How Many Hispanics are Catholic? A Review of Survey Data and Methodology* by Paul Perl, Jennifer Z. Greely, and Mark M. Gray Center for Applied Research in the Apostolate Georgetown University *An earlier version of this paper was presented at the 2004 meeting of the Association for the Sociology of Religion, San Francisco. We thank Michael Davern for valuable advice on analysis of telephone service interruption. The Pew Hispanic Center and the Kaiser Family Foundation bear no responsibility for the interpretations offered, or conclusions made based on analysis of the Pew Hispanic Center/Kaiser Family Foundation 2002 National Survey of Latinos data. Direct correspondence to Paul Perl, Center for Applied Research in the Apostolate, 2300 Wisconsin Ave. NW, Suite 400, Washington, DC 20007. E-mail: [email protected] How Many Hispanics are Catholic? A Review of Survey Data and Methodology Abstract The proportion of Hispanics who are Catholic and Protestant remains unclear, partly because of varying survey methodologies and limited understanding of how that variation affects estimates of Hispanic religious identification. We compare results and methodologies of 11 national surveys conducted since 1990. Our review suggests English-only interviewing artificially inflates Protestant identification among Hispanics. Additionally, defining Hispanic ethnicity based on national origin or ancestry may inflate Catholic identification. We also use survey data to explore effects of sampling bias, non-coverage bias, and weighting on religious identification. Analyses suggest post-stratification weighting is advisable, particularly for language use. However, weighting is not a substitute for extensive coverage of difficult to reach sub-populations such as recent immigrants and Spanish-only speakers. We conclude that 70 percent or slightly more is a reasonable estimate of the proportion of adult Hispanics who are Catholic, and 20 percent a reasonable estimate of the proportion Protestant or other Christian. Estimates of the proportion of Hispanics or Latinos1 in the United States who identify as Catholic vary considerably, from slightly over half to 90 percent (e.g., Froehle and Gautier 2000:17; U.S. Department of Health and Human Resources 2000). No clear consensus has emerged among scholars, and debate persists among church leaders and activists. Recently, one of the authors heard journalist and author Peter Steinfels say he had omitted in-depth discussion of Hispanics from his book on the state of the American Catholic Church, A People Adrift, in part because of conflicting information about the number of Hispanic Catholics and the rate at which this population is growing. In this paper, we sort through available evidence on the proportion of Hispanics who are Catholic. Although our primary focus is on Catholicism, we also examine the related issue of the proportion of Hispanics who are Protestant or other Christian. Our goal is – if not to arrive at exact figures – at least to narrow the range considerably and suggest what survey methods are best suited for addressing the topic of Hispanic or Latino religious identification. First we review existing survey evidence on the religious identification of adult Hispanics in the United States. Then we analyze data from a recent telephone poll to better understand how sampling bias, non-coverage bias, and post- stratification weighting are likely to affect estimates. The Problem: Difficulties Measuring Religious Identification of Hispanics The terms Hispanic and Latino refer to a particular ethnicity and are not race categories. For this reason, the U.S. Census Bureau asks about Hispanic ethnicity in a question that is separate from race. There were slight differences in the Hispanic origin questions used by the Census in 1990 and 2000. The 1990 Census questionnaire asked: “Is this person of Spanish/Hispanic origin?” and the 2000 Census questionnaire asked, “Is this person 1 Spanish/Hispanic/Latino?” Furthermore, the Census Bureau determined between the 1990 and 2000 censuses that presenting the race question after the question on Hispanic identification improved the reporting for both race and Hispanic ethnicity (Cresce, Schmidley, and Ramirez 2004). Consequently, in the 2000 Census form the Hispanic question and race question were reversed so that race identification now follows Hispanic origin. A plurality of Hispanics in the United States identify their race as white.2 Nine percent of the U.S. population in the 1990 Census and 12 percent in the 2000 Census identified as Hispanic or Latino. According to Census figures, the Hispanic population in the United States grew from 22.3 million in 1990 (Gibson and Jung 2002) to 39.9 million in 2003 (U.S. Census Bureau 2004), an increase of 78 percent. Over the same time span, the non-Hispanic population grew from 226.4 million to 250.9 million, an increase of 11 percent. The higher growth rate for Hispanics is due to a combination of immigration and a higher birth rate (Malone et al. 2003; Downs 2003). Data reported by Ramirez and de la Cruz (2003) indicate that 21 percent of the Hispanic population in 2002 had immigrated to the United States during the period of 1990 to 2002. Thirty-four percent of the Hispanic population in 2002 was under the age of 18, compared with just 23 percent of non-Hispanic whites (Ramirez and de la Cruz 2003). Given the relatively high rate of growth in the U.S. Hispanic population and the national origin of most Hispanics in predominately Catholic countries – two-thirds report their origin as Mexico and nearly one-quarter as Central/South America or Puerto Rico (Ramirez and de la Cruz 2003) – it is not a tremendous leap to infer a great deal of growth in Hispanic Catholics. In fact, virtually nobody disputes that the number of Hispanic Catholics in the U.S. is growing rapidly. However, 2 identifying the proportion of Hispanics who identify as Catholic, and the extent to which it is changing, is not a straightforward task. The high rate of immigration of Hispanics complicates this task, setting in motion two apparently countervailing trends. First, evidence indicates that over the course of recent decades the proportion of U.S. Hispanics who are Catholic has declined while the proportion who are Protestant has increased (Hunt 1999; Kosmin, Mayer, and Keysar 2001; Greeley 1988, 1997; see also Greeley 1990: 120-123). However, Espinosa, Elizondo, and Miranda (2003) note that second and third generation Hispanics are more likely than their first generation counterparts to identify as Protestant. This suggests that conversion from Catholicism to Protestantism often accompanies assimilation. Thus, while it is reasonable to expect continuing defection to Protestantism among the children and grandchildren of Hispanic immigrants, the sheer number of new Hispanic Catholic arrivals slows down the overall rate of growth in Hispanic Protestantism (Espinosa et al. 2003). These findings are, of course, derived from surveys. Because the Census Bureau does not ask about religion, random sample surveys of individuals are the best method for measuring religious identification in the U.S. population. Data taken from church records or surveys of congregations are occasionally used for this purpose but are ultimately inadequate. For example, one can count Hispanics affiliated with Catholic and Protestant congregations in a geographic area and divide by the total Hispanics residing in the area, as taken from the Census or Current Population Survey (CPS). One problem is that many religious adherents are not registered as congregation members, something that is particularly true of Hispanic Catholics (Davidson 2000). Our own unpublished analyses of survey data show that 53 percent of Hispanic Catholics 3 are registered parishioners compared to 71 percent of Anglo Catholics. Even among those who report attending Mass at least once a week, the registration rates are 65 and 86 percent, respectively.3 Congregation records are also fraught with accuracy problems. Some church members (particularly those residing in urban areas) are registered at multiple congregations. And many congregations are slow to clean their membership roles of those who have moved away or simply left the faith altogether. Because they require no base level of congregational involvement, surveys of individuals are superior to data from congregations for understanding rates of religious identification. However such surveys also have disadvantages. One is that most surveys are limited to adults. Thus it becomes necessary to make supplemental calculations about the proportion of children and teens who are Catholic or Protestant. In this paper we focus on adult Hispanics and leave the issue of Hispanic children and teens for future research. Another salient disadvantage is that all surveys miss some individuals. Most obviously, telephone polls miss people residing in households without telephones. Even face-to-face surveys are likely to provide incomplete coverage of certain difficult-to-reach populations. This is particularly pertinent to the study of Hispanics because recent immigrants and migrant workers are probably among those most likely to be excluded by even the best surveys. A final disadvantage is that polling methods vary greatly, probably contributing to varying results for religious identification. We consider three aspects of variation in polling methodology that are particularly relevant for surveying Hispanics. (1) Some surveys are conducted in English only while other surveys are bilingual. (2) Polls are inconsistent in how they define who is and who is not Hispanic. Some polls identify Hispanics based on questions 4 about ancestry or national origin rather than the Census approach of asking people directly if they are Hispanic or Latino. Even polls that do ask this directly often use different wording. (3) Surveys vary in whether or not sampling and post-stratification weighting are used. Surveys that do use weighting vary in whether post-stratification is based on population characteristics of Hispanics. In the following section, we explore the implications of variation in polling methodology by examining results from previous surveys. A Review of Existing Survey Evidence Table 1 lists 11 national surveys conducted since 1990 that provide an estimate of religious identification among Hispanics or Latinos. Three of the surveys are well-known recurring academic studies, the General Social Survey (GSS), the National Election Studies (NES), and the World Values Study (WVS). Others have been chosen because of large numbers of Hispanics (some are polls of Hispanics only) or because of a focus on religion. Most of the surveys are telephone polls. The exceptions are the GSS, the WVS, and the Latino National Political Survey (LNPS), which use in-person interviewing, and the NES, which employs both telephone and in-person interviewing. The surveys in Table 1 are divided into two categories: surveys of the general public – which have all been conducted primarily or exclusively in English – and bilingual surveys that focus on Hispanics or Latinos. The table presents each survey’s estimate of the percentage of Hispanic respondents who identify themselves as Catholic, as Protestant or other Christian, and as having no religion. The table also presents the number of Hispanic respondents in each survey. Percentages have been generated using the weighting variable (if any) recommended by the original data collectors. Wherever possible, percentages and Ns exclude invalid responses 5 (refusals and “don’t know”s). In every case where the data are publicly available4 we have run the numbers ourselves. For the others, we have relied on published information, occasionally supplemented by personal communications with the original researchers. The Appendix provides additional information about the polls in Table 1, including the question or method used to identify respondents as Hispanic or Latino, whether weighting has been used to generate the percentages, and the percentage of interviews conducted in Spanish. [Table 1 About Here] The lowest proportion Catholic – 47 percent – comes from the 1999-2000 wave of the World Values Study. The highest proportion – 76 percent – comes from the Latino National Political Survey. The proportion Protestant or other Christian5 ranges from 14 percent (LNPS) to 39 percent (1996 NES). The proportion who identify with no religion6 ranges, from 6 percent (e.g., 2004 National Survey of Latinos) to 25 percent (1999-2000 WVS). The latter two ranges should perhaps be qualified because of two outliers. The WVS produces a higher level of no religion than other surveys because of the wording of its religious identification question. The WVS asks, “do you belong to a religious denomination?” Perhaps because of confusion or discomfort with the concept of a “denomination,” this leads to a relatively high proportion of all respondents reporting no religion, including a fifth to a quarter of Hispanics. Excluding the WVS, the range for no religion is 6 to 15 percent. The NES estimate for Protestants or other Christians is higher than that of other surveys, including other surveys with low estimates for Catholics. The reason is not obvious. However, it should be noted that the NES is not truly a survey of the “general public” but of U.S. citizens; it excludes immigrants who are non-citizens.7 We speculate below about other possible reasons why so many in this survey identify as 6 Protestants/other Christians in this survey. Even when excluding the NES the range is 14 to 29 percent, with the latter estimate doubling the former. BILINGUAL INTERVIEWING What accounts for the variation in religious identification among the polls? Bilingual interviewing stands out as one source. In the last five or six years estimates for proportion Catholic from the predominately-English surveys (the first panel in Table 1) range from the upper forties to the mid 60s. The Religion and Public Life Survey, which uses English only, produces estimates in the range of 51 to 61 percent Catholic. The two most recent waves of the World Values Study, the second of which was entirely in English, put the proportion Catholic at 59 and 47 percent (albeit with low Ns). The estimate produced by the English-only American Religious Identification Survey is 57 percent. The National Election Studies, which have been conducted in entirely in English since 1994, have yielded estimates in the mid to upper 50s in the last decade, with the exception of 64 percent in 1998. The English-only General Social Survey produces some of the highest estimates in the first panel: 63 percent Catholic in 2000 and 67 percent in 2002. In contrast, all but one of the bilingual surveys in Table 1 yield an estimate in the upper sixties to mid 70s. Note that Spanish was used in at least half the interviews of the bilingual surveys for which information is available. (Forty-nine percent of respondents to the 2003 version of the National Survey of Latinos were interviewed primarily in Spanish, the lowest level of Spanish interviewing – see the Appendix.) An indication of the importance of Spanish interviewing in reaching Hispanics is that Hispanics make up only 4 to 8 percent of all respondents in most of the predominately-English surveys (see the Appendix). The finding of 7 Epinosa and colleagues (2003) that Hispanics of the second and third generations (and who most likely can be interviewed in English) identify as Catholic at a lower rate lends support to the inference that those who are excluded are disproportionately Catholic. IDENTIFYING HISPANICS Another possible source of variation among surveys in Table 1 lies in the way they identify Hispanics. All but two use an approach similar to that of the Census – directly asking respondents if they think of themselves as Hispanic or Latino. The two surveys in Table 1 that depart substantially from the approach of the Census are the WVS and the LNPS. It is difficult to imagine a more cumbersome measure of racial and ethnic identity than that used in the World Values Study. Respondents are asked: “Which of the following best describes you?” and presented a list of five statements, one of which reads, “Above all, I am an Hispanic American.” Other, parallel, statements substitute “white American,” black American,” and “Asian American.” Thus the question makes racial categories (such as white and black) mutually exclusive from Hispanic or Latino ethnicity. Perhaps an even greater difficulty is that another response category states, “I am an American first, and then a member of some ethnic group.” Roughly 30-40 percent of all respondents (depending on which wave of the WVS is examined) choose this option and remain essentially unidentified on race and ethnicity. World Values Survey interviewers separately record the race or ethnicity of respondents based on their own physical observation. In the 1999-2000 wave, one of the options for interviewers to record was “Hispanic.” This measure has its own problems, including interviewer effects, and we have deliberately chosen not to use it for Table 1. Still, the fact that 21 of 98 respondents coded as Hispanic by the interviewer chose to describe themselves as 8