Spillover Effects of Scandals on Exits from the Catholic and Protestant Churches in Germany

For decades, the annual numbers of exits of members from the Catholic Church and Protestant Church have been increasing. Closer inspection of diocese-level data shows that the time-series pattern of exits is nonlinear with a series of intermediate peaks. This paper uses an event study approach to model the impacts of various scandals on church exit rates. We find that sexual abuse and financial scandals located in the Catholic Church contribute to increased exit rates not just in that Church but also, by way of spillover effects, to higher exit rates of Protestant Church members.


Introduction
In Germany, the annual exit rate of members from the Catholic Church has been increasing for decades. This trend, however, is not linear, but is punctuated by several intermediate peaks that might have been caused by certain events or "shocks." Recent shocks like the public disclosure of abuse scandals might have driven exit rates. In Pennsylvania, thousands of cases of sexual abuse have been documented in the 40th Statewide Investigating Grand Jury report 1 and have received broad media coverage in the last decade. In Germany, the situation is not much different: Just recently, the Catholic Church in Germany published the results of a systematic investigation of sexual abuse cases. It revealed that 3,677 mostly male minors became victims of sexual abuse between 1946 and 2014. 1 In early 2010, a number of cases of sexual abuse of altar boys by Catholic priests became public for the first time. Since then, the Catholic Church has had to publicly deal with this scandal, the dimension of which still remains unknown.
However, the abuse scandals were not the only shocks hitting the Catholic Church as a number of financial scandals became public since 2014, when the bishop of the diocese of Limburg 2 JOURNAL FOR THE SCIENTIFIC STUDY OF RELIGION was found guilty of misallocating public funds of at least 31 million euros to renovate his residence. 2 Similar cases were revealed in other dioceses like Eichstätt, 3 where 50 million euros of public funds had been diverted in investments in real estate in the United States. Another case occurred in the diocese of Freiburg where the bishop's relevant staff ceased to pay social security contribution for the entire diocese's staff, the amount of which has been estimated to be as high as 160 million euros.
In this paper, we take such shocks as exogenous events and investigate their impact on member exits from the Catholic Church in Germany. While we are particularly interested in the effect the abuse and financial scandal have on exit rates, we also try to investigate these shocks in comparison to other events happening in the Catholic Church, such as the election of Popes or the announcement of encyclicals such as Humanae Vitae. Moreover, we also take into consideration political events like the fall of the Berlin Wall and the introduction of the solidarity tax or changes in the German tax system for reasons to be explained below. Further, we also analyze whether these shocks have spillover effects to the Protestant Church, resulting in higher Protestant exit rates as a response to the shocks occurring in the Catholic Church. The remainder of this paper is organized as follows: First, we present a brief review of the relevant previous literature. Second, we describe our theoretical and methodological framework, followed by descriptive statistics and econometric findings. We then discuss our findings. The paper ends with a conclusion.

Previous Literature
Economics of religion is a relatively young field (for reviews see, e.g., Gorski and Altinordu 2008;Hoffmann 2013;Iannaccone 1998;Iyer 2016). Nevertheless, various economic aspects have been investigated in different religious contexts already, for example, the role of religion for economic development, education, for labor market outcomes, 4 as well as for the political strategies pursued by different Popes. 5 As we focus on exogenous shocks in our analysis, we 2 https://www.hessenschau.de/gesellschaft/bistum-limburg-neue-geldverschwendung-bei-tebartz-wohnung,tebartzwohnung-umbau-100.html 3 http://www.spiegel.de/wirtschaft/soziales/bistum-eichstaett-verzockt-50-millionen-euro-in-usa-a-1191498.html 4 Guiso, Sapienza, and Zingales (2003) provide evidence that religious beliefs and convictions are likely to influence economic attitudes, which, in turn, lead to higher per capita income and economic growth. Especially Christian religions are found to be positively associated with attitudes conducive to economic growth (with similar findings 2010; Feess, Mueller, and Ruhnau 2014;Schaltegger and Torgler 2010). Basten and Betz (2013) find a significantly lower preference for leisure, redistribution, and government intervention among Protestants while Van Hoorn and Maseland (2013) find that Protestants suffer more from unemployment. Finally, Nunziata and Rocco (2018) document that Protestantism is associated with higher levels of entrepreneurship. However, Cantoni (2015) as well as Colvin and McCracken (2017) fail to find a statistically significant and/or economically relevant impact of a particular "Protestant ethic" on economic outcomes. Using county-level data from Prussia in the late 19th century, Becker and Woessmann (2009) find that Protestantism during that time was associated with higher rates of literacy (with similar findings for Switzerland; Boppart, Falkinger, and Grossmann 2014). In a companion paper, Becker and Woessmann (2010) analyze data from 300 Prussian counties in 1816 and find the number of schools as well as school enrollment to be significantly higher in (predominantly) Protestant regions. They interpret their findings as compatible with a human capital story instead of a distinct "Protestant work ethic." Using the German Socio-Economic Panel, Spenkuch (2017) finds a significantly positive impact of Protestantism on hours worked. Moreover, Sinnewe, Kortt, and Steen (2016) using the same data set identify a statistically significant pay premium for Catholics while Cornelissen and Jirjahn (2012)-again using the German Socio-Economic Panel (GSOEP)fail to find an effect of church membership on individual earnings. 5 For example, Barro and McCleary (2016) find that since 1590, the numbers beatified and canonized vary by Pope with the two Popes John Paul II and Benedict XVI being particularly active and that the number canonized rises with the stock of beatified, but not yet canonized as well as with Catholic-Protestant competition. This suggests that "leaders matter" in religious organizations at least as much as they do in profit-maximizing firms (see, e.g., Bandiera et al. 2011;Bennedsen, Perez-Gonzalez, and Wolfenzohn 2012;Goodall Kahn, and Oswald 2011;Goodall and Pogrebna 2015;Lazear et al. 2015). 3 restrict our review of the literature to two aspects of the economics of religion-the determinants of church attendance and of church membership, with the former showing a far greater decline than the latter in both Catholic and Protestant denominations.
The determinants of church attendance were first analyzed by Lazerwitz (1961) using data from the United States, who finds church attendance to be higher among women and upper class people. This holds for Protestants and Catholics as well as for Methodists and Baptists. Smith (1998) documents that "real" church attendance rates in the United States are significantly lower than reported rates due to a social desirability bias. Gruber and Hungerman (2008) find that in the United States, shopping malls opening on Sundays increases the opportunity costs of religious participation and, therefore, negatively affects church attendance (for similar findings, see Sullivan 1985). Becker and Woessmann (2017) find that the period of secularization in Germany that increased education ultimately led to a decline in church attendance.
So far, there are only a handful of studies that have investigated the impact of certain exogenous shocks on membership in the Catholic Church and these have almost exclusively used data from the United States. Hout and Greely (1987) document that in the United States, church attendance started to decline in 1968, the same year when Pope Paul VI published his highly controversial encyclical "Humanae Vitae," which strictly prohibits the use of contraceptives. Bottan and Perez-Truglia (2015) analyze the impact of abuse scandals on religious participation and religious beliefs in the United States in the period 1980 to 2010 when more than 3,000 scandals had been disclosed. Each scandal is associated with a persistent decline in church attendance as well as a decrease in church membership in the respective area. Analyzing similar data on abuse cases in the United States between 1971 and 2000 in the Catholic Church, Piazza and Jourdan (2018) find that over time the impact of sex abuse scandals on exit rates increased significantly at a local level. Hungerman (2013) finds that in the aftermath of sexual abuse scandals in the United States, non-Catholic faiths received significantly higher donations. Finally, Dills and Hernandez-Julian (2012) find that in the United States, enrollment in Catholic schools decreased significantly following disclosure of sexual abuse scandals. Lyytikäinen and Santavirta (2013) use data from Finland to investigate the relationship between church taxes and church membership. They find that church exits increase in the very last days of a year, as this is the last chance to avoid paying taxes for the next year. Using aggregate data for the Catholic and the Protestant Churches in Germany from 1953 to 2015, Frick and Simmons (2017) find that exogenous shocks like sexual abuse and financials scandals, German reunification as well as certain encyclicals by the reigning Pope all had statistically significant effects on exits in both denominations. These results are surprising insofar as encyclicals by the head of the Catholic Church are not binding for Protestants and should, therefore, be irrelevant to them.
With the exception of the latter study, the extant literature investigates only one event or "shock" but does not compare the effects of different events. Therefore, we include a variety of different (potential) determinants to explain the peaks of exit rates in certain years. In contrast to Frick and Simmons (2017) who use aggregated data from the entire Federal Republic of Germany for the Catholic and the Protestant Church, we analyze disaggregated data on member exits from 20 dioceses in each Church over an extended period of 60 years.
A large number of exits is problematic for Churches for two reasons: First, the financial situation of the church deteriorates as exiting members cease to pay taxes. This is a pertinent issue as the church is one of the important providers of health care, social services, and charitable activities (see Barth 2010: 780). Second, organizational legitimacy of the Church deteriorates too as exits reflect increasing levels of dissatisfaction and discontent among (previous) members. The existence, strength, and influence of the church as an organization strongly depends on the involvement of its members who can be considered its most valuable resource (see Iannaccone, Olson, and Stark 1995;Piazza and Jourdan 2018). The questions we ask in this paper are, therefore, as follows: Can we explain the observable variation in the annual exit rate of Catholics in 4 JOURNAL FOR THE SCIENTIFIC STUDY OF RELIGION Germany following sexual abuse and financial scandals? Which other shocks have affected exit rates? Finally, are there any spillover effects of these scandals to the Protestant Church?

Theoretical Framework
Assuming that rational individuals constantly compare the (expected) costs of and the (expected) returns to the activities they engage in and that these individuals withdraw from a particular activity as soon as the marginal costs exceed the marginal returns, we contribute to the discussion on the (potential) impact of various events (henceforth "exogenous shocks") on the annual number of exits from the Catholic and the Protestant Church in Germany in a particular way. Building on the influential work by Coleman (1990), we assume rational actors with resources linked by interests and control relations. Interests, that is, the motive governing actions, reflect a resource's impact on an actor's well-being while control refers to the rights to direct the use of resources. Moreover, actors are embedded in a specific social structure, that is, interdependencies, networks, organizations, and authority structures that limit them in pursuing their interests (e.g., Marsden 2005). Thus, the utility functions of individuals retaining and abandoning their church membership are different in the sense that their appraisal of costs and benefits of membership differs considerably.
From an economic perspective, social phenomena-such as church exits-are to be explained by showing how they result from individual actions, which, in turn, must be explained through reference to the intentional states motivating individual actors (e.g., Hechter and Kanazawa 1997). In this sense we explain the individuals' decision to either retain or abandon their church membership as a rational choice constrained by prices and incomes. As economists we continue to search for changes in relative prices or incomes to explain differences or changes in behavior. Although our data do not include any information on the actors' intentions, we derive from changes in their observable behavior that their individual cost-benefit relationship of church membership must have changed, most likely due to the occurrence of certain observable exogenous shocks.
Admittedly, the assumptions underlying rational choice theory have been and continue to be considered "reductionist" by many critics (Udehn 2002). They can, however, easily be justified by their analytical convenience, because the results of economic models typically remain robust even if some of these assumptions are relaxed. Economics in general and economics of religion in particular is not focused on studying individuals' mental life. Rather, it studies how changes in relevant parameters change individuals' behavior (and, eventually, collective outcomes). Thus, the attention is on how changes in the socioeconomic environment change the individual actors' choices. If the key explanatory variables change over time (due to, for example, one or more idiosyncratic events) and their consequences can be observed, we conclude that the changes have a causal impact on the observable outcomes. We conjecture that the explanatory power of economic models would not increase if we made our model more complicated by including additional parameters that might also be of interest (Ylikoski 2016).
Summarizing, utility-maximizing individuals join an organization when the expected returns to membership exceed the expected costs and eventually leave that organization when the expected costs rise above the expected benefits. However, membership in a particular religious denomination is usually passed over from parents to their children who usually do not decide themselves to join a church but are rather born into it. This suggests that the only decision that most individuals can make in their lifetime is to either remain a loyal member or to exit at some point in time-depending on how they perceive the relationship between costs and returns of continued membership. Exiting by officially leaving one of the mainline churches in Germany is a purely administrative move as one simply needs to request to have one's affiliation removed from the civil register. Once done, the individual no longer has to pay church taxes, which are SPILLOVER EFFECTS OF SCANDALS 5 collected by tax offices on behalf of the churches. In Germany, taxpayers, whether Catholic or Protestant, pay between 8 percent and 9 percent of their income tax to the church to which they belong. While the monetary costs of church membership have remained a constant proportion of income over time, the nonmonetary costs have certainly increased for many people due to a number of sexual abuse scandals that became public in early 2010 and the misuse of church taxes such as renovation cost of the former bishop of Limburg's residence that became public in early 2014. Moreover, the percentage of people attending religious services has declined over the last decades, suggesting that the benefits of church membership (such as access to the sacraments) are now less attractive than they used to be (Iyer 2016). Thus, there are different aspects that determine an individual's relationship with his/her denomination that we now describe in more detail in order to derive our hypotheses.

The Role of Scandals in the Catholic Church
Following disclosure of the sexual abuse scandals in the diocese of Regensburg in 2010, many of the leading representatives of the Catholic Church preferred to ignore the irritation and even horror that these events had caused among their members. Instead, many church officials started to neglect core aspects of their work, like the spiritual care and general welfare of the members of their church (Barth 2010: 784) in an attempt to prevent elucidation of the different scandals, to impede (presumably) "unjustified" critique and to avoid (further) reputational damages.
As individuals typically identify themselves with the social group(s) they belong to, they tend to experience their positive evaluation being challenged when a negative event -a scandaloccurs. As a result, their cost-benefit-relation of continued membership is likely to change (see Piazza and Jourdan 2018: 171). In case a particular scandal gains public attention, people may even decide they do not want to be associated with an organization with a deteriorating reputation and, therefore, leave that organization (see, e.g., O'Reilly and Chatman 1986; Piazza and Jourdan 2018; Pontikes, Negro, and Rao 2010;Pryor, Reeder, and Monroe 2012). In the present context, this means that as soon as a scandal-be it an abuse or a financial scandal-becomes public, members experiencing increasing (psychic or mental) costs of membership "resolve" their problem by deciding to exit from the Catholic Church (see Piazza and Jourdan 2018: 171). Thus, we hypothesize: H1: Scandals in both denominations, the Catholic as well as the Protestant Church, lead to increases in the annual exit rate.
In contrast to the election of a new Pope or the announcement of an encyclical by the incumbent Pope, which only affects Catholics, financial or abuse scandals might not be an idiosyncrasy of the Catholic Church. Thus, in contrast to the election of a new Pope, Protestants might react to scandals revealed in the Catholic Church similar to Catholics as they might assume that similar events might have simply gone undetected so far in their own denomination.

The Role of Leadership in the Catholic Church
The Catholic Church is a highly centralized institution with a pronounced hierarchical structure. Its head, the Pope, is elected by the College of Cardinals. Once elected, the Pope remains in office until he dies. The church's policy is mainly exercised by the Pope. As he also selects new members to the College of Cardinals, he also has quite some indirect influence on who will be elected as his successor. Wittman (2012) therefore argues that "electing a leader for life results in the electoral body electing older religious leaders than would be the case if the leader were elected for a term. This effect is magnified when the leader is an autocrat as there are then fewer ways to limit the damage from the leader" (Wittman 2012: 17). Thus, the Catholic Church can be considered an institution in which "leaders can become exceptionally powerful -indeed even totalitarian -because they are the embodiment of the organizational ideal" (Schwartz 1990, quoted in Barth 2010. Therefore, we hypothesize: H2: Given his peculiar role, the election of a new Pope has either a significantly positive or significantly negative impact on exit rates from the Catholic Church, depending on the public perception of the person as either "detached from the believers'reality" or "open to the people." As the Pope is the head of the Catholic Church, Protestants are not affected by this kind of change in the Catholic Church. Therefore, we do not expect to find an impact of the election of a new Pope on exit rates from the Protestant Church.
We further expect that the publication of Humanae Vitae in 1968 led to an increase in church exits. As the political and societal climate became more liberal in the mid 1960s in terms of sexual liberty and freedom, an encyclical strictly prohibiting the use of contraceptives was perceived a "fallback" of the Catholic Church, especially among young people. As the encyclical again affects Catholics only while being irrelevant for Protestants, we hypothesize: H3: The publication of Humanae Vitae had a positive impact on exits from the Catholic Church.

The Role of Political Changes and Tax Regimes
Further, we expect that political changes like the fall of the Berlin Wall in 1989 influenced exit rates, as members of both denominations in East Germany did not have to pay church taxes before reunification. However, as Gatz (2009) argues, this money was urgently required as the dioceses in the East of Germany had to be restructured following reunification. This additional financial burden might have induced people who struggled with their church membership anyway to leave the institution. As tax increases affect both the Catholic and the Protestant Churches, we expect reunification to have a positive impact on exit rates. Closely connected to the effects of German reunification is the introduction of the so-called "solidarity tax" (a tax introduced to finance the economic as well as the political reconstruction of East Germany) that was implemented in 1991 (see, e.g., Gatz 2009). We expect this to have a positive influence on exit rates, too, as people who are in doubt about their church membership-no matter if they are Catholic or Protestant-might decide against it, as they now have to pay another tax on top of their church tax. Moreover, the church tax that each member has to pay is directly proportional to the individual's income. Thus, with increases in the marginal tax rate, we expect more members of both Churches to leave as their church taxes also increase.

H4: The reunification of Germany, the introduction of the solidarity tax, as well as an increase in minimum/maximum tax rates lead to an increase in church exits in both denominations.
Before we proceed to the empirical tests, we first describe our data and variables, followed by the descriptive statistics and the econometric analysis. It is important at this point to mention that we investigate the impact of exogenous shocks on an individual's church membership, but not on the strength of his or her religious commitments, as we cannot observe the latter in our data.

Data
Our data include information on members, exits, church attendance, parishes, and priests for all 27 Catholic dioceses in Germany for the years 1946 to 2017 (Aachen, Augsburg, Bamberg, Berlin, Dresden-Meißen, Eichstätt, Erfurt, Essen, Freiburg, Fulda, Görlitz, Hamburg, Hildesheim, SPILLOVER EFFECTS OF SCANDALS 7 Köln, Limburg, Magdeburg, Mainz, München und Freising, Münster, Osnabrück, Paderborn, Passau, Regensburg, Rottenburg-Stuttgart, Speyer, Trier, and Würzburg). To avoid biases resulting from new dioceses, from reunification, and from mergers of already existing dioceses, we restrict our analyses to the 20 dioceses that have been in existence since 1946 already and exclude Berlin, Görlitz, Magdeburg, Erfurt, Dresden-Meißen, Hamburg, and Essen (see Figure A4 in the Online Appendix displaying the regional disribution of the 27 dioceses). Due to the fact that we include data on tax rates, which are available only from 1958 on, we had to further restrict our data set to the years 1958-2017. Thus, our data are a balanced panel with 1,200 diocese-year-observations coming from two sources: the "Kirchliches Handbuch", an annual collection of statistics of all German dioceses on different matters, and from the website of the German Bishops' Conference (https://www.dbk.de). The data on exits from the Protestant Church come from the main website of the denomination (https://www.ekd.de) also covering the period 1958-2017 (see Figure S5 in the Online Appendix for the geography of the administrative units of the Protestant Church).

Variables
The dependent variable in our econometric analysis is the annual exit rate, that is the number of exits per 1,000 members and year by diocese ( Figure A1 displays Kernel density estimations of the annual exit rates from the Catholic and the Protestant Churches).
The independent variables are, first, two "scandal dummies" that we consider potential determinants of the annual exit rate. To account for the impact of these scandals, we include in our estimations public disclosure of a series of sexual abuse scandals in 2010 (dummy = 1 if year = 2010) and a dummy for public disclosure of the financial scandal in the diocese of Limburg (dummy = 1 if year = 2014). As we want to analyze the effect of the scandals on exit rates over time, we generate a linear time trend variable for the abuse scandal in 2010 and for the financial scandal in 2014, which includes four years in case of the abuse scandal and three years in case of the financial scandal. Thus the abuse × time trend variable increases from 1 to 4 over the years 2011, 2012, and 2013 and 2014 while the finance × time trend variable increases from 1 to 3 over the years 2015, 2016, and 2017. This allows us to distinguish between one-shot effects in the year of the scandal and longlasting effects over time as discontent and disillusionment may increase among Church membership. This is to be expected if members get disenchanted with the responses of the Church to the scandal and therefore cancel their membership some time after the initial scandal.
Second, we include a series of dummies to reflect the potential impact of changes in the leadership of the Catholic Church. These are the  6 We further take into account German reunification in 1990 (dummy = 1 if year = 1990 and 1991) as well as the introduction of the solidarity tax. Finally, we also include variables reflecting changes in tax regimes by using the minimum and the maximum tax rate as potential determinants of exit rates ( Figure A2 displays the development of the German tax regime over time) (see Table 1 for descriptive statistics).  Note: Due to the long observation period of 60 years, the variation of the annual exit rate within dioceses/within regional churches is considerably larger than the variation between the dioceses/regional churches.  Figures 1, 2, and 3 show that the pattern of exits is quite similar across the dioceses with considerable differences in the magnitude: Exits are rather low in the dioceses of Passau, Regensburg, Trier, Würzburg, Münster, and Eichstätt and rather high in Osnabrück, Köln, München, Mainz, Limburg, Hildesheim, and Rottenburg-Stuttgart (the remaining dioceses-Augsburg, Aachen, Bamberg, Paderborn, Freiburg, Speyer, and Fulda-range somewhere in between).

Descriptive Evidence on Exit Patterns
It also appears from Figures 1, 2, and 3 that there are several peaks. Around 1990, the number of exits increased considerably (most likely due to reunification and the ensuing increase in taxes). Perhaps surprisingly, exits from the Catholic and the corresponding regional Protestant Church are highly correlated in periods surrounding specific scandals located in the Catholic Church, suggesting that considerable spillover effects may exist, although neither the abuse and the financial scandal nor the election of a Pope has any impact on the lives of Protestants (see Figue A3).

Econometric Findings
We estimate two different fixed effects models (one for the Catholic and one for the Protestant Church) and four Arellano-Bond Models (two for each denomination, with different lag lengths). Explanatory dummy variables are included to represent: (1) abuse and financial scandals, (2) elections of a new Pope and controversial declarations by the incumbent Pope, (3) political developments, and (4) changes in tax regimes.
Our prior view, to be tested, is that the dependent variable for a diocese in a given denomination, church exit rate, will depend in part on values of exit rate from previous years. We conjecture that this is due to habit persistence in church membership in each denomination. If habit persistence is important but is left out of our fixed effects models, then we have potential for omitted variable bias where the excluded lagged dependent variables are correlated with the error term. If we simply add lagged dependent variables to the fixed effects models, then this creates econometrically inconsistent estimates where the lagged dependent variables are endogenous.
The standard resolution to these econometric problems is to apply the Arellano-Bond estimator for dynamic panel data (see Cameron and Trivedi 2009: 287-295). This uses further lags of the dependent variable as instruments for the endogenous lagged dependent variable(s). From pretesting, we set the maximum lag length of the dependent variable at two and the maximum lag length of the dependent variable in the instrument set at four. The Arellano-Bond estimator offers consistent and more efficient estimation over OLS fixed effects models and is hence preferred. In our estimates of church exit rate models, we find mostly qualitatively similar results for effects of independent variables. But we also find some specific substantive differences in effects, which we highlight below.
Beginning with results where Arellano-Bond and OLS fixed effects estimates are similar, it appears from Table 2, Model (1.1) that public disclosure of the sexual abuse scandal in 2010 had an immediate positive, statistically significant, and economically relevant impact on exits from the Catholic Church. Moreover, this scandal affected church exits for a longer period of time, as the coefficient of the time trend interaction terms is also positive and highly significant.
Model (1.2) shows that the effects of the sexual abuse scandals were different in the Protestant Church. First, the sexual abuse scandal in the Catholic Church caused an initial statistically significant decline in exits from the Protestant Church, most likely because many protestants considered such an event (almost) impossible to occur in their church as protestant priests are allowed to marry and have children themselves. This negative effect, however, was short lived and was followed by an increase in Protestant exit rates in the subsequent three years as shown by the coefficient on the abuse-time trend interaction term. This indicates a delayed response of Protestant exits to the Catholic abuse scandal, perhaps as Protestant Church members took time to consider their positions.
With respect to changes in leadership in the Catholic Church, it appears that the election of both Pope John in 1958 and Pope Paul in 1963 had highly significant and positive impacts on exits, implying that both were not appreciated by many German catholics. The elections of the two succeeding Popes, John Paul in 1978 and Benedict in 2005, had the opposite effect, although both were also considered very conservative and reluctant to respond to claims for change with respect to, for example, the role of women in the Catholic Church. However, contrary to their two predecessors-and perhaps due to their personal characteristics (extensive travel activities in the case of Pope John Paul and German nationality in the case of Pope Benedict)-they were both considered "men of the people." The coefficients of the respective dummies indicate a highly significant negative effect on annual exit rates, supporting the conjecture that both persons were at the time-most likely for different reasons-appreciated by the German public. Perhaps surprisingly, the coefficient of the dummy variable indicating publication of Humanae vitae is negative and statistically significant, suggesting that the highly controversial publication was-contrary to our expectations-associated with a decrease in the annual exit rate from the Catholic church. We now turn to the results of the Arellano-Bond estimations (see Table 3). For Catholic exit rates, dependent variable lag lengths of one and two each have statistically significant coefficients supporting our prior of habit persistence in church membership. Given the choice of two years lag length shown in our preferred Model (3.1), the sexual abuse scandal had a statistically significant positive effect on church exits among Catholics. This effect is reinforced by the interaction of abuse scandal and time trend, suggesting that the way the Catholic Church handled the scandals induced many members to leave the church in the years following public disclosure of the first scandal as a delayed response.
For Protestant exits, only the one-year lag length of dependent variable has a significant coefficient, not the two-year lag, so Model (2.2) is preferred over Model (3.2) and also preferred to Fixed Effects Model (1.2). From the estimates of Model (2.2), we find a significant negative effect of public disclosure of sexual abuse scandals in the Catholic Church on Protestant exit rate. In years after the scandal, however, the Protestant Church too lost members due to the abuse scandal in the Catholic Church. The coefficient on time trend interaction term is larger than the coefficient on abuse scandal dummy variable such that the net effect of the scandals on Protestant exit rates is positive from one year after news of the scandal broke. 13 With respect to the finance scandal, the Arellano-Bond results show insignificant effects on exits from the Catholic Church. Thus, responses of Catholic Church members are very different according to the nature of reported scandal with sexual abuse scandals specifically incurring lack of favor by members. For the Catholic denomination, the effects of the finance scandal vary as between fixed effects and Arellano-Bond models but the latter are preferred following the econometric arguments made above. Among Protestants, however, the finance scandal had a statistically significant immediate as well as a long-term positive effect on the annual exit rate. Thus, the Protestant Church has apparently suffered in terms of losses of members from a scandal that occurred in the "competing" denomination.
From Model (2.2), it appears that some Pope elections have significant effects on membership in the Protestant Church. The reported effects differ as between fixed effects and Arellano-Bond models but, as argued above, we prefer the latter estimates. Specifically, we find statistically significant negative coefficients for the elections of Pope Paul and Pope John Paul though not for Pope John or Pope Benedict. This is surprising insofar as these personnel changes are of no relevance to members of the Protestant Church. However, in personal communications, the importance of spillover effects from the Catholic to the Protestant Church has been emphasized by leading representatives of both denominations.
Looking at the effects of the different changes in leadership in the Catholic Church from the Arellano-Bond estimates, considerable differences occur between the two denominations: First, while the elections of Pope John and Pope Paul increased the annual number of exits from the Catholic Church, these elections had zero (Pope John) or statistically significant negative (Pope Paul) effects on Protestant exits. The election of Pope John Paul reduced the exit rates in both denominations significantly, with the effect being slightly larger among Protestants. The election of Pope Benedict in 2005-the first Pope to resign voluntarily from office-was apparently highly welcomed by German Catholics as the exit rate declined significantly in the year of election as well as the subsequent year. Among Protestants, however, the latter election had no discernible impact on exit behavior. According to Table 3, Models (3.1) and (2.2), changes in the German tax regime have the expected impacts on exits from each church. While a lower minimum tax rate is associated with a decrease in exits, a higher maximum tax rate is associated with rising exit rates. Moreover, the coefficient of the solidarity tax variable has a statistically significant and positive effect on exit rates too, again in each church. German reunification significantly increased exit rates not only in East Germany, where people did not have to pay church taxes before reunification, but also in the Western part of the country. The estimates from Model (2.2) show that the tax effects are significantly stronger in the Protestant Church. According to the Arellano-Bond estimates, the reunification of Germany was associated with increases in exits from both churches. Here, the Arellano-Bond results are different from the fixed effects estimates in Model (1.2) above. In the latter model, there is a significant negative effect of German reunification on Protestant exit rate. On econometric grounds stated above, we prefer the Arellano-Bond estimates. These suggest that reunification led to rising exit rates from the Protestant Church.
Another source of difference between the fixed effects and Arellano-Bond models lies in consideration of the Humanae Vitae dummy variable for Protestant exits. The effect of publication of Humanae Vitae, directed of course at Catholics, appears to be negative in the fixed effects model for Protestant exits yet positive for the Arellano-Bond model, contrary to the finding for Catholic exits. Again we prefer the Arellano-Bond estimates but defer explanation for the intriguingly opposite results for the two denominations to further research.
Finally, the coefficient of the linear time trend is statistically significant and positive in Model (3.1) as well as in Model (2.2), suggesting that religious attachment has been gradually declining in both churches over the last 60 years (the linear time trend is strongly correlated with decreasing church attendance). 7 The trend impact is similar in magnitude for each church.

Conclusion
Summarizing, our estimations show that the various kinds of exogenous shocks that we distinguish here all have statistically significant and economically relevant impacts on the decision of individuals to either leave their church or retain their membership. First and most important, our preferred estimates show that the recent sexual abuse scandals had a significantly negative immediate as well as long-term impact, in the sense that, they increased the annual number of exits from the Catholic Church. Sexual abuse scandals in the Catholic Church also raised exit rates in the Protestant Church with a delay. In contrast, we find zero effects of the financial scandal on exits from the Catholic Church while there are positive impacts from these scandals on exits from the Protestant Church. Thus, especially for the financial scandal, we find a spillover effect from the Catholic Church onto the Protestant Church. Second, we find that the most recent elections of a new Pope had a positive impact on church membership as they reduced the annual number of exits among Catholics. Third, it is also striking that changes in the tax regime as well as the solidarity tax lead to significant increases in exit rates, where the magnitudes are somewhat greater for the Protestant Church. Finally, declining religious attachment, that is, secularization, has a statistically significant impact as exit rates increase over time to a similar extent in each church. Thus, our research questions can be answered generally in the affirmative-the shocks that we have identified can explain large parts of the observable variation in the annual exit rate in both denominations, with changes in taxes having the largest impact on people to give up their denomination. With regard to spillover effects, we mainly observe that the financial scandal in the Catholic Church had an impact on exit rates in the Protestant Church. This suggests that a scandal located in the Catholic Church had resonance as potential for malfeasance within the Protestant Church.
The exit patterns revealed here suggest that the leading representatives of the Catholic Church have done a rather poor job in dealing with the two scandals. While the bishop of Limburg was demoted from office only after a highly emotional public debate, the victims of the various sexual abuse scandals are still waiting for a conviction of their abusers. Thus, lack of transparency combined with a poor crisis management has been very costly to the Catholic Church not only in terms of loss of credibility, but also in terms of loss of financial resources.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.