Is Peer Delinquency in the Eye of the Beholder?

Abstract

Researchers have devoted significant attention to the measurement of peer delinquency, with recent work indicating that perceptual measures are plagued by various biases. Absent from this research is an inquiry into whether the manner in which perceptions are typically operationalized potentially contributes to these limitations. In this study, we report on a methodological quasi-experiment where the operationalization of perceptual peer delinquency was manipulated across two different versions of a survey questionnaire completed by a sample of young adults. Results indicated no significant difference in the strength of the association between perceptual peer delinquency items and self-reported delinquency items across the two survey conditions. As such, this study provides preliminary evidence that existing limitations of perceptual measures of peer delinquency cannot be overcome by altering the manner in which such items are operationalized within survey questionnaires.

Keywords

peer delinquency measurement perceptions quasi-experiment peer influence

Measures of peer delinquency occupy a central role within contemporary criminological research (e.g., Boman, Stogner, Miller, Griffin, & Krohn, 2012; Boman, Young, Baldwin, & Meldrum, 2014; Meldrum & Boman, 2013; Rebellon & Modecki, 2014; Young, Rebellon, Barnes, & Weerman, 2014), prompted in large part by Sutherland’s (1947) differential association theory and Akers’s (2009) social learning theory. For several decades, the empirical importance attributed to peer delinquency has rested upon the finding that perceptual measures of peer delinquency are one of the strongest known correlates of one’s own delinquency (Agnew, 1991; Pratt et al., 2010; Warr, 2002). Yet, the validity of these measures has been questioned by several researchers (e.g., Gottfredson & Hirschi, 1990; Haynie & Osgood, 2005; Rebellon, 2012). For example, Gottfredson and Hirschi (1990) contended that perceptions of peer delinquency are biased to the extent that respondents reference their own behavior or impute friendship to individuals who engage in behavior similar to their own. Informatively, a growing number of studies provide evidence that perceptual measures are contaminated by more than random measurement error, resulting in inflated estimates of the concordance between peer delinquency and respondent delinquency (e.g., Boman et al., 2012; Matsueda & Anderson, 1998; Rebellon & Modecki, 2014; Young, Rebellon, Barnes, & Weerman, 2013).

To the extent that perceptual measures reflect more than a respondent’s actual knowledge of the delinquency of his or her peers and random measurement error, studies based on the use of such measures may produce upwardly biased estimates of peer influence. Given the importance attributed to perceptions of peer delinquency by social learning theorists (Akers, 2009), finding ways to purge such indicators of bias warrants attention. In this regard, some researchers have attempted to achieve this by using methods to statistically account for errors in perceptual measures of peer delinquency (e.g., Matsueda & Anderson, 1998). For example, Rebellon and Modecki (2014) recently found that correlating error terms between indicators of perceived peer violence and respondent violence reduced the association between latent factors of perceived peer violence and respondent violence by 15%.

Beyond attempting to statistically rid perceptual measures of bias, a more fundamental question concerns the manner in which such indicators have been measured. As Young et al. (2013) recently observed, “Operationalization of key theoretical constructs is a crucial step in any empirical examination” (p. 22). Traditionally, perceptual peer delinquency is measured by having survey respondents report on the delinquent behavior of their generalized peer group, oftentimes with ambiguous response categories (e.g., Elliott, Huizinga, & Ageton, 1985). It is possible that such operationalizations could contribute to the biases contained within perceptual measures of peer delinquency identified through prior research. Yet, no studies have sought to systematically investigate alternative operationalizations of peer delinquency and compare them against traditional operationalizations. Therefore, the purpose of this study is to provide a methodological quasi-experiment in which the operationalization of perceptions of peer delinquency is manipulated across two different versions of a survey questionnaire. Prior to describing the design of the study and measures, we first discuss in greater detail prior research focused on perceptions of peer delinquency, with particular attention given to studies identifying limitations surrounding the use of such measures and reasons why traditional operationalizations of perceptual measures might contribute to these issues.

Measuring Perceptions of Peer Delinquency

Central to tests of social learning theory (Akers, 2009) and differential association theory (Sutherland, 1947), traditional measures of peer delinquency are based on having survey respondents report on the behaviors of their friends. Put differently, rather than being objective, such indicators are subjective, perceptual assessments of the delinquency of peers. The prototypical example of a perceptual measure of peer delinquency is that which has been used in the National Youth Survey (NYS; Elliott et al., 1985) and utilized in a number of studies (e.g., Agnew, 1991; Matsueda & Anderson, 1998; Warr, 1993; Warr & Stafford, 1991). In the NYS, respondents are first asked to name friends they “run around with.” Then, they are asked, “Think of the people you listed as close friends. During the past year, how many of them [insert act here]?” Response options include “none of them,” “very few of them,” “some of them,” “most of them,” and “all of them” (coded 1-5, respectively). Thus, a proportional measure of perceptual peer delinquency is generated in the NYS by having respondents report on their generalized peer group within single items.

Other data sets have included similar operationalizations. For example, the Rochester Youth Development Study asks respondents to indicate how many of their friends have committed delinquent acts within the past 6 months, ranging from “none of them” (coded 1) to “most of them” (coded 4; see Thornberry, Lizotte, Krohn, Farnsworth, & Jang, 1994). Likewise, The Netherlands Institute for the Study of Crime and Law Enforcement (NSCR) School Project (Weerman & Smeenk, 2005) asks respondents how many of their friends have engaged in behaviors such as robbery and theft, with response options of “no one,” “some,” and “most or all” (coded 0-2, respectively). In a slightly different manner, the Add Health Survey (see Perrone, Sullivan, Pratt, & Margaryan, 2004) asks respondents to indicate, “Of your three best friends, how many use [drug] at least once a month?” with “0 friends,” “1 friend,” “2 friends,” and “3 friends” (coded 0-3, respectively) offered as options. Thus, while there are slight differences in the wording of various operationalizations of perceptual measures of peer delinquency across different studies, what remains constant is that respondents are asked to provide their perceptions of the delinquency of multiple friends within single items on survey questionnaires.¹

Issues in the Measurement of Perceptual Peer Delinquency

Although perceptual measures of peer delinquency like those described are found to be strongly associated with respondent delinquency (Meldrum & Boman, 2013; Pratt et al., 2010; Warr, 2002), significant criticism of these measures has been levied by various researchers. For example, Gottfredson and Hirschi (1990) contended that when being asked to provide information on the behavior of peers, individuals may reference their own behavior, what is referred to as projection (Newcomb, 1961). Concordantly, Hymel (1986) argued that “individuals may selectively attend to, utilize, and interpret information to which they are exposed” (p. 432), resulting in disproportionate attention to behaviors and attitudes that match their own. This, too, may lead to upwardly biased estimates of perceptual peer delinquency stemming from “assumed similarity” or “false consensus” mechanisms (Marks & Miller, 1987), where individuals overestimate similarity between the attitudes and behavior of others and their own attitudes and behavior.

On these issues, a growing number of studies provide evidence of bias in estimates of perceptual measures of peer delinquency stemming from projection and false consensus mechanisms (e.g., Boman et al., 2012; Mullen et al., 1985; Prinstein & Wang, 2005; Rebellon & Modecki, 2014; Young & Weerman, 2013). For example, in analyzing three waves of data collected as part of the NSCR ‘School Project,’ Young et al. (2013) found consistent evidence that while earlier perceptions of peer delinquency had minimal effects on later respondent delinquency, earlier respondent delinquency was strongly correlated with later perceptions of peer delinquency. In the words of Young and his colleagues (2013), this finding “call[s] into question social learning explanations of the correlation between peer and personal deviance by revealing that much of the correlation may be due to projection” (p. 22). Thus, cross-sectional correlations between perceived peer delinquency and respondent delinquency are likely to be upwardly biased.

One method that has been used to try and account for such bias was developed by Matsueda and Anderson (1998). In particular, they proposed that if projection and false consensus biases are contained within perceptual measures of peer delinquency, then there should be a correlation between the measurement errors of indicators of respondent delinquency and indicators of perceptual peer delinquency. Testing this hypothesis using data from the NYS, Matsueda and Anderson found statistically significant error correlations between self-reported theft and perceptions of peers having committed theft. They interpreted this finding as supportive of the contention that perceptual measures are contaminated by projection, stating,

This finding implies that cross-sectional correlations between delinquent peer associations and delinquent behavior would be overestimated due to correlations among response errors. Moreover, regressions of delinquency on contemporaneous (cross-sectional) measures of delinquent peers, typically used in cross-sectional analysis, may overestimate the coefficients. (p. 291)

More recently, Rebellon and Modecki (2014) examined this same possibility when evaluating associations between perceptions of peer violence and respondent violence. Similar to what Matsueda and Anderson (1998) found, their analysis revealed evidence of statistically significant error term correlations between indicators of perceived peer violence and respondent violence, reinforcing the idea that projection may be occurring when perceptual measures are used. However, unlike Matsueda and Anderson’s analysis, which only included models with correlated error terms, Rebellon and Modecki (2014) presented and discussed latent factor models assessing the relationship between perceptual peer violence and respondent violence before and after correlated error terms were added. Their findings revealed that once the error terms were added to the model, the correlation between perceptions of peer violence and respondent violence was reduced by 15%, indicating that failing to account for biases in perceptual measures results in inflated estimates.²

The Need to Explore Alternative Operationalizations of Perceptual Peer Delinquency

As discussed, some researchers have attempted to statistically account for biases in perceptual measures of peer delinquency. However, the fundamental issue which remains is whether the wording and response options used to operationalize perceptual measures of peer delinquency contribute to inflated coefficients linking peer and respondent behavior. Recall that the typical operationalization asks respondents to think about a generalized set of peers and select from oftentimes ambiguous options. For example, what is meant by “some friends” and “most friends” is subjective—One respondent may interpret “some friends” differently than another respondent. Such vagaries may inadvertently lead some respondents to selectively focus on peers whose behavior is more like their own. Moreover, asking respondents to recall delinquent behaviors engaged in by their group of friends within single items may result in a situation where limited thought is given to behaviors committed by peers outside of the group context.

Given this potential, and the body of evidence indicating that perceptual measures are plagued by various biases, there is a clear need to explore alternative ways to operationalize perceptual peer delinquency that might result in less bias stemming from the manner in which perceptions are operationalized within surveys. For example, asking respondents to focus and separately report on the delinquent behavior of individual peers one at a time may partially address shortcomings that stem from the use of more typical operationalizations of perceptual peer delinquency. For example, if respondents are primed and asked to report on the behavior of specific friends one at a time, they may be more likely to accurately account for incidents and behaviors while attributing it to the accurate peer and less likely to be undiscriminating and general. In essence, if a more discriminating question and response design is used, it may prompt more discriminating and accurate respondent appraisal of peer behavior. This may help to avoid situations in which friends who behave in ways that are incongruent with the respondent’s behavior are excluded from thought, which may reduce potential false consensus biases.

In addition, using this strategy to measure perceptions may reduce projection effects to the extent that being asked to reflect on the behavior of individual peers one at a time reduces the potential for respondents to reflect on their own behavior. In other words, priming respondents to report on the behavior of specific individuals may result in a perceptual measure of peer delinquency that more accurately reflects their true perceptions of what their friends have or have not done. Such a strategy may increase the potential that respondents will be more discriminating in who they are thinking about rather than comparing their behavior with that of a generalized peer group.

The Current Study

To our knowledge, research investigating the relationship between perceptual measures of peer delinquency and respondent delinquency has yet to consider whether the strength of this association varies according to the manner in which perceptual measures of peer delinquency are operationalized. The current study provides a methodological quasi-experiment to consider this issue. Traditional measures of perceptual peer delinquency are evaluated against an alternative operationalization. In doing so, the traditional measure (having respondents report on the behavior of multiple friends within a single question) is administered to one sample of young adults and evaluated against an alternative operationalization of perceptual peer delinquency that is administered to a separate sample of young adults that has respondents report on the behavior of specific friends one at a time.

Given that prior research indicates projection, and false consensus biases tend to inflate the relationship between peer and self-reported delinquency, we hypothesize that the strength of the association between perceptual peer delinquency and respondent delinquency will be weaker when using the alternative operationalization. This prediction rests on the supposition that respondents will give greater thought and consideration to their responses and draw less on their own behavior when primed to report on the behavior of individual peers rather than a group of peers.

Method

Procedure

Data for this study come from approximately 500 young adults enrolled in 12 different courses in a criminal justice department in January 2013 at a large university located in the southeastern United States. After receiving institutional approval to conduct the study, a number of instructors were approached and asked if their students could be recruited to participate. All instructors who were approached consented to allow the study researchers to invite their students to participate during the first class meeting of the first week of the semester. Potential participants in each class were told that the topic of the study focused on things that explain deviant behavior and victimization, but were not told that two different versions of the survey questionnaires (one containing a traditional operationalization of peer delinquency and the other with the alternative operationalization) were being passed out.

Individuals were informed that the choice of whether to participate or not would in no way affect their course grade. In accordance with the approved research protocol, verbal consent rather than written consent was obtained to maintain the anonymity of the participants. With these considerations in mind, 535 individuals present on the first day of their respective classes were given the opportunity to participate in the study. Of these individuals, 42 declined participation, yielding a participation rate of 92%. Of the 493 who elected to participate, 18 had missing data on one or more of the items used in the analysis and were excluded. Thus, complete data were available for 475 young adults.

Perceptual Peer Delinquency Operationalizations

The survey questionnaires passed out to all participants were identical with the exception that the manner in which perceptual peer delinquency was operationalized differed across two versions of the survey. To minimize concerns that participants might realize there were two different versions of the survey being administered, the first three pages were the same across all of the surveys. It was not until the fourth page of each survey that the items used to measure perceptions of peer delinquency were presented. For participants receiving the version of the survey containing the traditional operationalization, they were asked, “In the past 12 months, how many of your FOUR CLOSEST FRIENDS have [insert act here]?” For each behavior, respondents could check one of five boxes: “none of them,” “one of them,” “two of them,” “three of them,” or “all four of them” (coded 0-4, respectively). In other words, this operationalization of peer delinquency is consistent with the typical manner in which perceptions are operationalized in that respondents were asked to report on the behavior of multiple friends within single survey items.

Participants receiving the version of the survey containing the alternative operationalization of peer delinquency were asked, “Think of your CLOSEST friend. In the past 12 months, has this specific friend committed [insert act here]?” For each behavior, respondents could check one of two boxes on the survey: “no” (coded 0) or “yes” (coded 1). After the respondent answered each of the items pertaining to antisocial behavior committed by their closest friend, identical sets of items were then presented on the survey for their “SECOND CLOSEST,” “THIRD CLOSEST,” and “FOURTH CLOSEST” friend.³ Thus, this alternative format for measuring peer delinquency presents respondents with an operationalization that is unlike the more traditional operationalization, as respondents are asked to consider and report on the delinquent behavior of specific, individual peers one at a time before moving on to report on the behavior of other friends.⁴

We should point out that the assignment of surveys to study participants was not the result of a truly random process, which limits our ability to claim a true methodological experimental design was used. However, the manner in which surveys were distributed helped to ensure that assignment resembled a random process. Specifically, the surveys passed out to participants alternated between the version containing the traditional measure of peer delinquency and the version containing the alternative measure. Because the first three pages of all of the surveys were identical, when a participant received a survey, they had no idea that individuals sitting next to them had a different version. Importantly, as we will discuss shortly, this method of distributing the surveys produced two samples that were statistically equivalent to one another on a variety of measures. And, in fact, there was also statistical equivalence with regard to the number of surveys included in the analyses (traditional n = 240; alternative n = 235).

Measures

Respondent Delinquency

To measure respondent delinquency, each participant was asked, “In the past 12 months, how many times have you [insert act here]?” A number of different behaviors were assessed, including marijuana use, cheating on exams, theft, carrying a hidden weapon, driving under the influence, damaging property, and trespassing. For each of the acts, response options were “0 times,” “1-2 times,” “3-5 times,” “5-9 times,” and “10+ times” (coded 0-4, respectively). Of the 12 different behaviors assessed, there was limited variability for a number of the items. For the present analysis, we chose to focus upon behaviors where at least 10% of participants responded that they had engaged in a particular behavior at least once in the past 12 months. This left us with four indicators that were the focus of the present analysis: marijuana use, driving under the influence, use of other illicit substances, and cheating on exams.

Perceptual Peer Delinquency

As previously discussed, the manipulated condition in this study was the operationalization of perceptual peer delinquency. The assignment of scores for participants completing the version of the survey containing the traditional operationalization was rather straightforward. For each of the items on this version of the survey, respondents reported on how many of their four closest friends had engaged in eight different behaviors in the previous 12 months using the response options previously described. The measurement of perceptual peer delinquency for participants completing the version of the survey containing the alternative operationalization required that we add together the four dichotomized scores for each of the four friends, because respondents were asked to report on the behaviors of their four closest friends one at a time (no = 0, yes = 1).

The end result, then, is that the scale and possible range of values for each perceptual peer delinquency indicator obtained from surveys containing the traditional operationalization are identical to the scale and range of values obtained from surveys containing the alternative operationalization. What is different is that the values generated using the traditional format are based on a single item, whereas the values generated using the alternative format are based on the sum of four separate dichotomized items (one for each peer). For the analysis, only the four behaviors of peer marijuana use, driving while intoxicated, other illicit drug use, and cheating on exams were included to keep symmetry between the perceptual peer measures and respondent measures of delinquency.

Control Variables

To assess whether the experimental and comparison groups generated from the two survey conditions were statistically equivalent, and to account for any differences that might exist between the two groups in multivariate models, the surveys included items measuring a number of variables. The inclusion of several different items on the survey was also a practical necessity to ensure that the surveys were long enough that participants would not discover there were two versions being administered. With regard to demographic variables, an open-ended item measuring age in years was included, along with items measuring gender (male = 1) and race/ethnicity. As a high proportion of participants indicated they were Hispanic, a single dummy-coded variable (Hispanic = 1, all others = 0) was used to measure race/ethnicity. To measure college grade point average (GPA), an ordinal-level measure (3.75-4.00 = 1; <2.00 = 9) was included, with higher scores indicative of lower grades. All participants also responded to an ordinal item measuring the highest level of education their mother completed (less than high school = 1; graduate degree or higher = 6), with higher values representing more education.

In addition to these variables, we also included items on the survey measuring attitudes toward delinquency, attachment to school, and low self-control. Delinquent attitudes were measured based on eight items that asked respondents, “How wrong do you think it is for someone to [insert act here]?” Items included smoking marijuana, cheating on an exam, stealing, destroying property, and getting into a serious fight. Responses to each of the items ranged from “not wrong at all” (coded 1) to “very wrong” (coded 5). All items were reverse-coded, and the eight items were averaged together for the analysis (α = .76). Next, a five-item measure of attachment to school was included by asking respondents, “How much does each of the following statements apply to you?” Items included “I feel proud of my school,” “I feel I am part of my school,” and “The professors at my school treat me fairly.” Responses for each of the five items ranged from “strongly disagree” (coded 1) to “strongly agree” (coded 5). The five items were averaged together for the analysis (α = .79). Last, to measure low self-control, we included eight items from the Grasmick, Tittle, Bursik, and Arneklev (1993) scale, including impulsivity (I often act on the spur of the moment without stopping to think), anger (I lose my temper pretty easily), and risk-seeking (Sometimes I will take a risk just because it’s fun to do so) items. For each of the items, respondents were asked, “How much does each of the following statements apply to you?” with responses ranging from “strongly disagree” (coded 1) to “strongly agree” (coded 5). The eight items were averaged together for the analysis (α = .75).

Results

The analysis began by conducting a check on whether the two groups of participants representing the comparison group and experimental group were equivalent to one another on each of the analysis variables other than the manipulated perceptual peer delinquency items.

Table 1 presents the descriptive statistics for each of these variables for the entire sample, as well as the mean and standard deviations for each of the two groups. A series of t tests was conducted to determine whether significant differences existed between the two groups. Informatively, across each one of the control variables and each of the four measures of self-reported delinquency included in the analysis, the two groups are statistically equivalent. The largest t value is for the measure of low self-control (t = −1.76), and many of the t values are smaller than 1.00. As such, the process used to assign participants to complete either the comparison or experimental version of the survey questionnaire, though not truly based on random assignment, produced two groups that are statistically equivalent to one another.

Table 1:

Descriptive Statistics for Entire Sample and by Survey Group

	Full Sample (N = 475)			Comparison (n = 240)		Experimental (n = 235)
Variable	M	SD	Minimum/Maximum	M	SD	M	SD	t Value*
Age	22.22	4.07	17/57	22.05	3.70	22.39	4.41	−0.91
Male	0.43	—	0/1	0.45	—	0.40	—	1.10
Hispanic	0.73	—	0/1	0.73	—	0.74	—	−0.24
Poor GPA	4.80	2.42	1/9	4.95	2.39	4.64	2.45	1.40
Mother’s education	3.30	1.42	1/6	3.33	1.38	3.27	1.46	0.46
Delinquent attitudes	1.71	0.53	1/5	1.69	0.54	1.73	0.52	−0.82
School attachment	3.34	0.80	1/5	3.29	0.80	3.40	0.80	−1.50
Low self-control	2.38	0.68	1/5	2.33	0.69	2.44	0.67	−1.76
Marijuana use	0.86	1.45	0/4	0.85	1.45	0.87	1.45	−0.14
Other illicit drug use	0.23	0.72	0/4	0.27	0.79	0.19	0.64	1.20
Driving intoxicated	0.79	1.16	0/4	0.76	1.14	0.82	1.18	−0.55
Cheating on exams	0.53	0.85	0/4	0.51	0.82	0.55	0.88	−0.52

Note. GPA = grade point average.

There were no significant differences across any of the variables between the two groups.

Having established that the process used to assign participants to either the experimental or comparison survey condition resulted in equivalent groups, we turned attention to the primary focus of the study—whether the strength of the association between respondent delinquency and perceptual peer delinquency is substantively weaker under the experimental condition relative to the comparison condition. To assess this, we began by examining bivariate correlations between respondent and peer delinquency separately for the experimental and comparison group participants. Recall that the analysis focuses on four separate behaviors: marijuana use, other illicit drug use, driving while intoxicated, and cheating on exams.

As shown in Table 2, for each of the four behaviors considered, the strength of the correlation between respondent delinquency and peer delinquency is quite large, regardless of whether a participant completed the experimental or comparison version of the survey questionnaire. In fact, the weakest correlation in the table is a sizable .699 between respondent self-reported exam cheating and perceptions that their four closest friends engaged in cheating among those who completed the comparison condition version of the survey. What is also evident from an examination of the correlations in Table 2 is that, there is little difference in the magnitude of the correlations across the two groups for each of the behaviors examined.

Table 2:

Correlations Between Perceptions of Peer Delinquency and Self-Reported Delinquency

	Comparison (n = 240)	Experimental (n = 235)
Variable	r	r
Marijuana use	.749	.764
Other illicit drug use	.763	.740
Driving intoxicated	.757	.737
Cheating on exams	.699	.726

Note. Polychoric correlations estimated due to the limited number of categories for each item.

For example, the correlation between respondent self-reports of driving while intoxicated and their perceptions of their four closest friends having driven while intoxicated for comparison group participants was .757, while the correlation for experimental group participants was .737. Of the four behaviors considered, two of the correlations are slightly stronger for experimental group participants, whereas two of the correlations are slightly weaker for the experimental group participants. However, none of the differences are substantively different between the two samples, and a series of Fisher Z tests indicated no statistically significant differences in the pairs of correlations across each of the four behaviors. Thus, this portion of the analysis fails to support our stated hypothesis.

To further evaluate our hypothesis, we estimated a series of logistic regression equations where we dichotomized each of the four respondent behaviors under consideration, such that a respondent was assigned a value of 1 if they reported engaging in the behavior at least once in the past 12 months. Respondents reporting that they did not engage in a particular behavior at all were assigned a value of 0. The percentage of respondents who reported committing each of the behaviors at least once in the prior 12 months was as follows: marijuana use 34%, other illicit drug use 12%, driving while intoxicated 42%, and cheating on exams 36%.

We included in these models a variable indicating whether an individual had completed the comparison or experimental version of the survey (experimental = 1), each of the control variables, and an interaction term produced by multiplying together the value for the respective peer delinquency item and the dichotomized item indicating group membership. The inclusion of the interaction term, then, allows us to further assess whether the effect of peer delinquency on each of the four behaviors examined is significantly different among participants who completed the experimental version of the survey. A coefficient for the interaction terms that is statistically smaller than zero would provide support for our hypothesis.

Consistent with the bivariate analysis presented in Table 2, the results of the logistic regressions presented in Table 3 fail to support our hypothesis. None of the interaction terms are statistically significant, and across each of the four models they are close to zero. Put differently, completing the version of the survey containing the experimental measures for perceptual peer delinquency had no influence on the size of the effect for peer delinquency relative to completing the version of the survey containing the more typical measures of peer delinquency. Thus, taken as a whole, the results of our analyses provide no support for the hypothesis that the alternative operationalization of peer delinquency has any discernible influence on the strength of the association between perceptual peer delinquency and respondent delinquency.

Table 3:

Logistic Regressions of Delinquency on Peer Delinquency and Interaction Term (N = 475)

	Model 1		Model 2		Model 3		Model 4
	Marijuana Use		Other Drug Use		Driving Intoxicated		Cheating on Exams
Variable	b	SE	b	SE	b	SE	b	SE
Experimental Group	−.11	.47	−.27	.56	.11	.38	−.20	.30
Age	−.04	.05	−.13	.09	.06*	.03	−.21**	.07
Male	−.26	.27	.15	.37	.04	.25	.07	.25
Hispanic	−.58	.30	.00	.44	−.11	.29	.33	.29
Poor GPA	.03	.05	.05	.08	.07	.05	−.04	.05
Mother’s education	.14	.09	.23	.13	.19*	.09	.12	.08
Delinquent attitudes	.69*	.27	.63*	.29	.32	.26	.27	.24
School attachment	−.13	.16	−.23	.24	−.12	.16	.03	.15
Low self-control	.84***	.21	−.12	.28	.32	.20	.52**	.19
Peer delinquency item	.92***	.14	1.23***	.21	1.31***	.18	.79***	.14
Experimental group × Peer delinquency item	.05	.20	−.15	.28	−.17	.23	.12	.20
Pseudo R²	.36		.37		.36		.28

Note. GPA = grade point average.

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

Discussion

Prior research indicates the use of perceptual measures of peer delinquency results in biased estimates of peer influence (e.g., Boman et al., 2012; Rebellon & Modecki, 2014; Young et al., 2013) owing to things such as projection and false consensus. Yet, lacking from this area of research is work evaluating whether the manner in which perceptual measures of peer delinquency are operationalized contributes to such limitations. In this study, we considered whether an operationalization giving respondents the ability to report on the behavior of specific friends one at a time produces results that differ from those obtained using a more typical operationalization where respondents report on multiple friends’ behavior within single items. To consider this issue, we conducted a methodological quasi-experiment where a sample of young adults was assigned to complete a survey questionnaire containing one of the two operationalizations. In the following paragraphs, we discuss the findings of this study and their implications for researchers seeking to use perceptual measures of peer delinquency in future research. We then discuss the limitations of the study and directions for future research.

The primary finding of this study is that the operationalization of perceptual peer delinquency where respondents reported on peer behavior one peer at a time correlated with respondent delinquency in a manner identical (statistically speaking) to the operationalization of perceptual peer delinquency where respondents reported on peer behavior for multiple friends within single items. We found this to be the case in both bivariate and multivariate models across four different behaviors (marijuana use, other illicit drug use, driving under the influence, and exam cheating). This finding runs contrary to our stated hypothesis, yet nonetheless is practical in its implications. First, from a methodological standpoint, the version of the survey containing the alternative operationalization took up 4 times as much page space relative to the version of the survey containing the more traditional operationalization. Given what was found in this study regarding the invariance in the results across the two samples, and that page space and concerns over respondent fatigue are two issues that are central to the design of survey instruments, this study benefits scholars seeking to carry out future research by informing them that a lengthier operationalization of perceptual peer delinquency like that considered here makes little to no difference with regard to study conclusions.

Second, this null finding provides further insight into the potential limitations of relying upon perceptual measures of peer delinquency. We examined an alternative way to measure perceptions, yet produced results identical to those obtained using a more typical measure. This provides preliminary evidence that regardless of the manner in which perceptions are measured, it may be particularly difficult to produce a perceptual measure of peer delinquency that is free of respondent-driven biases. Until this issue can be resolved, the empirical significance attributed to perceptual measures of peer delinquency will continue to be questioned (Young et al., 2013). Having said this, it is important to recognize that not all researchers agree with the argument that perceptual measures of peer influence are biased. Although we interpret the findings of this study as suggestive of the idea that the alternative operationalization is as equally problematic as the more traditional operationalization, others may contend that the alternative operationalization operates just as well as traditional operationalizations.

It is important to consider the findings of this study in light of its potential limitations, each of which point to avenues for future research. First, this study relied upon a sample of undergraduate students. Although work focusing on the measurement of peer delinquency has sometimes been based on college samples (e.g., Boman et al., 2012; Rebellon & Modecki, 2014), most studies investigating the influence of peers on respondent behavior have been based on adolescent samples. Reliance on college students may result in a sample that is more homogeneous with regard to antisocial behavior relative to a community sample of youth. As we noted, there was limited variation for a number of the behaviors included on the survey questionnaires, which led us to only focus on four behaviors for the analysis. Given this, a replication of this study based on an adolescent sample is warranted. Likewise, future work should also consider how the findings of this study might apply to violence and property offending, two behavioral domains which were not considered in the present analysis.

Second, it is important to consider that different results could have emerged if the focus of the peer delinquency items was on the frequency of offending rather than on the prevalence of offending. Recall that for each of the two survey conditions, respondents were only asked to report whether or not close friends had committed a given behavior in the prior 12 months, not how many times friends had committed the behavior in question; frequency of offending was not measured. Different results could emerge with items that ask about the frequency of offending by peers. This possibility provides another potential important avenue for future research. Third, although there is now growing evidence that projection and false consensus mechanisms contribute to inflated estimates of peer influence when using perceptual measures of peer delinquency, there are rival explanations, including the possibility that large correlations could instead reflect shared method variance or repeat co-offending.

Fourth, and perhaps most important, the possibility exists that the alternative operationalization we developed and implemented was not enough of a radical departure from the traditional operationalization to elicit different response patterns from participants. Having individuals report on the behavior of individual peers, it appears, does not result in response patterns that are any different from when they are asked to report on multiple peers within a single question. Thus, additional alternative operationalizations of perceptions of peer delinquency should be compared against one another. For example, Meldrum and Boman (2013) recently suggested asking respondents to report on and differentiate between direct observations of peer behavior versus behavior which is learned about through secondhand knowledge, hearsay, or rumor. Significant changes in language and priming could be the key to developing perceptual measures of peer delinquency that possess greater validity than those which are currently used by the field. As no study other than the current study has investigated this potential, ample opportunity exists for additional research on this subject.

Although far from being conclusive, the current effort opens an important dialogue that has remained largely absent from consideration by researchers engaged in the dialogue surrounding the operationalization of peer delinquency. Given the centrality of the peer delinquency construct, and the fact that perceptual measures have been and will continue to be used by researchers, it is important to consider how biases and limitations associated with the use of perceptual measures can be minimized. Although this study has suggested that it may be difficult to overcome such obstacles, additional attention to this issue should help to yield a more nuanced understanding of the strengths and limitations of relying upon respondents to provide perceptions of the behavior and attitudes of individuals other than themselves.

Footnotes

Acknowledgements

The authors would like to thank Jean McGloin and the anonymous reviewers for helpful comments on earlier drafts of this manuscript.

Notes

Ryan Charles Meldrum is an assistant professor in the Department of Criminal Justice at Florida International University in Miami, Florida. His research interests include theory testing and individual-level correlates of delinquent behavior. His recent publications have appeared in Criminology, Journal of Quantitative Criminology, Criminal Justice and Behavior, and Crime & Delinquency, among others.

Jamie L. Flexon is an associate professor in the Department of Criminal Justice at Florida International University in Miami, Florida. Her research interests involve juvenile delinquency, minorities and crime, and evaluating criminal justice policy. Her recent work appears in such journals as Crime & Delinquency, Journal of Quantitative Criminology, and American Journal of Criminal Justice, among others.

References

Agnew

(1991). The interactive effects of peer variables on delinquency. Criminology, 29, 47-72.

Akers

R. L.

(2009). Social learning and social structure: A general theory of crime and deviance. Brunswick, NJ: Transaction.

Boman

J. H.

Stogner

J. M.

Miller

B. L.

Griffin

O. H.

Krohn

M. D.

(2012). On the operational validity of perceptual peer delinquency: Exploring projection and elements contained in perceptions. Journal of Research in Crime & Delinquency, 49, 601-621.

Boman

J. H.

Young

J. T. N.

Baldwin

J. M.

Meldrum

R. C.

(2014). Specifying the sources of misperceptions of peer deviance: A tale of two levels. Criminal Justice and Behavior, 41, 91-113.

Elliott

D. S.

Huizinga

Ageton

S. S.

(1985). Explaining delinquency and drug use. Beverly Hills, CA: SAGE.

Gottfredson

M. R.

Hirschi

(1990). A general theory of crime. Stanford, CA: Stanford University Press.

Grasmick

H. G.

Tittle

C. R.

Bursik

R. J.

Arneklev

B. J.

(1993). Testing the core empirical implications of Gottfredson and Hirschi’s general theory of crime. Journal of Research in Crime & Delinquency, 30, 5-29.

Haynie

D. L.

Osgood

D. W.

(2005). Reconsidering peers and delinquency: How do peers matter? Social Forces, 84, 1109-1130.

Hymel

(1986). Interpretations of peer behavior: Affective bias in childhood and adolescence. Child Development, 57, 431-445.

10.

Marks

Miller

(1987). Ten years of research on the false-consensus effect: An empirical and theoretical review. Psychological Bulletin, 102, 72-90.

11.

Matsueda

R. L.

Anderson

(1998). The dynamics of delinquent peers and delinquent behavior. Criminology, 36, 269-308.

12.

Meldrum

R. C.

Boman

J. H.

(2013). Similarities and differences between perceptions of peer delinquency, peer self-reported delinquency, and respondent delinquency: An analysis of friendship dyads. Journal of Criminal Justice, 41, 395-406.

13.

Mullen

Atkins

J. L.

Champion

D. S.

Edwards

Hardy

Story

J. E.

Vanderklok

(1985). The false consensus effect: A meta-analysis of 155 hypothesis tests. Journal of Experimental Social Psychology, 21, 262-283.

14.

Newcomb

T. M.

(1961). The acquaintance process. New York, NY: Holt, Rinehart & Winston.

15.

Perrone

Sullivan

C. J.

Pratt

T. C.

Margaryan

(2004). Parental efficacy, self-control, and delinquency: A test of a general theory of crime on a nationally representative sample of youth. International Journal of Offender Therapy and Comparative Criminology, 48, 298-312.

16.

Pratt

T. C.

Cullen

F. T.

Sellers

C. S.

Winfree

L. T.

Jr. Madensen

T. D.

Daigle

L. E.

. . . Gau

J. M.

(2010). The empirical status of social learning theory: A meta-analysis. Justice Quarterly, 27, 765-802.

17.

Prinstein

M. J.

Wang

S. S.

(2005). False consensus and adolescent peer contagion: Examining discrepancies between perceptions and actual reported levels of friends’ deviant and health risk behavior. Journal of Abnormal Child Psychology, 33, 293-306.

18.

Rebellon

C. J.

(2012). Differential association and substance use: Assessing the roles of discriminant validity, socialization, and selection in traditional empirical tests. European Journal of Criminology, 9, 73-96.

19.

Rebellon

C. J.

Modecki

K. L.

(2014). Accounting for projection bias in models of delinquent peer influence: The utility and limits of latent variable approaches. Journal of Quantitative Criminology, 30, 163-186.

20.

Rees

Pogarsky

(2011). One bad apple may not spoil the whole bunch: Best friends and adolescent delinquency. Journal of Quantitative Criminology, 27, 197-223.

21.

Sutherland

E. H.

(1947). The principles of criminology. Philadelphia, PA: Lippincott.

22.

Thornberry

T. P.

Lizotte

A. J.

Krohn

M. D.

Farnsworth

Jang

S. J.

(1994). Delinquent peers, beliefs, and delinquent behavior: A longitudinal test of interactional theory. Criminology, 32, 47-83.

23.

Warr

(1993). Parents/peers, and delinquency. Social Forces, 72, 247-264.

24.

Warr

(2002). Companions in crime: The social aspects of criminal conduct. New York, NY: Cambridge University Press.

25.

Warr

Stafford

(1991). The influence of delinquent peers: What they think or what they do? Criminology, 29, 851-866.

26.

Weerman

F. M.

Smeenk

W. H.

(2005). Peer similarity in delinquency for different types of friends: A comparison using two measurement methods. Criminology, 43, 499-523.

27.

Young

J. T. N.

Rebellon

C. J.

Barnes

J. C.

Weerman

F. M.

(2013). What do alternative measures of peer behavior tell us? Examining the discriminant validity of multiple methods of measuring peer deviance and the implications for etiological models. Justice Quarterly. Advance online publication. doi:10.1080/07418825.2013.788730

28.

Young

J. T. N.

Rebellon

C. J.

Barnes

J. C.

Weerman

F. M.

(2014). Unpacking the black box of peer similarity in deviance: Understanding the mechanisms linking personal behavior, peer behavior, and perceptions. Criminology, 52, 60-86.

29.

Young

J. T. N.

Weerman

F. M.

(2013). Delinquency as a consequence of misperception: Overestimation of friends’ delinquent behavior and mechanisms of social influence. Social Problems, 60, 334-356.