Incentives, predestination and free will.

Glaeser, Edward L. ; Glendon, Spencer

I. INTRODUCTION

Max Weber's [1930] "The Protestant Ethic and the Spirit of Capitalism" argues that the Protestant reformation set the stage for post-Medieval economic growth. Using an occupational study of Germans from 1895, Weber shows that Protestants dominate the mercantile or capitalist occupations in that country at that time period. Weber also argues that between 1600 and 1900, Protestant countries advanced faster than their Catholic neighbors and that this was the result of Calvin's urging acceptance of material success and the dogma of predestination.(1)

An ocean of work beginning with Rachfahl [1909] and Sombart [1913] has since attacked Weber's theological generalizations (see Giddens, [1992], for a review). Tawney [1926] and Robertson [1933] argue Weber's causal interpretation is wrong and that economic growth laid the framework for Protestantism, not Protestantism for economic growth. Samuelsson [1961] disputes the very correlation between Protestantism and economic success which is the basis of the Weberian hypothesis.

The correlation between Protestantism and economic success is certainly weaker than Weber suggests, even in the German data that Weber himself uses. Furthermore, more modem data tends to find only a modest link between Protestantism and economic success, as in Winter [1974] and Tomes [1994].(2) However, for centuries there were gaps in economic success between Protestant and Catholic nations, and new findings on early modem data tend to confirm differences between Protestants and Catholics at least within the U.S. As Weber's arch-critic Samuelsson himself writes "there is some truth in the contention that the Protestant countries, and especially those adhering to the Reformed church, were particularly vigorous economically."

Given that there was some correlation between Protestantism and economic success before the mid to late 20th century, the question then becomes why this correlation existed. Some elements of Protestant theology clearly encourage capitalism. Usury doctrines became less rigid under Calvinism (Nelson [1972]); avarice was de-emphasized as a sin. Weber claims that Luther's concept of a "calling" includes worldly success and that this is a prominent difference with Catholic doctrine. Although this theological claim has been subject to much debate, e.g. Sombart [1915], there still appears to be some basic truth to this point.

One of the largest theological gaps between the denominations is that Calvinism accepts the dogma of predestination while Catholicism argues for a dogma of free will.(3) For the purposes of this paper we will follow Weber and accept broad brush definitions of doctrine rather than getting mired in the more subtle features of the two doctrines. Under predestination, a spiritual elite is preordained before the beginning of time and will receive eternal life.(4) Under free will, it is only through a lifetime of good actions that individuals are accepted into Heaven.(5)

Weber argues that predestination creates a more lasting basis for economic growth than the dogma of free will. In this paper, we present a model that formally treats Weber's claim under particular circumstances and show how predestination could create the correlation between Protestantism and growth that existed in the pre-modern era. Our model also suggests conditions under which a more successful society would favor the doctrine of predestination, which can be seen as a formalization of Tawney's and Robertson's view that economic growth created the reformation.

In this essay, both free will and predestination create incentives which spur good behavior.(6) Under the doctrine of free will, good behavior is motivated by an increased probability of getting into heaven. Under the doctrine of predestination, this incentive is irrelevant because individual actions do not determine spiritual outcomes. Instead an individual's actions are motivated by a desire to receive positive behavior from others by convincing them that he is part of the spiritual elite. Examples of this positive behavior could include everything from respect or generosity to basic civil liberties (individuals who were believed to be among the damned were brought under the law in 17th century Calvinist Switzerland).(7)

The formalization yields several results. Under free will, an individual's level of "moral" output is determined by the costs of this output and by belief in divine rewards. Under predestination, an individual's moral output is a function of costs and society-wide beliefs, not his own beliefs. Thus, individual behavior will be much more homogeneous under predestination than under free will and there will be a greater connection between an individual's behavior and the average behavior of his community. Furthermore, predestination will bias behavior towards observable moral actions and away from unobservable actions. A decline in belief among subsections of the population may be much less harmful under predestination (since such declines will not lead to declines in moral behavior), so the advantages of Calvinism will be particularly strong during periods of declining faith in the hereafter.

We also present a second model where a further advantage of Calvinism is that moral behavior provides a more accurate signal of an individual's costs of morality. Religious systems where outward religious behavior signals economically valuable information will help to generate efficient matches in society. Under free will, moral behavior is a function of beliefs in the hereafter and an individual's costs of behaving ethically. Under predestination, moral behavior is a function of ethical costs alone. If ethical costs are relevant for non-religious relationships, but beliefs in the hereafter are not, then predestination will yield better matches between individuals.

[TABULAR DATA FOR TABLE I OMITTED]

We test our model by examining an implication for behavior in late 20th century America. Unfortunately Calvinist denominations no longer believe in predestination as a cornerstone of their faith, so our tests do not capture a direct effect of predestination. Instead, we believe that looking at modern data has some value because current social norms may still be the legacy of prior religious beliefs. At the least, one would think that the temporal distance between our data and the time period when predestination was a critical part of Protestant life, should bias our results towards zero and lead us against finding support for the model.

Instead, our findings do generally support the model. Using data from the General Social Survey since the 1970s, we document a stronger connection between economic success and religiosity among Protestants than among Catholics. We also find that there is a greater connection between individual behavior and community behavior among Protestants than among Catholics. Both findings suggestively support our theory and Weber's basic hypotheses.

This paper should not be seen as wholesale attempt to resurrect Weber's viewpoint. We accept much of the work debunking Weber, and we believe that much more intense empirical work is needed before we truly understand the effects of religious doctrine on individual behavior. However, we also believe that there is more truth to Weber than many observers now acknowledge.

II. MOTIVATIONAL FACTS

This section is meant to motivate the following models by stressing that economic growth was faster in Protestant countries before 1950. Table I uses income per capita figures from Maddison [1982] and separates the sample of countries available from 1850 onward into predominantly Catholic and predominantly Protestant countries. We define Protestant countries as those with more than 50% of the population being Protestant in 1995, and Catholic countries as those with more than 50% of their population being Catholic at that date.(8)

When we split the time series into four natural time periods, the Protestant countries have a faster annual rate of income growth per capita during each of the three time periods before 1950. The differences between 1820 and 1950 are all statistically significant at the 10% level. After 1950, the growth rates are equal. We also find that the average income level is always significantly higher in Protestant than in Catholic countries. Expanding the sample when it becomes available does not change the results.

Micro-evidence on Protestant-Catholic differences also exists. As Tomes [1994] documents, [TABULAR DATA FOR TABLE II OMITTED] this difference has faded in the modem era, but during earlier time periods the difference is quite real. Despite the apt criticisms of Samuelsson [1961], the 1895 German data still confirms some differences between Protestants and Catholics. Furthermore, Goldin and Katz [1997] use the 1915 Iowa census to document a .11 log point percent difference between Catholics and Protestants in earnings, even controlling for age, immigrant status and years in the country. This difference is primarily the result of different schooling levels, as it contracts to a .04 log point difference when they control for years of schooling. There is certainly a correlation between denomination and economic welfare during this earlier period.

In table II, using the data from the World Values Survey, we follow La Porta et al. [1997] and examine the connection between Catholicism and a variety of country attributes, controlling for income per capita. We include only the 25 countries in the survey that are predominantly Christian. The definition of Catholicism in this table is the same as above. The first regression documents that there is still a gap between Protestant and Catholic countries in secondary school completion rates suggesting that the catch-up between denominations is still imperfect.

The correlation between Protestantism and economic success may have waned over the 20th century because Protestantism only sped up adoption of capitalist methods and eventually Catholic countries also adopted these technologies.(9) Alternatively, the secular decline in religiosity (particularly outside the U.S.) or in doctrinal differences between denominations may have led to the elimination of Catholic-Protestant differences. Overall, though, we believe that there has been more than enough difference between Protestant and Catholic economic success over the past four centuries to warrant continuing interest in why Protestantism was associated with economic growth.

The next three regressions are not related to the factual question as to whether Catholicism really does hinder growth. Instead, they examine social behavior and Catholicism. We connect Protestantism with membership in either professional (regression 2) or civic (regression 3) organizations, controlling for GDP per capita and schooling. In both cases, we find that individuals in Protestant countries are more likely to be members of these social groups. The final regression shows that individuals in Catholic countries are much less likely to answer yes to the question "Generally speaking would you say that most people can be trusted or that you can't be too careful in dealing with people?" This question is one index of the level of social capital in the community, as in Putnam [1996].

Our results do not mean that Protestant countries are more communal or have greater social connection generally. Instead, these findings should be taken to mean that Protestantism appears to be associated with highly visible, low intensity social interactions that rely on trust of strangers. Even if Catholicism was associated with highly communal, intimate relationships of long time duration, it could still be true that Protestantism is associated with formal, visible social associations that are much less intense, but possibly more important for economic growth. The model that we present will focus on the tendency of predestination to induce visible displays and these regressions help in suggesting that certain types of visible social activity are more common in Protestant countries (these findings are similar to those of La Porta et al. [1997] who document a connection between trust and hierarchical religions generally).

III. THE BASIC MODEL

Each individual, who is indexed i, chooses observable and unobservable externality-producing actions denoted [A.sub.1](i) and [A.sub.2](i) respectively, which are produced at a cost C([A.sub.1], [A.sub.2]). These moral actions are meant to include any positive externality producing deeds which increase other people's welfare: conventional good deeds, economic actions that lead to technological change or growth, or just the absence of theft, murder and arson.(10) These actions are assumed to bring individuals themselves no private material benefit, but they are assumed to benefit all of society. Individual j's utility from individual i's actions is denoted V([[Alpha].sub.1][A.sub.1](i) + [[Alpha].sub.2][A.sub.2](i)), where [[Alpha].sub.1] and [[Alpha].sub.2] reflect the utility weights on observed and unobserved actions respectively.

Thus the total welfare from other people's actions is

[integral over i] V([[Alpha].sub.1](i) + [[Alpha].sub.2](i))di

for each person. The social planner will be concerned with maximizing the externality producing actions or equivalently maximizing E(V([[Alpha].sub.1][A.sub.1](i) + [[Alpha].sub.2][A.sub.2](i))), where E2([center dot]) represents an expectation that is taken over i. The next sections show how this quantity differs between religious systems based on free will and religious systems based on predestination.

Under the doctrine of free will, moral effort is expended to ensure good behavior in the hereafter. Under free will individuals will maximize [Delta]B([A.sub.1], [A.sub.2]) -C([A.sub.1], [A.sub.2]), where [Delta] is the individual specific belief in heaven and B(.,.) represents the benefits in the hereafter as a function of moral actions. The first order conditions are:

(1) [Delta] [Delta]B([A.sub.1], [A.sub.2])/[Delta][A.sub.1] = [Delta]C([A.sub.1], [A.sub.2])/[Delta][A.sub.1]

and

[Delta] [Delta]B([A.sub.1], [A.sub.2])/[Delta][A.sub.2] = [Delta]C([A.sub.1], [A.sub.2])/[Delta][A.sub.2]

and we assume that B(.,.) is strictly concave. To get closed form solutions, and further comparative statics, we will assume that B([A.sub.1], [(A.sub.2]) = [B.sub.1][A.sub.1]+[[Beta].sub.2][A.sub.2] and [Mathematical Expression Omitted]). With this assumption, an individual's levels of moral actions are:

(2) [A.sub.1] = [Delta][[Beta].sub.1][C.sub.2] - [[Beta].sub.2]R/[C.sub.1][C.sub.2] - [R.sup.2]

and

[A.sub.2] = [Delta][[Beta].sub.2][C.sub.1] - [[Beta].sub.1]R/[C.sub.1][C.sub.2] - [R.sup.2].

Levels of moral actions are declining in own costs, rising in the level of beliefs and their own benefits, and with the costs of the other actions if the goods are substitutes in production (technically as long as [[Beta].sub.1][C.sub.2] [greater than] [[Beta].sub.2]R). We assume that none of these parameters, except for belief in the hereafter differ across the population. Social welfare is:

(3) E(V([[Alpha].sub.1][A.sub.1] (i) + [[Alpha].sub.2][A.sub.2](i)))

= E (V([Delta](i) [[Alpha].sub.1][[Beta].sub.1][C.sub.2] + [[Alpha].sub.2][[Beta].sub.2][C.sub.1] - R([[Alpha].sub.1][[Beta].sub.2] + [[Alpha].sub.2][[Beta].sub.1])/[C.sub.1][C.sub.2] - [R.sup.2]))

Social welfare is rising in the level of belief and declining in the variance of beliefs (by Jensen's inequality) as long as V([center dot]) is concave. Our social welfare function does not directly reflect religiosity, but since religion motivates good social behavior, stronger beliefs raise social welfare. Lower costs of good behavior and higher private benefits from moral actions raise social welfare except in certain strange cases which rely on the goods being substitutes in production.

When one action is not valued much in the social welfare function and when the actions are strong substitutes in production (technically, when [[Alpha].sub.1][C.sub.2] [less than] [[Alpha].sub.2]R), then if the first good becomes less attractive or more costly, this change will create so much more of the other actions, that the extra social welfare will make up for the losses from the decrease in the first action. Thus, society might benefit when a particular moral action becomes less attractive, which might occur if highly ritualistic, non-productive religiosity falls out of favor and thus leave more time for socially productive moral actions.

Under the doctrine of predestination, taking moral actions does not effect benefits received in the hereafter. The only incentive to undertake moral actions is that other people change their attitudes towards you if they observe you doing good deeds.(11) When individuals are believed to be members of the spiritual elite by other members of society then those individuals are given social advantages. These advantages can range from respect and authority, to material gifts, to freedom from imprisonment. Often in Calvinist societies ranging from Geneva to the Massachusetts Bay Colony, the damned (i.e. the non-elite) were treated with imprisonment and corporal punishment, as in Erikson [1966].

We assume that there is the same social welfare function, but under predestination individuals maximize

[integral over i] G([A.sub.1], [Delta](i))di,

where G([A.sub.1], [Delta](i)) reflects benefits that individual j receives from individual i based on the religious beliefs of individual i (again denoted [Delta](i)) and his own observable moral actions [[Alpha].sub.1].(12) We will also assume G([A.sub.1], [Delta](i)) = g[A.sub.1][Delta](i). Individuals will perform no unobservable moral actions, which might explain the allegations of hypocrisy often leveled against Calvinists.(13)

Agents maximize the benefits of morality minus the cost of the observed moral action, or

[Mathematical Expression Omitted],

which yields first order condition:

(4) [integral over i] [[G.sub.A].sub.i] ([A.sub.1], [Delta](i))di = [C.sub.1][A.sub.1]

or [Mathematical Expression Omitted]

where [Mathematical Expression Omitted]

Social welfare is [Mathematical Expression Omitted],

which is rising in the returns to good behavior and the average quantity of belief and falling in the cost of good behavior. Individual behavior will be strongly correlated with the behavior of others in the community (in fact under these assumption all individuals will behave identically), which will not be true under the doctrine of free will.

The Calvinist predestination system yields superior social outcomes to the Catholic system when:

[Mathematical Expression Omitted].

When unobservable actions are less important to social welfare or when more actions are observable (both of which are reflected in a lower value of [[Alpha].sub.2]), the predestination system becomes more appealing. Perhaps this result explains why Calvinism may have had more appeal in urbanized societies where monitoring was easier and more actions were observable. If public actions yield particularly valuable externalities then Calvinism may encourage more of these public actions. If the world is better off with less private morality (since private morality crowds out public morality) then Calvinism will be good because it discourages private morality.

While the overall level of belief affects both sides of the equation, diversity of beliefs hurts only the free will society. This means that diversity of ideas will be much more costly in a Catholic free will society than a Calvinist society. This may explain why Calvinism appears to have encouraged thought and rationality more than Catholicism. Furthermore, this explains why Calvinism may have been adopted by societies with more individuals who had pronounced doubt about divine providence.

Finally, Calvinism will be more appealing as v([center dot]) becomes more concave. The diversity of free will societies creates heterogeneous behavior. This heterogeneity is desirable if v([center dot]) is convex but costly when V([center dot]) is concave. Such concavity can be interpreted as meaning that there are extremely high social costs of extremely low levels of moral actions, but that moderate shirking in this regard is relatively costless from a societal level. If this is true, then Calvinistic societies will avoid extremes of morality and immorality and the homogeneity will be socially desirable.

IV. THE TWO DOCTRINES WITH SIGNALING

In this section, we follow Weber [1930] and examine the effects of predestination on the organization of the economy. In particular, we assume that moral behavior has a signaling use beyond its other values. For simplicity, we now assume that there is only an observable moral action (now denoted A, dropping the subscript) and that the costs of that action differ across individuals. Prospective employers, lenders, spouses and friends are interested in knowing an individual's cost of morality, because it relates to productivity in non-religious areas. For example, individuals with a low cost of ethical behavior may be less likely to steal from their employers or cheat on their wives.

We now assume that there are four types of individuals: high belief individuals with high costs of morality, high belief individuals with low costs of morality, low belief individuals with high costs of morality and low belief individuals with low costs of morality. The shares of these groups in the population are denoted HH, HL, LH, and LL, respectively. High cost of morality individuals face costs [Mathematical Expression Omitted] of the moral action; low cost of morality individuals face costs [Mathematical Expression Omitted] for the moral action. High belief individuals have beliefs [Mathematical Expression Omitted]; low belief individuals have beliefs [Mathematical Expression Omitted].

There are two occupations in the economy. The first, no responsibility occupation always yields productivity 0. The second occupation yields productivity, P, if the worker acts ethically and productivity -X if the worker does not act ethically. We assume that all low cost of ethical behavior individuals will act ethically and all high cost of ethical behavior individuals will not.

We do not allow contingent wages, and free entry ensures that firms will pay wages equal to expected productivity so that wages in the first occupation will always equal zero. Wages in the second occupation will equal zP - (1 - z)X, where z is the fraction of applicants for the second occupation who have high costs of moral actions. Efficient matching will lead to all low morality cost individuals performing the high responsibility occupation and all high morality cost individuals performing the no responsibility occupation.

In the world where predestination is the source of moral incentives, following equation (4), if agents do not try to imitate each other, they will choose their actions to maximize social approval, so [Mathematical Expression Omitted], which implies that only the cost of morality will determine the level of moral actions. There are possible suboptimal equilibria, but in this section and in the next section, we constrain ourselves to first best equilibria for the particular set of parameter values. High moral cost individuals may also perform more moral actions to try and imitate low moral cost individuals, but we assume parameter values so that this does not occur and individual signal their types perfectly through their choice of actions.(14)

Under free will, there are several possible scenarios. Whenever job related incentives are not strong enough to induce individuals attempt to imitate one another, then following equation (2), A(i) = [Delta](i)[Beta]/C(i). The first best equilibrium only exists as long as [Mathematical Expression Omitted] and P is sufficiently small. However, as long as there is sufficient heterogeneity of beliefs in the population, free will does not yield the first best outcome.

When free will does not yield optimal sorting, several possible equilibria exist depending on parameter values. As an example, we assume that [Mathematical Expression Omitted] so that in the absence of imitation, high belief, high cost individuals will perform the same amount of moral actions as low belief, low cost individuals. Again assuming individuals do not imitate, they will choose their moral actions so that A(i) = [Delta](i)[Beta]/C(i). High belief, high cost individuals and low belief, low cost individuals will choose [Mathematical Expression Omitted], and will receive wages of

[[Pi].sub.LL] P - [[Pi].sub.HH]X/[[Pi].sub.LL] + [[Pi].sub.HH]

which we assume is greater than zero.(15) For this no imitation equilibrium to exist, we must assume that no one wants to imitate high belief, low cost individuals to receive their wage of P and low belief, high cost individuals do not want to imitate low belief, low cost individuals.(16)

In this case, there are social losses because high belief, high cost types will also be given positions of responsibility which is a misallocation. Moral actions serve as a more efficient signal when they only reflect the costs of being responsible, not when they also reflect beliefs about the hereafter which will not necessarily transfer into responsibility in other domains.

This model closely follows Weber [1930]. Calvinist societies will be more efficient because moral behavior is a signal of traits that are desirable in many areas, while Catholic societies will be less efficient because moral behavior also reflects beliefs that are not important for non-religious activities. The model also predicts a stronger association between material success and moral behavior in Calvinist societies.

V. EMPIRICAL EVIDENCE

This section tests two primary implications of the model using modern American data from the General Social Survey (GSS). The GSS has been administered to a cross-section of the American population since 1972 and provides us with in depth knowledge of individuals' religious, social and economic characteristics. We separate the GSS sample into three specific groups: Catholics, Presbyterians and all other Protestants.(17) Among the Protestant groups specifically identified in the General Social Survey (which include Episcopalians, Baptists, Lutherans, and Methodists), Presbyterians are closest in doctrine and history to classical Calvinists. Therefore, we will examine whether Catholics differ from all other Protestants and whether Presbyterians particularly differ from Catholics.

There is unquestionably a problem using modern data to test theories about predestination. We do not claim that modern Presbyterians are in any sense still believers in traditional Calvinist dogma. Instead, we believe that predestination may have played a role in determining behavioral patterns and social norms among Presbyterians and their Calvinist ancestors. As time passed, the belief in predestination may have vanished, but the social norms and behavioral patterns remain. Our views and the approach of using modern data to test Weber's theory is hardly novel to this paper. Weber himself focuses on then-modern Protestant-Catholic data. He freely admitted that the importance of religious dogma has declined, but rather argued that the "origin [of Protestant-Catholic differences] is what really needs explanation." Weber [1930] argues that Puritan ideals "were carried out of monastic cells into everyday life and began to dominate world morality," at least in Protestant cultures.

Weber [1930] is not alone in taking the view that the disappearance of Puritan doctrine did not mean that end of Puritan habits or Puritan social norms. Many sociologists (e.g. Baltzell) have argued that individuals who are completely devoid of formal belief in Puritan dogma (for example, Unitarians or agnostics) are still strongly influenced by longstanding elements of Puritan behavior. In any case, the decline of belief in predestination should make it even more surprising if we find any results. Under most reasonable assumptions, the bias created by the decline in belief in predestination should be towards zero.

Table III gives the means of our primary variables for these three groups; Table IV gives the means of all variables used for the entire GSS. As Tomes [1994] and others have stressed there is no longer a positive earnings or schooling gap between Catholics and the overall group of Protestants, although Presbyterians are better educated and wealthier than either other group. The third row documents that Catholics are more likely to attend church, and are more likely to have been arrested than either other group. Catholics are members of slightly more civic organizations than all other Protestants in the U.S. but members of many fewer organizations than Presbyterians.

Our first test of the model is meant to examine whether there is a stronger connection between religiosity and worldly success among Protestants than among Catholics. This is a primary implication of the signaling model and is central to Weber's theory. We test this hypothesis in the first column of Table V by examining whether there is a stronger connection between schooling and church attendance among Catholics, Presbyterians and all other Protestants. We use schooling as our proxy for worldly success, rather than income, primarily because income is poorly measured in the General Social Survey as many respondents refuse to report their income.

This regression, and all other regressions in the table, includes controls for denomination, region, marital status, year of birth, age, income, city size, children, and gender. These coefficients are not reported to save space (the full list of controls is given in the table). In general, there is a positive connection between years of schooling and church attendance, as reported by Stark, Iannaccone and Finke [1996]. Glaeser and Glendon [1997] suggest that this connection is explained by the general desire of the more educated for social organizations. However this effect is 25% weaker among Catholics and 100% stronger among Presbyterians. It appears that the connection between religiosity and education (which is our proxy for worldly success) is stronger among the Calvinist Protestants [TABULAR DATA FOR TABLE III OMITTED] than among Catholics, just as the theory predicts.

Our second test examines whether there is a stronger connection between communal behavior and individual behavior among Protestants or among Catholics. The first model suggests that individuals who believe in a doctrine of free will face private incentives, where individuals who believe in a doctrine of predestination will behave well if their community has a high degree of belief on average. If this theory is correct, the connection between individual behavior and group behavior will be higher among those who believe in Predestination. To test this hypothesis, we construct a proxy for an individual's peer group in the GSS by examining the average behavior among individuals in the same education group (high school dropout, high school graduate and college graduate) in the same state and then examine the connection between this average (omitting the individual's own behavior in the average) and the outcome variable in question, again controlling for a full battery of individual characteristics as well as education group dummies.

We perform this test for the outcomes of church attendance, being arrested and membership in non-professional, non-religious organizations. For all three variables the connection with social group is weakest among Catholics. For church attendance and being arrested, the social effect is strongest among Presbyterians. To additionally test the theory, we examine the connection between peer group attendance and individual attendance for two other Protestant denominations: Episcopalians and Lutherans. We found that both of these groups lie between Catholics and Protestant. Lutherans lie closer to Catholics in that they are less influenced by peer group behavior; Episcopalians lie somewhat closer to Presbyterians. As Episcopalians also officially embrace the doctrine of Predestination (again, we do not claim that any modern Episcopalians [TABULAR DATA FOR TABLE IV OMITTED] believe in this doctrine, just that its presence may have influenced persisting patterns of social behavior) and as many former Puritans did eventually join the Episcopalian church, we are pleased to see that they are the group closer to Presbyterians.

The effect on non-religious memberships is not stronger among Presbyterians than all other Christians. While the differences between the groups are not always significant statistically, they are always large relative to the baseline effect of the group average.(18) For [TABULAR DATA FOR TABLE V OMITTED] [TABULAR DATA FOR TABLE VA OMITTED] these other variables, the effect of peer group outcomes on Episcopalians and Lutherans also falls between the effect of peer groups on Presbyterians and Catholics. In general, we consider these results to be quite supportive of the theory.

To further test our hypothesis, we examine the effect of the size of one's denomination within one's state on church attendance. Glaeser and Glendon [1997] argue that this variable should be positively related to church attendance if church attendance is driven by a social desire to meet one's neighbors. We test whether this social variable is more powerful for Presbyterians, Catholics or all other Protestants. Again, the basic pattern of the strongest effect for Presbyterians and the weakest effect for Catholics remains, although these differences are only marginally statistically significant. This variable again suggests that church attendance among Catholics is determined internally while church attendance among Protestants generally and Presbyterians in particular, is driven by the social norm of their community.

One possible problem with out findings is that they occur because Catholics are so much of a larger and more heterogeneous group than Presbyterians. According to this view, we would find that more homogeneous subgroups of the Catholic population resemble Presbyterians even though the Catholic population as a whole does not. In order to test this possibility, Table Va repeats the basic regressions of Table V using three different subgroups of the Catholic population: ethnic Irish, ethnic Italians and Mexican-Americans. In almost every case, our basic results are confirmed. Catholics appear to be less influenced by peer group attendance than Presbyterians. Their religious attendance is also less influenced by whether a large share of their state is Catholic. We also performed these tests where we divided the Catholic population up by income category and found similar results that support the basic finding.

VI. CONCLUSION

Despite Weber's [1930] debunkers, there was a substantial difference between Catholic and Protestant economic performance before the middle of the 20th century. This difference is an empirical puzzle that is worth understanding. We present a model of the costs and benefits of the Calvinist belief in predestination and find that under many conditions predestination is a more socially efficient belief system. The doctrine of free will creates incentives for individuals who actually believe in the hereafter since their behavior will affect their outcome after death. The doctrine of predestination creates incentives for everyone because of individuals' desire to convince others that they are members of a spiritual elect chosen by God.

Predestination is socially preferable in cases where (1) observable actions are more socially valuable than unobservable actions, (2) when diversity of views about the hereafter is high, (3) when homogeneity of actions is desirable and (4) when signaling is important. The model also predicts many of the stylized facts claimed by Weber. In predestination societies, (1) hypocrisy is more common, (2) moral behavior is a good predictor of material success, (3) rationality will be more pursued than in free will societies and (4) there will be a greater connection between individual and group behavior than in free will societies. The model suggests that predestination should have been adopted by societies when observability of actions was high (as in urbanizing Switzerland or Holland), when the heterogeneity of values was high (as in denser urban communities) or when belief in the hereafter was faltering.

We test two implications of the theory in late 20th century America. First, we find that there is a greater connection between education, which we use as a proxy for worldly success, and church attendance among Protestants, especially Presbyterians, than among Catholics. Second, we find a greater connection between group behavior and individual behavior among Protestants, again particularly among Presbyterians. Both of these findings support the idea that the social structure created by the reformation has had long-lasting effects on behavior that may have outlasted the effects of doctrine on economic differences between denominations.

Glaeser acknowledges financial support from the National Science Foundation and the Sloan Foundation. Justin Funches and Jacob Vigdor provided excellent research assistance. Conversations with Glenn Ellison, Benjamin Friedman, Claudia Goldin, Lawrence Katz and Jose Scheinkman were helpful to our thinking about this topic.

1. Throughout this essay, we will not be able to present a fair view of the complexities of both Protestant and Catholic thinking. The unfortunate simplification is necessary to make the models intelligible and we hope that later work will address more of the complexities of religious thought at this period.

2. Lenski [1961] does however find that Protestants tend to be somewhat more approving of capitalist norms.

3. In fact, predestination was not a new doctrine distinct from Catholicism. Augustine, himself, was an early, powerful advocate of predestination (especially of Saints).

4. As Pelikan [1984] writes [Calvinism] "made the divine will the cause of their [sinners'] damnation no less than of the salvation of the believers." The incarnation of this view in the 1801 Articles of Religion of the Protestant Episcopal Church (still a centerpiece of Episcopalian faith), "he [God] hath constantly decreed by his counsel secret to us, to deliver from curse and damnation those whom he hath chosen in Christ out of mankind..."

5. In fact, the doctrine of predestination allowed some degree of individual action in many of its versions.

6. Iannaccone [1997] provides a more thorough investigation of the incentives that religion can create in a range of behaviors.

7. Our distinction between predestination and free will is similar to distinctions made by Piers and Singer [1952] on shame vs. guilt societies. In their terminology shame societies are those where good behavior is created by external sanctions. Guilt societies, alternatively, operate through internal motivation. Kandel and Lazear [1994] also present a model of peer pressure which resembles the incentive model we use to capture predestination societies.

8. In fact, most of our countries are far from this margin. Only Germany, which is 37% Catholic is even close to our cutoff point. Our Catholic countries are France, Italy, Spain, Argentina and Mexico. Our Protestant countries are Canada, the U.S., Denmark, Sweden, the Netherlands, the U.K. and Germany (after WWII West Germany).

9. Iannaccone [1994] documents the strikingly low levels of religious attendance in most of Europe.

10. We use the word moral not because we wish to express any personal ethical feelings, but rather because this model asks about actions which are directly proscribed or suggested by religious systems.

11. An additional role of good deeds might be in convincing oneself that one is part of the elite. Modeling this type of behavior is beyond the scope of this paper.

12. We are only including here the positive reactions induced by agents that are justified on "spiritual" grounds. There are other reasons why morality might be advantageous, in particular morality might be desirable in employees or borrowers for purely financial reasons. These signaling uses of moral behavior will be dealt with in the next section.

13. As Tawhey [1926] writes "[Calvinism was] the attempt to crystallize a moral ideal in the daily life of a visible society." (Italics added).

14. Technically we assume [Mathematical Expression Omitted] so that high moral cost individuals do not attempt to imitate low moral cost individuals.

15. While this pooling equilibrium is not stable in the Kohlberg and Mertens [1986] sense, minor changes to functional forms, which would not change any other part of the model, would make this equilibrium stable.

16. The assumptions that are necessary for this are that

[Mathematical Expression Omitted].

which prevents imitation of the high belief, low cost type and

[Mathematical Expression Omitted]

so that high cost, low belief individuals don't imitate the other groups.

17. We exclude from our sample all GSS respondents who report themselves as Jewish, other Religion, No Religion, Protestant Other Denomination, or Protestant No Denomination.

18. Statistical significance tends to be low because our coefficients have all been correlated for state-education group random effects.

REFERENCES

Giddens, Anthony. "Introduction," in The Protestant Ethic and the Spirit of Capitalism. London: Routledge, 1992.

Glaeser, Edward L., and Spencer Glendon. "The Demand for Religion." Mimeographed, 1997.

Goldin, Claudia, and Laurence Katz. "Education and Income in the American Heartland: Iowa, 1915." Mimeographed, 1997.

Erikson, Kai T. Wayward Puritans: A Study in the Sociology of Deviance. New York: Wiley, 1966.

Iannaccone, Laurence. "The Consequences of Religious Market Structure: Adam Smith and the Economics of Religion." University of Western Ontario Working Paper, 1994.

-----. "Religion, Value and Behavioral Constraint." Mimeographed, 1997.

Kandel, Eugene, and Edward Lazear. "Peer Pressure and Partnerships." Journal of Political Economy, 100, 1992, 801-17.

Kohlberg, Elon and Jean-Francois Mettens. "On the Strategic Stability of Equilibria." Econometricia, 54(5), 1986, 1003-37.

La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer and Robert W. Vishny. "Trust in Large Organizations." American Economic Review, 87(2), 1997, 333-38.

Lenskni, Gerhard. The Religious Factor: A Sociological Study of Religion's Impact on Politics, Economic and Family Life. Garden City: Anchor Books, 1963.

Maddison, Angus. Phases of Capitalist Development. New York: Oxford University Press, 1982.

Nelson, Benjamin. The Idea of Usury: From Tribal Brotherhood to Universal Otherhood. Chicago: Charles C. Thomas, 1969.

Pelikan, Jaroslav. The Vindication of Tradition. New Haven: Yale University Press, 1984.

Putnam, Robert. "The Case of the Missing Social Capital." Mimeographed, 1996.

Rachfahl, F. in Internationale Wochenschrift fur Wissenschaft, Kunst und Technik, Nos 39-43, 1909.

Robertson, H. M. Aspects of the Rise of Economic Individualism, A Criticism of Max Weber and His School. Cambridge, England: The University Press, 1933.

Samuelsson, Kurt. Religion and Economic Action. New York: Basic Books, 1961.

Sombart, Werner. The Quintessence of Capitalism. London: T. Fisher Unwin, 1915.

-----. The Jews and Modern Capitalism. London: T. Fisher Unwin, 1913.

Stark, Rodney, Laurence Iannaccone and Roger Finke. "Religion, Science, and Rationality." American Economic Review, 86(2), 1996, 433-37.

Tawney, R. H. Religion and the Rise of Capitalism. New York: Harcourt, Brace and Company, 1926.

Tomes, Nigel. "The Effects of Religion and Denomination on Earning and the Returns to Human Capital." Working paper, 1994.

Weber, Max. The Protestant Ethic and the Spirit of Capitalism. New York: HaperCollins, 1930.

Winter, Jerry. "Quantitative Studies off the Applicability of the Weber Thesis to Post-World War II in USA: A Call for Redirected Efforts." Review of Religious Research, 16, 1974, 47-8.

Glaeser, Edward L.: Professor, Department of Economics, Harvard University and National Bureau of Economic Research, Cambridge, Mass., Phone 1-617-495-0575, Fax 1-617-495-7730, E-mail eglaeser@kuznets.harvard.edu

Glendon, Spencer: Doctoral Candidate, Department of Economics, Harvard University, Cambridge, Mass., Phone 1-617-588-0377, Fax 1-617-868-2742, E-mail glendon@nber.org