Survey Data and Computational Qualitative Analysis

Abstract

The available survey data sources can only give an incomplete access to the empirical problem under consideration for the reasons discussed in more depth in Sect. 12.1, but anyway three surveys can be used as sources for finding out which norm orientations prevail in the regions under consideration in GLODERS and which are the propensities and frequencies of norm-related actions.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Social Norms as Objects of Transnational Surveys

The available survey data sources can only give an incomplete access to the empirical problem under consideration for the reasons discussed in more depth in Sect. 12.1, but anyway three surveys can be used as sources for finding out which norm orientations prevail in the regions under consideration in GLODERS and which are the propensities and frequencies of norm-related actions.

To determine the initialisation of individual and normative weights for simulations of extortion racket systems , it is desirable to have some empirical information on the distribution of related traits in the populations of the regions which deliver scenarios for the design of the simulation model. The most easily usable source of information on this topic is the European Values Study (EVS, 2008) the data of which were documented on a very detailed regional level (down to NUTS-3, but for the ERS purposes NUTS-1 will be sufficient).

The questions asked in the EVS are not exactly aimed at finding out the interviewees’ orientation towards norms, but the set of questions listed in Table 9.1 can be used as a surrogate.

Table 9.1 Values which children should learn

Note that these variables are coded with 1 for mentioned, and 2 for not mentioned.

A factor analysis with varimax rotation yields three factors (Table 9.2), two of which seem to be at least loosely related to the individual (IW) and normative (NW) weights used in the agent’s decision process.

Table 9.2 Factor analysis of values which children should learn (varimax rotated component matrix)

Factor 1 can be interpreted as a trait of people who want their children to be able to make their decisions independently and according to their feeling of being responsible for their actions and not just according to what they are told to do by others or by religion; persons with high positive scores in this factor will have a high individual weight, and those with high negative scores will have a high normative weight.

The interpretation of factor 2 is in a way similar; it is a trait of people who want their children to have good manners and who are less interested in their independence and perseverance; persons with high positive scores in this factor will have a high normative weight.

Both factors are, according to the analysis method applied, uncorrelated; thus it might be difficult to decide which of the two is a better surrogate to the normative and individual weights.

Another way to find one dimension (except using only one factor from factor analysis) is a one-dimensional optimal scaling. The overall correlation of this scale with the two factors mentioned above is −0.595 and −0.747, respectively. Thus perhaps this is a variable whose distribution in the scenario regions could be used for initialising simulation runs.

To clarify the direction of the newly created scales, Table 9.3 shows their mean values for the two groups of respondents who mentioned and did not mention, respectively, “obedience” and “feeling of responsibility” among their five important “qualities which children can be encouraged to learn at home”.

Table 9.3 Relation between factors and important items

The result is that all newly created scales are positively correlated with the importance of children learning a feeling of responsibility as those who mentioned this have a positive mean score in all three scales whereas those who did not mention this item as important have negative mean scores. For the obedience item this is a little less clear, although for factor 1 and the dimension created by one-dimensional optimal scaling those who mentioned the item as important have a high negative mean score, but this does not hold for the second factor. One must also observe that a large proportion of the respondents seem to have characterised both items as important.

Taking into account the combinations of the respondents’ opinions with respect to these two items yields Table 9.4.

Table 9.4 Relation between the two important items and the factors

All differences are significant, but this is mostly due to the very large sample. If one only compares the means for those who mentioned exactly one of the two items as important, the t statistics for factor 1 is 256.075, for factor 2 it is −29.155, and for the dimension created by the one-dimensional optimal scaling it is 176.267. Thus factor one or the one-dimensional score should be preferred for further analyses.

The distribution of all three scores derived in the past subsection in the regions of interest is shown in Table 9.5.

Table 9.5 Factor scores in GLODERS regions

The differences between the means of the regions are highly significant. If one excludes the “Other” regions from the analysis and restricts the analysis to a comparison of the four regions of our interest, the etas are 0.304, 0.149, and 0.336, respectively (which is not extremely high as the variance within the groups is nearly as high as in the whole sample, but the differences between the four means are high enough). Thus one would initialise the normative drive weight (NDW) of Southern Italy with a distribution whose mean is greater than 0.5, the mean of Northern Italy (if it is used at all in later scenarios) would be approximately 0.5, whereas the Netherlands and particularly South-West Germany would start with a distribution whose mean is below 0.5.

Converting the scales discussed above (which are distributed around a mean of approximately 0 and range from about −3 to +3) into normative weights ranging from 0 to 1 is not straightforward as it is entirely unclear whether the individual and normative weights which one would find in the (fictitious) average European are 0.5 or any other number in the interval between 0 and 1. But it seems reasonable to start with 0.5 as the weight for the fictitious average European and then to convert the scale with a sigmoid function. This would lead to distributions in the four regions as shown in Fig. 9.1 (where the conversion function is the arcus tangens, to be exact: $N = \left( \arctan\ (s) + \pi /2\right)/\pi$). This yields estimates for the input parameter normative drive weight (NDW) of 0.31, 0.42, 0.51, and 0.59, respectively (see Fig. 9.1).

Fig. 9.1

Distribution of the main factor in GLODERS regions

Although the scores originating from factor analysis and from one-dimensional optimal scaling convey more information and less noise than the original observable items, it is interesting enough to analyse the two items selected above to support the interpretation of the scores with respect to the regions. Table 9.6 shows the result (F = 0.134).

Table 9.6 Most important items in GLODERS regions

The association is not extremely high, but as expected the percentage of respondents counting obedience (and not responsibility) among the five most important is highest in Southern Italy and the Islands and lowest in Baden-Württemberg. The percentages of those who count responsibility (and not obedience) among the five most important items have exactly the reverse order.

Unfortunately differences between Italian provinces cannot be analysed with the European Values Study as the Italian group has only provided region but not province codes for the interviewees. But it is at least possible to compare the two regions of Calabria and Sicily to each other and to the rest of Italy , which leads to the results reported in Table 9.7.

Table 9.7 Most important items in Northern and Southern Italy

Table 9.7 shows that Sicily is more similar to the rest of Southern Italy and Islands than Calabria , the main difference between the two being the importance of both items which is much higher in Calabria than in Sicily and the rest of Italy and Europe, and the extremely small percentage of interviewees who declare both less important in Calabria as compared to Sicily and the rest of Italy and Europe (all differences between Calabria and the rest are significant on a 5 % level, the differences between Sicily and the rest are not significant with an a of 0.27, whereas the differences between Calabria and Sicily are significant at a level a = 0.007, measured with c ²).

2 Prevalence of Crime and Propensity to Denounce Criminals

Another source is Eurobarometer 79.1 (European Commission & Brussels, 2013), carried out between February 23 and March 10, 2013. Although this survey did not address extortion racket systems , some of its results about the attitudes of the interviewees towards corruption can perhaps be used as indicators of the attitudes of people towards paying (or not paying) pizzo (i.e. extortion money) and towards denunciation. The situation of people from whom a bribe is demanded is at least similar to the situation of people from whom a pizzo is demanded, and the decision to denounce will be made according to the same rules in both cases. This is why the relation between the prevalence of corruption and the willingness to denounce corrupt practices can be taken as a surrogate of the respective relation between the prevalence of extortion and the willingness to denounce an extorter.

An analysis of a group of questions referring to corruption -related attitudes yielded three common factors which can be described with the matrix of factor loadings (the coding of the items is 1 = yes and 2 = no) reported in Table 9.8.

Table 9.8 Factor analysis of corruption types (varimax rotated component matrix)

If one compares different regions of Italy and the rest of the European Union along the lines of these factors (whose means are normalised to 0 and whose standard deviations are normalised to 1), this comparison yields the following result from which one can conclude that people in EU 15 minus Italy feel that “corruption is not part of business culture” on a level of one-third of a standard deviation above the mean of EU 28, whereas Northern Italy is one-third of a standard deviation below this mean, and people living in Southern Italy and the Islands have a mean which is even two-thirds of a standard deviation below the EU 15 minus Italy mean. Italy as a whole is approximately 0.46 standard deviations below the EU 28 mean (Table 9.9).

Table 9.9 Factor scores in different parts of Europe

The belief in the effectivity of anti-corruption measures and in the incorruptibility of political and administrative institutions is least in the middle of Italy (Toscana, Umbria, Marche Lazio), not in Southern Italy and the Islands (as one could have thought, perhaps this is an indicator or the resignation there).

The denunciation propensity can be estimated with the help of two questions from the same survey where interviewees were asked:

QB14: I am going to read out some possible reasons why people may decide not to report a case of corruption . Please tell me those which you think are the most important?

We list the proportions of the positive answers in Table 9.10. The percentages in bold show that the reporting culture in Italy differs from the respective cultures both in the EU 15 and in the member states which joined the EU after 1995. What is particularly interesting is that Italians would know where to report, even had no difficulties to prove, but find that “no one does”; they do not believe that reporting is not worth the effort, but they are afraid that there will be no protection. The reason “No one wants to betray anyone” (the omertà reason) is mentioned much less frequently than in the rest of the EU.

Table 9.10 Reasons for not reporting corruption in different parts of Europe

The number of interviewees who say that they have reported a case of corruption is too small to yield any significant general results, but at least the difference between Northern and Southern Italy is significant with a Pearson c ² = 7.477, a = 0.006 (Table 9.11).

Table 9.11 Frequency of corruption reporting in Northern and Southern Italy

Any regionally deeper analysis (as in the case of the European Values Study ) is impossible as the number of interviewees in both Calabria and Sicily was too small in this Eurobarometer .

Nevertheless, an analysis of all regions in Europe can give additional information on the relation between prevalence of corruption and willingness to denounce. For this purpose, the 289 regions represented in the Eurobarometer are weighted with the number of individuals interviewed. Then the correlation between the number of people believing that in their country corruption is fairly widespread and the proportion of interviewees having reported of all who experienced corruption is −0.238 (see also Fig. 9.2). In Fig. 9.2 this is particularly visible for the regions where more than 75 % of the interviewees believe that corruption is widespread in their country: Within these regions the percentage of those who experienced corruption and reported is smaller in regions with higher corruption prevalence—another indicator that victims of corruption do not believe that reporting corruption will have any effect. The interesting exception of this rule is the Italian province of Lombardia where 94.4 % (weighted) of the 159 (unweighted) interviewees believe that corruption is very or fairly widespread in Italy , where 2.2 % (unweighted: three interviewed persons) experienced that a bribe was demanded of which 80 % (unweighted two persons) reported this fact—hence this exception does not actually count as an exception.

Fig. 9.2

Scattergram of prevalence of corruption vs. willingness to report corruption

If one weights the region with the number of interviewees who have corruption experience, the correlation between the number of people believing that in their country corruption is fairly widespread and the proportion of interviewees having reported of all who experienced corruption is slightly stronger (−0.311).

From the sparse data available on the problem of the relation between falling victim to a crime such as bribe request and denouncing this attempt one can conclude that the prevalence of corruption and the readiness to denounce are negatively correlated, at least where the prevalence is high. For medium and low prevalence the Eurobarometer is not very helpful as the number of relevant cases is too small (2500 in total, only 36 regions with 20 and more interviewees who answered that they had experienced some kind of corruption, and 36 regions with none such interviewee at all).

3 Behaviour Related to Menace, Assault, and Aggression in Italy

The Italian National Institute of Statistics (ISTAT) has carried out several surveys about citizens’ security, the last wave having been in the field in 2008 (Istituto Nazionale di Statistica, 2008–2009). At least two sections (10A and 10B) seem appropriate to describe the setting in which victims of extortion decide whether they denounce or pay the requested amount—although the questions in the two mentioned sections are about unspecified menaces and aggressions.

In this survey 60,001 persons were interviewed by telephone. 2187 interviewees reported that they had experienced a case of menace or of assault or aggression during the past 3 years. 216 or 9.9 % of these cases were reported to the police. Two questions allow for a differentiation of both victims and offenders. Those interviews who reported to be entrepreneurs/imprenditori, professionals/liberi professionisti, self-employed/lavoratori in proprio, etc. could be called “typical targets”, while the others (employees of different categories) could be called implausible victims. Among the offenders there are those whom the interviewees described as strangers (“estraneo”) of loose acquaintances (“Una persona che conosceva di vista”) or whom they did not want to describe (“non risponde”)—these offenders could be called possible offenders, whereas the other offenders (i.e. members of the family, colleagues, or near neighbours) are rather implausible offenders.

If one analyses the data according to these criteria, Table 9.12 shows the results.

Table 9.12 Offences similar to extortion by regions, target type and offender type, and percentage of denounced offences

It becomes visible that those offences which can be interpreted as extortions are much more frequent in Campania, Puglia, Calabria , and Sicily than in the rest of Italy , and it is also true that in the mentioned regions denunciations are more frequent. Generally speaking, entrepreneurs and professionals seem to denounce slightly less often than the others (8.9 % vs. 10.6 %, but with Fisher’s exact test the significance level is only 0.056). Offences by strangers or loose acquaintances are much more often denounced than those committed by family members, colleagues, or close acquaintances (22.1 % vs. 4.4 %, significance level of Fisher’s exact test is less than 0.0000005). Denunciation propensity is slightly higher in the named regions than in the rest of Italy (11.1 % vs. 8.4 %, significance level of Fisher’s exact test is 0.212).

More detailed analyses are not very reasonable as the prevalence of all these offences is still moderate, and in spite of a very large sample of 60,001 cases the subsample of offences similar to extortion (i.e. menaces and aggressions committed by possible offenders towards typical targets as defined above) is with 214; see Table 9.12: Offences similar to extortion by regions, target type and offender type, and percentage of denounced offences—too small.

4 The Sicily and Calabria Database

This subsection will use a database compiled from 629 cases from all provinces of Sicily and Calabria (Frazzica et al., 2015)—see also Chap. 7—which allows for comparisons between provinces and for making more than one empirical case available for a comparison to simulation output (as each simulation run can be considered as a history produced in a fictitious province with its initial conditions and parameter settings).

A bivariate analysis also shows that about 80 % of all extortions reported in this database were never reported by the victims (see Table 9.13), but in those provinces where our data are more or less complete, nearly all of the denunciations led to a conviction.

Table 9.13 Frequency of extortion denouncement in provinces in Southern Italy

Table 9.14 also shows that only a minority of extortion requests were refused (53 out of 224, 23.7 %) and that out of these in only 16 cases (30.2 %) the information of the police came directly from the victim; in 14 more cases (26.4 %) information from the victim led to denouncement , but even in the refusal case 23 extortions (43.4 %) were found out by the police without any help from the victim. When the victim’s reaction was acquiescence or connivance, there was no direct denouncement (the eight cases with acquiescence where the first information came from the victim seem to have remained below the threshold of denouncement), of course; from these 161 cases, 138 were detected by police activities (84.7 %), and help from the side of the victims was only given in 15.3 % of these cases. Differences between the three provinces where we have data about both variables are small.

Table 9.14 Reaction of the extortion victim in provinces of Southern Italy

If one only considers the victim reaction and the fact whether there was a denouncement (here we have sufficient data for all the provinces in Southern Italy ), the analysis yields the following picture (Table 9.15): The association between the two variables is very high with the exception of the province of Catania (both Cramer’s V and the uncertainty coefficient are highly significantly different from 0.0).

Table 9.15 Relation between victim reaction and denouncement in provinces of Southern Italy

Another high correlation exists between the victim reaction and the fact whether the extortion was actually committed or only attempted (see Table 9.16).

Table 9.16 Relation between reaction of victim and success of the extortion attempt in provinces of Southern Italy

Again the association is very high except in the case of the province of Siracusa (all uncertainty coefficients with the exception of Siracusa’s and all Cramer’s V are again significantly different from 0.0).

If one goes further back in a presumable chain of causalities the following can be found (Table 9.17): Repeated payments are more often successfully requested (85.4 %) than one-off payments (63.5 %); the request for both at the same time is very often (83.7 %) successful. The effect sizes (Cramer’s V = 0.230, uncertainty coefficient with the success of the operation dependent = 0.047) are only moderate, but both are highly significant, but in most provinces the 0.05 significance level is not reached.

Table 9.17 Extortion types and success of extortion attempt

Surprisingly the reported intimidation only has a small effect on the success of the operation of extortion (Cramer’s V is 0.117, uncertainty coefficient is 0.015, both are significant, a is 0.004 and 0.001, respectively, but for such a small effect the significance is rather irrelevant).

The intimidation (Table 9.18) seems to have rather the opposite effect, but with only a very small number of extortions without intimidation this is not relevant either.

Table 9.18 Intimidation and success of extortion attempt

Whether only money is requested or something else (goods, services, staff, or any of these combined with money) makes a small difference (which is statistically significant at 0.0005 level, but the effect measured with Cramer’s V 0.173 is rather modest) (Table 9.19).

Table 9.19 Kind of demand and success of extortion attempt

For about one-third of the cases no information was available.

Given the fact that the empirical data do not cover the whole theoretical model, as much information backing the micro-specification is more or less unobservable (which is particularly obvious in the case of the extorters), validation of the theory and the simulation model derived from the theory is difficult. This will be discussed in more detail in Sect. 12.1.