Stop, Frisk, and Assault? Racial Disparities in Police Use of Force During Investigatory Stops

Rory Kramer Brianna Remster

Black civilians are more likely to be stopped by police than white civilians net of relevant factors. Less is known about whether or not racial inequalities exist in police use of force during stops. Using data on over 2 million police stops in New York City from 2007 to 2014 and drawing on literatures on race, policing, and the Black Lives Matter movement, we test hypotheses regarding the associations between race, civilian behavior, age, and police use of force. We also investigate whether recent reforms reduced any observed inequality in police violence during stops. Findings show that Black and White civilians experience fundamentally different interactions with police. Black civilians are particularly more likely to experience poten- tial lethal force when police uncover criminal activity and this disparity is greatest for black youth compared to white youth. Overall, if there were no racial disparities in police use of force, we estimate that approximately 61,000 fewer stops of black civilians would have included police use of force and 1,000 fewer stops would have included potential lethal force from 2007 to 2014. Furthermore, while reform efforts substantially reduced the num- ber of stops annually, inequalities in police use of force persist.

Police make contact with nearly 44 million Americans annually in the United States (Hyland et al. 2015). While the overall rate of contact remained stable from 2002 to 2011, urban residents around the country experienced a substantial increase in investi- gative police stops, known as stop-and-frisks. In New York City specifically, the number of stop-and-frisks increased threefold from 2003 to 2009 and were disproportionately concentrated among racial and ethnic minorities (Meares 2014). Indeed, black NYC residents are approximately 2.5 times more likely to be stopped than white residents, net of germane factors including neighborhood context and crime rates (Gelman et al. 2007). Yet beyond the act of being stopped, less is known about whether inequality exists in terms of what happens once individuals are stopped.

As the state’s legitimized form of physical coercion over citi- zens, racial disparities in police use of force are perhaps one of

We thank Laurie Krivo, Chris Smith, and members of the Penn and Rutgers Sociology Colloquia for helpful feedback and Denise Wilson for research assistance.

Please direct all correspondence to Rory Kramer, Department of Sociology and Crimi- nology, 800 Lancaster Avenue, Villanova, PA 19085. rory.kramer@villanova.edu.

Law & Society Review, Volume 52, Number 4 (2018) © 2018 Law and Society Association. All rights reserved.

960

the most extreme examples of racial inequality. This is, in part, why accusations of racial bias in police use of force have been and continue to be a common focal point of civil unrest in the United States. From the 1960s in cities such as Detroit, Philadelphia, and Chicago, to the recent protests that coalesced under the #BlackLi- vesMatter moniker in response to the deaths of young black vic- tims such as Michael Brown in Ferguson, Freddie Gray in Baltimore, and Laquan McDonald in Chicago, and others, accusa- tions of unjustified use of force against black victims persist. Indeed, recent investigations by the Department of Justice into those cities found evidence of civil rights violations by police, as did investigations into Albuquerque, Cleveland, and Seattle police among others. Protesters assert that the well-documented racial inequalities in the likelihood of being stopped are exacerbated by policing bias in the likelihood that force is used during stops, and that the bias is particularly harmful for black youth. Public dis- course focuses on civilian behavior during police encounters, with many suggesting that black people are more likely to be doing something wrong at the time than white people, thus precipitating police use of force. In this scenario, civilian behavior, not racial bias, is thought to drive police use of force. Despite these compet- ing explanations for police use of force, no systematic research testing these propositions exists.

The dearth of research on this topic is in part due to data limi- tations; however, this has recently begun to change. Since the 1990s, data collection by police has become increasingly common, but agencies only began disseminating data in the last few years. The New York Police Department (NYPD) was among the first to publicly release detailed data on investigatory stops as part of a legal settlement (Daniels et al. v. City of New York 1999). Subsequent analyses of this data helped convince a federal judge to declare NYPD’s stop-and-frisk policy unconstitutional due to racial bias (Floyd v. City of New York 2013). Even before that ruling, New York began to roll back the use of stop-and-frisk. Stops dropped from a high of 685,724 in 2011 to under 50,000 in 2014. While critics considered NYPD’s shift away from stop-and-frisk and the court’s ruling to be monumental victories, it is unknown whether this dramatic drop reduced racial inequality in police violence during stops.

We fill these voids in existing research, focusing specifically on New York City. We test the claim, re-energized by the Black Lives Matter movement, that black civilians, especially black youth, are more likely to be subject to physical force during a police encoun- ter than white civilians, after adjusting for other factors related to police use of force. We also examine whether or not black individ- uals are more likely to experience police violence during stops

Kramer & Remster 961

that end in arrest and/or the recovery of contraband or a weapon than whites, as criminal behavior is a common alternative expla- nation for high profile instances of police use of force against black civilians. Additionally, we assess whether recent NYPD reforms to the use of investigatory stops as a policing practice and changes in officer training affect any observed inequalities in police use of force.

New York City is a compelling research setting because it is widely viewed as a model for proactive policing. In response to a crime wave in the early 1990s, the NYPD implemented an espe- cially visible aggressive stop-and-frisk policy which was then expanded into the next decade. Moreover, there is little reason to expect New York to be an outlier in the broader pattern of police use of force in the United States. In fact, given that the NYPD was subject to some of the strongest early contemporary critiques of racial discrimination after the police shooting of Amadou Diallo in 1999, which the City settled via lawsuit, and that New York was under judicial oversight for racial disparities in stop-and-frisk dur- ing our observation period, the city may represent a conservative test of racial disparities in police use of force. On the other hand, New York was also the site of several large Black Lives Matter pro- tests after Eric Garner’s death and the subsequent acquittal of Officer Pantaleo. Regardless, a recent report analyzing police use of force in a multicity sample found similar patterns across juris- dictions ranging in size, demographics, and region. Although cit- ies differ in the degree of racial inequality in police use of force, there is a general pattern of racial inequality across localities (Goff et al. 2016).

Prior Research on Police Stops and Use of Force

As the legal rulings regarding stop-and-frisk focused on its efficacy in finding weapons and preventing crime, so too has most of the research (for a review, see Meares 2014). However, the original case involved police using force against an individual (Terry) who had not yet performed a criminal act (he was prepar- ing to commit armed robbery when stopped). Stuntz (1998) argues that the Supreme Court and researchers mistakenly focus on the legitimacy of the search and not when police use force against individuals. Stuntz’s critique remains apt today, particu- larly because foundational scholars conceptualize police capacity for use of force as the defining characteristic of police work (Bittner 1970).

Although deadly police encounters are rare compared to police use of force in general, public interest may have steered

962 Racial Disparities in Police Use of Force

researchers toward the most severe form of force. In the broadest sense, research finds that communities with greater racial inequal- ity have higher rates of lethal violence by police (Jacobs and O’Brien 1998, but for a recent exception, see Klinger et al. 2016). Although recently, in part due to research showing that structural factors explain only a portion of police behavior and the rise in public awareness of specific incidents of police use of force, researchers shifted focus to police use of lethal force at the indi- vidual level.

These latest studies, which adapt data from the Bureau of Jus- tice Statistics and/or crowdsourced data, show racial inequalities in who is killed by police (e.g., Guardian n.d.; mappingpoliceviolence. org, n.d.). For example, one study finds that unarmed black people are 3.5 times more likely to be shot by police than unarmed whites (Ross 2015). Such research is useful for determining the scope of the issue, but conclusions are limited. For example, without a com- parison group such as nonfatal encounters, scholars cannot identify how these fatal shootings differ from other police encounters. Moreover, as these data do not contain many details of the encoun- ter, it is impossible to ascertain whether black individuals are more likely to be shot because of racial bias or due to some other reason. Indeed, there are a variety of competing explanations for why racial disparities may exist other than racial bias. While data limita- tions have hampered empirical work, scholars have long been theo- rizing about why observed disparities in police use of force might occur, highlighting a series of behavioral, contextual, and organiza- tional factors (Engel and Calnon 2004; Friedrich 1980; Geller and Toch 1996; Smith 1986; Smith and Alpert 2007).

A recent study that adjusted for a range of these factors did not find racial disparities in police shootings in Houston, but did find disparities in police use of force in general (Fryer 2016). However, Fryer compared police shootings to police encounters involving arrests in which police could have been legally justified in shooting (e.g., a person resisting arrest) and research suggests that these arrests are racialized; officers are more likely to per- ceive people of color as verbally abusive or noncompliant than whites (Geller and Fagan 2010). As a result, Fryer’s analytic strat- egy likely obscured any racial disparity in police shootings.

Unfortunately, data limitations such as those described above are not new, rather, studying police use of force has always been challenging. Because of this, most of the literature debates how to measure it and achieve representative estimates of use of force (see Garner et al. 2002; Geller and Toch 1996). For instance, most early work relied on trained observers accompanying officers on their shifts and recording any use of force, which is susceptible to bias (e.g., Terrill and Mastrofski 2002). Similarly, other work relies

Kramer & Remster 963

on surveys of individuals in custody (Garner et al. 1995, 2002). Additional research using surveys of the public or data collected by the police also lack enough information to rule out competing explanations (Engel and Calnon 2004; Goff et al. 2016). Also, aside from crowdsourced studies, most research to date relies on data collected in the 1990s. Taken together, these limitations may explain why previous research on racial disparities in police use of force has historically been mixed (Goff et al. 2016; Sun and Payne 2004; Terrell and Mastrofski 2002). The current study builds on previous work by (1) using contemporary data to assess disparities in police use of force, (2) using a comparison group, and (3) adjusting for competing explanations that prior research has not adequately considered, such as civilian behavior. We detail these explanations next.

Contextual Differences

A common rationale for racially unequal rates of police use of force is that black residents live in neighborhoods with higher poverty rates than white residents (Smith 1986). Given that the risk of police violence is higher in disadvantaged neighborhoods (Terrill and Reisig 2003), black people may be more likely to experience police violence than white people because of where they live. Here, structural factors should explain any racial dispar- ities in police use of force. In short, the combination of racial seg- regation and poverty concentration may explain any racial inequality in police shootings (Massey and Denton 1993). These race and class differences can also affect organizational strategies, such as whether or not to aggressively use stop-and-frisk as a policing practice, which could lead to higher or lower use of force during stops. Recent research finds that context did not explain the racial disparity, but rather moderated it. Consistent with racial threat theory, the racial disparity in police use of force is greatest in segregated precincts (Levchak 2017).

Behavioral and Situational Differences

Racial disparities in crime rates are perhaps the default expla- nation for any observed inequalities in policing (Goff et al. 2016; MacDonald 2011). For example, if police shootings occur at ran- dom during police interactions and black individuals are three times more likely to interact with police than white individuals because of a higher rate of involvement in violent crime (Sampson and Lauritsen 1997), we would expect black individuals to be three times as likely to be shot as whites. In this case, observed racial disparities in police shootings would be due to dif- ferential involvement in crime. As evidence, recent work shows

964 Racial Disparities in Police Use of Force

that violent crime rates broken down by race decreases but does not eliminate the racial disparity in police use of force (Goff et al. 2016).

Furthermore, civilian behavior during the stop may also explain racial differences in police use of force. One way civilian behavior may influence this is whether the stop is productive or not. A productive stop is one that results in an arrest or finds a weapon or other contraband (i.e., drugs). Because civilians involved in criminal activity may resist the stop in the hopes of avoiding sanctions, productive stops may require police use of force more so than unproductive stops. In other words, racial differences in the stopped person’s involvement in crime may explain any observed racial disparities. For instance, if black people are more likely to be arrested during investigatory stops, then accounting for the stop outcome might explain disparities in police use of force. On the other hand, the “hit rate” for find- ing weapons or drugs during a stop is lower for black than white civilians in NYC, thus, adjusting for the “success” of the stop could exacerbate any observed racial disparity in police use of force (Gelman et al. 2007). Unfortunately, data limitations pre- vent researchers from ascertaining the sequencing of police locating criminal activity and police use of force; police might exert force after finding a weapon or, use force and then dis- cover contraband.

Other scholars emphasize different forms of civilian behav- ior (Durna 2011; Friedrich 1980). This line of work suggests that any observed racial disparity in police use of force is due to the way black civilians behave compared to white civilians during a police encounter. Put differently, some research finds that black civilians are more hostile and noncompliant toward officers and such behaviors increase the likelihood of police use of force (Engel 2003; Garner et al. 2002). Similarly, officers are more likely to use force when a civilian is suspected of a violent crime, and given racial disparities in violent crime, black civilians may be more likely to be suspected of such than white ones (Worden 1996). In this case, unlike “hit rates,” police are more likely to report that black civilians were sus- pected of violent crime or noncompliant when stopped, thus accounting for these behaviors may decrease any observed racial disparity. Moreover, because these behaviors are racialized—officers employ racialized “scripts” for these kinds of behaviors and as a result are more likely to perceive black civilians as, for instance, noncompliant—adjusting for these behaviors will produce conservative estimates of any police vio- lence disparities (Geller and Fagan 2010; Muhammad 2010). Like with the outcome of the stop, ascertaining the timing of

Kramer & Remster 965

JOURNAL OF APPLIED BEHAVIOR ANALYSIS

SOME CURRENT DIMENSIONS OF APPLIED BEHAVIOR ANALYSIS’

DONALD M. BAER, MONTROSE M. WOLF, AND TODD R. RISLEY

THE UNIVERSITY OF KANSAS

The analysis of individual behavior is a problem in scientific demonstration, reason- ably well understood (Skinner, 1953, Sec. 1), comprehensively described (Sidman, 1960), and quite thoroughly practised (Journal of the Experimental Analysis of Behavior, 1957 -). That analysis has been pursued in many settings over many years. Despite variable precision, elegance, and power, it has resulted in general descriptive statements of mecha- nisms that can produce many of the forms that individual behavior may take. The statement of these mechanisms estab-

lishes the possibility of their application to problem behavior. A society willing to con- sider a technology of its own behavior appar- ently is likely to support that application when it deals with socially important behav- iors, such as retardation, crime, mental illness, or education. Such applications have ap- peared in recent years. Their current num- ber and the interest which they create appar- ently suffice to generate a journal for their display. That display may well lead to the widespread examination of these applica- tions, their refinement, and eventually their replacement by better applications. Better applications, it is hoped, will lead to a better state of society, to whatever extent the behav- ior of its members can contribute to the good- ness of a society. Since the evaluation of what is a “good” society is in itself a behavior of its members, this hope turns on itself in a philosophically interesting manner. However, it is at least a fair presumption that behav- ioral applications, when effective, can some- times lead to social approval and adoption.

Behavioral applications are hardly a new phenomenon. Analytic behavioral applica-

‘Reprints may be obtained from Donald M. Baer, Dept. of Human Development, University of Kansas, Lawrence, Kansas 66044.

91

tions, it seems, are. Analytic behavioral ap- plication is the process of applying sometimes tentative principles of behavior to the im- provement2 of specific behaviors, and simul- taneously evaluating whether or not any changes noted are indeed attributable to the process of application-and if so, to what parts of that process. In short, analytic be- havioral application is a self-examining, self- evaluating, discovery-oriented research pro- cedure for studying behavior. So is all experimental behavioral research (at least, according to the usual strictures of modern graduate training). The differences are mat- ters of emphasis and of selection. The differences between applied and basic

research are not differences between that which “discovers” and that which merely “ap- plies” what is already known. Both endeavors ask what controls the behavior under study. Non-applied research is likely to look at any behavior, and at any variable which may con- ceivably relate to it. Applied research is con- strained to look at variables which can be effective in improving the behavior under study. Thus it is equally a matter of research to discover that the behaviors typical of re- tardates can be related to oddities of their

2If a behavior is socially important, the usual be- havior analysis will aim at its improvement. The so- cial value dictating this choice is obvious. However, it can be just as illuminating to demonstrate how a behavior may be worsened, and there will arise occa- sions when it will be socially important to do so. Dis- ruptive classroom behavior may serve as an example. Certainly it is a frequent plague of the educational system. A demonstration of what teacher procedures produce more of this behavior is not necessarily the reverse of a demonstration of how to promote posi- tive study behaviors. There may be classroom situa- tions in which the teacher cannot readily establish high rates of study, yet still could avoid high rates of disruption, if she knew what in her own procedures leads to this disruption. The demonstration which showed her that would thus have its value.

1968, 1, 91-97 NUMBER I (SPRING, 1968)

DONALD M. BAER et al.

chromosome structure and to oddities of their reinforcement history. But (currently) the chromosome structure of the retardate does not lend itself to experimental manipulation in the interests of bettering that behavior, whereas his reinforcement input is always open to current re-design.

Similarly, applied research is constrained to examining behaviors which are socially im- portant, rather than convenient for study. It also implies, very frequently, the study of those behaviors in their usual social settings, rather than in a “laboratory” setting. But a laboratory is simply a place so designed that experimental control of relevant variables is as easy as possible. Unfortunately, the usual social setting for important behaviors is rarely such a place. Consequently, the analy- sis of socially important behaviors becomes experimental only with difficulty. As the terms are used here, a non-experimental anal- ysis is a contradiction in terms. Thus, ana- lytic behavioral applications by definition achieve experimental control of the processes they contain, but since they strive for this con- trol against formidable difficulties, they achieve it less often per study than would a laboratory-based attempt. Consequently, the rate of displaying experimental control re- quired of behavioral applications has become correspondingly less than the standards typi- cal of laboratory research. This is not because the applier is an easy-going, liberal, or gen- erous fellow, but because society rarely will allow its important behaviors, in their cor- respondingly important settings, to be manip- ulated repeatedly for the merely logical com- fort of a scientifically sceptical audience. Thus, the evaluation of a study which pur-

ports to be an applied behavior analysis is somewhat different than the evaluation of a similar laboratory analysis. Obviously, the study must be applied, behavioral, and ana- lytic; in addition, it should be technological, conceptually systematic, and effective, and it should display some generality. These terms are explored below and compared to the cri- teria often stated for the evaluation of behav- ioral research which, though analytic, is not applied.

Applied The label applied is not determined by the

research procedures used but by the interest

which society shows in the problems being studied. In behavioral application, the behav- ior, stimuli, and/or organism under study are chosen because of their importance to man and society, rather than their importance to theory. The non-applied researcher may study eating behavior, for example, because it re- lates directly to metabolism, and there are hypotheses about the interaction between be- havior and metabolism. The non-applied re- searcher also may study bar-pressing because it is a convenient response for study; easy for the subject, and simple to record and inte- grate with theoretically significant environ- mental events. By contrast, the applied re- searcher is likely to study eating because there are children who eat too little and adults who eat too much, and he will study eating in exactly those individuals rather than in more convenient ones. The applied researcher may also study bar-pressing if it is integrated with socially important stimuli. A program for a teaching machine may use bar-pressing be- havior to indicate mastery of an arithmetic skill. It is the arithmetic stimuli which are important. (However, some future applied study could show that bar-pressing is more practical in the process of education than a pencil-writing response.3)

In applied research, there is typically a close relationship between the behavior and stimuli under study and the subject in whom they are studied. Just as there seem to be few behaviors that are intrinsically the target of application, there are few subjects who auto- matically confer on their study the status of application. An investigation of visual signal detection in the retardate may have little im- mediate importance, but a similar study in radar-scope watchers has considerable. A study of language development in the re- tardate may be aimed directly at an immedi-

“Research may use the most convenient behaviors and stimuli available, and yet exemplify an ambition in the researcher eventually to achieve application to socially important settings. For example, a study may seek ways to give a light flash a durable conditioned reinforcing function, because the experimenter wishes to know how to enhance school children’s responsive- ness to approval. Nevertheless, durable bar-pressing for that light flash is no guarantee that the obvious classroom analogue will produce durable reading be- havior for teacher statements of “Good!” Until the analogue has been proven sound, application has not been achieved.

92

SOME DIMENSIONS OF APPLIED BEHAVIOR ANALYSIS

ate social problem, while a similar study in the MIT sophomore may not. Enhancement of the reinforcing value of praise for the re- tardate alleviates an immediate deficit in his current environment, but enhancement of the reinforcing value of 400 Hz (cps) tone for the same subject probably does not. Thus, a pri- mary question in the evaluation of applied research is: how immediately important is this behavior or these stimuli to this subject?

Behavioral Behaviorism and pragmatism seem often to

go hand in hand. Applied research is emi- nently pragmatic; it asks how it is possible to get an individual to do something effec- tively. Thus it usually studies what subjects can be brought to do rather than what they can be brought to say; unless, of course, a verbal response is the behavior of interest. Accordingly a subject’s verbal description of his own non-verbal behavior usually would not be accepted as a measure of his actual be- havior unless it were independently substan- tiated. Hence there is little applied value in the demonstration that an impotent man can be made to say that he no longer is impotent. The relevant question is not what he can say, but what he can do. Application has not been achieved until this question has been an- swered satisfactorily. (This assumes, of course, that the total goal of the applied researcher is not simply to get his patient-subjects to stop complaining to him. Unless society agrees that this researcher should not be bothered, it will be difficult to defend that goal as socially important.)

Since the behavior of an individual is com- posed of physical events, its scientific study requires their precise measurement. As a re- sult, the problem of reliable quantification arises immediately. The problem is the same for applied research as it is for non-applied research. However, non-applied research typi- cally will choose a response easily quantified in a reliable manner, whereas applied re- search rarely will have that option. As a re- sult, the applied researcher must try harder, rather than ignore this criterion of all trust- worthy research. Current applied research often shows that thoroughly reliable quantifi- cation of behavior can be achieved, even in thoroughly difficult settings. However, it also suggests that instrumented recording with its

typical reliability will not always be possible. The reliable use of human beings to quantify the behavior of other human beings is an area of psychological technology long since well developed, thoroughly relevant, and very often necessary to applied behavior analysis. A useful tactic in evaluating the behavioral

attributes of a study is to ask not merely, was behavior changed? but also, whose behavior? Ordinarily it would be assumed that it was the subject’s behavior which was altered; yet careful reflection may suggest that this was not necessarily the case. If humans are ob- serving and recording the behavior under study, then any change may represent a change only in their observing and record- ing responses, rather than in the subject’s be- havior. Explicit measurement of the reliabil- ity of human observers thus becomes not merely good technique, but a prime criterion of whether the study was appropriately be- havioral. (A study merely of the behavior of observers is behavioral, of course, but prob- ably irrelevant to the researcher’s goal.) Alter- natively, it may be that only the experimen- ter’s behavior has changed. It may be reported, for example, that a certain patient rarely dressed himself upon awakening, and conse- quently would be dressed by his attendant. The experimental technique to be applied might consist of some penalty imposed unless the patient were dressed within half an hour after awakening. Recording of an increased probability of self-dressing under these condi- tions might testify to the effectiveness of the penalty in changing the behavior; however, it might also testify to the fact that the patient would in fact probably dress himself within half an hour of arising, but previously was rarely left that long undressed before being clothed by his efficient attendant. (The at- tendant now is the penalty-imposing experi- menter and therefore always gives the patient his full half-hour, in the interests of precise experimental technique, of course.) This error is an elementary one, perhaps. But it suggests that in general, when an experiment proceeds from its baseline to its first experimental phase, changes in what is measured need not always reflect the behavior of the subject.

Analytic The analysis of a behavior, as the term is

used here, requires a believable demonstra-

93

DONALD M. BAER et al.

tion of the events that can be responsible for the occurrence or non-occurrence of that be- havior. An experimenter has achieved an analysis of a behavior when he can exercise control over it. By common laboratory stan- dards, that has meant an ability of the ex- perimenter to turn the behavior on and off, or up and down, at will. Laboratory standards have usually made this control clear by dem- onstrating it repeatedly, even redundantly, over time. Applied research, as noted before, cannot often approach this arrogantly fre- quent clarity of being in control of important behaviors. Consequently, application, to be analytic, demonstrates control when it can, and thereby presents its audience with a prob- lem of judgment. The problem, of course, is whether the experimenter has shown enough control, and often enough, for believability. Laboratory demonstrations, either by over- replication or an acceptable probability level derived from statistical tests of grouped data, make this judgment more implicit than ex- plicit. As Sidman points out (1960), there is still a problem of judgment in any event, and it is probably better when explicit. There are at least two designs commonly

used to demonstrate reliable control of an important behavioral change. The first can be referred to as the “reversal” technique. Here a behavior is measured, and the measure is examined over time until its stability is clear. Then, the experimental variable is ap- plied. The behavior continues to be mea- sured, to see if the variable will produce a behavioral change. If it does, the experimen- tal variable is discontinued or altered, to see if the behavioral change just brought about depends on it. If so, the behavioral change should be lost or diminished (thus the term “reversal”). The experimental variable then is applied again, to see if the behavioral change can be recovered. If it can, it is pur- sued further, since this is applied research and the behavioral change sought is an im- portant one. It may be reversed briefly again, and yet again, if the setting in which the be- havior takes place allows further reversals. But that setting may be a school system or a family, and continued reversals may not be allowed. They may appear in themselves to be detrimental to the subject if pursued too often. (Whether they are in fact detrimental is likely to remain an unexamined question

so long as the social setting in which the be- havior is studied dictates against using them repeatedly. Indeed, it may be that repeated reversals in some applications have a positive effect on the subject, possibly contributing to the discrimination of relevant stimuli in- volved in the problem.)

In using the reversal technique, the experi- menter is attempting to show that an analysis of the behavior is at hand: that whenever he applies a certain variable, the behavior is pro- duced, and whenever he removes this vari- able, the behavior is lost. Yet applied behav- ior analysis is exactly the kind of research which can make this technique self-defeating in time. Application typically means produc- ing valuable behavior; valuable behavior usually meets extra-experimental reinforce- ment in a social setting; thus, valuable be- havior, once set up, may no longer be depen- dent upon the experimental technique which created it. Consequently, the number of re- versals possible in applied studies may be lim- ited by the nature of the social setting in which the behavior takes place, in more ways than one. An alternative to the reversal technique

may be called the “multiple baseline” tech- nique. This alternative may be of particular value when a behavior appears to be irre- versible or when reversing the behavior is un- desirable. In the multiple-baseline technique, a number of responses are identified and mea- sured over time to provide baselines against which changes can be evaluated. With these baselines established, the experimenter then applies an experimental variable to one of the behaviors, produces a change in it, and perhaps notes little or no change in the other baselines. If so, rather than reversing the just- produced change, he instead applies the ex- perimental variable to one of the other, as yet unchanged, responses. If it changes at that point, evidence is accruing that the experi- mental variable is indeed effective, and that the prior change was not simply a matter of coincidence. The variable then may be ap- plied to still another response, and so on. The experimenter is attempting to show that he has a reliable experimental variable, in that each behavior changes maximally only when the experimental variable is applied to it. How many reversals, or how many base-

lines, make for believability is a problem for

94

SOME DIMENSIONS OF APPLIED BEHAVIOR ANALYSIS

the audience. If statistical analysis is applied, the audience must then judge the suitability of the inferential statistic chosen and the pro- priety of these data for that test. Alternatively, the audience may inspect the data directly and relate them to past experience with simi- lar data and similar procedures. In either case, the judgments required are highly quali- tative, and rules cannot always be stated prof- itably. However, either of the foregoing de- signs gathers data in ways that exemplify the concept of replication, and replication is the essence of believability. At the least, it would seem that an approach to replication is better than no approach at all. This should be es- pecially true for so embryonic a field as be- havioral application, the very possibility of which is still occasionally denied. The preceding discussion has been aimed