2.1. Everyday crime

16 The subject of everyday crime has received little attention in the literature. Karstedt and Farral [2007] highlight how everyday crime differs from other types of crime. Such crimes are committed by what politicians call the “law-abiding majority”. For most people, they are due to a lack of civility and are part of individual experience; moreover, those people who commit them would condemn others who do likewise.

17 Everyday crime indicates the moral state of society: such behaviors can be the consequence of poor economic conditions or, more worryingly, people can feel victimized by organizations or services (see Karstedt and Farral [2006]) creating a situation of “anomie” regarding the law (following Sampson and Bartusch [1998]) leading to increasing cynicism regarding the law and a feeling of distrust.

18 Everyday crime can also be linked to the concept of economic morality. This concept was first defined by Thompson [1971] to explain the transition, in the eighteenth century, from old practices to the new market economy resulting in a fairer society. This concept reflects the resistance that is seen, and which characterizes, everyday crime when the importance of norms and regulations is neglected or not fully understood (Karstedt and Farral [2004]).

19 The desire to circumvent the law is reflected in deviant behavior, which is, notably, explained by the theory of “anomie”. The term was first defined by Emile Durkheim as a situation of social deregulation, absence, confusion or contradiction of social rules. This suggests that some societies are more subject than others to deviant behavior if their social structures function poorly. Merton [1938] suggests that there are certain phases in a social structure when circumstances justify deviant behavior, especially for people who have limited opportunities to reach social goals. In this sense, Karstedt and Farral [2006] depict the market anomie as “distrust, insecurity and cynical attitudes towards legal rules”. These elements increase the propensity to engage in everyday crime, especially when there is perceived inequality in the distribution of market opportunities.

20 In our study, national morals and economic conditions are controlled for by country fixed effects in our econometric model. This narrows the focus to individual characteristics such as the subjective probability of sanctions, norm- and legitimacy-based explanations of crime, and other control variables such as social capital and sociodemographic characteristics.

2.2. The main drivers of criminal behavior

21 Becker’s theory [1968] highlights the rational nature of criminal behavior: offenders compare the costs and the benefits of the illegal activity and decide to engage in it if they estimate that the advantages outweigh the costs. This theory led Becker to offer some recommendations regarding the resources and the type of punishment (fine, prison) that should be used to enforce legislation. Since its development, the theory has been subject to extensive empirical testing. Among others, Lauridsen [2010] investigated whether crime in Poland is governed by economic rationality; the study found a negative correlation between the propensity to commit a crime and the deterrent, confirming the hypothesis.

22 The subjective probability of sanctions raises the issue of the risk sensitivity of criminals. This question was studied by Langlais [2010] in a theoretical paper. He highlighted that “criminals willingly undertake a risky and dangerous activity both for other people and for themselves (although there exists also a risk of being caught and punished) – thus, they are prone to a risk seeking attitude.” Indeed, risk seeking behavior often appears when there is a high probability of gains or a low probability of losses. The effect of the risk of sanctions has been widely studied with respect to one specific everyday crime, namely traffic offenses. For instance, Lev et al. [2008] establish stable traits that characterize risky drivers, notably their extrovert personality and their tendency to emphasize gains rather than losses when they take decisions. Similarly, Michiels and Schneider [1984] studied the profiles of traffic offenders as a function of the type of infraction and found that risk-lovers were more likely to offend. Another important determinant of road traffic offenses seems to lie in the sensitivity to reward compared to punishment. Castella and Perez [2004] found a significant positive relationship between attitudes and behavior, consistent with the thesis that sensation-seeking incites risk and law-breaking (for a literature review see Jonah [1997]). These personality traits may also explain why someone will comply with a behavioral norm. For instance, Friehe and Schildberg-Hörisch [2017] use survey data to show how economic preferences (patience and attitude to risk), personality traits (the Big Five and locus of control), and a criminology self-control scale can complement each other in predicting self-reported norm enforcement behavior.

23 The norm-based explanation of compliance is increasingly used in the theoretical and empirical literature in various fields, such as legal sciences, economics, sociology and psychology (see e.g. Alm [2012]; Lederman [2003]; Tyler [1990]; Wenzel [2005]). Cooter [2000] explains that “a person who internalizes a social norm has a “taste” for obeying the norm.” (p. 7). This taste is reflected in a willingness to conform. In this situation, even if compliance is unobserved or sanctioned by an external observer, the individual is willing to comply even if it is costly (in terms of effort, time or money). This norm-based explanation also states that compliance relies on the alignment of personal norms with the values that are promoted in legislation. Thus, compliance is consistent with the perception that the rule is just. The empirical literature indicates that norm internalization is not independent of individual characteristics. For instance, Douhou et al. [2011] measure perceptions of “small crimes” based on a questionnaire administered to a large, representative sample of the Dutch population. They show that perceived severity varies with respondent characteristics such as gender, age, or education. This highlights the need to introduce individual characteristics into our econometric model to avoid endogeneity problems.

24 Following the seminal work of Tyler [1990], the legitimacy of the judicial system has been studied as an explanation of legal compliance. Thus, we consider the perceived legitimacy of the judicial system as a third driver of compliance. Broadly defined, legitimacy is the right to govern and the recognition by the governed of that right (Jackson et al. [2011]). This legitimacy-based explanation indicates that individuals are willing to comply with the law if it is produced or enforced by a legitimate system, even if the law itself is perceived as “unjust” or if the expected sanctions have little impact (see Jackson et al. [2013 a] and Tyler [1990]). Our study contributes to this literature as it helps us to understand whether these different theories of criminal behavior simultaneously apply to everyday crime.

2.3. Control variables

25 Empirical studies of crime usually control for sociodemographic variables and social capital. Here, we control for five sociodemographic factors (gender, household income, level of education, age, and professional status) and six social capital variables (involvement in an organization, degree of social relations, generalized trust in others, political interest, religiosity, and trust in domestic political institutions).

26 Sociodemographic factors In her study of whether, and under what conditions, feelings of economic hardship trigger everyday crime, Lopes [2010] carries out an empirical, cross-nation analysis based on data from the ESS (Round 2). She finds that a high degree of religiosity reduces the tendency to commit such crimes in eight European countries. On the other hand, being young and male had the opposite effect in all of the considered countries. Education and income level were found to only have a negligible, positive, effect in very few countries. Similarly, Karstedt and Farral [2006] compared surveys covering more than 4,000 individuals in three regions (England and Wales, Western, and Eastern Germany). They found that high-level victims and/or offenders came from what could be called the middle classes: both groups had a higher social status, higher incomes, were better educated and more likely to be employed. This result was consistent with the work of Rebovich et al. [2000] using American data. The latter study aimed to give a comprehensive picture of “what the average American thinks about white collar crime [1]” through a telephone survey of American citizens. They found that among the five, tested, demographic factors (age, sex, social views, education, and race), two were significant predictors of behaviors that could be linked to white collar crime: aging and being female.

27 In the “traditional” crime literature, several authors have studied the effect of sociodemographic factors. For example, Entorf and Spengler [2000] estimate supply-of-offenses functions for aggregated crime and eight crime categories using panel data from the German Laender (states). They find that being young and unemployed increases the probability of committing crime, however the results are ambiguous for general unemployment. In fact, there is no clear consensus regarding the effect of individual characteristics on crime. For example, the relationship between unemployment and crime is unclear. Raphael and Winter-Ebmer [2001] study state-level American panel data covering the period 1971-1997. They consistently find that unemployment increases crime, but only for property crime. For violent crime, the results are mixed (unemployment has a positive effect on robbery and assault, but negative effects on murder and rape). To explain this, Cantor and Land [1985] argue that “any inference about the effect of the unemployment rate has to account for both criminal opportunity and criminal-motivation effects. Neglect of both of these factors has led to finding inconsistent empirical results on this topic, since the total effect of the unemployment rate is the sum of positive motivational and negative opportunity impacts” (p. 329). Furthermore, although the link between education and crime may appear ambiguous, recent studies have found a negative relationship between them. For example, Buonanno and Leonida [2006] carried out an econometric analysis based on panel data for the 20 Italian regions over the period 1980-1995: their empirical results suggest that education (and especially high school graduation) has a significant negative effect on crime rates.

28 Social capital variables Regarding social capital, Putnam [1995] argues that the notion is multi-dimensional and refers to “features of social organization, such as networks, norms, and trust, that facilitate coordination and cooperation for mutual benefit”. More recently, Akçomak and Weel [2008] studied the link between social capital and crime in the Netherlands. They argue that criminal behavior is related to the behavior of peers or others in the individual’s environment. Consequently, they treat social capital as a latent construct based on the assumption that “social deviance reflects lower levels of social capital”. In their paper, social capital is defined as a function of participation in civic life, altruism, security and trust, and the extent of informal contacts/acquaintances. Exploiting data from several sources (Statistics Netherlands and ESS), the results of their empirical and cross-sectional analysis are consistent with other research: trust, civicness and electoral turnout have a negative effect on crime, while population heterogeneity and single parenthood are positively related. They also find that the higher the level of education the lower the crime rate, and the higher the percentage of young people, the higher the crime rate. Buonanno, Montolio and Vanin [2009] investigate the effects of civic norms and associational networks on crime rates across 103 Italian provinces. They find that both factors have a negative and significant impact on property crime, but the effects are the opposite for other types of crime such as theft and robbery. Messner, Baumer and Rosenfeld [2004] study the link between several dimensions of social capital and homicide rates in 40 American geographic areas. Using SCBS data (a large-scale telephone survey of households in the continental United States), their aggregate-level factor analysis yields 12 dimensions of social capital that are included as explanatory variables: social trust, religious participation, workplace connections, political engagement (voting, attending meetings, interest in politics), political activism (related to behaviors aiming to effect change such as participation in protests or demonstrations), community involvement (participation in clubs, community events, etc.), community social service, community activism (organizations to promote social change), team sports, informal socializing, altruism, and finally volunteering and charity. After controlling for socioeconomic conditions, they find that none of the dimensions has a significant effect on homicide rates, and only two dimensions (social trust and social activism) are significantly related to crime. The former has, as expected, a negative relationship with homicide rates, while the latter has a puzzling positive effect [2].

3.1. Brief overview of the data

Figure 1

Country-level propensity to participate in an everyday crime

Country-level propensity to participate in an everyday crime

Scatter graph of country-level participation in self-reported everyday crime. The graph is constructed from a binarized version of our dependent variables with “0: No crime” and “1: Crime”. Weighted individual observations are applied to compute national averages. Results of an OLS regression of criminal propensity are shown. Coefficients are significant at * 10 %, ** 5 % and *** 1 %. The gray area represents the 95 % confidence interval.

29 Everyday crimes in ESS5 The aim of this study is to evaluate the relevance of three common explanations of everyday crime: the subjective probability of sanction, norm internalization, and the perceived legitimacy of judicial institutions. Our empirical analysis is based on observations from 26 European countries provided by the ESS5. More than 50,000 individuals participated, divided into twenty-seven countries: Belgium, Bulgaria, Croatia, Cyprus, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Israel, Lithuania, the Netherlands, Norway, Poland, Portugal, the Russian Federation, Slovakia, Slovenia, Spain, Sweden, Switzerland, Ukraine and the United Kingdom. [3] Data were collected between September 2010 and July 2011. Observations with missing values were deleted leaving a final dataset of 26,109 individuals. The pattern of missing variables is shown in Appendix C., which confirms that our data satisfies the covariate-dependent missing assumption.

30 Our three dependent variables (false insurance claims, purchase of stolen goods and traffic offenses) are measured by three ordinal variables. They correspond to the answer to the following question: “How often have you done each of these things in the last five years? How often have you (...) made an exaggerated or false insurance claim? – bought something you thought might be stolen? – committed a traffic offense like speeding or crossing a red light?”. Disregarding missing values, there are five potential answers, encoded as follows: “0: Never”, “1: Once”, “2: Twice”, “3: Three or four times”, and “4: Five times or more”. Summary statistics for these variables and control variables are shown in Table 4.

31 Figure 1 shows the relation between the propensity to commit these offenses at country level, using binarized variables. Hence, “0.3” means that the propensity to commit the crime is 30 % at the country level. Figure 1 indicates a positive correlation between false insurance claims and the purchase of stolen goods. However, traffic offenses are unrelated a priori to false insurance claims. Surprisingly, we observe a negative correlation between traffic offenses and the purchase of stolen goods. False insurance claims are a relatively rare behavior in our final data set: 97.21 % of respondents declared that they had not exaggerated or made a false insurance claim during the preceding five years. The same holds for purchases of stolen goods: 93.06 % declared not having committed this offense. Finally, traffic offenses are the most common: 52.97 % of respondents declared being guilty.

32 Everyday crime and its traditional determinants We used our dataset to construct measurements of: (1) the individual subjective probability of being caught and sanctioned when committing a crime; (2) the degree of internalization of a norm related to a specific crime, and (3) the perceived legitimacy of judicial institutions. We describe the related ESS5 items in Appendix A. and we give summary statistics in Appendix B.

33 Subj. Probability_ij is the subjective probability of being caught and sanctioned for committing one of the crimes studied. There are three options, measured by the following questions: “tell me how likely it is that you would be caught and punished if you... made an exaggerated or false insurance claim?... bought something you thought might be stolen?... committed a traffic offense like speeding or crossing a red light?” These options were encoded as: “1: Not at all likely”, “2: Not very likely”, “3: Likely” and “4: Very likely”. In our analysis, we binarize this variable by dividing the subjective probability of being caught and sanctioned into high and low levels. “Low probability” is encoded as 0, and corresponds to “Not at all likely” and “Not very likely”. “High probability” is encoded as 1 and corresponds to “Likely” and “Very likely”.

34 Morality_ij measures norm internalization. There are three measures, one for each considered crime. Variables relate to the questions: “please tell me how wrong it is to... make an exaggerated or false insurance claim?... buy something you thought might be stolen?... commit a traffic offense like speeding or crossing a red light?” and were originally encoded as follows: “1: Not wrong at all”, “2: A bit wrong”, “3: Wrong” and “4: Seriously wrong”. We also binarized this variable, following the aforementioned methodology.

35 Fourteen variables measure judicial legitimacy. The corresponding questions are described in Appendix A. Together, Cronbach’s alpha is high: 0.81, indicating that a scale built on these questions is statistically reliable. We compute a composite index using factorial analysis. The binary variable Legitimacy_ij equals 1 if the individual has a composite index higher than the median, and 0 otherwise.

Figure 2

Regression coefficients – probit regressions of everyday crimes on the main variables of interest at the individual level

Regression coefficients – probit regressions of everyday crimes on the main variables of interest at the individual level

Probit regression coefficients of self-reported everyday crimes on each main variable of interest, including a constant and using Huber-White heteroskedatiscity-consistent standard errors. Nine regressions were performed, each containing only one explanatory variable. The dependent variables are encoded “0: No crime” and “1: Crime”. Coefficients are displayed with a 95 % confidence interval.

36 Figure 2 displays coefficients and 95 % confidence intervals for nine probit regressions on the data at the individual level. Each regression estimates the association between the declared decision to commit an everyday crime and one of the main variables. Dependent and explanatory variables are binary. Thus, to construct Figure 2, each explanatory variable is encoded as follows: “0: Low level” and “1: High level”. For each dependent variable, the following is used: “0: Did not commit the crime” and “1: Committed the crime at least once”. Figure 2 indicates that norm internalization has a negative association with reported crime for the three offenses under study. This is not the case for the subjective probability of sanction and the perceived legitimacy of judicial institutions. There is a negative relationship between the purchase of stolen goods and false insurance claims and the subjective probability of sanction and the perceived legitimacy of judicial institutions. Traffic offenses follow a different pattern as no significant correlation is found with the subjective probability of sanction. On the other hand and, surprisingly, we find a positive association between perceived legitimacy and traffic offenses. Consequently, Figure 2 shows that the main explanatory variables may differ in their associations with the three offenses under study. This first insight was explored in greater depth with further regressions.

3.2. Estimation strategy

37 Baseline model Given that our dependent variables are both discrete and ordered, we consider an ordered probit model with

0 if y_i^*_j ≤ c₁
{^{1 if c1 < yi*j ≤ c2}
y_ij = g (y_i^*_j)= 2 if c₂ < y_i^*_j ≤ c₃ [1]
3 if c₃ < y_i^*_j ≤ c₄
4 if c₄ < y_i^*_j

38 We denote y_ij as the choice of individual i in country j to commit a crime. If he or she does not engage in a criminal behavior, y_ij = 0. Otherwise the value is higher than one and indicates the frequency of this behavior. The dependent variable is encoded as follows in our analysis: “0: Never”, “1: Once”, “2: Twice”, “3: Three or four times”, “4: Five times or more”. In probit models, a latent continuous metric underlies the ordinal responses observed in the data. can be analyzed as the implicit utility derived from crime. Once the latent variable crosses a certain threshold c_L, the individual chooses the option L.

39 We consider the following specification for the latent variable

y_i^*_j = b₀ + b₁ Subj. Probability_ij [2]
+ b₂ Morality_ij
+ b₃ Legitimacy_ij
+ b₄ Subj. Probability_ij *Morality_ij
+ b₅ Subj. Probability_ij *Legitimacy_ij
+ b₅ Morality_ij *Legitimacy_ij
+ b₄ Subj. Probability_ij *Morality_ij *Legitimacy_ij
+ a₁ Soc. Capital_ij + a₂ Socio. Demog._ij + γ_j + ∈_ij

40 Equation 2 allows to study how the number of everyday crimes varies as a function of several characteristics. Soc. Capital_ij is a vector of variables measuring social capital, Soc. Demog._ij is a vector of sociodemographic variables, γ_j is the country fixed effect, and ∈_ij is the error term, i.i.d. and normally distributed. A complete description of the corresponding variables can be found in Appendix A., while summary statistics are shown in Appendix B.

41 By controlling for social capital and sociodemographic variables, we ensure that the individual socioeconomic background is held constant when measuring the effects of our three drivers of criminal behavior. Including a country fixed effect controls for knowledge of legal sanctions, institutions and cultural contexts that may vary between countries. We also introduce interaction terms between the three main explanatory variables to measure substitution or complementary effects.

42 Our sociodemographic control variables are gender (female, a binary variable that equals 1 if the individual is a woman and 0 otherwise), household income (household income, measured in decile), the level of education (years of education), age, and current professional status (employed, unemployed, retired or otherwise inactive).

43 Our analysis includes six measures of social capital: involvement in an organization (political or not, during the past twelve months); social meetings, a standardized variable that corresponds to the frequency of meeting friends, relatives or colleagues; generalized trust in others (a combination of three statements scored from 0 and 10: “Most people can be trusted [10] or you can’t be too careful [0]”, “Most people try to take advantage of you [0] or try to be fair [10]” and “Most of the time people are helpful [10] or mostly looking out for themselves [0])”; political interest, a standardized variable that relates to the question “how interested in politics are you”; religiosity, the degree of religious spirit (measured as the sum of the following variables, after standardization: “how often do you pray apart from at religious services”, “how often attend religious services apart from special occasions” and “how religious are you”); and finally, trust in domestic political institutions (defined as the sum of the variables measuring trust in country’s parliament, politicians and political parties, after standardization).

44 Further analysis: zero-inflated ordered probit model In our study 97.21 % of respondents stated that they had not committed insurance fraud. At the same time, 93.06 % of respondents declared that they had not purchased potentially stolen goods. This concentration of zeros is referred to as a zero-inflation problem. We treated the problem using an alternative model: the zero-inflation ordered probit regression.

45 The underlying idea is that zeros can arise from two aspects of individual behavior. First, declared compliance can be strongly determined by the individual’s ethical concerns, meaning that they will consistently claim to belong to an “always-zero” group, with “0: no crime” as the only possible value. This may be due to norm internalization and/or the perceived legitimacy of judicial institutions. Second, some compliant individuals may share similar characteristics with criminals. In this event, they respond to a potential decrease in the “price” of criminal behavior. This is more related to the subjective probability of sanction. Therefore, we have two groups of individuals: the first is an “always-zero” group, while the second is a “participation” group. The level of participation in this second group is conditional and the number of crimes includes the value zero.

46 Similarly, the zero-inflation ordered probit model has two parts. The first is a probit model predicting the “always-zero” group. Here, we include the subjective probability of sanction, norm internalization and perceived legitimacy as independent variables in order to reflect the fact that non-participation is related to not only ethical concerns, but also Beckerian motives. The second part is an ordered probit model describing the number of crimes in the participation group. To reflect the fact that individuals in this group are sensitive to price, we only include the subjective probability of sanction and control variables as independent variables.

4.1. Baseline model

47 We estimate an ordered probit model, as described in section 3.2. Table 8 in Appendix D. reports the corresponding results. Concerning the subjective probability of sanction, we find a counterintuitive result, as it is not negatively correlated with reported everyday crime. Nor is the coefficient significant for the reported purchase of potentially stolen goods, while it is negatively significant, at the 10 % level, for reported false insurance claims, as demonstrated by estimates (1) and (2) in Table 8. However, it is positively correlated with the propensity to commit a traffic offense at the 10 % level. Regarding traffic offenses, this finding is in line with previous studies and, particularly, the paper by Michiels and Schneider [1984] who indicate that traffic offenders tend be risk-lovers. It seems that a higher probability of sanction may increase the propensity to commit a traffic offense. Similarly, our non significant coefficient for the purchase of stolen items differs from common wisdom that claims that a higher probability of sanction is a deterrent.

48 Our data is consistent with the norm-based explanation of criminal behavior. We find that the degree of norm internalization is associated with a reduced level of reported criminal behavior. This coefficient for this variable is negative and significant at the 1 % level for the three offenses under study.

49 Regarding the perceived legitimacy of judicial institutions, we find that coefficients are non-significant for the three crimes under study. Consequently, our results cast into doubt Tyler’s seminal result [1990] that people are compliant primarily because they believe in respecting legitimate authority. We include interaction terms between the three main explanatory variables to detect if they are substitutes or complements. A positive and significant coefficient for an interaction term corresponds to a complementary effect, while a significant negative coefficient characterizes substitutes. The results show that only one interaction term is significant: the combination of the subjective probability of sanction and norm internalization for traffic offenses. In this case the coefficient is negative, but only with a 10 % level of confidence. This indicates that these two drivers of crime may be substitutes for this particular offense.

50 Hence, we find that the three main variables of interest have different associations with the declared propensity to commit crime, depending on the offense. Further analysis is needed to understand these differences.

4.2. Further analysis

51 We estimate a zero-inflation ordered probit model, which splits the data generation process into two. Results are reported in Table 9. First, we estimate the coefficients associated with an “always-zero” crime group. The results are described with estimates (5), (7) and (9). In these regressions, a significant and positive coefficient indicates that that the individual is more likely to declare having committed a crime. Second, conditional on reporting participation in a crime, we estimate the coefficients associated with the number of crimes, which can be found in estimates (4), (6) and (8).

52 “Always-zero” reported crime For the purchase of stolen goods and traffic offenses, the effect of the subjective probability of sanction is consistent with the findings from the baseline model and its coefficient is non-significant for the purchase of stolen goods. The coefficient for traffic offenses is positive at the 1 % level, meaning that an increase in the subjective probability of sanction is associated with a higher chance of declaring a traffic offense. However, the coefficient of the probability of sanction is non-significant for false insurance claims, which departs from the baseline model where a negative coefficient was found at the 10 % level.

53 Like the baseline model, norm internalization is significant for all three crimes. Coefficients are negative and significant at the 1 % level in estimates (5), (7) and (9). Thus, a high degree of norm internalization is associated with a higher probability of reporting compliance. Regarding perceived legitimacy, the results are also similar to the baseline model: there is no significant effect of perceived legitimacy in the “always-zero” reported crime group.

54 In the baseline model, there was only one significant (negative) interaction: the combination of subjective probability and norm internalization for traffic offenses at the 10 % level. In estimate (9) the sign is the same, but at the 1 % significance level.

55 Number of reported crimes Estimates (4), (6) and (8) address the drivers of reported crimes. Interestingly, we find that the subjective probability of sanction decreases the number of traffic offenses, conditional on participation. The coefficient is negative at the 1 % level. However, there isno effect on the purchase of stolen items or false insurance claims.