Abstract
The accuracy of measuring the prevalence of delinquency by means of self-reported questionnaires is difficult to evaluate. This study assesses the differential validity of self-reported delinquency in adolescents and, more specifically, self-reported police contacts because of suspected misconduct. This study was conducted as part of the Rotterdam Youth Monitor, a youth health surveillance system. Self-report data of pupils (mainly 12–15 years old) in the first or third grade of secondary school in the school years 2007–8 and 2008–9 (n = 23,914) were merged with police data. Of the pupils registered as a suspect, 62 percent admitted to having been interrogated at the police station. However, there were differences between groups. Multivariate analysis showed that Moroccan pupils and first-grade pupils were more likely to give an invalid response. Pupils who were registered for theft, vandalism or assault were more likely to give a valid response, whereas pupils who were registered for an offence involving fireworks were more likely to give an invalid response. We conclude that using only self-reported data to measure delinquency in an ethnically diverse population results in substantial bias. It is advisable to use multiple sources to measure the prevalence of delinquency.
Keywords
Introduction
Self-report questionnaires are an accepted means to measure all kinds of behaviour and are used in various areas of (social) research, such as health, psychology and education (for example, Hoyle et al., 2002). Although the self-report method is also widely used in the field of criminology, its use has been disputed (Thornberry and Krohn, 2000). Measuring delinquency with self-report questionnaires may introduce problems with validity because the aim is to measure behaviour that is illegal and punishable by law. There may be a tendency on the part of the respondents to conceal, or to fail to recall offences, out of fear of the consequences. When the self-report method was introduced to measure delinquency, respondents seemed more than willing to report this type of behaviour (Porterfield and Clifton, 1946; Wallerstein and Wylie, 1947). However, because studies repeatedly found discrepancies between self-report data and data from official records, the validity of these self-report measures to measure delinquency remains questionable (Farrington et al., 1996; Hindelang et al., 1981; Huizinga and Elliott, 1986; Maxfield et al., 2000).
In general, a measure is considered valid if it measures what it is intended to measure and if there is a strong relation between the theoretical concept and what one is actually measuring (Huizinga and Elliott, 1986; Thornberry and Krohn, 2000). There are several ways to evaluate validity. One is to investigate criterion validity, in which the relation between test scores and some external criterion that accurately indicates the quantity being measured is assessed. In the case of delinquency, ‘concurrent validity’ (a type of criterion validity) is often investigated by checking whether officially registered offences or arrests are admitted in a self-administered questionnaire (Huizinga and Elliott, 1986). When the degree of validity is different for different populations (for example, males versus females, black respondents versus white respondents), this is termed ‘differential validity’.
Self-report measures of delinquency have long been used and this method has been improved over time (Junger-Tas and Marshall, 1999). Because of these improvements and because the use of police records alone has several drawbacks (for example, the police rarely detect offences, bias in victims’ reports of crime, and police bias), self-reports of delinquency are now a widely used method to measure criminal behaviour (Junger-Tas and Marshall, 1999) in addition to police data. Nevertheless, the percentage of respondents who fail to report their delinquent behaviour is relatively high. For example, Maxfield et al. (2000) compared agreement between arrests found in official records and self-reported arrests for abused or neglected persons versus a control group, and also investigated gender and ethnicity. They found a similar degree of concurrent validity for abused or neglected persons (72 percent) versus controls (76 percent). However, more male (79 percent) and white respondents (77 percent) self-reported known arrests compared with female (63 percent) and black respondents (66 percent). In their National Youth Survey (1976–8), Huizinga and Elliott (1986) reported substantially lower percentages of concurrent validity. For their study, they searched police records in the towns of respondents and in all jurisdictions adjacent to the respondents’ residences: 36–48 percent of the respondents failed to acknowledge at least some of their offences. In line with the study by Maxfield et al. (2000), they also found variation by gender and race: 11–61 percent of blacks provided matched responses on all or a majority of their arrests, whereas among whites these percentages varied between 57 and 81 percent (Huizinga and Elliott, 1986). Hindelang et al. (1981) also compared self-reported delinquency with official police records by calculating correlations between the two. Overall, these correlations were reasonably high (around .60 for all subjects), but for African-Americans they were substantially lower than for other respondents. Not all studies have found lower validity for African-American groups. For example, Farrington et al. (1996) found that 65 percent of African-Americans admitted to having been apprehended, compared with 53 percent of Caucasians. Thus, most (but not all) studies found evidence of differential validity for different (ethnic) populations.
Most of the above-described studies were conducted in the USA among ethnic groups different from those residing in European countries. Each ethnic minority group has its own specific characteristics, which may affect the validity of self-reported police contacts in different ways. Therefore, the results of American studies cannot easily be generalized to the European situation.
In Europe, research on this topic is scarce. A European study by Junger (1989) investigated discrepancies between police data and self-report data for Dutch ethnic minorities. Junger found that Moroccan and Turkish youngsters were less likely to admit their delinquent behaviour than were Dutch youngsters. Factors accounting for these discrepancies included: a strong bond with traditional cultural attitudes; limited knowledge of the Dutch language; and no fear of being sent back to the home country by the police. Junger’s study was conducted among first-generation immigrants. Most of the explanations may not apply to the current groups of ethnic minorities because these are often born and raised in the Netherlands (second generation). Moreover, the population distribution by ethnic origin, especially in the large cities of the Netherlands, is not comparable to the situation 20 years ago. When Junger conducted her study, 9–13 percent of the population in one of the four large cities in the Netherlands consisted of ethnic minorities, whereas nowadays, in Rotterdam, for example, 48 percent of the population is of non-Dutch origin (Hoppesteyn, 2010).
Self-report questionnaires are still used on a large scale. In the field of criminology, self-report questionnaires are primarily used in addition to official police statistics to estimate the prevalence of delinquency. In contrast, in the field of public health, measures of delinquency serve as a marker of general well-being, because delinquent behaviour is associated with health and lifestyle. Within the Rotterdam Youth Monitor, for example, data on delinquency are used to detect problem behaviour at a collective level but also at an individual level. Therefore, it is important to investigate whether, and to what extent, differential validity may affect self-report data, because ethnic minorities comprise a large part of the population and especially because this has not been done before on a large scale.
Purpose of article
This study investigates the validity of self-reported police contacts as a measure of juvenile delinquency. This was done by comparing self-reported police contacts of pupils in secondary schools in the Rotterdam area with police data. In addition, we studied whether the agreement between self-report data and police data was influenced by gender, ethnicity, grade, educational level and the type of offence for which the pupils were registered.
Methods
Study design
Data from the Rotterdam Youth Monitor (a youth health surveillance system) were used. The Rotterdam Youth Monitor monitors general health, well-being, behaviour and related factors of youth aged 0 to 19 years living in Rotterdam and its environs in order to supply information for youth policy at school, neighbourhood and municipality level. Rotterdam Youth Monitor data are gathered within and as part of the routine government-approved health examinations of preventive youth healthcare.
Questionnaires are completed on a voluntary basis. Parents receive written information about these questionnaires and are free to object to their child’s participation. All questionnaire data are collected on a confidential basis. Pupils are told that all the information they provide is strictly confidential and that completed questionnaires will ‘never be shown to teachers, parents, police or anyone else, other than to the school nurse’.
For the present study, data were used from the 2007–8 and the 2008–9 Rotterdam Youth Monitor surveys of secondary schools located in Rotterdam and its surrounding municipalities. All secondary schools in this area were approached and most (77 percent) of the schools participated. All first- and third-grade pupils (mainly aged 12–13 years and 14–15 years, respectively) of participating schools were asked to fill in a questionnaire (n = 26,816). Of these, 23,914 pupils (89 percent) completed the questionnaire. Trained school nurses of the Municipal Public Health Service administered the questionnaire during regular class hours.
Police data from the Rotterdam-Rijnmond Police Department were used. In total, 66,728 persons born after 1 January 1982 were registered as a suspect for an offence between 1 January 2002 and 31 August 2009 in Rotterdam and its surrounding municipalities. Compared with the areas covered by the Rotterdam Youth Monitor, the Rotterdam-Rijnmond Police Department operates in nine additional municipalities around Rotterdam.
Questionnaire data were merged with police data. When police registration of a suspect took place after the survey had been conducted, matches were recoded as if the pupils were not registered as a suspect. In other words, only registrations that occurred before filling in the questionnaire were taken into account. After this procedure, 1427 pupils filled in a questionnaire and were also registered as a suspect. To combine questionnaire and police data, a strict procedure was followed in order to ensure the confidentiality of personal data in accordance with the Dutch Data Protection Act. First, we developed a key variable, which required merging personal data, such as names, date of birth and gender, from the Rotterdam Youth Monitor and the police data. This phase also enabled us to study the level of bias or misclassification. After an acceptable key variable was developed, all personal data were removed. Next, data from both sources were combined by means of the key variable. The Public Prosecution Service’s highest authority, the Board of Procurators General, gave written permission to conduct this study.
Questionnaire data
Police contact was measured by asking ‘Have you ever been interrogated at the police station because you did something wrong?’, based on a questionnaire developed by Garnefski and Diekstra (1993). Answers were recoded into ‘yes’ or ‘no’. In addition, information on the characteristics of the pupils were obtained from the questionnaires, including gender (boy or girl), ethnicity, age, grade (first or third grade) and educational level. Ethnicity was based on the country of birth of the pupils and the country of birth of their parents, and was categorized according to the classification of Statistics Netherlands (Statistics Netherlands, 2004). If the pupil and both his/her parents were born in the Netherlands, then the pupil was categorized as ‘Dutch’. If the pupil was born in the Netherlands and at least one of his/her parents was born abroad, the country of birth of the parent born elsewhere was designated as the pupil’s ethnicity. If the pupil and at least one of his/her parents was born abroad, then the country of birth of the pupil was designated as the pupil’s ethnicity. Pupils were classified into ‘pre-vocational secondary education’ or ‘senior general secondary education or pre-university education’ (Goodman, 1997).
Police data
Police data included information on suspects, the registration date and the type of offence. All offences were classified into 11 groups: burglary, theft, selling or receiving stolen goods, vandalism, arson, fireworks, assault, weapons, molestation, traffic or transport, and ‘other’.
Statistical analyses
The unadjusted percentage of agreement between self-reported police contacts and police data was calculated by dividing the sum of the number of positive and negative answers in both data sources by the total number of pupils in the study population. The degree of agreement was also investigated by calculating the kappa coefficient (κ), which is a measure of agreement adjusted for chance (Landis and Koch, 1977b). The κ takes into account the reality that agreement can be inflated owing to skewed distributions (Cicchetti and Feinstein, 1990; Feinstein and Cicchetti, 1990). The κ was interpreted using the Landis and Koch criteria, where: κ ≤ 0.2 indicates poor agreement, 0.2 < κ ≤ 0.4 fair agreement, 0.4 < κ ≤ 0.6 moderate agreement, 0.6 < κ ≤ 0.8 good agreement, and 0.8 < κ ≤ 1.0 excellent agreement (Landis and Koch, 1977a).
Based on self-reported police contacts and the police data, four response groups were distinguished:
Pupils who reported in the questionnaire that they had never been interrogated at the police station and who were not registered as a suspect in the police data.
Pupils who reported in the questionnaire that they had never been interrogated at the police station, although they were registered as a suspect for an offence in the police data.
Pupils who reported in the questionnaire that they had been interrogated at the police station, although they were not registered as a suspect for an offence in the police data.
Pupils who reported in the questionnaire that they had been interrogated at the police station, and who were registered as a suspect for an offence in the police data.
Logistic regression analysis was used to determine whether gender, ethnicity, grade, educational level or type of registered offence were associated with an invalid response. An invalid response was operationalized as having denied being interrogated at the police station but having a registration as a suspect in the police system (response group II). A valid response was operationalized as having reported in the questionnaire being interrogated at the police station and having a registration as a suspect in the system (response group IV).
Pupils who self-reported no police contacts and who were not registered as a suspect (response group I) were excluded from the analyses because this subgroup has no reason to make an invalid response. Furthermore, those who reported having been interrogated at the police station but who had no registration as a suspect (response group III) were also excluded. It seemed that this group over-reported their police contacts. However, after searching a sample of this group in the police registration system, it appeared that most pupils in this group had been in the police station but were not registered as a suspect. Youngsters are sometimes let off with a warning, or youngsters are interrogated as a group and it is not clearly specified who did what. In this latter situation, if these youngsters are still unknown to the police, the police only make a note and no one is registered as a suspect. Because it was impossible to determine exactly how many of this group made a valid or invalid response, this group was excluded.
The excluded response groups may also suffer from some misclassification, that is, some pupils may not be matched when they should have been. To determine the potential consequences of misclassification, a sample of 1700 cases were checked. Of these cases, 10 were unjust mismatches (0.6 percent). Of these 10 mismatches, 7 were due to differences between the person’s (first) name and the name registered in the municipal personal records database that is used in police systems (for example, the registered name is ‘Albert’ whereas the name used in school registers is ‘Bert’). Two mismatches were due to misplacement of prefixes (a prefix was sometimes registered as part of the last name when it should have been registered in a separate field), and one was due to other errors in the spelling of variables used to merge the data. Although the misclassification affects the absolute numbers in each response group, it is unlikely that this misclassification occurred differentially, that is, more in certain (ethnic) groups than in others. Consequently, substantial differences in validity in the misclassified and the correctly classified respondents are unlikely.
A series of analyses was conducted. First, for each variable we determined whether it was associated with an invalid response. Second, the determinants were divided into two blocks. The first block contained gender, ethnicity, grade and educational level. The second block consisted of the types of offence for which pupils were registered. The backward selection procedure was used to determine for each block which determinants were significantly associated with an invalid response. In the final step, only those determinants that remained significant after the backward selection procedure were included in one multivariate model. Only the result of the latter multivariate analysis is presented because results showed no meaningful change.
Non-response
For the non-response analysis, pupils who did not complete a questionnaire were compared with pupils who did: 11 percent of the pupils did not fill in a questionnaire. Pupils did not complete the questionnaire either because parents refused to give permission or because pupils were absent (owing to illness or truancy) when their classmates filled in the questionnaires. Pupils who failed to complete a questionnaire were statistically more often registered as a suspect (17 percent) than were pupils who completed a questionnaire (6 percent χ2 = 463.2, d.f. = 1, p < .001).
Results
Table 1 presents the characteristics of the study population: 51 percent of the participants were boys. Of all pupils, 54 percent were of Dutch ethnic origin. Of the non-Dutch group, 9 percent were Surinamese, 3 percent Antillean or Aruban, 8 percent Moroccan, 9 percent Turkish, 3 percent Cape Verdean and 14 percent had another ethnic background.
Sample characteristics for the school years 2007–8 and 2008–9
Note: n = 23,914.
The prevalence of police contacts according to the questionnaire data was higher (9 percent) than the prevalence based on the police registration system (6 percent χ2 = 4889.1, d.f. = 1, p < .001). The observed agreement between these two sources was 92 percent and the κ indicated moderate agreement (κ = 0.44, standard error (SE) = 0.0063). See Table 2.
Agreement on police contacts: Self-reported police contacts compared with police data
Notes: Observed agreement (886 + 21,104) / 23,914 (92 percent); Cohen’s kappa = 0.44 (SE = 0.0063).
Police contacts from self-reported questionnaire 2269/23,914 (9.5 percent).
Police contacts from police data 1427/23,914 (5.9 percent).
Pupils who reported having had police contacts or who were registered as a suspect in the police registration system comprised more boys than girls and more third- than first-grade pupils, compared with those who reported having had no police contacts and who had no registration as a suspect (see Table 3). Table 3 also shows that Moroccan pupils were overrepresented and Dutch pupils were underrepresented in the group that underreported compared with the other response groups. This suggests that Moroccan pupils more often than pupils with another ethnic background are dishonest about their police contacts.
Response groups by gender, ethnicity, grade and educational level (percent)
Notes:
Group I: No self-reported police contacts and no registration as a suspect.
Group II: No self-reported police contacts and registered as a suspect.
Group III: Self-reported police contact and no registration as a suspect.
Group IV: Self-reported police contact and registered as a suspect.
In Table 4, only the pupils who were registered as a suspect are selected (n = 1427) and the proportion of pupils who admitted to having had this police contact is presented according to gender, ethnicity, grade and educational level. In general, most pupils (62 percent) were honest about their police contacts. However, there were differences between ethnic groups (χ2 (6) = 45.93, p < .001) and between first- and third-grade pupils (χ2(1) = 8.71, p = .003). The Moroccan pupils and pupils in the first grade admitted their police contacts least often (46 percent and 57 percent, respectively) compared to the other ethnic groups and third-grade pupils, respectively.
Proportion of pupils who were registered as a suspect and who admit their police contact(s) in the self-report questionnaire, by gender, ethnicity, grade and educational level
Note:
In the self-report questionnaire.
Table 5 presents the associations between the determinants (pupil characteristics and type of offence) and giving an invalid response, that is, registered as a suspect but denying police contact in the questionnaire (n = 541), compared with a valid response, that is, registered as a suspect and confirming this in a self-report (n = 886). Because the results of the univariate analyses were very similar to those of the multivariate analysis, only the latter are discussed here. In this final multivariate analysis, we included all determinants that remained significantly associated after the backward selection procedure. Ethnicity, grade and certain types of offence appeared to be determinants of giving an invalid response. Moroccan pupils had a higher risk than pupils of Dutch origin of giving an invalid response (OR 2.24, 95 percent CI 1.63–3.08), and third-grade pupils had a lower risk compared with first-grade pupils of giving an invalid response (OR 0.78, 95 percent CI 0.61–0.98). Pupils who were registered as a suspect for theft (OR 0.62, 95 percent CI 0.48–0.80), vandalism (OR 0.73, 95 percent CI 0.55–0.97) or assault (OR 0.68, 95 percent CI 0.49–0.93) had a lower risk of denying police contacts than did pupils who were not registered for these offences. When pupils were registered for an offence involving fireworks, the odds ratio of giving an invalid response was 1.55 (95 percent CI 1.12–2.15).
Associations between pupil characteristics and risk of an invalid response (i.e. registered as a suspect but denying police contact in the self-report questionnaire)
Notes: OR = odds ratio, CI = confidence interval. Univariate analyses: logistic regression analyses in which the association is assessed between each covariate and the risk of an invalid response. Multivariate analysis: logistic regression analysis in which all covariates that remained significant after the backward selection procedure for each block were entered at the same time.
p < .01; ** p < .05
Discussion
Overall, we found a moderate degree of agreement between self-report questionnaires and police data. A majority of the pupils who were registered as a suspect reported that they had had police contact (62 percent). Differential validity was found for ethnicity and grade. Moroccan pupils and pupils from the first grade were most likely to give an invalid response regarding their police contacts compared with Dutch and third-grade pupils, respectively.
In line with Junger (1989), we found relatively high levels of denial of police contact for some ethnic minority groups (Moroccan 54 percent Turkish 40 percent) compared with the total (38 percent). Only for Moroccan pupils was this increased level statistically higher (compared with Dutch pupils). In contrast, for other ethnic minority groups we found lower levels of underreporting. For example, 24 percent of the Cape Verdean and 26 percent of the Antillean or Aruban pupils did not report their police contacts in the questionnaire. However, there were no statistical differences when compared with the Dutch reference group, and the groups were small (n = 72 and n = 62, respectively). In contrast to other studies, we found no statistical differences for gender. Another factor possibly affecting the validity of self-report data is the severity of the committed offence. Reports on this topic are relatively old and few in number. Petersillia et al. (1978), studying prison inmates, observed a tendency to report more serious crimes rather than less serious crimes, whereas others reported the opposite (see, for review, Huizinga and Elliott, 1986). In line with Petersillia et al. (1978), we found that pupils registered for the more serious crimes (for example, theft, vandalism and assault) were more willing to admit their police contacts than were pupils registered for less serious offences (for example, fireworks).
Strengths of the research
This study was conducted in a large, ethnically diverse population. To our knowledge, this is the first European population-based study that investigated whether and to what extent the validity of self-report data on police contacts differed across gender, ethnicity, grade, educational level and the type of offence. Furthermore, we chose not to compare self-reported delinquent acts from the questionnaire data with police registrations of offences, because the offences asked about in the questionnaire did not match with how offences were registered in the police system. Although police contacts reflect only part of delinquency, a major advantage of this approach is that the comparison is straightforward. If someone is registered as a suspect then he/she must have been interrogated at the police station, so if that person denies this in the questionnaire it means that he/she has given an invalid response.
Limitations on the research
We aimed to study differential validity by comparing self-reported police contacts with official police data on suspects. Several limitations need to be discussed. Whereas the first part of our self-report question enabled a straightforward comparison with police registration as a suspect, the second part of the self-report question may compromise the validity of the self-report measure. It could be that a pupil acknowledges that he/she was interrogated but (in contrast to the police) believes that he/she did nothing (wrong). In this case a pupil will justifiably give a negative response and consequently will be unjustly classified as having given ‘an invalid response’, that is, as an ‘under-reporter’.
Furthermore, police data may not be an ideal ‘external criterion’ but it is the best one available. Numerous studies have claimed that police bias or racial profiling exists, which implies that some (ethnic) groups may be disproportionately arrested because of their demeanour or ethnic background (Batton and Kadleck, 2004; Smith and Alpert, 2007). In the present study, Moroccan adolescents (easily recognizable by police officers) may have been disproportionately arrested/apprehended, or an interrogation at the police station might have disproportionately led to registration as a suspect. Assuming that this police bias exists and that Moroccan adolescents are disproportionately arrested/apprehended, this, in combination with the formulation of the second part of the self-report question, may partly explain the high level of invalid responses among Moroccan pupils.
It is difficult to estimate the extent to which this phenomenon distorted our results. For example, Junger (1989) discarded the possibility that police bias could explain the differential validity that she found in her study, because it would imply that some ethnic minority groups (Moroccan and Turkish boys) were discriminated against by the police while others were not (Surinamese boys). This line of reasoning is quite legitimate, given the era in which this study was conducted. However, others, such as Fergusson et al. (1993), have explained the large difference between the number of police arrests among the Maori/Pacific population compared with the self-reported number of arrests as resulting from bias in police arrest data.
Second, in the present study, a substantial number of pupils reported contact with the police although they were not registered as a suspect. We searched a random sample of this group in the police system. Although most of these pupils were known to the police (but not as a suspect), we cannot be sure whether all pupils who seemed to exaggerate their police contact in fact gave a valid response: therefore, we excluded this group in our analyses. In order to know more about this group, further research is necessary.
Selection and non-response may also have affected our results. Adolescents attending secondary schools in Rotterdam and the surrounding areas were approached to fill in a questionnaire. Because school dropouts and pupils attending other than regular schools (for example, special schools for children with learning disabilities or behavioural problems) are not included in this study, the results are probably not generalizable to all adolescents. Furthermore, our non-response analysis showed that pupils who did not fill in the questionnaire were more likely to be registered as a suspect than pupils who completed the questionnaire. Numerous studies have shown that truancy and school dropout are related to delinquency (see, for review, Weerman and Van der Laan, 2006). Our non-response group consisted at least partly of truants, but school dropouts were not included in our study. Although we could not investigate whether non-responders reported differently in terms of validity than responders, it is plausible that this may have affected our findings. A qualitative study by Kleijer-Kool (2006) showed that, compared with first offenders, recidivists more often denied their offences. If we infer from these results that more delinquency is associated with more underreporting, then in this study agreement is probably overestimated and the true level of agreement between self-report data and police data is probably lower.
Finally, questionnaire data were collected on a confidential basis but not anonymously. Van de Looij-Jansen et al. (2006) observed that boys who completed a questionnaire in conditions of confidentiality reported less vandalism, stealing and violent delinquent behaviour than boys completing a questionnaire in conditions of anonymity. Thus, the method of data collection may have resulted in an overall underestimation of the prevalence of police contact. However, this is problematic only if this method of data collection affected reporting of the various ethnic groups in dissimilar ways.
In this study, we aimed to evaluate the utility of self-reported police contacts in ethnically diverse populations. We found that most pupils admitted their police contacts. However, particularly for Moroccan pupils (compared with Dutch pupils), agreement between self-reported police contacts and police registration as a suspect was low. It is difficult to know what explains this result. Methodological issues – such as the formulation of the self-report question and bias in the police data – may be a part of the explanation. Furthermore, Junger (1989) suggested that cultural categories (for example, Islamic versus non-Islamic) rather than ethnic group, as well as factors associated with the immigrant situation, are important in predicting invalid responses. However, such explanations refer to both Moroccan and Turkish respondents, but we found significant underreporting only for Moroccan pupils. Thus the question remains: what factors differentiate between Moroccan and Turkish respondents?
First, Berber Moroccans (who form the majority of the Moroccan community in the Netherlands) are traditionally suspicious of the authorities. Second, more Moroccans than Turks experience high levels of stigmatization as a Moroccan, which may be a threat to their self-esteem (Pels, 2008). Third, the denial of delinquent acts by Moroccan adolescents has also been attributed to the Mediterranean shame-culture, in which concepts such as honour, status, decency, pride and respect are highly valued (Yeşilgöz and Harchaoui, 2003). The idea is that a person will always deny that he/she committed an offence as long as it is not proven; in this way, the family’s honour remains untouched. Finally, Harchaoui (2001) suggests that the high level of denial among Moroccan adolescents is the result of increased experience and insight (probably owing to high recidivism) in the Dutch justice system. For these reasons, Moroccans, more than other ethnic minority groups, may be reluctant to report problems in a questionnaire (Pels, 2008; Rood-Pijperset al., 1995). These are just a few factors that may explain the discrepancy between self-reports of delinquency and the official delinquency data found in the present study.
To understand why a specific group provides fewer valid responses than other groups and to find solutions for this problem of differential validity, further research is necessary. For example, qualitative research and, more specifically, interviews with focus groups and key informant interviews may help deepen our knowledge.
Based on the results of this study, we conclude that using self-report data alone is insufficient to capture the true prevalence of delinquency in an ethnically diverse population. We recommend using multiple sources to achieve a more complete picture of delinquency patterns.