Measuring the Problem

Abstract

This research examined socioeconomic variables, offending behavior, and prior police contact, as predictors of self-reported police contact (questioning or arrest). Utilizing multilevel models and eight waves of National Youth Survey Family Study data, the predictors were examined in a national sample of individuals, over 24 years. Results indicate that police contact is predominately predicted by sex, delinquent peers, and offending behavior. This suggests that several of the variables commonly discussed in police contact literature, including race, are not predictors of police contact at the national level in the United States. Implications of national-level findings on police policy are discussed.

Keywords

police culture/accountability law enforcement/security inequality and crime other race and crime/justice

According to the Bureau of Justice Statistics (BJS) over 25.7 million people in the United States were contacted by police in 2005 because of either traffic-related stops or because the police suspected them of wrongdoing (Durose, Smith, & Langan, 2007). Those numbers translate to approximately 11% of the U.S. population age 16 or older. Other research indicates that not all groups of people in the United States have an equal probability of being contacted by police. Many studies through time have indicated that non-White individuals are contacted more frequently than are White individuals, even when controlling for criminal offending (Fagan & Davies, 2000; Geiger-Oneto & Phillips, 2003; Ousey & Lee, 2008; Piliavin & Briar, 1964; Smith & Visher, 1981). Though race is arguably the most commonly used predictor of police contact in the literature, it is far from the only variable that divides the population and creates possible disproportionality where police contacts are concerned. Some predictors have been examined persistently, and frequently, through time. These variables appear in literature from the mid-1960s through the late 2000s, and include: race (Penn, 2006; Piliavin & Briar, 1964), sex (Lundman & Kowalski, 2009; Terry, 1967), socioeconomic status (SES; Miller, 2008; Werthman & Piliavin, 1967), criminal history (Miller, 2008; Werthman & Piliavin, 1967), and concurrent criminal offense (McAra & McVie, 2007; McEachern & Bauzer, 1967). Other correlates of police contact appear more recently (and generally less frequently) in the literature, including age (Miller, 2008), involvement with delinquent peers (Patterson, Forgatch, & Yoerger, 1998), drug and alcohol usage (McAra & McVie, 2007), and IQ/grade point average (GPA¹; Fergusson, Horwood, & Ridder, 2005).

Though many important pieces of research have been done on this subject, much of it with persistent findings through time, several aspects of this topic remain unexamined. First, no study, to date, has been conducted with a sample that would allow for a developmental, longitudinal, examination of respondents. Second, of the 23 studies reviewed, only 6 studies examined data that was beyond the city or local level (D’Alessio & Stolzenberg, 2003; Durose et al., 2007; Lundman & Kowalski, 2009; Ousey & Lee, 2008; Penn, 2006; Pope & Snyder, 2003). This is not particularly surprising considering that policing in the United States is generally considered a localized concern. The multitude of local police agencies can produce poor communication and collaboration between agencies (Thibault, Lynch, & McBride, 2011). In addition, the populations that police serve differ widely by location. Still, local-level studies can indicate predictors of disproportionate police contact on a local level only. These localized findings suggest localized causes. A study conducted on a national-level data set, will make it possible to see whether predictors of police contact are prevalent enough that they are, in fact, not local problems. In addition, it would suggest that any problems found are caused by more than just a few police officers or even one policing agency. In order to effectively address a problem, it is always helpful to first know how large the problem is. This could be useful in determining whether policing policies and/or policy changes should be implemented on a larger scale (nationally) or remain on a more localized scale.

The current study will utilize national level, longitudinal, self-report data from the National Youth Survey Family Study (NYSFS) to examine the extent to which relevant predictors affect negative police contact. Two level nonlinear (Bernoulli) models, using the statistical software hierarchical linear modeling (HLM), will be used to examine direct effects, as well as any possible interaction effects, of independent variables on police contact. Because this data set has self-reported information on over 1,000 individuals across the nation, for more than two decades, the current study will be able to fill several important gaps in a field of research that could disproportionately affect more than 11% of the U.S. population (Durose et al., 2007). Perhaps most importantly, it will provide new and more rigorous evidence of the extent of disproportionate police contact in the United States.

Review of Relevant Literature

Twenty-three studies published over a 45-year time period were reviewed. Though this time frame may be longer than typically required, it is necessary in this case because it covers the majority of the respondents’ life courses. The studies are organized below by the relevant potential predictor of police contact (see Appendix A for chronological overview of all studies in tabular form).

Race

Race effects have been reported in several studies (Brown, 1981; Brunson, 2007; Fagan & Davies, 2000; Ousey & Lee, 2008; Piliavin & Briar, 1964; Terry, 1967; Werthman & Piliavin, 1967). In the early 1960s, Piliavin and Briar (1964) performed a qualitative analysis of juvenile–police contact situations in a metropolitan city of approximately 450,000 people. Where race was concerned, they found that African Americans were stopped and interrogated more frequently and were often given more severe sanctions for the same offenses. They attribute this to two causes: (1) ingrained prejudices held by many officers and (2) the fact that these boys more often illustrated the noncompliant demeanor which commonly led to arrest. While 18 of the 27 officers interviewed openly admitted dislike for Black individuals, many simply claimed that African American individuals were more likely to commit crime or delinquent acts and, therefore, it just made sense to focus attention on these youths (Piliavin & Briar, 1964).

Nearly 20 years after Piliavin and Briar, Brown (1981) performed another qualitative analysis of officers in three southern California police departments. He found that the police “justify singling out blacks on the grounds that they are simply using those criteria which are most likely to lead to the discovery of a crime’” (Brown, 1981, p. 111). A more recent qualitative analysis, published in 2007, which examined 40 “at-risk” men in St. Louis, MO, found similar results (Brunson, 2007). Specifically, the men interviewed in this research felt that Black people with gold teeth or caps, or with nice clothes, are targeted as selling drugs, because the police never assume that they might have legitimate means (e.g., good jobs) for purchasing these items (Brunson, 2007). In short, these men felt that police unfairly target them because of their race and cultural indicators of their race (e.g., clothing styles). Despite the time differences in these studies, interestingly, the results remain fairly similar.

Quantitative analyses also indicate that race is a factor in police contact and/or arrest situations (Fagan & Davies, 2000; Ousey & Lee, 2008; Shannon, 1988). Shannon (1988) compared birth cohorts from 1942, 1949, and 1955 in Racine, Washington. Through the use of correlations and trend lines, Shannon found that “minorities made up a disproportionate number of those referred” (p. 165). This was because minorities were subjected to more contact with the police, especially when the suspected criminal event was a serious crime, and they were also disproportionately arrested for crimes that individuals of other races were not. Fagan and Davies (2000) utilized a Poisson regression analysis to examine street stops in New York City. They found that despite controlling for crime participation, police still appear to target the city’s minority population, specifically Blacks and Hispanics. Durose, Smith, and Langan (2007) reported the results of the 2005 police–public contact survey performed by the BJS. They found a correlation between race and arrest in traffic stop situations. Specifically, they found that “Black drivers (4.5%) were twice as likely as White drivers (2.1%) to be arrested during a traffic stop” (p. 6).

Even more recently, Kochel, Wilson, and Mastrofski (2011) performed a meta-analysis on 40 pieces of quantitative research that were, in turn, based on 23 different data sets.² They specifically examined the effect size of the suspect’s race on the probability of arrest. Results of this study indicated that race is still a factor, even when controlling for demeanor, the severity of the offense, the presence of a witness, the quality of evidence, the discovery of a new offense during the police encounter, drugs or alcohol, prior record, and requests made by victims to arrest (Kochel, Wilson, & Mastrofski, 2011).

Still, some studies that examined police contact and/or arrest, did not find race to be an important factor (Black & Reiss, 1970; D’Alessio & Stolzenberg, 2003; Lundman & Kowalski, 2009; Lundman, Sykes, & Clark, 1978; McEachern & Bauzer, 1967; Pope & Snyder, 2003). Black and Reiss (1970) systematically observed two police precincts each, in Boston and Chicago, and four police precincts in Washington, DC. In total, they recorded 281 cases that contained one or more juvenile suspects. Through the use of percentages and qualitative descriptions, these researchers found that “evidence that the police behaviorally orient themselves to race as such is absent” (Black & Reiss, 1970, p. 76). To justify this assessment, they give three reasons: (1) 72% of their observed police contacts were citizen initiated and therefore, cannot be the result of police prejudices, (2) African American suspects were more often suspected of a felony offense at the time of contact, and (3) the presence of a complainant made it more likely an arrest would be made. African American suspects more often had African American complainants who demanded that an arrest be made (Black & Reiss, 1970). Lundman, Sykes, and Clark (1978) did a replication of the Black and Reiss study by observing 200 cases of juvenile–police contact, in a large Midwestern city. Their results mirrored Black and Reiss’s findings. While Durose et al. (2007) found that race is correlated with arrests in a traffic stop situations, when they examined all police contacts, including those performed because the police suspected the individual of wrongdoing, they found that Whites were contacted by police more frequently than either African American or Hispanic individuals. Taken together, these studies indicate that through time, there is still some debate as to the effects of race on police contact.

Sex

Though sex, specifically being male, may seem intuitively related to police contact, some research indicates small, inverse, or indirect relationships between the variables (Lundman & Kowalski, 2009; Miller, 2008; Smith & Visher, 1981; Terry, 1967). Terry (1967) examined 9,023 cases of juvenile–police contact, from official police records in a Midwestern community of less than 100,000 people. He found a relatively small inverse relationship between being male and being referred to an official agency (Terry, 1967). This can be explained, according to Terry (1967), by the fact that girls represented half of the sex offense and incorrigibility cases, and nearly 70% of the referrals to social agencies were for those offenses or issues (1967). Smith and Visher (1981) examined 742 cases of adult and juvenile contacts with the police. Data were gathered by trained civilians who rode on 900 patrol shifts, in 24 police departments in St. Louis, MO, Rochester, NY, and Tampa-St. Petersburg, FL. Through the use of a probit model, Smith and Visher found that sex has an indirect effect, through victims’ preferences, on the severity of the outcome of the police contact. If the suspect was male, victims were more likely to request that an arrest be made. If a suspect were female, however, the victim’s request could fall equally into three possible categories: (1) arrest, (2) leniency, and (3) no preference (Smith & Visher, 1981).

Still other studies do indicate that being male is an important predictor of police contact or arrest (Durose et al., 2007; McAra & McVie, 2007). Durose et al. (2007) utilized cross sectional, national-level survey data, performed by the BJS during the last 6 months of 2005, as a supplement to the National Crime Victimization Survey. These data indicated that males were more likely than females to come into contact with the police. In traffic stops, males were more likely to be ticketed and three times more likely to be arrested than females were. In addition, males accounted for 53.6% of all police contacts, but 72.4% of all contacts involving police use of force (Durose et al., 2007, p. 8). McAra and McVie (2007) examined the impact of a number of variables, including sex, on the severity of sanctions received as a result of juvenile–police contact situations, in Edinburg, Scotland. Utilizing forward stepwise logistic regression, they found that being male made the individual more likely to face relatively severe consequences from police (arrest as opposed to a warning; McAra & McVie, 2007). As is the case with race, studies also indicate that sex may have an effect on the outcome of police contact situations; however, there is still some debate as to exactly how this relationship functions.

Age

Smith and Visher (1981), discussed above, examined age as an influence in police contact situations; they found only an indirect effect. After creating three age categories (under 19, 19–35, and over 35), they discovered that (1) suspects over 35 are more antagonistic toward police and that they commit more serious offenses, both of which increase the probability of arrest and (2) suspects over 35 are more likely to know their victims, which greatly reduces the possibility of an arrest. As a result, they found that these underlying factors cancel each other out (Smith & Visher, 1981). Still, Durose et al. (2007), in their report of the national-level survey performed by BJS, found that face-to-face contacts with the police were most common with suspects aged 18–24 years. In addition, they reported that individuals aged 16–29 years were more likely than those 30 years of age or older, to experience the use of force during contact with the police. This research indicates that the relationship between age and police contact/arrest may be mediated through other variables.

SES

The majority of studies that use some measure of SES as a predictor of police contact or of the severity of sanctions received from police after contact, have found SES to be significant (Fagan & Davies, 2000; Geiger-Oneto & Phillips, 2003; McAra & McVie, 2007; Werthman & Piliavin, 1967). Werthman and Piliavin (1967) interviewed and observed police officers for 18 months in Oakland and San Francisco. They also interviewed and observed 56 core gang members and over 100 general gang members in San Francisco. Through a qualitative analysis, they found that police used SES of the parents (specifically whether they were on welfare), as well as of the neighborhoods, as reasons to arrest many gang members (Werthman & Piliavin, 1967). More recent studies have produced similar results. Fagan and Davies (2000) found that “for all suspects, after controlling for crime, stops within the sub-boroughs were predicted by their poverty rates” (p. 495). McAra and McVie (2007) found that living in areas with high levels of neighborhood deprivation and being entitled to free school lunches increased the chances that individuals would be arrested as opposed to receiving a warning from police. Only one reviewed study did not find SES to be a significant predictor of arrest. Terry (1967) did initially find that being lower class increased the chances of a severe sanction from police (arrest), however, after controlling for the seriousness of the offense and the number of previous offenses, that relationship was reduced. Terry (1967) interprets his results as indicating that SES is most likely not a specific focus at the police level.

Delinquent Peer Group Involvement

Having friends or associating with people who are delinquent has also produced mixed results as far as its impact on arrest (McAra & McVie, 2007; Patterson et al., 1998). Patterson, Forgatch, and Yoerger (1998) examined two successive birth cohorts of forth-grade boys and their families (N = 72) in a Pacific Norwest city in Oregon. Using logistic regression, they found that deviant peer group involvement contributes to the progression of juvenile offending behavior, including the increased chances of early arrest (Patterson et al., 1998). McAra and McVie (2007) also found an initial bivariate link between having friends who are perceived as trouble makers and arrest, in a police contact situation. In their multivariate analysis, however, this link disappeared (McAra & McVie, 2007).

Substance Use

Utilizing a multiwave, longitudinal design, McAra and McVie (2007) were one of the few to examine drug use in relation to the severity of police dispositions in juvenile–police contact situations. Though this study consisted of more than one wave, the total time span of the study was only 6 years.³ Still, this design allowed the authors to use illegal drug usage in the previous year as a predictor of arrest. Using forward stepwise logistic regression, these authors did find that illegal drug usage in the previous year was a significant predictor of arrest in a juvenile–police contact situation. While alcohol usage was also measured, drinking weekly was not found to be a significant predictor of arrest.

IQ/Academic Achievement

Though not directly related to police contact, intelligence and academic performance have been noted as possible influences of delinquency (Hirschi & Hindelang, 1977; Menard & Morse, 1984). Menard and Morse (1984) argue that both IQ and academic performance (GPA) may be used in schools, as a criterion by which to administer differential treatment to children. Therefore, neither IQ nor GPA is a direct cause of delinquency, but the differential treatment of children based on these 2 items, may influence delinquency (Menard & Morse, 1984).

Fergusson, Horwood, and Ridder (2005) are among the only researchers to examine IQ as a predictor of later arrest. Data for this study came from a longitudinal sample of 1,265 children born in Christchurch, New Zealand in 1977. They found that after controlling for early conduct problems and social backgrounds, IQ at ages 8 and 9, was no longer a significant predictor of arrest and conviction after the age of 15 (Fergusson et al., 2005).

Criminal Offending/Criminal History

Research indicates that the magnitude of the current suspected offense, the individual’s criminal history, and the number of previous contacts that a person has had with police, can all adversely affect the outcome in a police contact situation (Lundman et al., 1978; McAra & McVie, 2007; Miller, 2008; Piliavin & Briar, 1964; Werthman & Piliavin, 1967). Werthman and Piliavin (1967) found that, “if a boy confesses to a crime and is not known to the police, he is often released. If he is caught for the third or fourth time, however, the sum total of previous contacts may be enough to affect a judgment about his moral character adversely” (p. 73), regardless of the individual’s current suspected offense, or the nature of his past offenses. Studies performed by McEachern and Bauzer (1967), Black and Reiss (1970), and Lundman et al. (1978), all found that the chances of arrest are positively related to the seriousness of the current suspected offense. The more recent study by McAra and McVie (2007), found that being a serious offender, having self-reported arrests at earlier waves, and having self-reported contacts with police at earlier waves, were all statistically significant predictors of arrest.

Using Self-Report Data

It is important to take a moment to discuss the differences between the use of official data sources and self-report data to examine police contact. These differences could affect the interpretation of research conclusions. Three studies used official data sources such as the Uniform Crime Reports or the National Incident Based Reporting System (D’Alessio & Stolzenberg, 2003; Ousey & Lee, 2008; Pope & Snyder, 2003). These data sources contain information on criminal offenses reported to police. In regard to using this type of data, McEachern and Bauzer (1967) state, “a youngster who commits or participates in a crime may or may not be contacted by the police. We have no information about delinquent activities in which police (or some other agency) do not contact the youngster. If there is a juvenile-police contact, a record may or may not be made of the event. We have no information about contacts for which no record is made or kept” (p. 149). Unfortunately, many types of police contact do not result in an official record being made.

Williams and Gold (1972) compared the 1967 National Survey of Youth (self-report data) with police and court records from the counties in which the individuals had resided. Overall, they found that only 3% of the total chargeable acts were detected by the police. In a more recent study, Elliott (1995) compared the NYS, Waves 1 through 8 (1976–1990), to official records from each jurisdiction in which an NYS respondent lived for at least 1 year. In this study, the correlation between the frequency of index arrests and self-reported index offending was .30. The correlation between all arrest rates and individual offending rates was .38 (Elliott, 1995). Elliott concludes that “arrest, conviction, and incarceration data, are most appropriately viewed as measures of official response to criminal behavior ... findings from arrest samples should not be generalized to offenders or offenses in the general population” (p. 8).

There are occasions, however, in which there is more agreement between self-report and official records. Hardt and Peterson-Hardt (1977) studied 914 males, seventh- through nineth-grade students, in a city in a Middle Atlantic state. They found that 78% of official arrests were accurately self-reported, when students were asked about their arrest and ticketing history. Taken together, these studies show that in some cases, self-report data may include most of what is available in official data; however, official data does not always include much of the information available through self-report data.

Finally, it is also important to note that several studies reviewed use observational data (Black & Reiss, 1970; Boydstun, 1975; Brown, 1981; Brunson, 2007; Lundman et al., 1978; Piliavin & Briar, 1964; Smith & Visher, 1981; Werthman & Piliavin, 1967). With observational data, all police contact events that occur while the observer is present are recorded and the data tend to be rich taking into account the suspect, the officer, and characteristics of the situation. Unfortunately, due to the cost and time requirements for obtaining observational data, sample sizes and/or the number of locations are often limited, resulting in a nonprobability sample. It would be extremely costly and time consuming to get a nationally representative data set based on observations of police contacts with citizens. Therefore, for the current task, self-report data is the best option.

The present study adds to the literature on disproportionate police contact by examining this issue using a new data source involving national-level self-reported police contact data, and by incorporating a broader range of predictors than has been typical in previous studies.

Data and Methodology

Data for this study comes from the NYSFS. The NYSFS, originally known as the National Youth Survey or NYS, began in 1977⁴ as a probability sample of U.S. households, based on a self-weighting, multistage, cluster sampling design (Elliott, Huizinga, & Menard, 1989). The original sample consisted of 1,725 individuals between the ages of 11 and 17, and was representative of the total 11–17-year-old population in the United States in 1976 (Elliott et al., 1989). Subsequently, 10 additional waves have been completed, the most recent of which was completed in 2003 and contained 1,174 individuals.

Waves 1 through 5 were completed annually from 1977 to 1981, Wave 6 was completed in 1984, Wave 7 in 1987, Wave 8 in 1990, Wave 9 in 1993, and Waves 10 and 11 were completed in 2002 and 2003, respectively. A 12th wave was completed in 2004 however it covered only the spouses and offspring of the original NYSFS respondents and is therefore not relevant in the current analysis.

Overall completion rates for Waves 2 through 4 were above 94% and Wave 5 had a completion rate of 87%. Comparisons were done between participants and nonparticipants for Waves 2 through 5. These comparisons did reveal some losses based on sex, race, class, age, and residence. Still, it was determined that these losses “did not substantially influence the underlying distributions on these variables” (Elliott et al., 1989, p. 3). Completion rates for Waves 6 through 9 were 87%, 80%, 83%, and 78%, respectively (Menard, Mihalic, & Huizinga, 2001). At Wave 9, there still appeared to be no systematic losses based on race, class, age, or residence that would affect the underlying distribution (Menard et al., 2001). In addition, a study done by Bosick (2009) examined selective attrition based on criminal offending in NYSFS data Waves 1 through 7. She determined that “the average offending across crime types was ultimately similar for dropouts and nondropouts” (Bosick, 2009, p. 487). This finding is relevant to the current study because it could be argued that those at highest risk for police contact (those who commit more crime) have left the sample over time. Bosick’s results indicate that, at least in the first seven waves, this does not appear to be the case. For Waves 10 and 11, the total eligible sample of original respondents was reduced to 1677, due to additional deaths and hard refusals. Of this eligible sample, 75% participated in Wave 10 and 70% participated in Wave 11 (N = 1,174; Menard et al., 2001).⁵

The current study will examine data on the original respondents beginning in Wave 4 and extending through Wave 11. While the dependent variables are not fully measured until Wave 6, the independent variables are available from the first wave forward. As a result, Waves 4 and 5 are included in the study in order to measure the impact of previous occurrences of selected behaviors or characteristics, on the later measurements of the dependent variable.

Variables

Police contact is the dependent variable in this study. This variable was measured in two ways: (1) arrest in the previous year (1 = yes, 0 = no) and (2) questioning by police about suspected involvement in a crime in the past year (1 = yes, 0 = no). These measures are analyzed separately. Respondents began being asked about arrest in Wave 5 (1981), and they were first asked about being questioned by police in Wave 6 (1984). Unlike a few of the studies reviewed above (Black & Reiss, 1970; Lundman et al., 1978), there is no information available through NYSFS on citizen-initiated contacts. While these contacts would be interesting to explore, the current study is attempting to examine police contacts in which the police may have some control or discretion. Accordingly, arrest is being used as the more serious form of police contact and questioning as a less severe form of police contact.

Independent variables in this study include race, sex, age, SES, delinquent peers, substance use, academic achievement, offending behavior, and previous police contact. Race is a dichotomous measure in which respondents are classified as either White or non-White. Sex is measured simply as either male or female. Age is the chronological age of the respondent, measured in years.⁶

SES is an individual measure of SES that is based on Hollingshead and Redlich’s (1958) Two Factor Index of Social Position and is described in detail by Bonjean, Hill, and McLemore (1967). This measure is constructed using respondents’ education and occupation. In the current data, educational attainment was rated on a 7-point scale ranging from (1) graduate degree to (7) under 7 years of education. Occupation was also rated on a 7-point scale which ranges from (1) high level executives of large concern to (7) Unskilled employees. The class measure was obtained by multiplying occupation by seven, multiplying education by four, and adding the two together (7 × Occupation + 4 × Education).

Involvement with delinquent peers is measured using a scale that consists of eight indicators. Respondents were asked how many of their friends had: (1) damaged property that was not theirs, (2) used marijuana, (3) stole something worth less than $5, (4) hit or threatened to hit someone, (5) broke into a building or vehicle to steal something, (6) sold hard drugs, (7) stole something worth more than $50, eight suggested that the respondent do something against the law. Responses were measured on a 5-point scale, ranging from none (low) to all (high) of their friends.

Three forms of substance use were analyzed in this study. All three forms of substance use were measured as prevalence (1 = use, 0 = nonuse).⁷ To measure licit substance use, alcohol use in the previous year was used. Marijuana use in the previous year was used as a soft illicit drug measure, and a composite measure consisting of the sum of hallucinogen, amphetamine, barbiturate, cocaine, and heroin use was used as a measure of hard drug use. For substance use measures, as well as for offending behaviors (discussed below), a one-wave lag was used to ensure the correct temporal order between the predictors and the outcome variable. This was done because respondents were asked about drug use and about police contact (questioning and arrest) that had occurred in the previous year. Without more distinction, it is impossible to tell which behavior came first. Therefore, substance use was examined one wave prior to the outcome, to ensure that it preceded the police contact.

Academic achievement (which, as noted by Menard & Morse, 1984, is closely related to IQ) is a self-report measure in which respondents were asked to estimate their grades on a 5-point scale ranging from mostly F’s (1) to mostly A’s (5). The average was then taken for respondents between the ages of 15 and 17 years. These years have been selected because they are the latest ages in which the highest numbers of respondents were attending school, thereby allowing for a measurement of academic achievement for most of the respondents.

Involvement in offending behavior was broken into two measures. Each of these measures was dichotomous, with zero indicating no involvement in any of the behaviors in the prior year, and one indicating involvement. Minor delinquency includes involvement in any of the following: minor theft (theft of something worth less than $50), hitting or threatening to hit someone, selling marijuana or hard drugs, being loud/rowdy in a public place, buying stolen property, carrying a hidden weapon, prostitution, panhandling, or joyriding.⁸ Index offending includes involvement in any of the following behaviors: felony assault (sexual assault, aggravated assault, and gang fighting), robbery (using force or threat of force to take something), and felony theft (motor vehicle theft, burglary, and theft of something worth more than $50). As noted above, these measures were lagged by one wave to ensure correct temporal order with the outcome variable. Finally, previous police contact employs two measures: previous questioning and previous arrest, both measured one wave prior to the outcome. In short, this is looking at whether arrest or questioning in one wave affects the likelihood of being questioned or arrested in the next wave. Table 1 illustrates the descriptive statistics for all variables used in the current analysis.

Table 1.

Descriptive Statistics for Variables in Waves 4 Through 10

	No. of Observations	Mean	Standard Deviation
Arrest	7,739	.0403	.1967
Question	7,037	.0352	.1844
Sex (male)	13,800	.5322	.4990
Race	13,800	.7890	.4080
Age	13,800	27.2492	8.9174
Alcohol use	11,566	.8059	.3955
Marijuana	11,570	.3360	.4723
Hard drugs	11,574	.1463	.3534
SES	12,984	43.92	16.546
Academic achievement	13,768	3.6758	.8235
Index offending	11,579	.0896	.2856
Minor delinquency	11,579	.4164	.4930
Delinquent peers	10,777	10.7987	3.8200
Previous arrest	7,445	.0372	.1893
Previous question	6,913	.0179	.1327

Note. SES = socioeconomic status.

Research Hypotheses

Based on previous findings, it is hypothesized that the following will result in a higher likelihood of police contact throughout the life course of the respondents: (1) being non-White, (2) being male, (3) being young, (4) a low SES, (5) increased exposure to delinquent peers, (6) the use of alcohol, (7) the use of marijuana, (8) the use of hard drugs, (9) low-academic achievement, (10) involvement in minor delinquency, (11) involvement in index offending, and (12) prior police contact.

These hypotheses are based on previous research which suggests that a majority of the above-indicated factors lead to an increased chance of police contact as well as an increased chance of more severe sanctions as a result of police contact (Durose et al., 2007; Geiger-Oneto & Phillips, 2003; McAra & McVie, 2007; Miller, 2008; Patterson et al., 1998; Penn, 2006; Shannon, 1988). It should be noted that both substance use and delinquent peers are extremely rare in literature examining predictors of police contact (McAra & McVie, 2007; Patterson et al., 1998). Further, alcohol use and academic achievement have not previously been found to be significant predictors of police contact (Fergusson et al., 2005; McAra & McVie, 2007; Menard & Morse, 1984). These measures are being included to check the consistency with previous findings.

Analytical Strategy

Multilevel Bernoulli models were chosen to analyze the current research hypotheses. This technique was selected because it is appropriate when the dependent variable is dichotomous and it works well with all types of independent variables (Menard, 2010). In addition, multilevel models can account for the possible clustering of events within individuals. Logistic regression, which is similar, but performed on one level, requires that observations in the data set be independent of each other (Menard, 2010). Because these data are longitudinal in nature, it is possible that events that occur within each person, such as incidents of police contact, may not be independent of other events experienced by the same individual at a different time point. Without accounting for this lack of independence, estimates of statistical significance would be inflated. Bernoulli models will be used to determine the amount of variation in the dependent variable that is explained by the independent variables, and to rank the relative strength of the independent variables.

Multilevel Bernoulli models predict the classification of cases into one of two categories of the dependent variable using a logit link, or by transforming the Level-1 outcome variable into a logit variable (Raudenbush, Byrk, Cheong, Congdon, & Toit, 2004). For the current analysis, Laplace estimation was used. This approach is an alternative to using the penalized quasi-likelihood estimation of parameters that is the default in HLM6. It is based on a Laplace transformation and produces results that approximate the maximum likelihood estimation used in logistic regression (Raudenbush et al., 2004).

In the present analysis, the first level consists of the time varying variables such as substance use, involvement with delinquent peers, offending behaviors, and prior police contact. The Level-1 structural model is expressed as follows:

η_{i j} = log \frac{ϕ i j}{[1 - ϕ i j]} = β_{0}_{j} + β_{1}_{j} (delpeer)+ β_{2}_{j} (Palcohol_1) ... β_{k}_{j} (previous questioning),

where η _ij is the log odds of success and

φ i j

is the probability of success.

The Level-2 model consists of variables that are sociodemographic in nature and primarily fixed. In the current analysis, Level 2 includes variables such as age, race, sex, SES, and academic achievement. According to Raudenbush, Byrk, Cheong, Congdon, and Toit (2004) the Level-2 structural model is expressed as follows:

β_{0 j} = γ_{00} + γ_{01} {(age)}_{i j} + γ_{02} {(male)}_{i j} {+ u}_{0 j .}

β_{1 j} = γ_{10} .

β_{2 j} = γ_{20} .

…

β_{k j} = γ_{k 0} .

Results produced from running Bernoulli models with Laplace estimation in HLM6 can be interpreted as logistic regression results. Accordingly, the model statistical significance will be tested using the model chi-squared (χ²) statistic which will be referred to as G_M . G_M is simply the difference in the −2 log-likelihood statistic between a model which contains only an intercept, and a model which contains an intercept plus one or more independent variables. If the G_M is statistically significant, then the addition of the predictors has allowed the model to better predict the odds of a certain event occurring (in the current case, this would be the odds of police contact).⁹ The model’s substantive significance will be measured using the likelihood ratio R ², also known as McFadden R ² (

R_{L}^{2}

R_{L}^{2}

indicates how much the inclusion of the independent variables in the model reduces the variation in the model’s −2 log-likelihood statistic (−2LL), or D ₀. D ₀ is the −2LL statistic with none of the independent variables entered into the equation.¹⁰

Statistical significance of the predictors will be established using the Wald statistic. The Wald statistic can be computed as: $W_{k}^{2} = {[b_{k} / (standard error o f b_{k})]}^{2}$ , where b_k is the unstandardized coefficient.¹¹ Finally, substantive significance of the predictors will be established using fully standardized regression coefficients (Menard, 2002, 2010). Standardized coefficients allow for an examination of coefficients which have been calculated using variables in common (standard deviation) units. Recall, however, that in multilevel models that work with dichotomous outcome variables, the dependent variable is the logit of Y (police contact). Because of this, the direct calculation of means or standard deviations (for logit (y)) is not possible. Standard deviations can be calculated indirectly, however, using predicted values of logit (Y) and R ², and once the standard deviation is calculated, standardized logistic regression coefficients can be calculated from the following formula: b*_yx = (b_yx)(S_x)(R)/(S _logit(Ŷ)), where b_yx are the unstandardized regression coefficients, S_x is the standard deviation of the independent variable, R is the Pearson’s correlation between the observed and predicted values for the dependent variable, and S _logit(Ŷ) is the standard deviation of the logit(Ŷ). The logit(Ŷ) is calculated using the following formula: logit(Ŷ)= ln[Ŷ/ (1−Ŷ)], where Ŷ is the predicted value for Y.¹²

Findings

Table 2 illustrates the results of the model in which “arrest” was used as the dependent variable.¹³

Table 2.

Bernoulli Model for Arrest (Waves 5–11)

Dependent Variable	Predictive Efficiency	Independent Variables	Unstandardized Coefficients (b)	Standardized Coefficients (b*)	Significance (Wald Statistic)
Arrest	G_m = 1286.75	Sex (male)	1.159	.156	.001
	Df = 13	Race	−.370	−.038	.167
	(p = .000)	Age	−.047	−.025	.420
		Alcohol use	−.460	−.041	.210
	$R_{L}^{2}$ = .136	Marijuana use	.670	.086	.011
		Hard drug use	−.144	−.015	.621
		SES	.015	.067	.043
		Academic achievement	−.082	−.018	.607
		Index offending	.368	.026	.232
		Minor delinquency	.422	.056	.138
		Delinquent peers	.103	.092	.000
		Previous arrest	1.018	.049	.003
		Previous question	−1.192	−.048	.103

Note. SES = socioeconomic status.

The arrest model is statistically significant (p = .000), and it explains 13.6% of the variation in the dependent variable (arrest). Arrest has a positive statistically significant relationship with sex (p = .001), delinquent peers (p = .000), marijuana use (p = .011), SES (p = .043), and previous arrest (p = .003). The positive relationships indicate that arrest is more likely for individuals who are: male, have more exposure to delinquent peers, use marijuana, have a lower SES, and have been previously arrested.¹⁴ There are no statistically significant negative relationships in this model. Sex has the largest relative impact on arrest, followed in order by delinquent peers, marijuana use, SES, and previous arrest. This suggests that sex is the strongest statistically significant predictor of arrest, while previous arrest is the weakest.

Table 3 illustrates the results of the model in which “questioned” was used as the dependent variable.¹⁵

Table 3.

Bernoulli Model for Questioning (Waves 6–11)

Dependent Variable	Predictive Efficiency	Independent Variables	Unstandardized Coefficients (b)	Standardized Coefficients (b ^*)	Significance (Wald Statistic)
Questioned	G_m = 92.7782	Sex (Male)	1.183	.095	.000
	Df = 13	Race	−.183	−.011	.470
	(p = .000)	Age	−.045	−.014	.429
		Alcohol use	−1.153	−.063	.001
	$R_{L}^{2}$ = .108	Marijuana use	−.269	−.020	.398
		Hard drug use	.724	.044	.023
		SES	.003	.008	.680
		Academic achievement	−.358	−.047	.004
		Index offending	.458	.018	.189
		Minor delinquency	.222	.017	.397
		Delinquent peers	.065	.033	.012
		Previous arrest	−.529	−.014	.366
		Previous question	.412	.009	.434

Note. SES = socioeconomic status.

The model with “questioned” as the dependent variable is statistically significant (p = .000). The model explains 10.8% of the variation in the dependent variable. Being questioned has a statistically significant positive relationship with sex (p = .000), hard drug use (p = .023), and delinquent peers (p = .012). Being questioned has a statistically significant negative relationship with alcohol use (p = .001), and academic achievement (p = .004). The direction of the relationships suggests that being questioned is more likely for individuals who: are male, do not use alcohol, have lower academic achievement, use hard drugs, and have more exposure to delinquent peers. The standardized regression coefficients in Table 4 indicate that sex has the largest relative impact on being questioned, followed in order by alcohol use, academic achievement, hard drug use, and delinquent peers. Put differently, this means that the strongest statistically significant predictor of being questioned by police is sex and the weakest statistically significant predictor is exposure to delinquent peers.

Table 4.

Statistically Significant Predictors of Police Contact

	Arrest	Question
Sex (male)	(+)	(+)
Race
Age
Alcohol		(−)
Marijuana	(+)
Hard drugs		(+)
SES	(+)
Academic achievement		(−)
Index offending
Minor delinquency
Delinquent peers	(+)	(+)
Previous arrest	(+)
Previous questioning

Note. SES = socioeconomic status.

(+) Indicates a positive statistically significant relationship in the multivariate model; (−) Indicates a negative statistically significant relationship in the multivariate model.

Discussion

Table 4 provides a side-by-side comparison of the significant predictors of both forms of police contact. When considered together, police contact is predominately predicted by sex, exposure to delinquent peers, and offending behavior (in this case in the form of substance use). Several important aspects of these findings deserve discussion, including an interpretation of the significant findings, limitations, future research, and what the findings suggest as far as police policy.

Where findings are concerned, it is particularly notable that race had no significant relationship with either arrest or questioning by police.¹⁶ This information is consistent with previous literature that either studied disproportionate police contact on a local level or found specifically that the problem of disproportionately focusing enforcement on non-White individuals was really caused by the acts of a relative few officers (Black & Reiss, 1970; Kerner et al., 1968; Piliavin & Briar, 1964). In addition, it is consistent with a growing body of research that is finding race to be not significantly related to or not correlated with police contact (Durose et al., 2007; Tracy, 2002).

It may also be relevant to compare the results (specifically in regard to race) of the current study with results from previous studies that used national-level official data. Of the three studies that used official data, all found a relationship between race and police contact, but the findings were inconsistent with respect to the direction of the relationship. D’Alessio and Stolzenberg (2003) and Pope and Snyder (2003) actually found that White individuals were more likely to be arrested than African American individuals, while Ousey and Lee (2008) found that African Americans were more likely to be arrested for soft crimes or those where police can exercise more discretion. The current study did not find a relationship between race and police contact at all. The differences in results may be explained by some of the methodological differences. First, all three studies that used official data sources examined specific offenses, for which an arrest was made. The current examination focused on arrests made for any offense. Also, because of the use of self-report data, the current study could look at arrests controlling for offending behavior, even if that offending behavior was not known to the police. Finally, because of the use of official data, those studies could only examine police contact in which an official record was made. Using self-report data, the current examination could include incidents of being questioned by police for suspected involvement in a crime, separately from incidents of arrest.

Returning to the discussion of results, it is also interesting that offending behavior is not a significant predictor of police contact in any model. This has at least two possible explanations. Recall that marijuana use was a statistically significant predictor in the arrest model, and that hard drug use and alcohol use were significant in the model regarding being questioned by police. Substance use (particularly marijuana and hard drug use) is a form of offending behavior since possession of these substances is illegal. When all forms of illegal behavior are examined with crosstabs, by year, it becomes clear that different forms of illegal behavior are significant in different years. When examined together, these effects may be canceling each other out. As a result, it may be more informative to look at whether any form of illegal behavior (index offending, minor delinquency, marijuana use, or hard drug use) is present, as opposed to focusing on the lack of a specific one. While a simple canceling out effect is one option, it is also possible that this finding is being caused by police emphasis on drug-related enforcement during the life course of these respondents. In that case, because police are focused on drug-related offending, it makes sense that police contacts are more consistently predicted by this particular type of offense as opposed to other, more general, offending behavior.

The relationship between questioning and alcohol use is also counter to expectation and, therefore, deserves some discussion. First, it is important to look at the legality of alcohol use for the current sample. Until 1984, it was legal, in most states in the United States, to drink at the age of 18 (NHTSA Fact Sheet; NHTSA.gov, 2011). After that time, the legal drinking age moved to 21 years of age. This means that, by Wave 6, even the youngest in the data set could legally drink alcohol (18 years old) and by Wave 7, when the drinking age had risen, the youngest respondents were 21 years old. Because this study starts when respondents are between the ages of 15 and 21 (Wave 4) and contains only two waves where some of the respondents would be underage, drinking, in and of itself, is largely legal for the current sample. As to why there is a significant negative relationship, there are two possible explanations. The first is that this may be an artifact of the current sample. Put differently, in the current sample this relationship exists, but it does not exist in the larger population. The second possibility is that police do, in fact, question people that use alcohol proportionately less frequently than they question those that do not use alcohol. The argument could be made that individuals who use alcohol are going to be the individuals who are more commonly under the influence of alcohol when they come into contact with police, and therefore questioning them does little good. In addition, if they are under the influence of alcohol, there may exist enough probable cause to arrest them without questioning.

Another interesting finding is that academic achievement (used here as a proxy for intelligence) is a significant predictor of being questioned by police, but not of arrest. This finding is consistent with research done by Fergusson et al. (2005) which said that IQ was not a predictor of arrest. Still, it appears that police are disproportionately questioning individuals with low academic achievement, on a large enough scale that it is evident on a national-level examination. This relationship is one that, to date, has not been well studied in literature on police contact.

A few limitations should be noted. While the data used in this analysis were rich and longitudinal in nature, the varying time lags between data collection are troublesome. Recall that in order to ensure correct temporal ordering between the outcome and the predictor variables, a one-wave time lag was used for predictors. If concurrent predictors were used, there would be no way to know if the predictor (e.g., substance use) occurred before the police contact or as a result of the police contact. In the earlier waves of data collection, this is not a concern. A one-wave time lag meant a 1-year time lag. At later waves however, it meant as much as a 9-year time lag. Events that occurred 1 year prior may be stronger predictors of the outcome variables than events that occurred 9 years prior.

Age in this study is assumed to have a roughly linear relationship with police contact. This means that the difference between a 15 and a 21-year-old individuals is treated the same as the difference between a 36 and a 42-year-old individuals. Where arrest is concerned, this roughly linear relationship is confirmed through scatterplots and crosstabs. The relationship is less clear where questioning is concerned. Future research may address age differently; however, it does not appear to be a significant factor in police contact, even in the arrest model.

In addition to the varying amounts of time between waves and the assumed linear relationship between age and police contact, information on the dependent variables (arrest and questioning) did not begin to be collected until Waves 5 (arrest) and 6 (questioning). This means that this information was only available for five to six waves of data. While this is far more extensive than other data sources, it does limit what can be done with statistical models, such as multilevel change models, that examine change over time.

Several findings here do suggest future research. The relationship between delinquent peers and police contact should be examined further. Specifically, if this relationship is mediated by offending behavior, it may suggest that individuals are learning offending behavior from delinquent peers and are then being contacted by police for that offending behavior. If, on the other hand, there is not a mediating effect, this suggests that these individuals are being labeled by association. In other words, it suggests that individuals are being questioned or arrested by police merely because they are in the presence of friends who are involved in delinquent behavior. The relationship between previous police contact and current police contact presents a similar area of research exploration. If the impact of previous police contact is mediated by offending behavior, it suggests that individuals who were previously contacted by police are also the individuals who are currently involved in offending behavior. If that is the case, then it is that offending behavior that is attracting police contact. If, on the other hand, there is not an indirect effect, it suggests that police are merely rounding up the usual suspects. These indirect effects could be seen through the use of path analysis with logistic regression (Menard, 2010). It would also be informative to compare this generation of respondents to the next generation, to see if these predictors are consistent across generations.

Finally, it is also important to discuss what these findings mean for police managers and police departments. This study clearly indicates that police cannot implement policies or changes based on what works in other areas. The research done on a local level and/or for specific types of offenses (such as speeding) may be useful in explaining police behavior in those contexts. As a result, local level or offense-specific studies could also be beneficial in guiding police policies within the boundaries of those locations or for those offense types. It would be a mistake, however, to generalize the results of local-level studies beyond the context in which they were performed. Where policy is concerned, this means that it would be a mistake to create any broad policing policies regarding disproportionate police contact, based on studies with limited samples or offense types.

More specifically though, it is not particularly concerning that police are contacting individuals because of illicit substance use. Since the possession of these substances is illegal nationwide, we would expect police to question or arrest people who regularly violate the law in this manner. Police should be cautious about contacting individuals solely because of previous police contact, or because these individuals keep bad company. But even if police are contacting people who are known to have prior police contact or those who spend time with delinquent peers, this is still not surprising. Research indicates that these factors are predictors of criminal offending and it may be the case that police are over using them in their efforts to find potential criminal offenders (Menard & Elliott, 1990). More concerning is the finding that police are contacting males disproportionately, even when controlling for criminal offending. This finding is true across the nation, for both questioning and arrest. Also concerning is the finding that police across the nation, are arresting individuals with low SES at disproportionately higher rates, and questioning individuals with low academic achievement at disproportionately higher rates. These findings indicate that across the life course of individuals who were 11–17 years old in 1977, sex, SES, and academic achievement are sources of disproportionate police contact that cannot be explained by disproportionate criminal offending.

As stated previously, policing in the United States is generally seen as a localized phenomenon. The citizens that are served by each agency vary widely in culture, practices, and in what they need or want from their local police department. To adjust for these differences, policing agencies also vary widely in their policies and practices, within the boundaries of state and federal law. While this research suggests that police should continue to develop policies on a more localized level in order to address issues of disproportionate police contact, it may also suggest some room for nationalized policy where sex, SES, and academic achievement are concerned. While there are no doubt variations in different agencies across the United States in levels of police contact based on these factors, these problems do appear to be prevalent enough that they need to be addressed by police officers nationwide.

Footnotes

Appendix A

Summary of Literature on Police Contact Across Time

Author(s)	Year of Publication	Methodology	Level of Measurement	Relevant Finding(s)
Piliavin and Briar	1964	Qualitative	City	African Americans are stopped and interrogated more frequently and given more severe sanctions
McEachern and Bauzer	1967	Quantitative	City	Chances of arrest increase as the seriousness of the offense increases
Terry	1967	Quantitative	City	Juvenile females are more likely to be referred to an official agency, when compared to juvenile males. Low SES did not increase the chances of severe sanction from the police
Werthman and Piliavin	1967	Qualitative	City	Prior police contact, SES of parents, and SES of neighborhoods increase chances of arrest as opposed to release
Black and Reiss	1970	Both	City	Police do not target race
Boydstun	1975	Both	City	Officers that were given specific training to do field interrogations did not have differences where race, age, or gender are concerned, between who they field interrogated and arrested. Officers not trained arrested more African Americans than they field interrogated
Lundman et al.	1978	Both	City	Police do not target race
Brown	1981	Qualitative	City	Police claim that African Americans disproportionately commit crime, which justifies increased attention
Smith and Visher	1981	Quantitative	City	If offender was male, the victim was more likely to request that an arrest be made. They also found that older individuals commit more serious offenses and are more antagonistic toward police; however, they are more likely to know their victims. These two factors cancel out the affect of age on arrest
Shannon	1988	Quantitative	City	Minorities are disproportionately contacted and arrested by police
Patterson et al.	1998	Quantitative	City	Deviant peer group involvement contributes to juvenile offending behavior and increases the chances of early arrest
Fagan and Davies	2000	Quantitative	City	Police appear to target minorities (particularly African American and Hispanic individuals). The area poverty rates act as a significant predictor of police stops
D’Alessio & Stolzenberg	2003	Quantitative	National (United States)	African American individuals were not disproportionately arrested for the offenses examined (robbery, aggravated assault, simple assault, & forcible rape)
Geiger-Oneto, and Phillips	2003	Quantitative	City	African American and Hispanic males experience more social control in traffic stop situations, than do females of all races
Pope and Snyder	2003	Quantitative	National (United States)	African American juveniles were not disproportionately arrested for the offenses examined (robbery, aggravated assault, simple assault, & forcible rape)
Fergusson et al.	2005	Quantitative	City	IQ at ages 8 and 9 is not a predictor of arrest after age 15
Penn	2006	Quantitative	National (United States)	Black individuals are being arrested at numbers disproportionate to their numbers in the general population
Brunson	2007	Qualitative	City	Men felt they were targeted by police as suspected drug dealers because of their race and cultural indicators of their race (e.g., clothing style)
Durose et al.	2007	Quantitative	National (United States)	Black drivers are twice as likely as White drivers to be arrested during a traffic stop. Across all police contact categories, Whites were contacted by police at higher rates than other minority groups. Males are more likely to be contacted by police than females. They are also more likely to be ticketed or arrested during a traffic stop and to have force used against them. Younger individuals are more likely to be contacted by police and to have force used against them. Note: Authors explicitly state that these reported associations are not taking into account many factors that would be needed to make causal claims
McAra and McVie	2007	Quantitative	City	Being a juvenile male, having parents with low SES, living in neighborhoods with high levels of deprivation, illegal drug usage in the previous year, being a serious offender, and prior police contact lead to an increased chance of arrest. Drinking alcohol regularly and delinquent peers were not significant factors
Miller	2008	Quantitative	City	Race (being African American) was a predictor of both warning and ticket stops, while being young was a predictor of ticket but not warning stops. Gender was not a predictor of either type of stop
Ousey and Lee	2008	Quantitative	National (U.S.)	For “soft” crimes (drug and weapon violations), there is an unexplained arrest disparity (higher for Blacks than for Whites)
Lundman and Kowalski	2009	Quantitative	State	African American individuals, males, and individuals under the age of 25 are more likely to be high rate speeders; therefore, their disproportionate ticketing may be due to disproportionate offending

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Black

Reiss

(1970). Police control of juveniles. American Sociological Review, 35, 63–77.

Bonjean

Hill

McLemore

(1967). Sociological measurement: An inventory of scales and indices. San Francisco, CA: Chandler.

Bosick

(2009). Operationalizing crime over the life course. Crime & Delinquency, 55, 472–496.

Boydstun

J. E.

(1975). San Diego field interrogation: Final report. Washington, DC: Police Foundation.

Brown

(1981). The allocation of justice and police-citizen encounters. In Goodsell

(Ed.), The public encounter: Where state and citizen meet (pp.102–125). Bloomington: Indiana University Press.

Brunson

R.K

. (2007). “Police don’t like black people”: African-American young men’s accumulated police experiences. Criminology & Public Policy, 6, 71–100.

D’Alessio

Stolzenberg

(2003). Race and the probability of arrest. Social Forces, 81, 1381–1397.

Durose

Smith

Langan

(2007). Contact between police and the public, 2005. Washington, DC: Bureau of Justice Statistics.

Elliott

(1995). Lies, damn lies and arrest statistics: The Sutherland award presentation. Boulder, CO: Center for the Study and Prevention of Violence.

10.

Elliott

Huizinga

Menard

(1989). Multiple problem youth: Delinquency, substance use, and mental health problems. New York, NY: Springer-Verlag.

11.

Fagan

Davies

(2000). Street stops and broken windows: Terry, race, and disorder in New York City. Fordham Urban Law Journal, 28, 457–504.

12.

Fergusson

D. M.

Horwood

L. J.

Ridder

E. M.

(2005). Show me the child at seven II: Childhood intelligence and later outcomes in adolescence and young adulthood. Journal of Child Psychology and Psychiatry, 46, 850–858.

13.

Geiger-Oneto

Phillips

(2003). Driving while Black: The role of race, sex, and social status. Journal of Ethnicity in Criminal Justice, 1, 1–25.

14.

Hardt

Peterson-Hardt

(1977). On determining the quality of the delinquency self-report method. Journal of Research in Crime and Delinquency, 14, 247–259.

15.

Hirschi

Hindelang

(1977). Intelligence and delinquency: A revisionist review. American Sociological Review, 42, 571–587.

16.

Hollingshead

Redlich

(1958). Social class and mental illness: A community study. New York, NY: Wiley.

17.

Kerner

Lindsay

Harris

Abel

Brooke

Thornton

Jenkins

. (1968). Report of the national advisory commission on civil disorders. Washington, DC: Special report to the President.

18.

Kochel

Wilson

Mastrofski

(2011). Effect of suspect race on officers’ arrest decisions. Criminology, 49, 473–512.

19.

Lundman

Kowalski

(2009). Speeding while Black? Assessing the generalizability of Lange et al.’s (2001, 2005) New Jersey turnpike speeding survey findings. Justice Quarterly, 26, 504–527.

20.

Lundman

Sykes

Clark

(1978). Police control of juveniles: A replication. Journal of Research in Crime and Delinquency, 15, 74–91.

21.

McAra

McVie

(2007). Youth justice? The impact of system contact on patterns of desistance from offending. European Journal of Criminology, 4, 315–345.

22.

McEachern

Bauzer

(1967). Factors related to disposition in juvenile police contacts. In Klein

(Ed.), Juvenile gangs in context: Theory, research, and action (pp. 148–160). Englewood Cliffs, NJ: Prentice Hall.

23.

Menard

(2002). Applied logistic regression analysis (Sage University Paper Series on quantitative applications in the social sciences, 07-106). Thousand Oaks, CA: SAGE.

24.

Menard

(2010). Logistic regression: From introductory to advanced concepts and applications. Thousand Oaks, CA: SAGE.

25.

Menard

Elliott

D. S.

(1990). Longitudinal and cross-sectional data collection and analysis in the study of crime and delinquency. Justice Quarterly, 7, 11–55.

26.

Menard

Mihalic

Huizinga

(2001). Drugs and crime revisited. Justice Quarterly, 18, 269–299.

27.

Menard

Morse

(1984). A structuralist critique of the IQ-delinquency hypothesis: Theory and evidence. American Journal of Sociology, 89, 1347–1378.

28.

Miller

(2008). Police stops, pretext, and racial profiling: Explaining warning and ticket stops using citizen self reports. Journal of Ethnicity in Criminal Justice, 6, 123–149.

29.

NHTSA Fact Sheet: Minimum drinking age laws (2011). Retrieved December 13, 2011, from http://www.nhtsa.gov/people/injury/alcohol/Community%20Guides%20HTML/PDFs/Public_App7.pdf

30.

Ousey

Lee

(2008). Racial disparity in formal social control: An investigation of alternative explanations of arrest inequality. Journal of Research in Crime and Delinquency, 45, 322–355.

31.

Patterson

Forgatch

Yoerger

(1998). Variables that initiate and maintain an early-onset trajectory for juvenile offending. Development and Psycopathology, 10, 531–547.

32.

Penn

(2006). Black youth: Disproportionality and delinquency. In Penn

Greene

Gabbidon

(Eds.), Race and juvenile justice (pp. 47–61). Durham, NC: Carolina Academic Press.

33.

Piliavin

Briar

(1964). Police encounters with juveniles. American Journal of Sociology, 70, 206–214.

34.

Pope

Snyder

(2003). Race as a factor in juvenile arrests. Washington, DC: Office of Juvenile Justice and Delinquency Prevention.

35.

Raudenbush

Byrk

Cheong

Congdon

Toit

(2004). HLM6: Hierarchical linear & nonlinear modeling. Lincolnwood, IL: Scientific Software International.

36.

Shannon

(1988). Police contact referrals and their use in predicting continuity in careers. In Shannon

(Ed.), Criminal career continuity: Its social context (pp. 152–166). New York, NY: Human Sciences.

37.

Smith

Visher

(1981). Street-level justice: Situational determinants of police arrest decisions. Social Problems, 29, 167–177.

38.

Terry

(1967). Discrimination in the handling of juvenile offenders by social control agencies. Journal or Research in Crime and Delinquency, 4, 218–230.

39.

Thibault

Lynch

McBride

(2011). Proactive police management (8th ed.). New York, NY: Prentice Hall.

40.

Tracy

(2002). Decision making & juvenile justice: An analysis of bias in case processing. Westport, CT: Praeger.

41.

Werthman

Piliavin

(1967). Gang members and the police. In Bordua

(Ed.), The police: Six sociological essays (pp. 56–98). New York, NY: Wiley.

42.

Williams

Gold

(1972). From delinquent behavior to official delinquency. Social Problems, 20, 209–229.