Specifying the Sources of Misperceptions of Peer Deviance

Abstract

“Peer deviance” is normally measured through one’s perceptions of the deviant behavior of friends. However, recent research suggests that peer deviance perceptions may be inaccurate and unreflective of a peer’s actual deviance. Using dyadic data, the current study addresses the potential for three distinct sources of misperceptions of peer deviance stemming from (a) the actor who generates the perception, (b) the friend about whose deviance is perceived, and (c) the friendship between the actor and the friend. Using multilevel regression alongside analyses of variance (ANOVAs), results demonstrate that misperceptions, overperceptions, and underperceptions of peer deviance occur frequently and systematically covary with the deviant behavior of the perceiver, the friend, and the total amount of deviance within the friendship.

Keywords

peer deviance perceptions misperception friendship dyads differential association measurement

Peer deviance is traditionally measured with an individual’s perception of his or her peer’s or peer group’s deviant behavior (e.g., Elliott, Huizinga, & Ageton, 1985). This measure, which is referred to as a “perceptual” or “indirect” measure of peer deviance (Weerman & Smeenk, 2005), is an important theoretical and pragmatic correlate of one’s criminal activities (see Akers, 2009). However, in the past two decades, criticisms of the indirect measure have appeared and challenged this seemingly “established phenomenon” (Merton, 1987). One argument questioning the causal mechanism of deviant peer influence stems from Gottfredson and Hirschi (1990), who claim that perceptions of peer behavior are analogous to an individual’s own deviant behavior. As a consequence, the validity of findings regarding the nature of peer influence procured from studies using such measures is a matter of debate (see Thornberry & Krohn, 1997).

Critiques of past findings resultant from indirect measures have redirected attention toward examining the validity and operationalization of deviant peer influence (e.g., Gottfredson & Hirschi, 1990; Kandel, 1996), and different ways to measure peer deviance have been developed. A growing number of studies have begun to use a measure of peer deviance reported directly by the peer himself or herself, a measure which is termed “direct” peer deviance. In this strategy, respondents self-report their behavior and nominate peers who also self-report their own behavior (Boman & Gibson, 2011; Boman, Stogner, Miller, Griffin, & Krohn, 2012; Haynie, 2001; Iannotti & Bush, 1992; Kandel, 1996; Meldrum, Young, & Weerman, 2009; Prinstein & Wang, 2005; Weerman & Smeenk, 2005; Young, Barnes, Meldrum, & Weerman, 2011).

Findings regarding the measurement of peer deviance have emerged in the literature and bear on the understanding of peer influence. Perhaps the main discovery in this area of research is that the perceptions of peer deviance may not accurately measure the peer’s deviant behavior because they might actually be misperceptions reflecting one’s own deviant actions. Individuals have been found to exaggerate the prevalence of behavior by “projecting” their own behaviors onto others (e.g., Davies & Kandel, 1981; Fisher & Bauman, 1988; Jussim & Osgood, 1989; Krueger & Zeiger, 1993; Ross, Greene, & House, 1977), resulting in a positive relationship between self-reported deviance and the misperceptions of peer deviance (Fisher & Bauman, 1988; Prinstein & Wang, 2005; Young et al., 2011; Young & Weerman, 2013). This relationship is continuing to be understood more extensively. Rebellon (2012) recently concluded that an actor’s perceptions of a peer’s drug use are extremely highly correlated to his or her own self-reported drug use. Research suggests that this finding holds true in cross-sectional (Matsueda & Anderson, 1998) and longitudinal frameworks (Zhang & Messner, 2000). This is perhaps because a respondent may tend to assume that his or her friends are behaving similar to himself or herself in the absence of absolute knowledge of what deviant acts the peers are committing (see Boman, Stogner, et al., 2012).

Another important recent discovery is that individuals with higher levels of deviant behavior may be more likely to overreport peer deviance, while nondeviants (and those who rarely engage in deviance) may underreport the deviance of peers (Young & Weerman, 2013; see Pratt et al., 2010). Although research in this area is just emerging, it appears that there may be differences between factors that affect general misperception (overall misperception) and those that influence overperceptions and underperceptions. For example, Young et al. (2011) constructed an “egocentric network” measure of misperception by examining the difference between a respondent’s report of peer delinquency with the delinquency reported by his or her nominated peers. Using the absolute value measure, they found that misperception was lower in dense networks but higher in networks where individuals spend a great deal of time together. In contrast, for the directional measure, self-control predicted underperception, while self-reported delinquency predicted overperception, suggesting that individual traits may be related to specific types of misperception.

While research that has produced these findings is informative, three critical features of prior studies prevent a complete understanding of misperception. First and foremost, prior research largely examines how the characteristics of the individual respondent may relate to misperception. This practice is alarming—perceptions of a peer’s deviance are perceptions of a peer’s behavior, yet research has investigated whether the characteristics of the individual respondent relate to misperception with little to no acknowledgment that the peer can influence a tendency to misperceive. For this reason, it is likely that the characteristics of the respondent, the friend(s), and the friendship itself will have a substantive relationship with misperception. Second, in past research on misperception, respondents perceive the deviance of their “peers” as opposed to the deviance of each individual peer. This lack of precision is problematic because in an abstract peer group, the perceptions cannot be verified based on the behavior of individual friends. Third, and finally, the majority of studies collapse multiple behaviors into a single profile of “deviant behavior.” Because different types of deviance are qualitatively different from one another (e.g., violence is quite different from theft), it is likely that there is a variability in misperception across distinct types of behaviors.

This study advances past literature on misperceptions of peer deviance in several important ways. First, we use a dyadic data set where indirect, perceptual measures of individualized peer deviance can be linked to the direct reports of the individualized behavior of specific friends. Second, our research design allows for an examination of the sources of misperception stemming from the individual respondent, the friend, and the dyadic friendship itself. Third, the data are suitable for an examination into general misperception and overperceptions and underperceptions specifically. Fourth, we use a wide range of behaviors that form five distinct latent constructs of deviance (theft, vandalism, violence, alcohol use and related behaviors, and drug use and sales) and examine the misperception of these constructs individually and cohesively as one general scale of “deviant behaviors.”

Current Study

Using a large dyadic data set of friends in emerging adulthood, the goal of the current study is to evaluate how self-reported deviance is related to general and directional misperceptions of five latent constructs of peer deviance. We seek to answer three questions. First, how does respondent deviance relate to general misperceptions, overestimations, and underestimations of peer deviance? Following Young et al. (2011) and Prinstein and Wang (2005), we hypothesize that higher amounts of self-reported deviance will be related to a greater misperception of peer deviance, in general, and more overestimation, specifically.

Second, how does peer involvement in deviant behaviors relate to the perceiver’s tendency to misperceive generally and, specifically, overestimate and underestimate? We hypothesize that greater peer deviance will be positively related to general misperceptions as well as underestimations on the part of the perceivers. We base this hypothesis on the idea that not all deviant behaviors committed by an individual are discussed with or brought to the attention of friends (Gottfredson & Hirschi, 1990). As such, the more deviant a peer is, a greater potential may exist that an actor will underestimate the deviance of that specific peer.

Third, how does the overall amount of deviance within dyadic friendships relate to an actor’s tendency to misperceive and over- and underestimate peer deviance? Because deviance seems to be positively related to overestimation-type misperceptions (Prinstein & Wang, 2005), we hypothesize that higher amounts of dyadic deviant behavior will be related to more misperceptions and overperceptions, but less underperceptions. We further investigate the relationship between misperception and dyadic deviance alignment. Using the same logic, we expect that the most deviant member of the dyad will misperceive and overestimate more than the least deviant member (see Young et al., 2011). Inversely, we also expect the least deviant member of the dyad to misperceive the least, but underestimate the most.

Method

Data and Sample

Data from the current study come from a sample of 2,154 individuals nested within 1,077 dyads (or friendship pairs). The respondents in the sample are undergraduate students at a large university in the southeastern United States. The data were collected as part of a larger project primarily focused on friendships and the operationalization and construct validation of the peer deviance construct. To gather respondents, the research team contacted the instructors of the 50 largest undergraduate courses offered at the university during the spring of 2009. Twenty-four instructors responded that they were interested in offering extra credit for study participation, and the principal investigator made in-class visits and described the study. The 24 classes had enrollments ranging from 50 to more than 1,500 students with a combined enrollment of approximately 5,000 students.

Although not randomly drawn from the university’s population, the sample demographics closely approximated the university’s characteristics. The sample is 62% White (60% of the population), 9% Asian (9% of the population), and 14% Black (15% of the population); 18% of the sample is Hispanic (16% of the population); and the sample has a mean age of 19.4 years (the population’s mean age is 20.0 years). The largest difference between the sample and population demographics was a slight overrepresentation of females in the sample (66% of the sample, 59% of population).

Each potential respondent was asked to come to the dyadic research project’s headquarters during designated operating hours with “one of [his or her] five best friends” currently enrolled in undergraduate studies. After agreeing to voluntarily participate in the study, the friends were separated and sent to different rooms where each was administered a paper-and-pencil survey. The surveys were identical in design and content and were pre-coded with a matching dyadic identification number to categorize the friends as a linked pair. The research team members monitored the rooms to ensure that no contact (e.g., texting) occurred between the friends during survey administration. Following the completion of the survey, each participant was individually debriefed and exited the project’s headquarters. The separation and monitoring of communication between the dyad members was designed to eliminate data contamination. Respondents in one (or more) of the selected courses received the designated amount of extra credit for the course(s) in which they were enrolled. Because there were more than 5,000 students in the participating classes, approximately 20% of the friends were also given extra credit for a course(s). The other 80% of the peers attended the study with the sole incentive of helping their friend receive extra credit.

Dataset Structure

In accordance with dyadic literature (e.g., Campbell & Kashy, 2002; Kenny, Kashy, & Cook, 2006), the current data set is structured as a double-entry (or pairwise) file. In a double-entry file, the units of analysis are individuals within dyads instead of the dyads themselves, and each individual receives his or her own line of data. There are two distinct pieces of information on each line of data. First, the respondent’s self-reported deviance and perceptions of the peer’s deviance are indicated (the indirect measure). Second, at the end of each line of data, self-reported deviant behavior measures from the respondent’s friend are inserted (the direct measure). The person whose information is listed first on the line of data is referred to as the “actor,” and his or her friend is referred to as the “partner” or “peer.” The end result is a data set where each person is nested as an actor and as a peer, meaning each respondent is the focal individual in analyses and a friend for another focal individual. Stated differently, Person 1 is an actor who has a friend (Person 2), and Person 2 is also an actor who has Person 1 as a friend. For more information on the double-entry dyadic data set, readers are encouraged to refer to the work of Campbell and Kashy (2002) and Kenny et al. (2006).

Measures

Individual Respondent Deviance Measures

Perceptual Peer Deviance

Each actor responded to 25 questions that asked about the deviant behavior of the specific friend with whom they attended the study. These perceptions inquired about a range of deviant behaviors that the friend may have committed over the past 12 months. Each of the items was originally scaled on the National Youth Survey (NYS; Elliott et al., 1985) frequency metric of 0 (“never”) to 8 (“two to three times a day”). The metric on which the items are measured is of considerable importance when conceptualizing “misperception.” Misperception using the 9-point NYS metric has a substantively different meaning than it would if the items were to be measured using a different scale; survey respondents seemingly will misperceive more when they have more ordinal choices from which to choose. That is, a metric with nine response options will produce more misperception than a scale with four response options because of there being eight potential incorrect responses versus only three, respectively. Almost all metrics, however, allow an investigation into whether a crime was perceived or committed or not. For this reason, each perceptual peer deviance item’s frequency was collapsed into a binary measure where “0” indicated that the actor perceived that the peer did not commit the act and a “1” indicated that the actor perceived the partner did commit the act.

Using these binary items, a single-order confirmatory factor analysis (CFA) with close fit (comparative fit index [CFI] = .940; Tucker–Lewis index [TLI] = .977; root mean square of approximation [RMSEA] = .041) revealed that five constructs were tapped: theft (six items), vandalism (five items), violence (four items), alcohol-related behaviors (five items), and drug-related behaviors (five items). This CFA model served as the basis of constructing five variety indices of theft, vandalism, violence, alcohol-related behaviors, and drug-related behaviors. A description of the individual items is available in Table 1.

Table 1:

Descriptive Statistics of Self-Reported and Perceptual Peer Deviance Measures by Latent Construct (N = 2,148)

		Self-Reported Deviance^a		Perceptual Peer Deviance
Item Description	Item No.	M	SD	M	SD
Theft construct
Stolen things worth less than US$5	1	.197	.398	.104	.305
Stolen things worth between US$5 and US$50	2	.080	.271	.045	.207
Stolen things worth more than US$50	3	.047	.213	.035	.185
Stolen from neighbors, friends, or roommates	4	.076	.266	.040	.195
Avoided paying for goods and services	5	.258	.438	.185	.389
Cheated on school tests	6	.329	.470	.231	.422
Vandalism construct
Vandalized property of neighbors or roommates	7	.106	.307	.065	.247
Vandalized school property	8	.097	.296	.040	.196
Vandalized other property	9	.136	.343	.053	.223
Broken into a building	10	.033	.179	.027	.161
Acted rowdy and unruly in public	11	.399	.490	.295	.456
Violence construct
Been involved in a group fight	12	.104	.305	.085	.279
Forced sexual contact	13	.011	.105	.012	.109
Hit another person	14	.249	.433	.142	.349
Thrown objects at cars or people	15	.146	.353	.101	.302
Alcohol construct
Drank alcohol	16	.844	.363	.797	.403
Drank more than five drinks at once	17	.644	.479	.621	.485
Lied about age to purchase alcohol	18	.313	.464	.279	.449
Bought alcohol for a minor	19	.225	.418	.174	.379
Been drunk in public	20	.523	.500	.455	.498
Drugs construct
Used marijuana	21	.330	.470	.263	.440
Used hard drugs, such as heroin, cocaine, and LSD	22	.041	.199	.028	.165
Used salvia (Salvia divinorum)	23	.043	.204	.023	.149
Sold marijuana	24	.041	.197	.031	.174
Sold hard drugs, such as heroin, cocaine, and LSD	25	.013	.111	.011	.105

Descriptive statistics reflect the deviance of the actor and the peer because of the structure of the double-entry data file (see Kenny, Kashy, & Cook, 2006).

Self-Reported Deviance

All respondents also answered 25 questions about their personal deviance over the past 12 months. These questions inquired about personal participation in the same behaviors they perceived for their peers. The actor’s and the partner’s self-reported deviance measures are used in this study. Originally scaled on the same NYS 9-point metric, the self-reported deviance items were collapsed so that a “0” indicated that the respondent had not committed the act in question in the past 12 months and a “1” indicated that he or she did commit the act.

Using these binary items, a single-order CFA was estimated for the self-reported deviant behavior measures. This factor analysis showed that the five factors of self-reported deviance (theft, vandalism, violence, alcohol-related behaviors, and drug-related behaviors) also fit the data closely (CFI = .940; TLI = .965; RMSEA = .043).¹ Most importantly, this CFA established that the factor structure of the perceptions and self-reports corresponded with one another—that is, the item structure of the perceptual peer vandalism construct, for example, is identical to the item structure of the actor and peer self-reported vandalism constructs because the constructs are measuring the same perceived and self-reported behaviors, respectively. In accordance with the CFA’s five latent constructs, five variety indices were created that summed the actor’s reports of his or her involvement in theft, vandalism, violence, alcohol, and drug-related behaviors. In addition, we estimated a higher order CFA model that loaded the five lower order constructs of self-reported deviance onto one latent construct of “self-reported deviant behavior.” This model (CFI = .913; TLI = .949; RMSEA = .052) also showed the evidence of a close fit to the data. As such, a 25-item self-reported deviance variety index was created. Descriptive statistics of these variety indices are presented in Table 2.²

Table 2:

Descriptive Statistics of Variety Indices Used in Analyses (N = 2,148)

	Self-Reported Deviance	Misperception	Overperception	Underperception	Range
	M (SD)	M (SD)	M (SD)	M (SD)	Min.-Max.
Full measure
General (25 item)	5.287 (3.892)	4.020 (3.115)	1.437 (2.172)	2.583 (2.796)	0–25
Latent constructs
Theft (6 items)	.987 (1.291)	1.152 (1.267)	.402 (.856)	.750 (1.102)	0–6
Vandalism (5 items)	.772 (1.070)	.837 (1.071)	.272 (.643)	.565 (.946)	0–5
Violence (4 items)	.510 (.805)	.607 (.828)	.218 (.549)	.388 (.687)	0–4
Alcohol (5 items)	2.550 (1.578)	1.082 (1.013)	.429 (.738)	.653 (.915)	0–5
Drugs (5 items)	.468 (.796)	.343 (.701)	.115 (.446)	.227 (.562)	0–5

Dyadic Deviance Measures

Dyadic Deviant Behavior

To examine how the overall amount of deviance within the dyadic friendship is related to the actor’s tendency to misperceive peer deviance, a measure of dyadic deviance was created by summing the self-reported deviant behaviors of the actor and the partner. The measure has a range of 0 to 50 different acts of deviance committed over the past 12 months.

Most Deviant Dyadic Member

We also hypothesized that the most deviant member of a friendship would overestimate and misperceive the most but underestimate the least. To evaluate this hypothesis, we created a measure based on information from the self-reported deviant behavior of the actor and the partner to distinguish whether the friends were equally deviant: whether the actor was more deviant than the partner or the partner was more deviant than the peer. Three values—“0,” “1,” and “2”—were assigned to this nominal variable. The actor’s and partner’s deviance scores had to be precisely equivalent for an equally deviant score of “0” to be assigned. If the actor’s deviance score was greater than the partner’s deviance score, a score of “1” was assigned. Likewise, if the actor’s deviance score was less than the partner’s deviance score, a value of “2” was assigned.

Misperception Measures

General Misperception

Using information from the actor’s perceptions of peer deviance and the peer’s self-reported deviance measures, a series of misperception measures were created. All misperception measures are actor variables. Three measures (general misperception, overperception, and underperception) were constructed from a similar root set of items. To construct this root set of items, the peer’s self-reported deviance items were subtracted from their corresponding actor perceptions. This created a trichotomy of scores (−1, 0, and 1) for each item. The score of “0” is a correct perception. This indicates that either the respondent perceived the behavior and the peer self-reported it or, alternatively, the respondent perceived no incidence and the peer self-reported none. On the other hand, scores of “−1” and “1” on these variables are misperceptions.The first set of misperception measures created was a general misperception measure. To construct the general misperception measures, the root items were replicated and recoded to dichotomous values where “0” indicates no misperception (or a correct perception) and a “1” indicates that the respondent misperceived the item. Dichotomous general misperceptions were summed to create a series of variety indices in accordance with the prior CFAs. The descriptive statistics of these measures are listed in Table 2.

Overperception

Using the same trichotomous root items, dichotomous variables were created that distinguished whether an actor overperceived (or overestimated) each deviant act of the peer. Overperceptions are indicated by a score of “1” on the individual trichotomous root items. In these instances, the actor perceived that the peer committed the deviant act, but the peer did not self-report the act. Accordingly, the root items were replicated and recoded so that scores of “1” indicated an overperception and “0” indicated no overperception (thus, grouping together accurate perceptions and underestimations). From these binary overperception items, a series of variety indices were created for the full measure of deviance and the five latent constructs drawn out by the CFA (see Table 2).

Underperception

The values of “−1” on the trichotomous root items indicate underperceptions. Specifically, this value represents the situation when an actor perceived no deviance while the peer indicated that he or she had committed the act in the past 12 months. To create the underperception dichotomies, the root items were again replicated and recoded so that a “1” indicates an underperception and a “0” indicates no underperception (grouping together accurate perceptions and overperceptions). These binary items were summed together to create a similar set of variety indices described in Table 2.³

Analytic Strategy

Several methodological options are available to explore how deviant behavior from the actor and the peer is related to the misperceptions of peer deviance. Because the data are nested, we initially used a dyadic multilevel mixed modeling technique called the “actor–partner interdependence model” (see Kenny et al., 2006). These models regressed the misperceptions of peer deviance onto actor self-reported deviant behavior variables and statistical controls (age, gender, and race) at Level 1 and used the dyadic identification number as a grouping variable at Level 2. While the models estimated successfully and showed no signs of methodological troubles such as multicollinearity, the results were complex. In an effort to further interpret what the actor–partner interdependence models were indicating, we employed a more descriptive analysis of variance (ANOVA) approach to validate the mixed models.

The ANOVA approach proved especially useful as it not only validated the Level 1 and 2 results of the mixed models but also eliminated a considerable amount of complexity. Because the multilevel actor–partner interdependence modeling approach was adding a level of complexity without adding any substantively different results, we integrated the two approaches and present the more straightforward ANOVA models coupled with Bonferroni’s comparison tests where possible.

To begin the presentation of results, we initially use the multilevel modeling approach to determine where the variance in the actor’s misperceptions lies. To accomplish this, we present intraclass correlations for misperceptions to determine if the variance in the actor’s deviance is related to the characteristics of the actor and the partner (at Level 1) or the friendship itself (at Level 2).

Next, we move on to investigate how the amount of deviant behavior committed by the actor is related to the amount of his or her misperception. Using the actors’ self-reported deviant behavior, a classification variable was created that placed the actors into quartiles representing the amount of deviance they committed (ranging from the least deviant quartile to the most deviant quartile). A series of ANOVAs paired with Bonferroni’s comparison tests were estimated to evaluate whether significant differences exist in misperceptions based on the amount of deviant behavior the actor has committed. Subsequently, the general misperception variable is replaced with overestimation and underestimation variables to see how the deviance of the actor relates to overperceptions and underperceptions of a peer’s deviant behavior. A conceptually identical process follows to address our second research question, with the important caveat that the focus is no longer on the deviant behavior of the actor but rather that of his or her friend (the partner).

Finally, we conduct our friendship-level investigation by estimating a series of ANOVAs that evaluate the total amount of misperception by the actor based on the amount of deviant behavior within the dyad, which was broken into quartiles to create a classification variable. We investigate how overperceptions and underperceptions relate to the quartiled measure of total dyadic deviance. We then conclude our friendship-level investigation by determining whether the most deviant member of a dyad generates the most misperceptions as hypothesized.

Results

Where Is the Variance in Misperception?

We first sought to determine whether the variance in misperception is the result of individual-level or dyadic-level characteristics. A series of six dyadic actor–partner interdependence null mixed models grouped across the dyad at Level 2 were estimated. These six models used the full 25-item general misperception measure and each latent construct’s (theft, vandalism, violence, drugs, and alcohol) general misperception measure as dependent variables. Figure 1 illustrates the intraclass correlations (represented by the point estimates on the graph) calculated from these models alongside their 95% confidence intervals. Generally, anywhere from one third to one half of the variation in an actor’s misperception stems from dyadic characteristics. Dyadic traits are responsible for almost half of the variance in misperceptions of a peer’s theft behavior (ρ_ICC = .467) and for over a third of the variance in misperceptions of a peer’s vandalism and violent deviance (vandalism: ρ_ICC = .411; violence: ρ_ICC = .388). Approximately one third to one half of the variance in the misperceptions of substance use and associated behaviors stems from the dyad (alcohol: ρ_ICC = .394; drugs: ρ_ICC = .464). Overall, half of the variation in misperception for the 25-item index is coming from the characteristics of the dyad (ρ_ICC = .510). Collectively, Figure 1 provides strong evidence that the characteristics of the actor and peer (at Level 1) and the dyad itself (at Level 2) are accounting for variance in an actor’s general tendency to misperceive peer deviance. As such, these models validate the further exploration of our hypotheses because significant amounts of variation exist at both the individual and dyadic levels.

Figure 1:

Intraclass correlations for null models

The Relationship Between Actor and Partner Deviance and Misperception

A series of ANOVAs presented graphically in Figure 2 address the amount of misperception, overperception, and underperception based on the amount of deviant behavior committed by the actor and the partner. The total degrees of freedom for each model are the sample size minus one. To begin the presentation of results, the overall 25-item scale of deviant behavior is used.⁴ Deviance of both the individuals (the actor and the partner), which is found on the x-axis, is classified by quartiles with the least deviant individuals in Quartile 1 (Q1) and the most deviant in Quartile 4 (Q4). The point estimates are the means of misperceptions, overperceptions, and underperceptions based on the quartile of deviance of the actor and the friend. For the full measure of misperception, the ANOVA reaches high levels of statistical significance (F = 65.21, p ≤ .001). As the amount of misperception increases, the amount of deviance by the actor also increases. Unreported Bonferroni’s comparisons reveal that actors in all the four quartiles of deviant behavior misperceive significantly different amounts of peer deviance, with the exception of the middle quartiles—actors in Q2 and Q3 of deviant involvement do not differ significantly in the amount to which they misperceive. The same general trend is true for overperceptions; more deviant actors overestimate peer deviance significantly more than less deviant actors (F = 190.14, p ≤ .001). Bonferroni’s comparisons indicate that the mean levels of overperception are significantly different between all the quartiles of deviant involvement. However, the overall ANOVA model investigating the relationship between the amount of deviance by the actor and underperceptions of the peer’s deviance barely reaches significance (F = 3.05, p ≤ .05). While actors in Q2 have underperceived the peer’s deviance significantly more than actors in Q4, no other differences are significant, and the amounts of underperceptions are relatively unchanging across the range of actor deviance.

Figure 2:

ANOVA results: misperception, overperception, and underperception by amount of deviance by the actor and the peer

Figure 2 also addresses the amount of an actor’s misperceptions, overperceptions, and underperceptions based on the amount of deviance committed by his or her peer. The first model (F = 446.61, p ≤ .001) addresses general misperceptions and shows a clear and robust trend. As the peer’s deviance increases, the amount of misperception increases, and Bonferroni’s comparisons indicate that all increases are significant. Actors who have the least deviant peers misperceive 1.8 deviant acts out of 25 on average, while actors with the most deviant peers misperceive more than seven deviant acts on average. The next model assesses how the overperceptions of peer deviance are related to the peer’s self-reported deviance. The model barely reaches significance (F = 2.77, p ≤ .05) and shows that there is little variation in the amount of overperceptions based on the amount of deviance of the peer. However, the final model looks at the relationship between the underperceptions of peer deviance and the peer’s self-reported deviance and shows a large amount of variance in underperceptions across the amount of peer deviance (F = 637.03, p ≤ .001). As the peer’s deviance increases, the amount respondents have underperceived increases dramatically and significantly. Actors with the least deviant peers underperceive an average of only 0.56 items out of 25, but for the most deviant peers, the number of underperceptions spikes to almost six items out of 25. Thus, a peer’s self-reported deviance appears not to have a great influence on the amount an actor overestimates peer deviance but instead on the extent to which an actor underestimates peer deviance.

Table 3 reports overestimations and underestimations across the five latent constructs of deviance, based on the amount of the actor’s deviant behavior. Because the number of offenses within each latent construct is limited, the quartiled measures of deviant behavior were replaced with a classification scheme based on the amount of deviant behavior the actor committed per construct. The classification has four groups representing actors who committed none of the offenses in the construct (“0 acts”), one of the offenses (“1 act”), two of the offenses (“2 acts”), or three or more of the offenses (“3+ acts”). The overperception results yield a general trend; as the actor’s deviance increases, so does his/her tendency to overestimate the deviance of his or her friend. Without exception, this trend is true across all the five latent constructs of deviance. Among the five latent constructs, the highest average overestimation comes from actors who commit very high amounts of theft; these individuals have overestimated an average of 1.325 of the six total peer theft items to which they responded.

Table 3:

ANOVA Results Showing Amount of Overperception (OP) and Underperception (UP) from the Actor Based on the Amount of Deviant Behaviors Committed by the Actor (Means and Standard Deviations Reported [N = 2,148])

		Number of Deviant Acts per Construct Committed by Actor					Bonferroni’s Comparison
Overperception	Range	0 Acts	1 Act	2 Acts	3+ Acts	F	Bonferroni’s Comparison
Theft OP	0–6	.089 (.383)	.396 (.656)	.699 (.891)	1.325 (1.500)	209.44***	All comparisons significant***
Vandalism OP	0–5	.071 (.316)	.310 (.541)	.654 (.927)	1.006 (1.192)	170.57***	All comparisons significant***
Violence OP	0–4	.083 (.314)	.316 (.575)	.677 (.861)	.921 (1.168)	143.76***	All comparisons significant***
Violence OP	0–4	.083 (.314)	.316 (.575)	.677 (.861)	.921 (1.168)	143.76***	(2 < 3**)
Alcohol OP	0–5	.171 (.459)	.308 (.611)	.342 (.682)	.555 (.814)	29.08***	0 < 2* and 3***
Alcohol OP	0–5	.171 (.459)	.308 (.611)	.342 (.682)	.555 (.814)	29.08***	1 and 2 < 3***
Drugs OP	0–5	.049 (.260)	.169 (.473)	.283 (.605)	.841 (1.405)	82.59***	All comparisons significant***, except 1 and 2
Underperception	Range	0 Acts	1 Act	2 Acts	3+ Acts	F	Bonferroni’s Comparison
Theft UP	0–6	.792 (1.145)	.779 (1.132)	.688 (1.043)	.587 (.898)	2.91*	0 > 3*
Vandalism UP	0–5	.584 (.944)	.562 (.920)	.508 (.994)	.514 (1.002)	0.56	None
Violence UP	0–4	.400 (.695)	.353 (.629)	.394 (.745)	.349 (.699)	0.62	None
Alcohol UP	0–5	.706 (1.096)	.737 (.952)	.912 (1.069)	.538 (.774)	17.18***	0 < 2*
							0 > 3*
							3 < 1 and 2*
Drugs UP	0–5	.218 (.552)	.227 (.562)	.283 (.526)	.349 (.786)	1.51	None

Note. OP = overperception.

p ≤ .05. **p ≤ .01. ***p ≤ .001.

The underestimation ANOVAs at the bottom of Table 3 show that, for the most part, underperceptions are unrelated to the deviant behavior of the actor. The exceptions to this are the constructs of theft and alcohol behaviors, which do reach the levels of statistical significance. Actors who refrain from theft underperceive significantly more than actors who engage in three or more acts of theft. For the alcohol model, the least underperception comes from those who self-report three or more of the alcohol behaviors in the past 12 months.

A series of ANOVAs presented in Table 4 investigate the relationship between over- and underperceptions of peer deviance and the amount of lower order construct deviance self-reported by the peer. Peer deviance is classified with a conceptually similar four-category (0 acts through 3+ acts) classification variable that was previously used in Table 3 for the actor’s deviance. The overperception ANOVAs show that, for the most part, the overestimations of the peer’s deviant behavior are unrelated to the amount of deviance the peer actually commits. Again, the theft and alcohol behavior constructs are the exceptions. Actors with peers who have not committed any acts of theft overestimate the theft behaviors of the peer significantly more than peers who have committed two or more acts of theft. Although the alcohol behaviors latent construct model shows several significant between-group differences, these differences are sporadic and do not display a clear progression or digression of overestimations. For example, the least overestimation is found when peers self-report three or more alcohol behaviors, which is in accord with our expectations. However, the second lowest amount of overperceptions for alcohol behaviors is found for peers who self-report zero acts, which is not one of our expectations.

Table 4:

ANOVA Results Showing Amount of Overperception (OP) and Underperception (UP) from the Actor Based on the Amount of Deviant Behaviors Committed by the Peer (Means and Standard Deviations Reported [N = 2,148])

		Number of Deviant Acts per Committed by Peer					Bonferroni’s Comparison
Overperception	Range	0 Acts	1 Act	2 Acts	3+ Acts	F	Bonferroni’s Comparison
Theft OP	0–6	.473 (.956)	.385 (.814)	.309 (.737)	.258 (.571)	6.13***	0 > 2* and 3***
Vandalism OP	0–5	.293 (.648)	.261 (.686)	.251 (.598)	.207 (.470)	1.15	None
Violence OP	0–4	.231 (.593)	.211 (.482)	.193 (.428)	.079 (.272)	1.80	None
Alcohol OP	0–5	.444 (.911)	.512 (.912)	.709 (.842)	.316 (.550)	28.47***	0 < 2***
							0 > 3*
							1 < 2**
							1 and 2 > 3***
Drugs OP	0–5	.114 (.437)	.110 (.458)	.115 (.477)	.206 (.481)	0.91	None
Underperception	Range	0 Acts	1 Act	2 Acts	3+ Acts	F	Bonferroni’s Comparison
Theft UP	0–6	.000 (.000)	.777 (.416)	1.528 (.649)	2.801 (1.371)	1879.64***	All comparisons significant***
Vandalism UP	0–5	.000 (.000)	.652 (.477)	1.581 (.609)	2.754 (1.261)	2037.49***	All comparisons significant***
Violence UP	0–4	.000 (.000)	.831 (.375)	1.462 (.733)	2.048 (1.263)	1853.00***	All comparisons significant***
Alcohol UP	0–5	.000 (.000)	.343 (.475)	.550 (.691)	.938 (1.055)	118.75***	All comparisons significant***
Alcohol UP	0–5	.000 (.000)	.343 (.475)	.550 (.691)	.938 (1.055)	118.75***	(1 < 2**)
Drugs UP	0–5	.000 (.000)	.424 (.495)	1.159 (.727)	1.921 (1.324)	879.48***	All comparisons significant***

p ≤ .05. **p ≤ .01. ***p ≤ .001.

While the overestimation models show mostly nonsignificant and sporadic significant differences, the underperception ANOVAs at the bottom of Table 4 show a robust and clear pattern. As the deviance of the peer increases for each latent construct, the actor’s estimates increasingly become incorrect in the underestimation direction. As indicated by extremely large model statistics, this trend is robust; actors who have peers who self-report three or more acts of theft, vandalism, and violence tend to underperceive over two acts per construct on average. For the latent construct of drug behaviors, actors with peers who self-report three or more acts tend to misperceive almost two drug-related acts. It would appear then, that the two most important sources of individual-level (Level 1) misperception are (a) having a peer that is actually deviant and (b) deviant behavior from the perceiver.

The Relationship Between Dyadic Deviance and Misperception

With the individual-level results completed, the focus of the analysis turns to investigating how misperception is related to the total amount of deviance within the friendship. Recall that a significant portion of variance in an actor’s misperception is attributable to dyadic characteristics (anywhere from 33% to 51%, depending on latent construct). A series of ANOVA models visually depicted in a series of plots in Figure 3 evaluate the amount of an actor’s general misperception (top left), overperception (top right), and underperception (bottom left) based on latent constructs, as well as the aggregated, higher order measure of dyadic deviance (bottom right). Dyadic deviance is plotted on the x-axis and is classified by quartiles with the least deviant dyads in Q1 and the most deviant dyads in Q4.

Figure 3:

ANOVA results: misperception, overperception, and underperception by amount of deviance within the dyad

The top left figure shows ANOVAs investigating the amount of misperception based on the five latent constructs of deviance. All five ANOVAs reach statistical significance (theft: F = 557.99, p ≤ .001; vandalism: F = 510.43, p ≤ .001; violence: F = 777.39, p ≤ .001; alcohol: F = 73.47, p ≤ .001; drugs: F = 276.54, p ≤ .001). As the amounts of theft, vandalism, violence, and drug usage within the dyads increase, the actors’ misperceptions also significantly increase (because of 0-inflated incidence rates, Q1 encompasses Q2 for violence and drug use). For these four models, Bonferroni’s comparisons indicate that the amount of misperception increases significantly based on which quartile of deviance the dyad is situated. The exception to this general increase in misperception as the dyadic deviance increases is found in the alcohol model. The significantly highest amount of misperception about a peer’s alcohol use comes from actors within dyads that have relatively high alcohol use (Q3).

The top right graph in Figure 3 shows ANOVA results comparing overperceptions to the amount of deviance within the dyad. All ANOVAs reach significance (theft: F = 59.50, p ≤ .001; vandalism: F = 54.24, p ≤ .001; violence: F = 82.86, p ≤ .001; alcohol: F = 13.26, p ≤ .001; drugs: F = 27.83, p ≤ .001) and show a weak, positive relationship between dyadic deviance and the amount to which the actors overestimate the peers’ deviance. However, the Bonferroni’s comparisons reveal only consistent significant differences in overestimation between actors in the least (Q1) and the most (Q4) deviant dyads. No model shows significant differences between the middle quartiles of deviance (Q2 and Q3). Again, the alcohol construct shows different visual results from the other four latent constructs. The most overestimations of a peer’s alcohol behaviors come from actors within dyads that are moderately deviant in terms of their alcohol behaviors.

The bottom left graph in Figure 3 depicts a series of ANOVAs that investigate the amount of underperceptions of a peer’s deviance based on the amount of deviance within the dyad. For four of the five latent constructs (theft: F = 303.26, p ≤ .001; vandalism: F = 311.80, p ≤ .001; violence: F = 510.17, p ≤ .001; and drugs: F = 256.86, p ≤ .001), actors tend to increasingly underperceive their peers’ deviance as the amount of deviance within the dyad increases. Comparatively, the changes in underperceptions as dyadic deviance increases are much more dramatic than the changes observed in the overperception models. Again, however, actors within dyads containing members who at least commit moderate amounts of alcohol use and subsequent behaviors tend to underperceive their friend’s alcohol behaviors the most (F = 43.16, p ≤ .001).

The bottom right graph in Figure 3 uses an aggregated, higher order measure of dyadic deviance to summarize the results. The misperception (F = 364.67, p ≤ .001), overperception (F = 64.64, p ≤ .001), and underperception (F = 156.77, p ≤ .001) models all reach high levels of statistical significance and all Bonferroni comparisons reveal statistically higher amounts of misperceptions, overperceptions, and underperceptions as deviance increases. Thus, significantly higher amounts of misperceptions generally (and over- and underperceptions specifically) are found in more deviant dyads. Overall, the findings from the graphs in Figure 3 show that more deviant behavior within the dyad is related to the members of the dyad both overestimating and underestimating a peer’s deviance.

Another series of ANOVAs presented in Table 5 investigate the relationships between an actor’s general misperceptions based on who in the dyad is the most and least deviant. Results indicate that an actor tends to misperceive the most when he or she is the least deviant member of the dyad. When the actor is the least deviant member of the friendship, misperceptions occur at a significantly higher rate than when he or she is the most deviant member of the dyad. This is true for the overall measure of misperception as well as the latent constructs of misperception. Thus, the most misperception comes from the member of the dyad who participates in deviant behavior the least—not the member who is the most deviant.

Table 5:

ANOVA Results Showing Amount of General Misperception (MP) from the Actor Based on Whether the Actor Is the Most or Least Deviant Member of the Dyad (Means and Standard Deviations Reported [N = 2,148])

	Range	Actor and Peer Are Equally Deviant	Actor Is Less Deviant Than Peer	Actor Is More Deviant Than Peer	F	Bonferroni’s Comparison
Gen. misperception
Full measure MP	0–25	2.242 (1.878)	5.040 (3.205)	3.439 (2.918)	117.87***	All comparisons significant***
Latent constructs
Theft MP	0–6	.406 (.734)	1.990 (1.233)	1.124 (1.242)	385.01***	All comparisons significant***
Vandalism MP	0–5	.254 (.571)	1.702 (1.124)	.810 (.987)	491.37***	All comparisons significant***
Violence MP	0–4	.242 (.553)	1.392 (.731)	.634 (.877)	494.33***	All comparisons significant***
Alcohol MP	0–5	.694 (.865)	1.426 (1.050)	1.037 (.964)	97.24***	All comparisons significant***
Drugs MP	0–5	.111 (.396)	1.029 (.865)	.379 (.798)	366.55***	All comparisons significant***

p ≤ .05. ** p ≤ .01. ***p ≤ .001.

Ancillary Analysis: The Relationship Between Misperceptions and Deviance for the Least Deviant Dyadic Member

Results presented to this point paint a rather complex picture. On one hand, actor deviance is positively related to over- and underperceptions, and more deviance within a dyad is also positively related to overperceptions and underperceptions. However, it is the least deviant member of the dyad who tends to misperceive the most. To further aid in the interpretation of why these different effects are present, we estimate a final set of ANOVAs that explore the relationships between actor deviance and misperception for actors who are the least deviant member of the dyad.

Table 6 depicts results from a series of ANOVA models that investigate how the amount of deviance by the least deviant actor within dyads is related to his or her tendency to generally misperceive, overestimate, or underestimate peer deviance. Because the misperception of only the least deviant dyadic member is under investigation, the sample size is now restricted to 956 (i.e., those who committed less deviance on the whole relative to their peer).⁵

Table 6:

ANOVA Results Showing Amount of General Misperception (MP), Overperception (OP), and Underperception (UP) from the Actor, if the Actor is the Least Deviant Member of the Dyad (Means and Standard Deviations Reported [N = 956])

		Amount of Actor Deviance, Classified by Quartile				F	Bonferroni’s Comparison
	Range	Least Dev. Actors (Q1)	Mid–Low Dev. Actors (Q2)	Mid–High Dev. Actors (Q3)	Most Dev. Actors (Q4)	F	Bonferroni’s Comparison
Full measure MP	0–25	4.066 (3.125)	5.004 (2.868)	5.621 (2.835)	8.185 (3.283)	45.64***	All comparisons significant***, except Q2 and Q3
Full measure OP	0–25	.438 (.936)	.660 (1.063)	.986 (1.295)	1.765 (1.832)	33.50***	Q1 < Q3 and Q4***
							Q2 < Q3*
							Q2 and Q3 < Q4***
Full measure UP	0–25	3.628 (3.104)	4.344 (2.916)	4.636 (2.838)	6.420 (3.657)	20.15***	Q1 < Q2*
							Q1 < Q3 and Q4***
							Q2 and Q3 < Q4***

p ≤ .05. **p ≤ .01. ***p ≤ .001.

For the general measure of misperception, actors who have misperceived the most also tend to be involved in deviant behavior (despite being the least deviant member of the friendship). When the least deviant dyadic member has committed very low amounts of deviance, he or she tends to misperceive approximately four out of 25 peer deviance items; when he or she is very deviant, he/she tends to misperceive more than eight acts. This finding carries some substantive importance because it indicates that the more deviant the least deviant dyadic member is, the more he or she misperceives. When the overall misperception measure is disaggregated into overperception and underperception measures, several relationships become evident. First, the most common type of misperception committed by the least deviant actor within the dyad is an underestimation of the peer’s deviance. Underestimations account for 89.2% (3.628 underestimations/4.066 total misperceptions) of perceptual error for those who are least deviant (Q1), 86.8% of errors in actors in Q2 deviance, 82.5% of error for actors in Q3, and 78.4% of misperceptions for Q4.

Second, despite not being the most common type of misperception for the least deviant dyadic members, the overestimations of peer deviance still increase in conjunction with the self-reported deviance of the least deviant dyadic member. This suggests that the least deviant dyadic member still estimates more deviance than the peer commits in some cases, and this proclivity increases as the actor’s deviance increases.

Third, and perhaps most importantly, the least deviant actor’s tendency to underperceive escalates as their deviance increases. That is, as the least deviant dyadic member’s self-reported deviant behavior increases, the more he or she overperceives and underperceives a peer’s deviance. For the least deviant member of the dyad—the person who misperceives significantly more than the most deviant member of the dyad—it appears that participation in deviant behaviors cripples what ability the respondent may have to correctly perceive a peer’s deviance.⁶

Discussion and Conclusions

Researchers are becoming increasingly aware of the limitations of relying upon perceptual measures of peer delinquency because of concerns that the use of such measures potentially leads to biased estimates of peer influence (e.g., Boman, Stogner, et al., 2012; Davies & Kandel, 1981; Gottfredson & Hirschi, 1990; Haynie & Osgood, 2005; Kandel, 1996; Meldrum et al., 2009). Part of this concern is the result of a limited, but important, body of emerging research that has started to consider the extent to which individuals misperceive peer deviance and the factors that predict misperceptions (Prinstein & Wang, 2005; Weerman & Smeenk, 2005; Young et al., 2011). The current study sought to build on and extend this literature in several important ways. In the following paragraphs, we summarize the findings, the implications they hold, limitations of the study, and provide closing remarks.

One set of findings pertains to the portion of the analysis that examined how the actor’s and the peer’s deviance is related to general misperceptions, overperceptions, and underperceptions of peer deviance. The first of these main findings is that respondents who are more often involved in delinquency are more likely to misperceive peer deviance. This was found not only for the general deviance measure that included all 25 behaviors but also for each of the five latent constructs. This is consistent with our first hypothesis and previous research (Prinstein & Wang, 2005; Young et al., 2011) but goes well beyond these previous studies by revealing that this relationship appears to hold across a wide variety of deviant behaviors. In addition, consistent with our first hypothesis and prior work (Young et al., 2011), the analyses further revealed that respondents who are more involved in deviance are also more likely to overperceive peer deviance. This was true for the general deviance measures and each of the latent constructs. However, the underperceptions of peer deviance were not found to be strongly systematically related to the perceiver’s deviance.

Unique to this study, an additional finding pertaining to individual actor and partner deviance in relation to misperceptions was that respondents are more likely to misperceive peer deviance when peers are more deviant. This is a particularly informative finding that previous research has not addressed. Namely, this result supports our second hypothesis and reveals that it is not just respondent characteristics that influence misperceptions but also the behavior of peers themselves. A nuanced understanding of this association emerged when examining the direction of these misperceptions. Specifically, underperception, but not overperception, was consistently related to greater levels of peer deviance, which is consistent with our research hypothesis dealing with this particular relationship.

The second set of findings pertains to the relationship between absolute and relative levels of deviance within dyads and an actor’s misperceptions of peer deviance. Being that the intraclass correlations reported in Figure 1 demonstrated that one third to one half of the variation in respondent misperceptions is attributable to the characteristics of friendship dyads, continuing to ignore the friendship as a source of misperception could potentially limit the field’s understanding of the sources of misperception.⁷ Three main findings were revealed at the dyadic level and provide mixed support for our stated research hypothesis pertaining to the relationship between misperception and dyadic deviance. First, in support of what was hypothesized, respondents are more likely to misperceive peer deviance when there are higher absolute levels of deviance within friendship dyads. This finding held for the general deviance measure as well as the individual latent construct deviance measures. Further investigation into the directionality of misperceptions in relation to absolute levels of dyad deviance provided less support for our research hypothesis as greater underperception and overperception are related to greater deviance within friendship dyads.

Second, the dyadic analyses revealed that general misperception for each of the latent constructs of deviance (and the general measure of misperception) is most likely to occur when respondents are less delinquent than their peers, a finding that runs counter to what we hypothesized. Perhaps if one is not involved in delinquency but his or her friend is, the relative deviance distance between the individuals might result in the less delinquent friend being more likely to misperceive peer deviance. In fact, subsequent analyses revealed this to be the case; the least deviant individual within each friendship dyad was far more likely to underestimate peer deviance than overestimate.

Finally, in the ancillary analysis, we found that the more deviant the least deviant member of the friendship is, the more likely he or she will misperceive peer deviance. This suggests that the absolute and relative levels of deviance within friendship dyads are important for understanding the extent to which individuals accurately (and inaccurately) report the behavior of their friends.

Collectively, these findings hold significant implications for research on peer influence and deviance or crime. Individuals who are quite deviant and who also associate with individuals who are heavily involved in deviant behavior have the greatest potential to provide systematically erroneous perceptions of peer deviance. Of course, crime researchers generally place a primary focus on these individuals because they commit high amounts of crime. Considered in conjunction with prior findings (Boman, Stogner, et al., 2012), it appears that the amount and accuracy of perceptions of peer deviance change systematically with the actor’s self-reported deviance (a dependent variable) and the peer’s deviance itself (an independent variable). In this regard, future work will find it worthwhile to consider why misperceptions take place. Though this study is primarily a descriptive one and stops short of conducting predictive modeling, the present analyses do suggest that future work should look into how characteristics of the actor, peer, and the friendship itself may cause misperceptions. It may be that individual traits, such as low self-control, explain misperceptions (Young et al., 2011). It might also be reasonable to speculate that a dyad member’s estimate of friendship quality to a peer may be related to the accuracy of his or her perceptions of peer deviance. Furthermore, information sharing could vary across dyads based on how close (e.g., proximity, emotional closeness) the two friends are, which in turn could influence (mis)perceptions. Information sharing could also function differently based on the type of deviant behavior under consideration. While these possibilities are well beyond the focus of the current study, they are important issues that should be the focus of future research.

Although we caution that the current study alone is not sufficient to definitively inform intervention strategies, a rapidly growing body of literature on peer groups and perceptions of peer deviance is moving toward a point where policy recommendations could be formulated. Recent studies have recommended that programs aiming to change behavioral patterns should focus not only on individuals but also on friends and friendships (e.g., Kapadia et al., 2012; Lyons, Giordano, Manning, & Longmore, 2011). The findings of this study may perhaps indicate that it could be worthwhile to try to alter individual perceptions of peer behavior to help individuals understand that deviant behavior might not be as normative and/or widely accepted by friends as actors may believe. Specifically, correcting overestimates of peer deviance, as opposed to underestimates, may serve to reduce the offending patterns of actors. However, to alter perceptions or misperceptions, programs and policy makers must be made aware of the sources of misperception (such as those identified in this study) and develop an understanding of how misperception influences individual involvement in delinquency (Young & Weerman, 2013). Continued research is necessary in these areas. Once identified through research, policy makers and program creators may consider attempting to target the factors or mechanisms that influence the formation of perceptions of peer deviance, such as self-control (see Boman, Stogner, et al., 2012; Young et al., 2011).

Although this study advances knowledge on how the deviant behavior of respondents, peers, and deviance within friendship dyads relates to the several forms of misperception, there are limitations that should be acknowledged. First, as previously discussed, the design of our study cannot address the potential for a causal link between misperceptions of peer deviance and respondent’s self-reported deviant involvement. A longitudinal study of perceptual processes of this magnitude is an important future direction of research (see Young & Weerman, 2013). In addition, while relatively unique, the current sample is potentially limiting in three ways. First, college students probably do not commit the same types or amounts of deviance that individuals would in other, more high-risk samples, perhaps limiting the generalizability of the current results. Second, the design of the current study asked the individuals to identify only one friend. Although “friendships” are necessarily dyadic (Hartup, 1993), individuals are typically nested within multiple friendships and we cannot speak to how multiple friends could potentially alter perceptions of a specific friend’s deviant behavior. Third, we are incapable of tapping cooffending between the friends in the current data. It very well could be that cooffending is a very important cause of accuracy in perceptions of peer deviance. For instance, high amounts of cooffending could potentially facilitate correct perceptions of peer deviance, and vice versa. This is a particularly worthy future direction of exploration.

In considering the applicability of the current results to the larger body of extant and future literature, it may be of importance to keep in mind that the observed patterns emerged with binary measures of perceptual peer deviance and self-reported deviance. Because we do not address the accuracy of the frequency of peer deviance measures, the dichotomization of the measures has served the purpose of classifying perceivers as correct when they accurately indicate whether a peer committed or did not commit a behavior at all. If the 9-point NYS metric had been used, the absolute value of misperception would have been much higher. This study, then, has provided a conservative test where perceptions were given the best operational chance of being correct. Despite this, our analyses reveal that there is a considerable amount of systematic (i.e., nonrandom) inaccuracy in peer deviance measures stemming from Levels 1 (individuals) and 2 (friendships).

Based on these results, is an indirect or direct peer deviance measure more appropriate for use in multivariate modeling? The answer to this question has yet to be definitively answered. Given the inaccuracies revealed here as well as the research discussed earlier (e.g., Haynie & Osgood, 2005; Young et al., 2011), we send a strong word of caution to researchers: Relying exclusively on perceptual measures of peer deviance may result in systematic inaccuracies that inflate correlations between and among (a) the peer deviance construct, (b) independent variables, and (c) dependent variables. To play devil’s advocate, criminologists must also come to realize that this issue is not a “one-way street” because the direct measures of peer deviance are not always “better.” According to social learning theory, perceptual measures are valid and appropriate regardless of inaccuracies. While this philosophy hardly resolves empirical problems encountered with perceptions (see, for example, Rebellon, 2012), it does theoretically indicate that relying strictly on direct measures of peer deviance will result in a model that is not as correctly specified as possible. The middle ground in the ongoing debate between indirect and direct peer deviance measures would be the discovery of an index or scale of perceptual peer deviance that is not significantly different from the peer’s self-reported deviance. However, to date, researchers attempting to find such a measure have been unsuccessful (see Boman, Ward, Gibson, & Leite, 2012). If such a measure were identified and validated across several different sources of data, it would be one of the most important methodological (and potentially theoretical) contemporary advances in criminology.

The deviant behavior of peers, respondents, and friendships all appear to be important when considering why individuals may misperceive the behavior of their peers. Coupled with prior work using social-network (Young et al., 2011) and dyadic data (Prinstein & Wang, 2005), a generalized understanding of the factors linked to the misperceptions of peer deviance is starting to take shape. Continued investigations are necessary to definitively determine the impact that misperceptions have not only on estimates of peer deviance and differential association and social learning theories but also on other theoretical covariates frequently used in multivariate models (see Boman & Gibson, 2011; Meldrum et al., 2009; Weerman & Smeenk, 2005). In the meantime, we advise researchers to be cognizant that the amount of peer deviance perceived and the accuracy of these perceptions appear to systematically covary with the dependent variable (self-reported deviance), the deviance of friends, and the characteristics of superordinate friendships, which very few studies are able to effectively capture outside of stochastic error terms.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest 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.

Notes

John H. Boman IV, is an assistant professor at the University of Wyoming in the Department of Criminal Justice. He recently graduated from the University of Florida with a PhD in criminology (2013). His research focuses on the influence of friends on crime, and especially on the influence of friends over the life-course, theory, gender, conceptualization, and measurement.

Jacob T. N. Young is an assistant professor in the School of Criminology and Criminal Justice at Arizona State University. He received his doctorate in Sociology from the University of Washington. His research examines how social structure affects the diffusion and accuracy of information in friendship networks and the consequences for health related behaviors.

Julie Marie Baldwin is an assistant professor at the University of Arkansas at Little Rock. She received her PhD from the University of Florida in 2013 and was named Outstanding Graduate Student 2012–2013 by the Southern Criminal Justice Association. Her primary research interests include court systems and processes, substance abuse, evaluation, and policy.

Ryan C. Meldrum is an assistant professor in the Department of Criminal Justice at Florida International University in Miami. His current research focuses on the tests of criminological theory, with a particular emphasis on the interrelationships between self-control, peers, and involvement in delinquency. His recent publications have appeared in journals such as Criminology, the Journal of Quantitative Criminology, and Youth Violence and Juvenile Justice.

References

Akers

R. L.

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

Boman

J. H.

Gibson

C. L.

(2011). Does the measurement of peer deviance change the relationship between self-control and deviant behavior? An analysis of friendship pairs. Journal of Criminal Justice, 39, 521–530.

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 on Crime & Delinquency, 49, 601–621.

Boman

J. H.

Ward

J. T.

Gibson

C. L.

Leite

W. L.

(2012). Can a perceptual measure of peer deviance accurately measure a peer’s self-reported deviance? Journal of Criminal Justice, 40, 463–471.

Campbell

Kashy

D. B.

(2002). Estimating actor, partner, and interaction effects for dyadic data using PROC MIXED and HLM: A user-friendly guide. Personal Relationships, 9, 327–342.

Davies

Kandel

D. B.

(1981). Parental and peer influences on adolescents’ educational plans: Some further evidence. American Journal of Sociology, 87, 363–387.

Elliott

D. S.

Huizinga

Ageton

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

Fisher

L. A.

Bauman

K. E.

(1988). Influence and selection in the friend-adolescent relationship: Findings from studies of adolescent smoking and drinking. Journal of Applied Social Psychology, 18, 289–314.

Gottfredson

M. R.

Hirschi

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

10.

Hartup

W. W.

(1993). Adolescents and their friends. In Laursen

(Ed.), Close friendships in adolescence: New directions for child development (pp. 3–22). San Francisco, CA: Jossey-Bass.

11.

Haynie

D. L.

(2001). Delinquent peers revisited: Does network structure matter? American Journal of Sociology, 106,1013–1057.

12.

Haynie

D. L.

Osgood

D. W.

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

13.

Iannotti

R. J.

Bush

P. J.

(1992). Perceived vs. actual friends’ use of alcohol, cigarettes, marijuana, and cocaine: Which has the most influence? Journal of Youth and Adolescence, 21, 375–389.

14.

Jussim

Osgood

D. W.

(1989). Influence and similarity among friends: An integrative model applied to incarcerated adolescents. Social Psychology Quarterly, 52, 98–112.

15.

Kandel

D. B.

(1996). The parental and peer contexts of adolescent deviance: An algebra of interpersonal influences. Journal of Drug Issues, 26, 289–315.

16.

Kapadia

Frye

Bonner

Emmanuel

P. J.

Samples

C. L.

Latka

M. H.

(2012). Perceived peer safer sex norms and sexual risk behaviors among substance-using Latino adolescents. AIDS Education and Prevention, 24, 27–40.

17.

Kenny

D. A.

Kashy

D. A.

Cook

W. L.

(2006). Dyadic data analysis. New York, NY: Guilford.

18.

Krueger

Zeiger

J. S.

(1993). Social categorization and the truly false consensus effect. Journal of Personality and Social Psychology, 65, 670–680.

19.

Lyons

Giordano

P. C.

Manning

W. D.

Longmore

M. A.

(2011). Identity, peer relationships, and adolescent girls’ sexual behavior: An exploration of the contemporary double standard. Journal of Sex Research, 48, 437–449.

20.

Matsueda

R. L.

Anderson

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

21.

McGloin

J. M.

(2009). Delinquency balance: Revisiting peer influence. Criminology, 47, 439–477.

22.

Meldrum

R. C.

Young

J. T. N.

Weerman

F. M.

(2009). Reconsidering the effect of self-control and delinquent peers: Implications of measurement for theoretical significance. Journal of Research in Crime & Delinquency, 46, 353–376.

23.

Merton

R. K.

(1987). Three fragments from a sociologist’s notebooks: Establishing the phenomenon, specified ignorance, and strategic research materials. Annual Review of Sociology, 13, 1–29.

24.

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.

25.

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.

26.

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.

27.

Ross

Greene

House

(1977). The “false consensus effect”: An egocentric bias in social perception and attribution processes. Journal of Experimental Social Psychology, 13, 279–301.

28.

Thornberry

T. P.

Krohn

M. D.

(1997). Peers, drug use, and delinquency. In Stoff

D. M.

Breiling

Maser

J. D.

(Eds.), Handbook of antisocial behavior (pp. 218–233). New York, NY: Wiley.

29.

Weerman

F. M.

Smeenk

W. H.

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

30.

Young

J. T. N.

Barnes

J. C.

Meldrum

R. C.

Weerman

F. M.

(2011). Assessing and explaining misperceptions of peer delinquency. Criminology, 49, 599–630.

31.

Young

J. T. N.

Weerman

F. M.

(2013). Delinquency as a consequence of misperception: Overestimation of friend’s delinquent behavior and mechanisms of social influence. Social Problems, 60(3).

32.

Zhang

Messner

S. F.

(2000). The effects of alternative measures of delinquent peers on self-reported delinquency. Journal of Research in Crime & Delinquency, 37, 323–337.