Keywords

Several analyses were conducted to reach different explanatory models and thereby to determine the most critical variables in the datasets. This quantitative part is correlational, not experimental (causal). Therefore, the importance of a variable is based on its explanatory power, not on causal relations.

The competitiveness GCI index contains numerous sub-indices, which are aggregated into 12 different categories (World Economic Forum, 2016). Among the sub-indices, GCI comprises corruption (CPI) (or its opposite, transparency perception), as well as some of its determinants. However, the interactions between the variables of interest and corruption, on the one hand, and between the variables of interest and competitiveness, on the other, were analysed and modelled separately.Footnote 1

This chapter has three sections. The first overviews the process of modelling competitiveness (GCI) and corruption (CPI). The second corresponds to the methodological protocols and limitations. The third presents the empirical results of the regression analysis for GCI and CPI.

1 Modelling Competitiveness and Corruption

1.1 Modelling Competitiveness (GCI) (Stages 1 and 2)

I modelled several variables associated with theories of prosperity (see Chap. 5 and Appendices 1–2, and Supplementary Materials) as possible exogenous determinants of competitiveness in most countries worldwide. For example, the Environment Performance Index (EPI) is linked to geographical and environmental determinants, while religion (adherents) is related to cultural and Weberian theory. State religion and legal origins are proxies of the institutional influence of religion. I tested the possible different contributions of those variables to explain the variation in competitiveness (GCI).

The next generation of models focused on countries in Europe and the Americas, i.e. those two continents under study here. Finally, the subsequent generation of models isolated the influence of variables such as state religion and share of population adherents on different religious affiliations. Such isolation ought to benefit the model fitness as these variables are related (Barro & McCleary, 2005) and might overfit the model.

1.2 Modelling Corruption (CPI Stage 3)

Analysing corruption as a criterion variable (stage 3) associates several common predictor variables frequently used in the literature to explain corruption: GDP per capita, religion, or political liberties (Rose-Ackerman, 2006) (see Part III, Chap. 6, and Appendix 2). The modelling of corruption focused on Europe and the Americas. However, unlike the modelling of competitiveness, it bypassed the first generation of models (worldwide).

2 Methods of Regression Analysis

2.1 Methodological Limitations

2.1.1 Latest Available Data Measured at Different Time Periods

To draw the most accurate picture of existing reality, this study is based on the latest data available for all the investigated variables (most data come from 2016 reports). However, 2016 data were not available for all variables (care was taken to ensure that data were not older than 20 years). Saying this, slightly older data would not amount to a significant limitation. The prosperity performance of most Western countries exhibits comparatively long-term persistence (of a century or more) and thus indicates historical inertia (Michalopoulos & Papaioannou, 2017; Acemoglu & Robinson, 2012; Acemoglu & Robinson, 2008). The same applies to some environmental variables such as geography, latitude (Diamond, 1997), or state religion (Barro & McCleary, 2005).

2.1.2 Regression Analysis Is Not a Causal Approach

A significant correlation does not always indicate causation (Arruñada, 2010, p. 908). Moreover, causal approaches to the present dataset and research design (i.e. based on the latest available data) might not be desirable for two reasons: First, this would lead to endogeneity problems. Testing plain variables as evident in the model would render regression results inconsistent due to the likely presence of endogeneity (i.e. because of simultaneous causality). Reverse causality is a widespread issue in regression models. Likewise, testing variable causality blindly (i.e. merely statistically, without any historical or theoretical consideration) would not be appropriate either. Thus, the historical approach to the variables under investigation is explained and addressed in detail (Figs. 1.1 and 8.1; Chaps. 6–12). Second, adopting a causal approach would require Instrumental Variable (IV) estimation, a complex experimental design, and a definition of independent, dependent, and control variables. It would also need instrumental variables, which might not exist in reality (or might be highly challenging to identify and define). While a time series approach would partially resolve the issues arising from a causal quantitative approach, it is not addressed in this study (see below). However, one other method in this research (QCA) uses a causality approach.

2.1.3 This Regression Analysis Excludes a Time Series Approach

A time series would be an ideal approach to test whether variables are related over short or rather over long periods. This approach would allow identifying whether relationship patterns between variables exist over time. Conceptually, it would make sense to relate performance in a period t with explanatory factors measured in period t-1. One example would be if increasing or decreasing religiosity or the influence of a Christian denomination coincided with a prosperity variable (i.e. whether changes in a population’s religious affiliation relate to changes in corruption).

The relationship between prosperity and religion in Protestant versus Catholic societies could be tested in time and space. Regarding time, for example, countries that were Roman Catholic before the Protestant Reformation (e.g. England, Germany or Switzerland) were pre-industrial (mainly primary sector economies) in the Middle Ages. In geographical space, for instance, countries currently under pervasive Roman Catholic influence (e.g. Spain or Latin America) tend to exhibit lower performance than Protestant ones.

Analysing Christian religion reveals that the Reformation initiated gradual changes, including its slow institutionalisation and influence on democracy, literacy, and the state (Becker et al., 2016; Woodberry, 2012; Witte, 2002). Consequently, a time series approach ought to cover longer periods (at least several centuries) based on the assumption that the prosperity of most Western countries maintains almost the same relative position over long-term time periods (a century or more) (Michalopoulos & Papaioannou, 2017; Acemoglu & Robinson, 2008).

A time series approach might also alleviate endogeneity issues, as current performance cannot (directly) influence past conditions. Still, relevant reverse causality might exist via indirect correlation. Nevertheless, a time series approach also has the disadvantage that the same available data does not exist for all variables. Some variables are measured yearly, some every 5 years, and some every 50 years (most have only been measured during the last 50 years or less). For some essential variables, no information and proxies exist, thus making it necessary to create new databases to cover particular periods (e.g. Pre-Reformation and Reformation). Undoubtedly, the time series approach is promising, yet requires considerable time and resources (which, however, exceed the scope of the present book). Notwithstanding this reservation, future research might consider applying this approach.

The current approach uses “legal origin” (La Porta et al., 1999) and “state religion” (Barro & McCleary, 2005) to capture some variable relationships over long periods. Barro and McCleary (2005) proved that a growing fraction of adherents to a country’s main religion increases the probability of a “state religion”. This gradual process means that few Western countries changed their status of having a “state religion” in the twentieth century.

2.2 Data and Empirical Strategy

Population

The countries studied here were selected based on data availability (the same consideration applies to QCA, see Chap. 16). Data stem from constructed and from secondary (existing) data available in public databases. Further, the data consist of censuses of currently available indicators for the following cross-country analyses:

Stage 1. Worldwide (107 countries): The database comprises data series available on competitiveness, corruption, social inequality, as well as other social, environmental, and economic indicators. It also includes indicators of denominational tradition and religious background for most of the surveyed countries.

Stages 2 and 3. Europe and the Americas (66 countries—competitiveness; 61 countries—corruption): Cases with straightforward data access for most variables in Europe and the Americas.Footnote 2 Unlike in other regions, Christianity has been the dominant religion for at least the past two centuries in Europe and the Americas. The populations of these continents currently adhere largely to Roman Catholic, Orthodox, or Protestant churches of various denominations (Johnson & Zurlo, 2016).

2.3 Protocol

I followed an 11-step protocol (summarised below):

Stage 1: Competitiveness in the World

An illustration of stage 1, competitiveness in the world presents 6 steps as follows. 1, data preparation. 2, variables selection. 3, model generation. 4, model selection. 5, model preparation. 6, further modeling. These are further detailed.

Step 1 aimed to include as many countries as possible in the database. However, not all the variables had the same data available for all the observations. Variables such as the GLOBE cultural index (House et al., 2004) were excluded from the analysis as the corresponding data were only available for a sample of three dozen countries. If less than 25% of values were still missing in the remaining dataset, they were imputed using Multivariate Imputation by Chained Equations in R (mice).

Step 2 involved using automated functions to help select from a high number of variables (see diagram above) from different theoretical backgrounds (see Appendix 1). Although the results of one automated model search may be arbitrary, about 40 of these models were created using different approaches to eliminate path-dependence and bias.

Step 3 concentrated on meeting linear regression assumptions in all the models. Assumptions were tested in a continuous process whenever a new model was created. Among other assumptions, I tested for multicollinearity by checking the Variance Inflation Factor (VIF) and the correlation value. VIF results refused multicollinearity and, along with low correlation value, suggest that these variables are orthogonal. Likewise, all significant correlations among the predictor and criterion variables were controlled for.

Step 4 involved using cross-validation based on the prediction quality of shuffled groups of observations. Those model and variables exhibiting the best prediction ability were selected. “Step” is an automatic method based on the R function step (), which is also used to perform variable selection. These methods are useful when the number of explanatory variables is large, and when fitting all possible models proves unfeasible (University of Columbia, 2017).

Step 5 used Beta standardisation, which is necessary for comparing influence among variables. This indicates the significance of each variable in the context of others (and thus facilitates comparison). The goal, however, is not to test significance. Nor can Beta standardisation be computed for interaction variables (see second column “Standardised” in the Results).

In Step 4, Model 1 (see Results), two selected variables that might potentially outshine each other were analysed: religious denomination (proportion in population) and state religion. Consequently, I isolated these variables and ran separate models (see Step 6).

Stage 2: Modelling for Europe and the Americas (competitiveness).

An illustration of stage 2, modeling for Europe and the Americas, competitiveness, presents 3 steps as follows. 7, dataset modification. 8, variables and model selection. 9, model interpretation. These are further detailed.

Europe and the Americas exhibited higher data quality and availability. Consequently, there was no need to impute data as in the world models (Stage 1: Models 1, 2, and 3). Therefore, stable regional models with higher reliability (i.e. avoiding data imputation and thus reducing bias) were created (Stage 2: Models 4 and 5).

Step 7. Due to data availability issues, some countries were excluded (e.g. most of the small island states in the Caribbean). Step 8 was similar to steps 2, 3, and 4, while step 9 was similar to step 5.

Stage 3: Modelling for Europe and the Americas (corruption).

An illustration of stage 3, modeling for Europe and the Americas, corruption presents 2 steps as follows. 10, dataset modification. 11, variables and model selection. These are further detailed.

The competitiveness index (GCI) includes corruption (or its opposite, transparency) perception (CPI). However, to analyse whether transparency has the same direction as competitiveness, I ran a model that considers corruption as a criterion variable applicable to both Europe and the Americas.

Step 10. The corruption index (CPI) lacks data in several observations; therefore, fewer countries were selected from the dataset than in stage 2 (Step 7: Competitiveness).

Step 11. Regression trees are used for classification purposes or to predict outcomes when the response variable is numerical or continuous as in this situation (Morgan, 2014).

3 Empirical Results of Regression Analysis

Some parts of this section were originally published as “Empirical Results (Correlational Analysis)” in: Garcia Portilla, J. (2019). “Ye Shall Know Them by Their Fruits”: Prosperity and Institutional Religion in Europe and the Americas. Religions, 10(6), 362. MDPI AG. Retrieved from https://doi.org/10.3390/rel10060362

© 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.

Appendices 1–2 provide a detailed description of the variables, theories, categories, and sources used here, as well as of the interactions between the variables in the general research model. Presented below are the results for competitiveness (Models 1–5 and conclusions) and for corruption (Model 6 and conclusions).

3.1 Competitiveness

3.1.1 Stage 1: Competitiveness in the World

The analysis of competitiveness as a criterion variable associates different predictor variables usually corresponding to theories of prosperity. Model 1 was chosen from more than 30 other models as it has the highest cross-validations and satisfactorily explains GCI variability.

3.1.1.1 Model 1
$$ {GCI}_i={\beta}_0+{\beta}_1{\mathrm{EPI}}_i+{\beta}_2{Mulatto}_i+{\beta}_3{Asian}_i+{\beta}_4\mathrm{Protestant}\_\mathrm{St}.{\mathrm{Rel}}_i+{\beta}_5\mathrm{Catholic}\_{\mathrm{pop}}_i+{\beta}_6 Orthodox\_{pop}_i+{\beta}_7\mathrm{Protestant}\_{\mathrm{pop}}_i+{\beta}_8\mathrm{Muslim}\_{\mathrm{pop}}_i+{\beta}_9\mathrm{German}\_{\mathrm{LO}}_i+\varepsilon i $$
3.1.1.1.1 Positive Correlations

The most significant variable is the Environment Performance Index (EPI), which is highly positively correlated with competitiveness GCI (0.72). If EPI was removed from the model, R-squared would drop by 17%. However, the same model without EPI exhibits similar results to Model 1 (Table 15.1) for the remaining variables. Here, an increase of EPI by one point is associated with a GCI growth of 0.026 points. EPI occupies larger units (mean value 70) than GCI (mean value 4) (The scale is standardised, meaning that these values refer to the non-standardised ranges; see Appendix 1 and Supplementary Materials).

Table 15.1 Model 1. Competitiveness in the world
Full size table

A second important variable positively related to competitiveness (GCI) is the German legal origin: “GERMAN (LO)”. Such countries have on average a higher GCI by 0.229.

Third, if Protestantism was the state religion “PROTESTANT (S.R)” (e.g. UK, Sweden, Denmark), then the overall effect would be positive on GCI. Establishing Protestantism as a state religion leads to a GCI increase of 0.223, which is far greater (almost double) than for a highly Protestant population. This confirms the higher importance of the institutional influence of religion compared with that of the proportion of adherents (in line with Sect. 7.1.)Footnote 3

3.1.1.1.2 Negative Correlations

All variables related to religion distribution in a population—Catholics (%), Orthodox (%), Protestants (%), Muslims (%))—are negatively correlated with GCI although these correlations are only marginally significant. Nonetheless, the Orthodox population causes the most substantial negative effect. An increase by one percentage point of the Orthodox population leads to a decrease of 0.0078 units in GCI. The same goes for changes in the Roman Catholic population, where an increase in one percentage point of their population would mean a GCI decrease by 0.0057. Likewise, if the share of the Protestant population increased by one percentage point, GCI would decrease by 0.0054 units. Finally, if the share of the Muslim population increased by one percentage point, GCI would decrease by 0.0047 units.

On the other hand, Mulatto and Asian ethnical values both have a negative effect. A growth of one percentage point in the mulatto/Asian population means a 0.00117/0.00194 decrease in GCI, respectively.

Models Analysing Major Religious Population Groups or State Religions Separately

Model 1 (Table 15.1) has shown a differential influence of the religion distribution in a population and of state religion on GCI. Models 2 (Table 15.2) and 3 (Table 15.3) were run to separate any potential differential influence. Separate analysis of the two variables eliminates the possibility that a religious population distribution and a state religion would “overfit” the model.

3.1.1.2 Model 2 with Population Percentage

Model 2 shows the relation between the percentage of religious adherents and competitiveness while excluding state religion variables.

$$ {GCI}_i={\beta}_0+{\beta}_1{\mathrm{EPI}}_i+{\beta}_2{\mathrm{Caucasian}}_i+{\beta}_3{Mulatto}_i+{\beta}_4{Asian}_i+{\beta}_5{Dogmas}_i+{\beta}_6{Socialist}_i+{\beta}_7{\mathrm{German}}_i+{\beta}_8\mathrm{Catholic}\_{\mathrm{pop}}_i+{\beta}_9\mathrm{Muslim}\_{\mathrm{pop}}_i+{\beta}_{10} Orthodox\_{pop}_i+\varepsilon i $$

Model 2 (Table 15.2) confirms the findings of Model 1 (Table 15.1) in the same order:

  1. 1.

    EPI is the most significant variable.

  2. 2.

    Increasing Orthodox, Catholic, and Muslim populations negatively influence GCI. However, the Protestant population and other Christian adherents are not significant in Model 2.

  3. 3.

    Model 2 confirms the direction and influence of most other variables in Model 1. However, in Model 2, Caucasian ethnic values positively influence GCI. Likewise, Dogmas and Socialistlegal origin negatively impact GCI.

Table 15.2 Model 2: Competitiveness in the world including the percentage of religious adherents and excluding State religion variables
Full size table
3.1.1.3 Model 3 (Including State Religion)

Model 3 (Table 15.3) excludes the percentage of religion adherents variables. State religion variables alone explain most of the variability otherwise explained by religious population.

Table 15.3 Model 3: Competitiveness in the world including state religion and excluding the percentage of religion adherents
Full size table
$$ {GCI}_i={\beta}_0+{\beta}_1{\mathrm{EPI}}_i+{\beta}_2{\mathrm{Caucasian}}_i+{\beta}_3{Mulatto}_i+{\beta}_4 Asia{\mathrm{n}}_i+{\beta}_5{\mathrm{Catholic}}_i+{\beta}_6{\mathrm{Protestant}}_i+{\beta}_7{Socialist}_i+{\beta}_8{\mathrm{German}}_i+\varepsilon i $$

The results of Model 3 (Table 15.3) ratify those of the previous models. The most significant positive influence on GCI comes from EPI, Protestant state religion, Caucasian ethnicity, and German legal origin. In contrast, Asian and Mulatto ethnicities, Roman Catholic state religion, and Socialist legal origin would negatively affect GCI.

3.1.2 Stage 2: Modelling Competitiveness (Europe and the Americas)

The following models were produced explicitly for Europe and the Americas, excluding the noise existing in the world database. Likewise, Models 4 and 5 serve to compare whether the same variables chosen in the previous “world” models are still significant in Europe and the Americas.

3.1.2.1 Model 4: Results of Cross-Validation
$$ {GCI}_i={\beta}_0+{\beta}_1{Hostilities}_i+{\beta}_2\mathrm{German}\_{\mathrm{lan}}_i+{\beta}_3\mathrm{Catholic}\_{\mathrm{pop}}_i+{\beta}_4\mathrm{Orthodox}\_{\mathrm{pop}}_i+{\beta}_5{\mathrm{EPI}}_{\boldsymbol{i}}+\varepsilon i $$

This result shows the high robustness of this model (Table 15.4), in that it explains almost 66% of GCI variability with four significant variables at a 99% confidence level:

  1. 1.

    EPI is the most important variable because it accounts for most of the GCI variability. The increase in EPI by one percentage point is related to a “GCI” growth of approximately 0.038 percentage points.

  2. 2.

    The second most important variable is the Orthodox population, which exerts the most substantial negative effect in the model. If the Orthodox population increased by one percentage point, GCI would decrease by approximately 0.010 percentage points.

  3. 3.

    Similarly, the effect of the Roman Catholic population also negatively influences competitiveness. If the Roman Catholic population increased by one percentage point, GCI would drop by approximately 0.005 percentage points. This finding, along with the previous one (2), is consistent with La Porta et al.’s (1999) conclusions about the negative influence of hierarchical religions on prosperity.

  4. 4.

    Finally, the proportion of the German-speaking population positively affects GCI. An increase of 1 percentage point in the German-speaking population would mean a GCI increase by approximately 0.008 percentage points. This finding would confirm a positive relationship (as outlined in Sect. 11.1).

Table 15.4 Model 4: Competitiveness in Europe and the Americas (cross-validation method)
Full size table
3.1.2.2 Model 5: Results with Step
$$ {GCI}_i={\beta}_0+{\beta}_1{Hostilities}_i+{\beta}_2\mathrm{GERMAN}\_{\mathrm{LEGAL}}_i+{\beta}_3\mathrm{Catholic}\_{\mathrm{pop}}_i+{\beta}_4\mathrm{Orthodox}\_{\mathrm{pop}}_i+{\beta}_5{\mathrm{EPI}}_i+\varepsilon i $$

“Step” is a fundamentally different method, yet mostly exhibits the same results (Table 15.5) as those of cross-validation (Table 15.4). Step analysis further confirms the choice of the right variables (based on reality, not on random data effects). Here, the only difference is that Step analysis chose GERMAN (legal origin) instead of German language (percentage of German-speaking population) with 99% confidence. The chapters that discuss the theoretical framework and conclusions further discuss these variables and findings.

GERMAN (legal origin) is a binomial variable, meaning that if a country is of German legal origin, then its GCI is 0.759 higher. If it is not of German legal origin, then the variable does not affect GCI. Only these two extremes exist.

On the other hand, the variable Social Hostilities due to religion also appears in Models 4 (Table 15.4) and 5 (Table 15.5). It exhibits the lowest value of standardised beta and also low values. More importantly, this variable appeared only at 90% confidence value (all other results had a 99% confidence value). Therefore, this variable is not sufficiently significant.

Table 15.5 Model 5: Competitiveness in Europe and the Americas (Step method)
Full size table

Finally, standard deviation confirmed the results of both cross-validation (Table 15.4) and Step (Table 15.5) and found the same, most robust coefficients. Consequently, only four variables in the models had statistical significance.

3.1.3 Conclusions for Competitiveness

3.1.3.1 Conclusions for Competitiveness in the World (All Models)

The consistent results of the five previous models are:

  1. 1.

    A positive influence of EPI on GCI;

  2. 2.

    A positive influence of a German legal origin (or German language) on GCI;

  3. 3.

    A negative influence of an Orthodox population on GCI;

  4. 4.

    A negative influence of a Roman Catholic population (or Roman CatholicState Religion) on GCI.

These results are valid for the world—Models 1 (Table 15.1), 2 (Table 15.2), and 3 (Table 15.3)—, as well as for Europe and the Americas—Models 4 (Table 15.4) and 5 (Table 15.5)—.

Partially conclusive findings:

Ethnic influence appeared with some degree of importance in the world models but disappeared in Europe and the Americas.

Socialist legal origin negatively influenced GCI—Models 2 (Table 15.2), and 3 (Table 15.3)—.

The influence of the share of Protestants in the population is not conclusive. In Model 1 (Table 15.1), Protestants negatively affected GCI, after Orthodox and Roman Catholics. However, in Model 2 (Table 15.2), which analyses religious population, the influence of Protestants disappeared.

Neither Protestant population (1) nor Protestant State Religion (2) are significant variables for competitiveness in Europe and the Americas—Models 4 (Table 15.4) and 5 (Table 15.5)—. Hypothetical reasons for such findings include, respectively, (1) the high influence of Pentecostalism in the Protestant population today might neutralise the possible positive effect of historical Protestantism. Pentecostalism impacts little on human capital and institutions (Becker et al., 2016; McCleary, 2013; Woodberry, 2012) and has often fallen into established practices “of corporatism and clientage” (Martin, 1999, p. 40; Schäfer, 1997) (Sect. 10.4.3). (2) While Switzerland and the USA are the most competitive (GCI) countries, they do not have Protestantism as their state religion, despite being historically Protestant (Barro & McCleary, 2005; Inglehart & Baker, 2000).

Nonetheless, Protestant State Religion exhibited a positive significance on Model 3 (State Religion). This suggests that Protestant State Religion is more important for GCI than the proportion of Protestants in the population. The effect of Protestant State Religion on higher GCI might also be related to its influence in diminishing the institutional power of the Roman Church (Sect. 8.3.4). The latter conclusions both confirm Fanfani’s claim as early as 1936 (as cited in Grier, 1997) that the separation of state and church is the critical prosperity trigger. Such a separation occurred mainly in Protestant countries for anti-clerical reasons. Fanfani argued that religion per se harms prosperity unless it leads to the separation of ecclesiastical and political/economic powers in a country (as historical Protestantism did).

3.1.3.2 Conclusions for Competitiveness in Europe and the Americas

In Europe and the Americas, Models 4 and 5 discover a combination of variables that largely explains GCI variability in the following order of importance:

Environmental Performance Index (EPI)

I found a high correlation of EPI with competitiveness (GCI), as suggested by the environment-and-geography theory of prosperity (Diamond, 1997; Sachs, 2001; Brown & Lall, 2006). This index has by far the highest positive influence on GCI of all the variables considered and explains most of the GCI diversity. Higher EPI strongly implies higher GCI.

Legal Origin

As predicted in the variables description sections (Chap. 3–12), I found a highly positive relation of German legal origin and German language with GCI. The influence of the Reformation on the German legal system has been widely discussed (Witte, 2002; Berman, 2003) (Sect. 8.3.4). Likewise, several studies (Besch, 1999; Greenslade, 1963) have discussed the influence of the Reformation on the dissemination and standardisation of the German language (Sect. 11.1).

3.2 Corruption

3.2.1 Proxy Predictor Variables for the Regression Model of Corruption

The quantitative model used here tested some proxy variables of selected main determinants of corruption (Rose-Ackerman, 2006; Alesina et al., 2003). Such predictor variables are:

  1. 1.

    Proxies of economic prosperity: the log GDP per capita (GDPPC) and the Index of Economic Freedom (IEF);

  2. 2.

    Proxies of democracy: the Political Rights and Civil Liberties Index (PRCL) and the Political Stability Index (STABIL).

  3. 3.

    Proxies of culture: the percentage of Roman Catholics (CATH), and Protestants (PROT) in the population. This also includes ethnic, linguistic, and religious fractionalisationmeasures (ETHF, LING and RELF, respectively).

For each variable, the latest available data were used (one observation per country). Please refer to Annex 1 for details on the variables and sources.Footnote 4

$$ CP{I}_i={\beta}_0+{\beta}_1 GD{PPC}_i+{\beta}_2{IEF}_i+{\beta}_3{PRCL}_i+{\beta}_4{STABIL}_i+{\beta}_5{PROT}_i+{\beta}_6{CATH}_i+{\beta}_7{ETHF}_i+{\beta}_8{LINGF}_i+{\beta}_9{RELF}_i+\varepsilon i $$

3.2.2 Regression Results for Corruption in Europe and the Americas

Data analysis includes a regression tree (see Fig. 15.1). The Political Rights and Civil Liberties Index (PRCL) is the most critical factor in determining the Corruption Perception Index (CPI). Lower corruption levels (CPI = 81.5) are evident in countries with a (PRCL) ≥44 and that also have a higher log GDP per capita (GDPPC > 10.6). Such results are entirely consistent with the respective theoretical predictions (Sects. 3.1, 3.3 and 8.1 indicate that these variables are part of the same phenomenon). In line with theory (Sect. 6.1), transparency increases even more if the country in question also has a high proportion of Protestants (PROT > 0.24) in its population (CPI = 84.75).

Fig. 15.1
A tree diagram. The root node is labeled, x P R C L, left-angled bracket, right-angled bracket, 44. It has 2 child nodes, which further divide into nodes labeled 1, 2, 3, 4, and 5.

Regression tree, with the mean value of CPI, the number of observations, and percentage deviance. The minimum number of observations per node for a split to be attempted is set to 15. Any split not decreasing the overall lack of fit by a factor of 0.005 is not attempted. (Source: Author’s collection)

Full size image

This regression tree confirms that PRCL; GDPPC, and PROT are all positively related to transparency (or vice versa, negatively related to corruption). GDPPC and PROT do not seem to play an essential role in countries with low PRCL (democracy). This result is also in line with theory. For instance, Treisman (2000) found that only stable democracy (before 1950) reduces corruption. Most of the Protestant upsurge in the developing world has stemmed from Pentecostalism after 1950 (Johnson & Zurlo, 2016), which only marginally impacts human capital and institutions (Becker et al., 2016; McCleary, 2013; Woodberry, 2012).

The estimates in Table 15.6 corroborate the relation between lower corruption (or higher transparency) and a high proportion of Protestants in the population. Similarly, other predictor variables, above all GDPPC, PRCL, LING correlate with lower corruption levels (i.e. higher CPI). The negative impact of the share of Catholics in the population is less visible, with a lower coefficient in magnitude. These results are entirely compatible with theory, in particular with Treisman (2000).

Table 15.6 Model 6: Corruption in Europe and the Americas
Full size table

As indicated by the positive coefficients for LING, linguistic fractionalisation tends to imply lower corruption. This trend is illustrated by countries (e.g. Luxembourg, Canada, or Switzerland) facing low corruption and featuring in the top ten countries of the present sample regarding language fractionalisation. Conversely, a higher level of ethnic fractionalisation (ETHF) tends to be associated with higher levels of corruption, as established by other empirical studies (Sect. 10.1). This observation confirms that linguistic heterogeneity does not necessarily coincide with ethnic heterogeneity. For instance, most Latin American countries are relatively homogenous concerning language but less so regarding “ethnicity” or “race”.

The regressions do not display any significant relation between religion fractionalisation RELF and corruption. Consequently, low corruption is present in both multicultural, free societies with high measured religious fractionalisation (e.g. USA, United Kingdom, Canada), and countries with low religious fractionalisation (e.g. Luxembourg or Belgium). A “packed” index of fractionalisation measures works like a “black box” as analysing interactions between the components (i.e. each language or ethnicity) is impossible.

3.2.3 Conclusions for Corruption in Europe and the Americas

Generally speaking, the conclusions reached here about the selected variables resemble those of previous research (Treisman, 2000; Paldam, 2001; Arruñada, 2010; Chase, 2010; and to a less extent La Porta et al., 1999). Results were validated using different databases and methods.

3.2.3.1 Proportion of Protestants

The findings confirm my hypothesis that corruption levels are directly (i.e. negatively) related to the proportion of Protestants in countries in Europe and the Americas. Thus, the share of the Protestant population might play an essential role in reducing corruption. This result is consistent throughout the regressions and the regression tree and confirms previous empirical studies (e.g. La Porta et al., 1997; Treisman, 2000; and Chase, 2010). The influence of Protestantism is further explored using the QCA method, which enables one to consider other variables also playing an essential role besides population share.

3.2.3.2 Proportion of Roman Catholics

Much of the previous empirical literature report that Roman Catholicism, Orthodox, and Islam religions correlate negatively with transparency (high corruption or low CPI) (Paldam, 2001; La Porta et al., 1997; Chase, 2010). However, the negative relation between Roman Catholics adherents in the population and corruption (CPI) is not conclusive in my results. While Roman Catholics adherents are present with a negative yet not particularly significant coefficient in the regression model, they are absent in the regression tree. Treisman (2000) reported comparable results when including proxy measurements for democracy and economic development in the model similar to mine. Those factors are, as discussed, interrelated rather than competing explanations.

As stated in the theoretical part (Sects. 3.3 and 8.1), prosperity, democracy, and transparency might all be part of the same phenomenon. When regressing all those variables together, the model might dull the influence of other variables (e.g. Proportion of Roman Catholics). QCA further analyses this issue along with different variables.

3.2.3.3 Fractionalisation Measurements

No clear pattern exists in ethnic, linguistic, or religious fractionalisation. Only linguistic fractionalisation is significant in the regression analysis, but no fractionalisation measurement appears in the regression tree analysis. Fractionalisation indices express “how diverse” countries are regarding language, religion, and ethnicities. Further, the indices aggregate and “pack” such diversity into one figure (Alesina et al., 2003). Consequently, fractionalisation measurements prevent analysing the interactions of each specific religion, language, or ethnicity units using criterion (prosperity) variables. Such aggregation constitutes the main limitation of fractionalisation measurements. Therefore, the QCA section uses a disaggregated value to reflect the shares of the main ethnicities and languages spoken in a given country or region (Central Intelligence Agency of the United States [CIA], 2016).

3.2.3.4 Political Rights and Civil Liberties

As observed in the regression tree analysis, the political rights and civil liberties (PRCL) variable is the most significant factor relative to CPI. This finding was expected, on account of its theoretical relation with Protestantism (Sect. 8.2.2). Woodberry (2012) and Becker et al. (2016) have provided ample empirical and historical evidence for the essential role of Protestantism (i.e. Protestant missionaries) in the rise and spread of stable democracies around the world.

3.2.3.5 Gross Domestic Product (GDP) Per Capita

As explained, GDP per head was expected to be a significant predictor variable and is negatively related to corruption in the tree regression method.

3.3 General Conclusion: Models of Religious Population (Competitiveness and Corruption)

Generally, the models presented here suggest that Orthodox and Roman Catholic populations correlate negatively with competitiveness in Europe and the Americas. However, their role in corruption appears less evident in the corruption model. In contrast, a Protestant population might play an essential role in reducing corruption. Nevertheless, the role of the Protestant population in competitiveness is less clear. The next component (QCA) expands on the understanding of these variables along with other proxies.