, are the most available segment of thepopulation. While biographical availabilityhas proven to be an extremely useful variable

in social movement and collective behaviorresearch (particularlywith regard to studentprotest movements), it is less useful in this

context. Many studies conducted on riot par-ticipants have found no significant differ-ences in employment status between rioters

and non-rioters. For example, Opp (1989)

found that unemploymenthad no relationship

to illegal protest and a negative relationship

to legal protest. Similar relationships havebeen found in a numberof surveys of partici-

pants and nonparticipantsof the 1960s riots(Rodgers and Bullock 1974; Caplan 1970;Mason and Murtagh 1985). Furthermore,many studies have established that urban ri-ots occur "attimes when (e.g., after workinghours or on weekends) people are available"

(McPhail 1994:9; also see McPhail and Mil-ler 1973; Burbeck, Raine, and Stark 1978).Because the unemployed are not constrainedby working hours, their availability for riot-

ing is continual.Therefore, f the unemployedwere largely responsible for rioting, this tem-

poral patternof rioting would not exist.A third explanation of the importance of

non-White unemploymentto rioting reasons

that the unemployed are the most discontent

and therefore would be the more likely thanthe employed to lash out in violence. Againevidence that the unemployed no more con-

stitute the body of rioters than they do thegeneral population contradicts this explana-

tion. Furthermore, few studies have estab-lished any attitudinal differences betweenthose who participatein riots and those who

do not (Rodgers and Bullock 1974).In short, the reason that the numberof un-

employed is so importantto riot rates is notpresently clear. It may be that even thoughthe unemployed are not overrepresentedamong riot participants, they may be over-

represented among those who initiate rioting.

If this were true, biographical availability,

severe discontent, or both could be operat-ing. On the other hand, it may be thathighernumbers of unemployed Blacks make

Blacks' economic problems highly visible

and thus highly salient. Such a condition

would increase the likelihood of riots inde-

pendent of biographical availability or par-

ticipation by the unemployed specifically. Atpresent, available empirical data are not ad-equate to assess these two possibilities. Thisstudy, however, strongly suggests that the

role of unemployment is one key to under-standing riot outbreaks, and that further re-search should focus on determining exactlyhow unemployment contributes to civil un-rest.

Daniel Myers is a Ph.D. candidate in Sociologyat the University of Wisconsin-Madison. His re-search interests include collective violence, for-inal models of collective action, game theory,andmedia coverage of protest activity. His disserta-

tion research focuses on developing and testingmodels of the diffusion of collective violence.

REFERENCES

Alford, Robert and Michael Aiken. 1970. Gov-ernmental UnitsAnalysis, 1960 [MRDF]. Uni-versity of Wisconsin, Data and Program Li-brary Service, Madison, WI.

Allison, Paul D. 1984. Event History Anahlysis:Regression for Longitudinal Event Data. New-bury Park,CA: Sage Publications.

Berkowitz, Leonard. 1968. "The Study of UrbanViolence: Some Implications of LaboratoryStudies of Frustrationand Aggression." Ameri-can Behavioral Scientist 2:14-17.

Bonacich, Edna. 1972. "A Theory of Ethnic An-

tagonism: The Split Labor Market."AmericanSociological Review 37:547-59.

Brown, Lawrence A. 1981. InnovationDiffusion:A New Perspective. London, England: Meth-uen.

Bryan, Marguerite. 1979. "The Social Psychologyof Riot Participation."Research in Race and

Ethnic Relations 1:169-89.Burbeck, Stephen L., Walter J. Raine, and M. J.Abudu Stark. 1978. "The Dynamics of RiotGrowth:An Epidemiological Approach."Jour-nal of Mathematical Sociology 6:1-22.

Caplan, Nathan. 1970. "The New Ghetto Man: AReview of Recent Empirical Studies." Journalof Social Issues 26:59-73.

Carter,Gregg Lee. 1986. "The 1960s Black RiotsRevisited: City Level Explanationsof Their Se-verity." Sociological Inquiry 210-28.

1990. "Collective Violence and the Prob-

lem of Group Size in Aggregate-Level Stud-ies." Sociological Focus 23:287-300..1992. "HispanicRioting Duringthe Civil

Rights Era."Sociological Forum 7:301-22.

Cox, D. R. 1972. "Regression Models and Life

Tables." Journal of the Royal Statistical Soci-ety Series B 34:187-202.

Diekmann, Andreas. 1979. "A Dynamic Stochas-








tic Version of the Pitcher-Hamblin-MillerModel of 'Collective Violence."' Journal of

Mathematical Sociology 6:277-82.Dotson, A. Bruce. 1974. "Social Planning and

Urban Violence: An Extension of McElroy and

Singell." UrbanAffairs Quarterly9:283-301.Downes, Bryan T. 1968. "Social and Political

Characteristicsof Riot Cities: A ComparativeStudy." Social Science Quarterly49:504-20.

Flinn, Christopher J. and James J. Heckman.

1982a. "Models for the Analysis of LaborForce Dynamics." Advances in Econometrics

1:35-95.. 1982b. "New Methods for Analyzing In-

dividual Event Histories." Sociological Meth-

odology 11:99-140.

Gurr,Ted R. 1968. "UrbanDisorder: Perspectivesfrom the Comparative Study of Civil Strife."American Behavioral Scientist 2:50-55.

. 1970. Why Men Rebel. Princeton, NJ:Princeton University Press.

Hamblin, Robert L., R. Brooke Jacobsen, and

Jerry L. L. Miller. 1973. A Mathematical

Theory of Social Change. New York: John

Wiley and Sons.Heckman, James J. and Burton Singer. 1982.

"The IdentificationProblemin Economic Mod-els for Duration Data."Pp. 35-95 in Advances

in Econometrics,editedby J. Hildebrand.Cam-bridge, England: Cambridge University Press.. 1984. "A Method for Minimizing the Im-

pact of Distributional Assumptions in Econo-

metric Models for Duration Data." Econo-

metrica 52:271-320.

Hedstr6m, Peter. 1994. "Contagious Collectivi-

ties: On the Spatial Diffusion of Swedish Trade

Unions, 1890-1940." American Journal of So-

ciology 99:1157-79.Heskin, Ken. 1985. "Political Violence in North-

ern Ireland."Journal of Psychology, 119:481-

94.Jenkins, J. Craig and Craig M. Eckert. 1986.

"ChannelingBlack Insurgency:Elite Patronageand Professional Social Movement Organiza-tions in the Development of the Black Move-

ment." American Sociological Review 51:812-

29.Jiobu, Robert M. 1971. "City Characteristicsand

Racial Violence." Social Science Quarterly55:52-64.

Kelly, William R. and Larry Isaac. 1984. "The

Rise and Fall of Urban Racial Violence in the

U.S.: 1948-1979." Research in Social Move-ments Conflict and Change 7:203-33.

Koomen, Willem and ErikG. Frankel. 1992. "Ef-

fects of Experienced Discrimination and Dif-

ferent Forms of Relative Deprivation amongSurinamese, a Dutch Ethnic Minority Group."Journal of CommunityandApplied Social Psy-chology 2:63-7 1.

Koopmans, Ruud. 1993. "The Dynamics of Pro-test Waves: West Germany, 1965 to 1989."American Sociological Review 58:637-58.

Lichbach, Mark. 1985. "Protest: Random or Con-tagious: The PostwarUnited Kingdom."ArmedForces and Society 11:581-608.

Lieberson, Stanley. 1980. A Piece of the Pie:Blacks and WhiteImmigrants since 1880. Ber-keley, CA: University of California Press.

Lieberson, Stanley and Arnold R. Silverman.1965. "The Precipitants and Underlying Con-ditions of Race Riots." American SociologicalReview 30:887-98.

Mahajan, Vijay and Robert A. Peterson. 1985.Models for InnovationDiffusion. Beverly Hills,CA: Sage Publications.

Mason, David T. and Jerry A. Murtagh. 1985."Who Riots? An EmpiricalExaminationof the'New UrbanBlack' versus the Social Margin-ality Hypotheses." Political Behavior 7:352-73.

McAdam, Doug. 1982. Political Process and theDevelopment of Black Insurgency, 1930-1970.Chicago, IL: University of Chicago Press.

McElroy, Jerome L. and Larry D. Singell. 1973."Riot and Nonriot Cities: An Examination ofStructuralContours." UrbanAffairs Quarterly8:28 1-302.

McPhail, Clark. 1991. The Myth of the MaddingCrowd. New York: DeGruyter.. 1994. "Presidential Address-The Dark

Side of Purpose:Individual andCollective Vio-lence in Riots." The Sociological Quarterly35:1-32.

McPhail, Clark and David L. Miller. 1973. "The

Assembling Process: A Theoretical and Em-

pirical Examination." American SociologicalReview 38:721-35.

McPhail, Clark and Ronald T. Wohistein. 1983."Individual and Collective Behavior within

Gatherings, Demonstrations and Riots." An-nual Review of Sociology 9:579-600.

Morris, Aldon D. 1984. The Origins of CivilRights Movement: Black Communities Orga-nizing for Change. New York: Free press.

National Advisory Commission on Civil Disor-ders. 1968. Report of the National AdvisoryCommission on Civil Disorders. New York:BantamBooks.

Nielsen, Francois. 1985. "Toward a Theory of

Ethnic Solidarity in Modern Societies." Ameri-can Sociological Review 50:133-49.

Oberschall, Anthony. 1978. "The Decline of the1960s Social Movements." Research In Social

Movements, Conflict and Change 1:257-89.

Oliver, Pamela E. 1989. "Bringing the Crowd

Back In: The Nonorganizational Elements of

Social Movements."Research in Social Move-ments, Conflict and Change 11:1-30.

Olzak, Susan. 1987. "Causes of Ethnic Protest








and Conflict in Urban America, 1877-1889."Social Science Research 16:185-210.

. 1989a. "Analysis of Events in the Study

of Collective Action." Annual Review of Soci-

ology 15:119-41.. 1989b. "LaborUnrest, Immigration, and

EthnicConflict in UrbanAmerica, 1880-1914."

American Journal of Sociology 94: 1303-33.

. 1990. "The Political Context of Competi-

tion: Lynching and Urban Racial Violence,1882-1914." Social Forces 69:395-421.

. 1992. The Dynamics of Ethnic Competi-

tion and Conflict. Stanford, CA: StanfordUni-

versity Press.

Olzak, Susan andJoane Nagel. 1986. CompetitiveEthnic Relations. Orlando, FL: Academic.

Olzak, Susan andSuzanne Shanahan.1996. "Dep-rivation Race Riots: An Extension of Spiler-

man's Analysis." Social Forces 74:931-61.Olzak, Susan and Elizabeth West. 1991. "Ethnic

Conflict and the Rise and Fall of Ethnic News-

papers." American Sociological Review 56:

458-74.Opp, Karl-Dieter. 1989. TheRationality of Politi-

cal Protest: A ComparativeAnalysis of Ratio-

nal Choice Theory. Boulder, CO: Westview.Paige, Jeffrey M. 1971. "PoliticalOrientationand

Riot Participation."American Sociological Re-

view 36:810-20.Park, Robert. 1950. Race and Culture. Glencoe,

IL: Free Press.Pitcher, Brian L., Robert L. Hamblin, and Jerry

L. L. Miller. 1978. "Diffusion of Collective

Violence." American Sociological Review 43:

23-35.Polletta, Francesca. 1992. "Politicizing Child-

hood: The 1980 Zurich Burns Movement."So-

cial Text 33:82-102.Rodgers, Harrell R. and Charles S. Bullock III.

1974. "Political and Racial Attitudes: Black

versus White."Journal of Black Studies 4:463-85.

Rogers, Everett M. 1983. Diffusion of Innova-

tions. New York: Free Press.

Rud6, George. 1964. The Crowd in History: A

Study of Popular Disturbances in France and

England, 1730-1848. New York: Wiley..1972. Europe in the Eighteenth Century:

Aristocracy and the Bourgeois Challenge. New

York: Praeger.Schneider, Hans Joachim. 1992. "Criminology of

Riots."InternationalJournal of OffenderTher-

apy and Comparative Criminology 36:173-86.Shorter, Edward and Charles Tilly. 1974. Strikes

in France: 1830-1968. New York: Cambridge

University Press.

Smelser, Neil. 1962. Theory of Collective Behav-

ior. New York: Free Press.

Spilerman, Seymour. 1970a. "The Causes of Ra-cial Disturbances: A Comparison of Alterna-tive Explanations."American Sociological Re-view 35:627-49

. 1970b. Governmental Units AnalysisData, 1960: Urban Racial Disorders, 1961-1968 [MRDF]. University of Wisconsin, Dataand Program Library Service, Madison, WI.

. 1971. "The Causes of Racial Distur-bances: Test of an Explanation."American So-ciological Review 36:427-42.

. 1972. "Strategic Considerations in Ana-lyzing the Distributionof Racial Disturbances."American Sociological Review 37:493-99.

. 1976. "StructuralCharacteristics of Cit-ies and the Severity of Racial Disorders."

American Sociological Review 41:771-93.Strang, David. 1991. "Adding Social StructuretoDiffusion Models: An Event History Frame-work." Sociological Methods and Research19:324-53.

Strang, David and Nancy Brandon Tuma. 1993."SpatialandTemporal Heterogeneity in Diffu-sion." American Journal of Sociology 99:614-39.

Thernstrom,S. 1973. The Other Bostonians: Pov-erty and Progress in the American Metropolis,1880-1970. Cambridge,MA: HarvardUniver-

sity Press.Tilly, Charles, Louise Tilly, and Richard Tilly.1975. The Rebellious Century: 1830-1975.Cambridge, MA: HarvardUniversity Press.

Tolnay, StewartE. and E. M. Beck. 1995. A Fes-tival of Violence: An Analysis of Southern

Lynchings, 1882-1930. Urbana, IL: Universityof Illinois Press.

Tomlinson, T. M. 1968. "The Development of a

Riot Ideology among Urban Negros." Ameri-

can Behavioral Scientist 11:27-3 1.

Trussell, James and Toni Richards. 1985. "Cor-

recting for UnmeasuredHeterogeneity in Haz-ard models Using the Heckman-Singer Proce-

dure."Sociological Methodology 14:242-76.

Tuma, Nancy Brandon. 1985. "Effects of LaborMarket Structure on Job-Shift Patterns." Pp.327-363 in Longitudinal Analysis of Labor

Market Data, edited by J. J. Heckman and B.

Singer. Cambridge, England: Cambridge Uni-

versity Press.Tuma,. Nancy Brandon and Michael T. Hannan.

1984. Social Dynamics: Models and Methods.San Diego, CA: Academic.

Yamaguchi, Kazuo. 1986. "Alternative App-roaches to Unobserved Heterogeneity in the

Analysis of Repeatable Events." SociologicalMethodology 16:213-49.

. 1991. Event History Analysis. NewburyPark,CA: Sage.

racial rioting in the 1960's: an event history analysis of local conditions

Documents