Abstract
From demographics, to technology, to attitudes, the U.S. population has changes since the 1970s. Over the past 40 years, policing has also changed to include more individuals who are female and non-White. Despite all of the changes, no study has yet been conducted to determine whether predictors of police contact, including factors such as race and gender, are consistent over time. The current study used multilevel Bernoulli models and logistic regression to examine two generations of respondents from the National Youth Survey Family Study. Results indicate some consistency in predictors of police contact between the two generations, with two notable exceptions: gender and socioeconomic status. Implications for police policy and practice are discussed.
Introduction
Between 1970 and 2010, the U.S. population expanded by more than 105 million people, resulting in an increase of nearly 30 people per square mile of land (U.S. Census Bureau history, 2011). In addition to more people, U.S. culture has also changed. In 1981, the first woman joined the United States Supreme Court, followed in 1991 by the first African American justice, and in 2009 by the first minority female justice (Members of the Supreme Court of the United States, 2011). In 2008, U.S. citizens watched as the first African American president was sworn into office. From demographics, to technology, to attitudes, the U.S. population has changed since the 1970s.
Policing has also changed. As of 2011, the Federal Bureau of Investigation (FBI) reports that their employees are 43.7% female and 25.7% non-White (Federal Bureau of Investigations careers, 2011). Local agencies vary but have also seen a push to include non-White and female individuals on their respective forces. Despite this attempt at diversity, factors such as race and gender, among others, remain points of contention where police/public interactions are concerned (Fagan & Davies, 2000; Lundman & Kowalski, 2009). Studies beginning in the 1960s and extending into the 21st century indicate that police may be disproportionately initiating contact with members of the public, for reasons other than criminal offending (Boydstun, 1975; Durose, Smith, & Langan, 2007; D’Alessio & Stolzenberg, 2003; Fagan & Davies, 2000; McAra & McVie, 2007; Ousey & Lee, 2008; Shannon, 1988). Furthermore, some of these studies indicate that disproportionate police contact leads to a decreased belief that U.S. law enforcement officers are legitimate agents of the government, deserving of respect (Fagan & Davies, 2000). This, in turn, leads to a corresponding increase in the chances of violence during police/public contact (Fagan & Davies, 2000; Kerner et al., 1968).
Because disproportionate police contact can lead to fear, anger, and frustration by the public and to the potential of violence and a reduction of efficiency for police, it is an area that has been studied frequently through time (Boydstun, 1975; D’Alessio & Stolzenberg, 2003; Fagan & Davies, 2000; Kerner et al., 1968; McAra & McVie, 2007; Piliavin & Briar, 1964; Shannon, 1988). Despite the large quantity of studies on police contact, important aspects of this topic remain to be examined. One such aspect is the continuity of predictors of police contact over time. Though there are studies performed from the 1960s through the 2010s, these studies are difficult to compare because of the differences in respondents that comprise the samples, the diversity of the locations where the studies occurred, and the differences in the types of police contact that may have been examined (such as police contact specifically for traffic offenses). No study has yet been able use a longitudinal sample that contains data on more that one generation of respondents, to examine the change or consistency of predictors of police contact in the United States.
The current study will use multilevel Bernoulli models and logistic regression to examine two generations of respondents from the National Youth Survey Family Study (NYSFS). The original respondents of the NYSFS will be compared with their youth (11-17 years of age) and adult (older than 18 years) offspring to determine whether predictors of police contact (being questioned or arrested) are consistent over time, across the United States. This data source provides a unique opportunity to examine police contact in a longitudinal national level probability sample, where police contact is not limited to any specific offense type. Society has changed in the United States since the 1970s; this research will determine whether predictors of police contact are changing as well. If they are, it will also help to determine whether the change is in a direction that makes police interactions with the public more safe and efficient for all involved.
Literature Review
There are many variables that are commonly studied as predictors of police contact. These include race (Brown, 1981; Chambliss, 1994; Fagan & Davies, 2000; Lundman & Kowalski, 2009; Piliavin & Briar, 1964), gender (Geiger-Oneto & Phillips, 2003; McAra & McVie, 2007; Shannon, 1988), age (Black & Reiss, 1970; Durose et al., 2007), socioeconomic status (SES; Miller, 2008; Terry, 1967), criminal history (McAra & McVie, 2007; Werthman & Piliavin, 1967), and the seriousness of the criminal event that led to the police contact (Black & Reiss, 1970; McEachern & Bauzer, 1967; McAra & McVie, 2007). There are also a few predictors that are less commonly studied in reference to police contact but have appeared in the literature. These include involvement with delinquent peers (Patterson, Forgatch, & Yoerger, 1998), intelligence, as measured by an intelligence quotient (IQ; Fergusson, Horwood, & Ridder, 2005), and drug or alcohol use (Crawford & Burns, 1998; Terrill & Paoline, 2007).
Despite the large number of studies on the topic, the impact of many of these predictors is still open to debate. In addition, out of the 29 studies examined for the current purposes, only 7 looked at police contact over any period of time (Chambliss, 1994; Fergusson et al., 2005; McAra & McVie, 2007; McEachern & Bauzer, 1967; Menard & Morse, 1984; Patterson et al., 1998; Shannon, 1988), and many of these studies did not focus on the change (or continuity) of predictors of police contact. The following is an examination of relevant literature, subdivided by whether the study was clearly a cross-sectional examination or whether it was longitudinal in nature.
Cross-Sectional Examinations of Police Contact
A large number of cross-sectional studies focus on race as a predictor of police contact, and a majority do indicate that race is an important factor (Brown, 1981; Brunson, 2007; Durose et al., 2007; Fagan & Davies, 2000; Geiger-Oneto & Phillips, 2003; Miller, 2008; Ousey & Lee, 2008; Piliavin & Briar, 1964; Smith & Visher, 1981; Werthman & Piliavin, 1967). These studies also indicate some consistency across time and different study methodologies. For example, Werthman and Piliavin (1967) observed and interviewed police officers and gang members in Oakland and San Francisco, California. Police officers who were interviewed reported that “Negros were more likely to cause public disturbances than whites” (p. 75). They also stated that being a “Negro” who is in a White neighborhood or is “out of place,” is an automatic indicator of suspiciousness, deserving of contact. African American gang members interviewed in this study reported that the clothes they wore and/or the hair styles that were popular among African Americans at the time were seen by police as a sign that the individuals wearing them were trying to look tough or that they were looking for trouble. These symbols automatically aroused suspicion and caused more police contact (Werthman & Piliavin, 1967). A more recent qualitative analysis, published in 2007, which examined 40 “at-risk” men in St. Louis, Missouri, had results similar to Werthman & Piliavin’s 1967 findings (Brunson, 2007). Specifically, the men interviewed in this research felt that Black people with gold teeth or caps, or with nice clothes, were automatically suspected by the police of selling drugs (Brunson, 2007).
Quantitative studies have also been performed on the relationship between race and police contact. In 2003, Geiger-Oneto and Phillips studied police contact for traffic offenses in Houston, Texas. They noted an interaction between race and gender when they found that African American and Hispanic males experience more social control in traffic stop situations than do females of all races. Ousey and Lee (2008) examined police contact on a larger scale when they examined city level data for 136 different U.S. cities with a population over 100,000. They found that for “soft” crimes (drug and weapon violations), in which police have more discretion, there is an unexplained arrest disparity (higher for Blacks than for Whites). Again, despite the time span between the studies and the different methodologies used, race is consistently found to be a significant predictor of police contact.
However, a minority of cross-sectional studies have indicated that race is not a predictor of police contact (Black & Reiss, 1970; D’Alessio & Stolzenberg, 2003; Horowitz & Pottieger, 1991; Lundman & Kowalski, 2009; Lundman, Sykes, & Clark, 1978; Miller, 2008; Pope & Snyder, 2003). Black & Reiss (1970) systematically observed two police precincts each, in Boston and Chicago, and four police precincts in Washington, D.C. They 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: (a) most police contacts were citizen-initiated and, therefore, cannot be the result of police prejudices, (b) African American suspects were more often suspected of a felony offense at the time of contact, and (c) African American suspects more often had African American complainants who demanded that an arrest be made (Black & Reiss, 1970).
More recently, Lundman and Kowalski (2009) performed a study regarding police contact with speeders in Massachusetts. They found that in 65 mile per hour (MPH) zones, Black drivers were more likely to be high-rate speeders (15+ MPH over the speed limit) than White drivers. These authors suggest that their findings might explain why Black drivers get ticketed at levels disproportionate to their numbers in the population.
Gender is also a fairly commonly studied predictor of police contact. In reporting data from a police–public contact survey, which was distributed by the Bureau of Justice Statistics (BJS) during the last 6 months of 2005, Durose et al. (2007) found that males were more likely than females to come into contact with the police. In traffic stop situations, males were more likely to be ticketed and 3 times more likely to be arrested than females were (Durose et al., 2007). As noted previously, Geiger-Oneto and Phillips (2003) who studied traffic stops in Houston, Texas, found an interaction between race and gender when they found that African American and Hispanic males were more likely to experience social control than females of any race.
Other studies have found that gender does not predict police contact (Horowitz & Pottieger, 1991; Smith & Visher, 1981). Horowitz and Pottieger (1991), who studied 391 youths, 14 to 17 years old, in Miami, Florida, found that being male had no significant impact on arrest. Their research indicated that males are more likely to commit serious offenses and are, therefore, arrested in larger raw numbers. Lundman and Kowalski (2009) had similar results in Massachusetts when they found that, in 55 and 65 MPH zones, men were significantly more likely than women to be high-rate speeders (15+ MPH over the speed limit). Like Horowitz and Pottieger (1991), they suggest that this discrepancy in illegal behavior could be the reason for the disproportionate contact between males and the police.
Age is even more mixed regarding whether it is considered a significant predictor of police contact. Durose et al. (2007) reported that face-to-face contact with the police was most common with suspects aged 18 to 24 years. These findings are supported by Miller (2008) who found that age (being younger) was a significant predictor of warning and ticket stops for local police and a significant predictor of ticket stops (but not warning stops) for state police in North Carolina.
Other researchers disagree. Smith and Visher (1981), who used data collected by trained civilians who rode on 900 patrol shifts with 24 police departments in metropolitan St. Louis, Missouri; Rochester, New York; and Tampa-St. Petersburg, Florida, examined age as an influence in police-contact situations; they found only an indirect effect. They discovered that (a) suspects above 35 years are more antagonistic toward police and commit more serious offenses, both of which increase the probability of arrest, and (b) suspects above 35 years 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). Lundman and Kowalski (2009) who studied police contact with drivers in Massachusetts found that in 55 and 65 MPH zones, individuals who were speeding at 15 or more MPH above the speed limit were likely to be younger than 25 years. This disparity in illegal behavior may explain why drivers below 25 years of age are being contacted more often by officers, in traffic stop situations.
The majority of cross-sectional 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 it to be significant (Fagan & Davies, 2000; Geiger-Oneto & Phillips, 2003; Werthman & Piliavin, 1967). Werthman and Piliavin (1967), in their qualitative analysis, found that police used the SES of the parents, 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), who studied street stops in New York City, found that “for all suspects, after controlling for crime, stops within the sub-boroughs were predicted by their poverty rates” (p. 495). Geiger-Oneto and Phillips (2003), who studied drivers in Houston, found that individuals of low SES experienced more social control from police in traffic stop situations than did those of higher SES.
Only two of the reviewed studies did not find SES to be a significant predictor of arrest (Miller, 2008; Terry, 1967). After controlling for the seriousness of the offense and the number of previous offenses, Terry (1967) found no significant relationship between being lower class and arrest. Similarly, Miller (2008) found that driver SES was not a statistically significant predictor of warning or ticket stops, by local or state police in North Carolina.
Substance use is something that has been examined fairly rarely and also only fairly recently in the study of police contact. Terrill and Paoline (2007), who observed police contacts in Indianapolis, Indiana, and St. Petersburg, Florida, noted that individuals who did not show any outward signs of alcohol or drug use were less likely to be arrested. Similarly, Crawford and Burns (1998), who studied arrests in Phoenix, Arizona, found that individuals who showed signs of chemical impairment were more likely to have force used against them during an arrest. Because both of these studies focused on outward indications of substance use, none of them separated out whether police contact differed by the legality of the substance being used (alcohol vs. illicit drugs). However, it appears that substance use is generally a significant predictor of police contact.
Research indicates that the magnitude of the current suspected offense and the individual’s criminal history are likely to adversely affect the likelihood and outcome of police-contact situations (Black & Reiss, 1970; Lundman et al., 1978; Miller, 2008; Werthman & Piliavin, 1967). Werthman and Piliavin (1967) found that juveniles who have no criminal history are more likely to be released than those with a criminal history, regardless of the individual’s current suspected offense, or the nature of his past offenses (Werthman & Piliavin, 1967). Miller (2008) found that traffic conviction history was a significant predictor of ticket stops by local and state police in North Carolina. Black and Reiss (1970) found that the likelihood of arrest increases as the seriousness of the suspected offense increases. Lundman et al. (1978), who did a replication of Black and Reiss’s study, found that 100% of felony offenses in their study ended in arrest or in attempted arrest. There appears to be little debate that criminal history and the seriousness of the offense at hand do predict police contact.
Longitudinal Examinations of Police Contact
In 1967, McEachern and Bauzer examined factors that affected the disposition of police contacts with juveniles. They used data from the Central Juvenile Index, maintained by the Los Angeles County Sherriff’s department, as well as records of all juvenile-police contacts in Santa Monica, California, between 1940 and 1960. Though they were not examining change over time specifically, they did note some consistency when they concluded that there was a statistically significant positive relationship, in every case, between the seriousness of the offense at the time of police contact, and whether a formal petition was requested.
Menard and Morse (1984) and Fergusson et al. (2005), who independently used longitudinal data, were among the only studies to examine the impact of intelligence and academic achievement on crime and police contact. Through the use of data from a longitudinal random subsample of San Diego high school youths (N = 257), Menard and Morse (1984) found a statistically significant, direct relationship between IQ and academic achievement but did not study either as a predictor of police contact. Instead, they examined IQ and academic achievement as predictors of delinquency. Fergusson et al. (2005) did look at 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. In their multivariate model, they found that IQ at ages 8 and 9 was not a significant predictor of arrest and conviction after the age of 15 (Fergusson et al., 2005).
Shannon (1988) compared birth cohorts from 1942, 1949, and 1955 in Racine, Washington. He found that, regardless of the birth cohort, “minorities made up a disproportionate number of those referred” because they were subjected to more contact with the police and were disproportionately arrested for crimes that individuals of other races were not arrested for (p. 165). He also found that, across all three cohorts, females were counseled and released in higher percentages than were males (Shannon, 1988).
Chambliss (1994) did a longitudinal qualitative analysis. After riding with officers in Washington, D.C., for several years, he observed that consistently, police do unfairly contact racial minorities (African Americans and Latinos) and that the unfair contacts help to define these groups as criminal. The repeated arrests of racial minorities are used to justify longer prison sentences, which destroy the possibility of normal community and family relationships.
Patterson et al. (1998) and McAra and McVie (2007) are among the few researchers who have examined delinquent peers. Patterson et al. (1998) examined two successive birth cohorts of forth grade boys and their families (N = 72) in a city in Oregon. They found that deviant peer group involvement contributed to the progression of juvenile offending behavior, including the increased chances of early arrest (Patterson et al., 1998). McAra and McVie (2007) examined the impact of a number of variables, including delinquent peers, on the severity of sanctions received as a result of juvenile-police contact situations, in Edinburgh, Scotland. In their multivariate analysis, they did not find a significant relationship between delinquent peers and arrest (McAra & McVie, 2007). Like intelligence, this may be an area that is deserving of more study before any conclusions could be made regarding its impact on police contact.
When considering all the currently reviewed studies on police contact together, it appears that only criminal history and the seriousness of the current offense are agreed upon predictors of police contact across time, sample variation, and differing methodologies. The impact of some predictors such as race and gender remains contested, whereas other predictors such as intelligence/academic achievement and delinquent peers have been studied so rarely that a debate has not truly been opened. The current study will further the literature on this topic by examining all of these predictors across two generations of a national level sample of individuals. This will illustrate which of these variables are predictors of police contact on a large enough scale to be considered problems in policing across the United States. It will also give some insight as to whether significant predictors of police contact are consistent or changing across generations.
Data and Method
Data for this study comes from the NYSFS. The NYSFS, originally known as the National Youth Survey or NYS, began in 1976/1977 1 as a probability sample of U.S. households, based on a self weighting, multistage, and 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- to 17-year-old population in the United States in 1976 (Elliott et al., 1989). Because these individuals are the original respondents of the NYSFS, they will be referred to as the OR sample. Subsequently, 11 additional waves have been completed, the most recent of which was completed in 2004. Waves 1 through 5 were conducted annually from 1976/1977 through 1980/1981. Waves 6 through 9 were conducted every 3 years from 1983/1984 through 1992/1993, and Waves 10 through 12 were conducted annually from 2001/2002 through 2003/2004. The last wave completed on the OR sample, Wave 11, was completed in 2003 and contained 1,174 individuals. 2 In addition, for Wave 11, a total of 464 (66 %) adult offspring (AO) were interviewed, and 802 or 77% of the youth offspring (YO) were interviewed. Only the offspring samples were interviewed in Wave 12. Seventy percent, or 491, of the AO were interviewed and 815 or 78% of the YO were interviewed in this wave. The current study will examine OR data from Waves 4 through 11, as well as data from both types of offspring in Waves 11 and 12. 3
Variables
Police contact is the dependent variable in this study. This variable was measured in two ways: (a) arrest in the previous year (1 = yes, 0 = no) and (b) questioning by police about suspected involvement in a crime in the past year (1 = yes, 0 = no). These measures will be analyzed separately, with arrest being the more serious form of police contact and questioning being less severe. Respondents began being asked about arrest in Wave 5 (1980/1981), and they were first asked about being questioned by police in Wave 6 (1983/1984).
Where independent variables are concerned, race is a dichotomous measure in which respondents are classified as either White (1) or African American (2). This race classification was chosen for several reasons. First, in accordance with the U.S. population in 1976 (when NYSFS began), the majority of the sample is White, with a reasonable number of African Americans also included. Together, these two racial categories make up over 90% of each sample. 4 No other racial group (including a combined Hispanic measure) contained enough respondents to analyze separately. Second, it was determined that a White/non-White measure of race might confound too many racial categories that are each unique in the context of police contact, and deserving of their own evaluation. Finally, it is common in police-contact literature to see the comparison of White and African American racial categories (Brown, 1981; Brunson, 2007; D’Alessio & Stolzenberg, 2003; Piliavin & Briar, 1964; Pope & Snyder, 2003). This suggests that the current examination will be a useful addition to this large area of research, even if individuals of other races cannot be examined at this time. 5
Gender 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 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, for which higher scores indicate lower SES. For the AO and YO samples, the Hollingshead index was calculated for the OR parent and, if available, the spouse of the OR parent, then the score for the principal wage earner was assigned as the score for parental SES.
Involvement with delinquent peers is measured using a scale that consists of eight indicators. Respondents were asked how many of their friends had (a) damaged property that was not theirs, (b) used marijuana, (c) stolen something worth less than US$5, (d) hit or threatened to hit someone, (e) broken into a building or vehicle to steal something, (f) sold hard drugs, (g) stolen something worth more than US$50, and (h) 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), and all three were lagged by one wave to ensure correct temporal order with the dependent variable. 7 Alcohol use was included as a measure of licit substance use (at least for those older than the legal drinking age). Marijuana use was included as a “soft” illicit drug measure, whereas a composite measure consisting of the sum of hallucinogen, amphetamine, barbiturate, cocaine, and heroin use was used as a measure of “hard” drug use. Academic achievement is a self-report measure in which respondents were asked to estimate their grades on a 5-point scale ranging from mostly Fs (1) to mostly As (5). This measure is available for the OR and YO samples, but not for the AO. For the OR, the average was taken for respondents between the ages of 15 and 17 years. 8 For the YO, academic achievement was measured only once, when the individuals were 11 to 17 years of age.
Involvement in offending behavior was broken into two measures. Each of these measures was dichotomous, with 0 indicating no involvement in any of the behaviors, and 1 indicating involvement. As with substance use, these measures are also lagged by one wave to ensure correct temporal order. Minor delinquency includes involvement in any of the following: theft of something worth less than US$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. 9 Index offending includes involvement in any of the following behaviors: sexual assault, aggravated assault, gang fighting, robbery, motor vehicle theft, burglary, and theft of something worth more than US$50. Finally, previous police contact uses two measures: previous questioning and previous arrest, both measured one wave prior to the outcome. Table 1 illustrates the descriptive statistics for all variables included in the analysis.
Descriptive Statistics for All Variables.
Note: OR = odds ratio; AO = adult offspring; YO = youth offspring; SES = socioeconomic status. n represents number of observations.
Research Hypotheses
Based on previous findings, it is hypothesized that the following will result in a higher likelihood of police contact in the original respondents as well as in the offspring of the original respondents: (a) being non-White, (b) being male, (c) being young, (d) a low SES or parental SES, (e) increased exposure to delinquent peers, (f) the use of alcohol, (g) the use of marijuana, (h) the use of hard drugs, (i) low academic achievement, (j) involvement in minor delinquency, (k) involvement in index offending, and (l) prior police contact.
These hypotheses are based on previous research, which indicates that being non-White, being male, being young, having a low SES, increased exposure to delinquent peers, substance use, being involved in offending behavior, and having previous police contact all lead to an increased chance of police contact (Durose et al., 2007; Geiger-Oneto & Phillips, 2003; McAra & McVie, 2007; Miller, 2008; Patterson et al., 1998; Shannon, 1988). It should also be noted that substance use, IQ/academic achievement, and delinquent peers are rare in literature examining predictors of police contact (Fergusson et al., 2005; McAra & McVie, 2007; Patterson et al., 1998). Therefore, academic achievement, substance use, and delinquent peer measures have been included in the current study to check the consistency with previous findings. Table 2 illustrates the zero-order or bivariate correlations between the dependent variables and each of the independent variables.
Bivariate Correlations in AO and YO Data Sets.
Note: OR = odds ratio; AO = adult offspring; YO = youth offspring. Correlations are weighted to account for the number of waves the DV is present. All substance use and offending variables are lagged by one wave.
Analytical Strategy
Multilevel Bernoulli models were chosen to analyze the current research hypotheses in the OR data set. This technique was selected because it works well with a dependent variable that is dichotomous and with all types of independent variables (Menard, 2010). In addition, multilevel models can account for the possible clustering of events within individuals. Because the data is longitudinal in nature, it is possible that events that occur in one wave may affect the occurrence of the same events in future waves, within each individual. Without accounting for this possible clustering, estimates of statistical significance would be inflated. It should also be noted that a Laplace estimation was used to produce results that approximate the maximum likelihood estimation used in logistic regression (Raudenbush, Byrk, Cheong, Congdon, & Toit, 2004). In the present analysis, the first level consists of time varying measures, including substance use, involvement with delinquent peers, offending behaviors, and prior police contact. The Level 2 model consists of variables that are sociodemographic in nature and primarily fixed. In the current analysis, Level 2 includes age, race, gender, SES, and academic achievement.
Binary logistic regression was chosen to analyze the current research hypotheses in the YO and AO data sets. Because these two data sets are cross-sectional in nature (predictors in one wave and the dependent variables in the latter wave), it is unnecessary to account for the possible clustering of events. Therefore, binary logistic regression is an appropriate choice because each of the two dependent variables is dichotomous and binary logistic regression works well with any type of independent variable (Menard, 2010).
Results produced from running Bernoulli models with Laplace estimation in the statistical program HLM6 can be interpreted as logistic regression results. Accordingly, for both forms of regression, the model statistical significance will be tested using the model chi-square (χ2) statistic, which will be referred to as GM. 10 The model’s substantive significance will be measured using the likelihood ratio R2, also known as McFadden R2 (RL2). 11 Statistical significance of the predictors will be established using the Wald statistic, whereas substantive significance will be established using fully standardized regression coefficients (Menard, 2002, 2010). 12
Findings
Table 3 illustrates the arrest models for all three data sets (OR, AO, and YO).
Multivariate Models with Arrest as a Dependent Variable.
*Note: OR = odds ratio; AO = adult offspring; YO = youth offspring; SES = socioeconomic status.
All three models are statistically (p = .000) and substantively significant. The models explained 13.7% of the variation between being arrested and not being arrested in the OR sample, 26.4% in the AO sample, and 61.7% in the YO sample. In the OR model, arrest is statistically significantly more likely for individuals who are male (p =.000), or lower SES (p = .021), have more association with delinquent peers (p = .000), use marijuana (p = .006), and have been previously arrested (p = .004). Looking at the standardized coefficients, it is clear that gender has the biggest impact on arrest, followed in order by delinquent peers, marijuana use, SES, and previous arrest.
In the AO model, previous arrest is the only predictor that has a statistically significant relationship with arrest (p = .000). The direction of the relationship indicates that those who have been previously arrested are more likely to be arrested again. In the YO model, arrest is statistically significantly more likely for individuals who have more association with delinquent peers (p = .006), are more involved in index offending (p = .023), and who have been previously arrested (p = .004). The standardized coefficients indicate that involvement with delinquent peers has the greatest impact on arrest in the YO sample, followed in order by previous arrest and index offending.
Table 4 illustrates the results of all three models with “questioned” as the dependent variable.
Multivariate Models with Questioning as a Dependent Variable.
Note: OR = odds ratio; AO = adult offspring; YO = youth offspring; SES = socioeconomic status.
As with the arrest models, all three models are statistically (p = .000) and at least weakly substantively significant. The model explains 1% of the variation between being questioned and not being questioned in the OR sample, 26% in the AO sample, and 16.7% in the YO sample. In the OR model, being questioned is statistically significantly more likely for individuals who are male (p = .000), have more association with delinquent peers (p = .007), do not use alcohol (p = .001), and for those who use hard drugs (p = .007). Standardized coefficients indicate that gender has the largest impact on whether an individual was questioned by police, followed in order by the nonuse of alcohol, hard drug use, and delinquent peers.
In the AO model, being questioned is statistically significantly more likely for individuals who have more association with delinquent peers (p = .001) and who were previously questioned by police (p = .006). According to the standardized coefficients, the association with delinquent peers had the greatest impact on whether a person was questioned by police, followed by having been questioned previously. For the YO model, being questioned by police is statistically, significantly more likely for individuals who are younger (p = .045), have lower academic achievement (p = .011), are involved in minor delinquency (p = .030), and who have previously been questioned by police (p = .003). The standardized regression coefficients indicate that low academic achievement has the greatest impact on whether an individual is questioned by police in the YO sample, followed in order by age, involvement in minor delinquency, and previous questioning. 13
Discussion
For the arrest models, none of the findings are counter to expectation. There are, however, two findings in the questioning models that are counter to the relationships expected in the research hypotheses, and therefore are deserving of some discussion here. The first finding of interest is the relationship between questioning and alcohol use in the OR model. According to Table 4, being questioned by police is more likely for individuals who do not use alcohol. Though the zero-order correlation between alcohol use and being questioned by police did not reach statistical significance in the OR sample, the direction of the relationship is consistent with the multivariate finding. This could suggest that the multivariate finding is merely a statistical anomaly of the sample. Another possibility is that police do actually question nondrinkers more than they question drinkers of alcohol. It could be argued that those who do drink alcohol are more likely to be intoxicated when they come into contact with police. If that is true, then questioning them about involvement in a crime may be either difficult or fruitless.
Another counter intuitive finding is the relationship between age and being questioned by police in the YO model. Although research generally suggests that it is younger people who are more likely to come into contact with the police (Durose et al., 2007), recall that in the YO sample, ages are capped between 11 and 17 years. With this age range, one would expect it to be the older individuals in the sample that would be more likely to be questioned. The zero-order correlation supports this idea. To investigate this finding, several things were done. First, models were run with all possible alcohol interaction terms to check for the possibility of an interaction affect that might be changing the direction of the relationship between the predictor and the outcome in the multivariate model. None of the interaction terms were statistically significant. Second, a line graph and cross tab were run to examine the pattern of questioning across the age of the respondents. Although questioning does consistently increase with age (as suggested by the zero-order correlation), the cross tab and line graph indicate that the pattern is not a perfectly straight line. The largest increase in being questioned by police occurs between the ages of 11 and 12 (younger individuals) and then increases much more gradually thereafter. While this early increase in questioning may explain the finding in the multivariate model, the exact reason for the increase in questioning at this age is uncertain. Near 11 or 12 years of age, juveniles are shifting from elementary school to middle school. It is possible that the current finding is merely reflecting corresponding changes in juvenile behaviors at this age, which may lead to increased need for police to question them.
With the counter intuitive findings addressed, it is now important to discuss the change or consistency of predictors across two generations of respondents. To that end, there appears to be some consistency in predictors between generations. When both forms of police contact are considered jointly, involvement with delinquent peers, previous police contact, and some form of offending behavior are consistent predictors across generations. In addition, low academic achievement might also be considered a consistent predictor specifically of being questioned by police. Although it does not reach the more strict standard of statistical significance (p < .05) in the OR model, it does reach a standard of marginal statistical significance (p = .075) and, as noted in the findings, it is also statistically significant in the YO model. Where offending behaviors are concerned (marijuana use, hard drug use, minor delinquency, or index offending), it may be more important to note the presence of any of these behaviors in each model as opposed to the absence of any one of them. In the OR sample, when examined with crosstabs by year, it becomes clear that different forms of illegal behavior are significant in different years. When put together, these effects appear to be canceling each other out.
Notable differences in significant predictors between the two generations include gender and SES. Because there is no recent research that suggests that women are committing more crime on a wide scale, the loss of gender as a predictor of police contact between the two generations suggests a change in police behavior. In the most recent generation, police do not appear to be questioning or arresting males disproportionately, when controlling for involvement in offending behavior. Police also no longer appear to be disproportionately contacting individuals of a lower SES, again when controlling for offending behavior. Both of these changes are in a positive direction where fairness in police/public contact is concerned.
Also interesting to note is that race does not appear to be a predictor of police contact in any generation. Despite previous research, which indicates that this variable is an important predictor of disproportionate police contact in smaller areas within the United States (Brunson, 2007; Miller, 2008; Piliavin & Briar, 1964; Smith & Visher, 1981), it does not appear to be prevalent enough to be significant in a national level examination of respondents. On the other hand, academic achievement and delinquent peers, both of which have only been studied rarely in literature on police contact (Fergusson et al., 2005; McAra & McVie, 2007), were significant predictors of some form of police contact, across generations.
Where police policy is concerned, the results of this study appear largely favorable. At least on a large scale in the United States, police are contacting individuals because they are involved in some form of offending behavior, they have been previously involved in police contact, and/or because they hang out with delinquent peers. Though police should be cautioned against judging people solely based on past behavior or the friends they keep, it should also be noted that both of these variables are commonly found to be predictors of criminal offending (Menard & Elliott, 1990), and therefore, police may simply be overusing them in their efforts to catch offenders. The most troublesome finding is that police do appear to be questioning individuals who have lower academic achievement (and arguably lower intelligence) in a disproportionate manner. Although these individuals are not arrested disproportionately, this finding represents an area that is in need of further study to determine why police are doing this and whether it can be justified in the name of crime control.
As with any study, this study has some limitations that should be noted. Unfortunately, information on police contact was not collected in Waves 1 through 4 of the OR data. Having that information would allow for a direct comparison of the OR with their juvenile offspring at the same age. In addition, results of the analysis indicate that intelligence may be an important factor where police contact is concerned. To that end, it would be preferable to have IQ scores on all respondents, as opposed to the measure of academic achievement. Even without IQ scores, it would have been beneficial if the academic achievement measure was also available for the AO sample. Again, this would allow for a more direct examination of academic achievement.
Beyond issues noted in the previous discussion, it should also be noted that, for the OR sample, there were varying time lags between the waves of data collection. Because offending behaviors were lagged by one wave to ensure correct temporal order, this could mean a lag of 1 year (optimal) to a lag of 3 or more years. Finally, more data points in both generations would allow for an examination of change over time, within each generation. However, the current data does allow for a multigenerational examination of predictors of police contact in a national level sample of individuals. This provides important insight into the positive changes in predictors of police contact over time, and the possible need for more change. In addition, it suggests several predictors (specifically academic achievement/intelligence and delinquent peers), which perhaps should be included in future studies of police contact.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interests with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.