NBER WORKING PAPER SERIES RELIGIOSITY AND STATE WELFARE Angela K. Dills Rey Hernández-Julián Working Paper 19169 http://www.nber.org/papers/w19169 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 June 2013 We thank Brian Duncan, Melanie Guldi, Dan Hungerman, Leo Kahane, and participants at the 2012 Southern Economic Association meetings and 2013 Eastern Economic Association meetings for their useful comments. We are grateful for the feedback provided by the participants at the 2013 NBER conference on the Economics of Religion and Culture. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. At least one co-author has disclosed a financial relationship of potential relevance for this research. Further information is available online at http://www.nber.org/papers/w19169.ack NBER working papers are circulated for discussion and comment purposes. They have not been peer- reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications. © 2013 by Angela K. Dills and Rey Hernández-Julián. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. Religiosity and State Welfare Angela K. Dills and Rey Hernández-Julián NBER Working Paper No. 19169 June 2013 JEL No. H41,I38,Z12 ABSTRACT The Catholic sex abuse scandals reduced both membership and religiosity in the Catholic Church. Because government spending on welfare may substitute for the religious provision of social services, we consider whether this plausibly exogenous decline in religiosity affected several measures of the public taste towards government spending on welfare between 1990 and 2008. In places where there were more scandals, individuals state a preference for less government provision of social services. In contrast, a higher level of abuse is also associated with an increase in voting for Democratic candidates for President and state legislatures, and an increase in per capita government welfare spending, although this increase is insufficient to replace the decrease in Catholic-provided charity. Angela K. Dills Providence College 1 Cunningham Square Providence, RI 02918 [email protected] Rey Hernández-Julián Department of Economics, CB 77 Metropolitan State University of Denver Denver, CO 80217-3362 [email protected] I. Introduction Many religious institutions provide services to the members of their congregations with in-kind transfers, such as providing poor families with Christmas toys or Thanksgiving dinner. In fact, one of the benefits of religious participation is the insurance it provides against income shocks through these transfers and similar practices (Dehejia, DeLeire, and Luttmer, 2007). Religious institutions also provide services for the general public such as soup kitchens, medical assistance, inexpensive resale clothing, and shelters for the homeless. Government provides similar services through programs such as food stamps, Medicaid, and unemployment insurance that offer protection against income loss. The literature strongly supports that government welfare spending crowds out religious, charitable giving: as government provides more social services, individuals donate less to the charities that provide these services. Gruber and Hungerman (2007) estimate that the New Deal reduced church charitable spending by 30 percent. Hungerman (2005) demonstrates that the reduction in public welfare spending resulting from the 1996 Welfare Reform was partially offset by the Presbyterian Church (USA) increasing their charitable endeavors for the affected populations. His estimates of crowd-out effects ranged between twenty and thirty-eight cents on the dollar. Comparing across countries, Gill and Lundsgaarde (2004) find that increased government-provided welfare weakens support for religions. This paper explores the converse. We know that as government programs expand, private charity decreases. But if religious charity diminishes, does government provision of social services grow? We examine whether a negative information shock about one religion increases the demand for the government provision of social services. Previous research documents similar behavior by government. Becker and Lindsay (1994) show that private philanthropic contributions to public institutions of higher education crowd-out government support of these institutions dollar-for-dollar. In the wake of the Catholic sex abuse scandals, religiosity in the U.S. declined, particularly among Catholics (Hungerman, forthcoming and Perez-Truglia and Bottan, 2011). We re-establish this fact in the analysis below. Bottan and Perez-Truglia (2011) find that the abuse scandals reduced private charitable giving as well as the number of charitable organizations, particularly among Catholics. As church and state appear to be substitutes in providing social insurance, one would expect the decline in religiosity to strengthen support for government-provided welfare. 2 The evidence in support of this expectation is mixed. Responses in the General Social Survey suggest that individuals living where the reports of abuse were more prevalent increase their stated opposition to government provision of social services. We then examine whether voting patterns reflect this stated preference. We find the contrasting result that the scandals increased support for Democratic Party candidates in presidential and state legislative elections. In addition, actual welfare spending increased despite the stated preferences. Using state-level data on transfers per capita for family aid, Medicaid, Social Security Insurance, and food stamps, we find that the Catholic sex abuse scandals are associated with an increase in per capita government welfare spending of 7.9 percent for every one standard deviation increase in abuse. Most of the increase in spending appears to stem from Medicaid. The outline of the paper is as follows. First, we review the history of the Catholic sex abuse scandals and discuss the construction of our data on these scandals. In Section III we outline the relationship between private charity and state welfare and discuss the implied relationship between religion and government. In Section IV, we describe the data on religiosity, welfare spending, and presidential voting. The results are presented in Section V. We discuss the results and their policy implications in the conclusion. II. Catholic sex abuse scandals in the United States1 In 2002, Cardinal Bernard Law resigned over his mishandling of the sex abuse scandals wracking the Catholic Church in the Boston archdiocese (see the Boston Globe’s coverage of the scandals for more detailed information). Much of this abuse occurred in the 1970s, although many victims did not report it until much later; indeed, much of the reporting occurred in the wake of the extensive 2001 coverage of the scandals in Boston. The U.S. Conference of Catholic Bishops commissioned a report summarizing information provided by the Catholic Church from its archives on perpetrators and victims of abuse. The John Jay Report, published in 2004, found that 4,392 priests (about 4 percent) participated in abuse. Settlements related to sex abuse cases have cost the Church over three billion dollars (The Economist, 2012). 1 Most of this section draws from Dills and Hernández-Julián (2011). 3 Although the data made available to the researchers of the John Jay report (2004) by the Catholic Bishops of the United States have not been made public, alternate measures of the publicity related to these scandals are available. Like previous work, including Dills and Hernández-Julián (2011), Hungerman (forthcoming), and Bottan and Perez-Truglia (2011), this paper focuses on negative publicity about the abuse. Scandals can be damaging regardless of their veracity: a marriage may fall apart based on allegations of infidelity and the suspicion of plagiarism can lead to the end of an academic career. A false positive on a drug test can lead an employer to fire a worker.2 Allegations of a sexual nature may be particularly damaging and difficult to repudiate. Expanding on the work in Dills and Hernández-Julián (2011), we generate a measure of negative publicity that tabulates the number of priests and nuns in each diocese involved in sexual abuse cases. The website www.bishopaccountability.org compiles and publishes the names of priests and nuns involved in sexual abuse cases as well as their current diocese, former dioceses, and current status within the Church (still with the Church, convicted, retired, or deceased). The site is run by a small staff as an educational enterprise, and its goal is to collect documents that allege abuse within the Catholic Church, using broad requirements for including documents. The website cites and includes allegations and the documents reporting the allegations. Notes about each offender include the dates the Church was informed about an incident, whether cases were filed or settled, and information on arrests, indictments, confessions, and convictions of clergy. We use the dates that a priest or nun was arrested, was convicted, confessed, or settled his or her case. These dates mark events likely to capture public attention through newspaper articles, press releases, or court documents; we refer to these as public notice dates.3 As a result, we have measures of when notice was brought to each offender and at which diocese. Typically, each offender has multiple dates—separate ones for their dates of arrest, indictment, confession, etc.—sometimes more than one of these happens in a single year. We consider only the earliest such event as the initial public notice; each 2 Barnum and Gleason (1994) estimate that one third of those identified as drug users may be falsely accused. 3 We assume that these public events accurately convey the timing and degree of the public information, though we realize that information on allegations may be known by the parishioners prior to being made public. 4 offender thus has one initial public notice date.4 We aggregate these to the diocese level and have an annual total count of initial public notices for each diocese. The impact of these notices is likely to compound over time. A single notice in a community may not be as meaningful as when it follows a long string of bad press in an area with a long history of abuse. We use the initial public notices to generate two measures accounting for the importance of historical information. The first sums all of the initial notices that we have on record for the entire history that bishopaccountability.org covers; this is the cumulative number of initial notices documented in that place up to that year. We also generate a second measure that aggregates the initial notices for the previous four-year period, as the effect of the notices may expire after some time. In the analysis below, this offender-level data is aggregated to either the diocese- or state-level. The Catholic Church is organized into dioceses and archdioceses, each administered by a bishop or archbishop. There are 175 of these in the United States, with each state and the District of Columbia having at least one. Dioceses, for the most part, follow county lines.5 Texas, at 14, is the state with the most dioceses. The average state has approximately 3.5 dioceses. Figure 1 presents both the cumulative sum and 4-year sum of initial public notices by year from 1990-2008 for the U.S. In the mid-1990s, a spike in public notices steepens the trend. Public notices sharply increase again in 2002 and continue rising. These data measure the timing of the allegation rather than the timing of abuse. Allegations occur around the time of abuse as well as many decades afterwards, and many cases of abuse likely remain unreported. The pattern in Figure 1 accords with the distribution of dates presented in the John Jay report (their Figure 5.2.1). Table 1 summarizes the cumulative initial public notices between 1990 and 2008. The median value of cumulative initial public notices is 22. If we split the sample in two at the median, those states with high abuse have a mean value of 113.3 notices, while those below the median have a mean value of 16.8. Figure 2 maps the variation of public notices across the United States, with the darker states having a higher number of notices. Some dioceses, such as Los Angeles, Chicago, and Boston, experience many events; nineteen of the 175 dioceses experienced no 4 Estimates using all of the public notice dates results in qualitatively and quantitatively similar estimates. Estimates using a three year sum of initial public notices also results in qualitatively and quantitatively similar results. 5 Ten counties cross diocese lines. In these cases, we halve the county demographic information between the two dioceses. 5 recorded events.6, 7 A potential weakness of our measure is that it depends, to varying degrees, on the local level of interest. Interested locals may be more likely to participate on the bishopaccountability.org website, more completely reflecting the level of abuse. Our measures conflate public interest with the actual level of abuse in an area. This is not problematic, as long as it is clear that we are measuring scandal more than actual abuse. In fact, since our outcome likely responds to the scandal, a measure that captures some information on its severity may be beneficial. The scandal data described above is measured at the diocese level. Some regressions are estimated at the diocesan-level. In most regressions, we aggregate these to the state level, depending on the unit of measurement available for our other variables. This is described in more detail in the following section.8 III. Estimation and identification strategy A. Religious and Government provision of social services Huber and Stanig (2011) model the competition between religious redistribution and governmental redistribution from the rich to the poor. In their model, the ‘religious’ poor and the secular poor compete for redistribution.9 The ‘religious’ poor and the rich elect officials who favor low taxes and limited redistribution contra the preferences of the secular poor. A decline in ‘religious’ poor weakens support for redistribution through the church and strengthens support for governmental redistribution. They test this model using international variation in religiosity and their results support the model’s implied comparative statics. Their model does not rely on religion affecting charitableness; church-based redistribution and governmental redistribution are substitutes. Similarly, Scheve and Stasavage (2006) model the substitutability between religious and governmental welfare. Using international, individual-level data, they empirically test their model. They 6 These include Amarillo, Beaumont, Biloxi, Birmingham, Colorado Springs, Dodge City, Gary, Gaylord, Grand Island, Kalamazoo, Knoxville, Lake Charles, Las Cruces, Lubbock, Rapid City, Saginaw, Shreveport, and Victoria. 7 Excluding Los Angeles, Chicago, and Boston from the regression does not change the pattern of our results and tends to make our results more statistically significant. 8 We generate an additional measure of the severity of the scandal based on a Lexis-Nexis count of published news items that include the words “sex,” “abuse,” and the name of each Diocese. Results from these estimates are largely similar to, albeit less statistically significant than, those presented and are available upon request. 9 The ‘religious’ poor refers to those willing to receive charity from religious organizations, not necessarily those with religious beliefs. Hence, we place religious in quotation marks. 6 find that more religious countries have lower social welfare spending, though it could be the case that in places where there is more welfare spending, individuals are less motivated to join a religious organization. Much of the literature estimating the substitutability of private and state charity relies on exogenous changes in government programs (Hungerman, 2005 and Gruber and Hungerman, 2007). These papers examine how charitable provision responds to changes in government welfare spending. Here, we consider how government welfare spending responds to a plausibly exogenous change in religiosity and its attendant decline in charitable provision. Our data permit us to estimate how religiosity responded to the scandals in the Church. We rely on Bottan and Perez-Truglia (2011) as an intermediate step in the logic of our paper. They demonstrate how charitable donations responded to the degree of information about scandals in a particular diocese. We then extend that research by documenting how individuals’ stated preference for government responds to the scandals and to what degree, as the charitable contributions diminishes, state welfare spending increases. B. Scandals and religiosity First, we verify the decline in religiosity in the wake of the Catholic sex abuse scandals. We estimate the effect of scandals in state j in year t on the religiosity of person i: (1) Scandals are measured using the two variations of initial public notices. The vector X contains a variety of the individual respondent’s characteristics: marital status, a quadratic in age, sex, race, ethnicity, education, real household income, number of children, and labor force status. The vector Z contains state- level characteristics including average real income per capita, the unemployment rate, and the fractions of the population that are black, white, female, aged 18 to 24, aged 25 to 34, aged 35 to 64, and aged 65 and over. We estimate equation (1) using ordinary least squares or, in the case of binary dependent variables, linear probability models.10 10 Estimates using a logit or probit are similar to the linear probability models for the Catholic and raised Catholic regressions. 7 Religiosity, the dependent variable, is either an indicator for whether the respondent identifies as Catholic or a measure of religious participation. The impact of the abuse scandals on religiosity has been examined previously in Hungerman (forthcoming) and Bottan and Perez-Truglia (2011), where the authors address the identification issues. Hungerman (forthcoming) in particular addresses pre-existing trends and argues that these do not drive the results. State fixed effects account for state-specific levels of religiosity and degree of Catholicism. Year dummies capture national changes in attitudes towards the Church and religion. Standard errors are clustered by state. C. Scandals and stated preferences for government If government and religious provision of social services are substitutes, then as people move away from the Church, they might desire a higher level of government support. The next models consider estimate several measures of the stated preference for the size of government as a function of the abuse scandals. (2) The vectors X and Z remain the same as in the religiosity regression. The individual-level variables include the respondent’s marital status, a quadratic in age, sex, race, ethnicity, education, real household income, number of children, and labor force status. These are necessary as these traits are correlated both with the typical preferences for government and with religious identification and intensity, and thus may also be correlated with the level of scandal. The state-level variables include real per capita income, the unemployment rate, and the fractions of the population that are black, white, female, aged 18 to 24, aged 25 to 34, aged 35 to 64, and aged 65 and over. We continue to include state fixed effects and year dummies. These fixed effects control for, for example, the possibility that states with better social services may also be more likely to have more redistributive government spending and stronger desires to out offenders. Standard errors are clustered by state. D. Scandals, Voting Patterns, and Welfare Spending 8 We supplement the results from people’s stated preferences with information on how behavior responds to the abuse. If individuals prefer an increase in provision of government services, then we would expect them to change their voting patterns. We estimate a diocesan-level version of the regression from part C above, replacing the dependent variable with the percentages that vote for the Democratic candidate in presidential elections and elections of representative to state legislatures and the U.S. House of Representatives. Although presidential contests only happen every four years, voters evaluate the same candidates nationwide. The data include results from every presidential election from 1992 to 2008. If, as predicted by the model, individuals respond to the scandals by preferring an increase in the government provision of social services, they would be more likely to vote for the Democratic candidate (Kiewiet, 1981; Rodríguez, 1999). We also consider elections to state legislatures and the U.S. House, which take place every two years during the same period. These elections allow for more observations. We estimate the following for diocese d in year t. (3) Here we seek an unbiased, causal estimate of . This estimate could be biased if scandals were more likely to take place in those areas where the abusers believed the community was weakening in its responsiveness to abuse. Diocesan fixed effects subsume any state fixed effects. The diocesan fixed effects capture information on each diocese’s unobserved traits that correlate to its allegiance to a party, but not any diocesan-specific responsiveness in political affiliation to new information. If political attachments are weak and, in the unlikely case that potential abusers care about changing political attachment, then the estimate is biased. In other words, we assume that both the choice of the abuser to engage in crime and the choice of the victim to go public are independent of expected political responses. The vector W includes diocesan characteristics: the real per capita income; the unemployment rate; and the percentages of the population that are Hispanic, black, white, female, aged 18 to 24, aged 25 to 34, aged 35 to 64, and aged 65 and over. 9