What the Polls Don't Show: A Closer Look at U.S. Church Attendance C. Kirk Hadaway; Penny Long Marler; Mark Chaves American Sociological Review, Vol. 58, No. 6 (Dec, 1993), 741-752. STOR ® Stable URL: http://links.jstor.org/sici?sici=0003-1224%28199312%2958%3A6%3C741%3AWTPDSA%3E2.0.CO%3B2-W American Sociological Review is currently published by American Sociological Association. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/asa.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact support@jstor.org. http://www.j stor.org/ Sun Apr 10 20:24:42 2005 WHAT THE POLLS DON'T SHOW: A CLOSER LOOK AT U.S. CHURCH ATTENDANCE* C. Kirk Had away United Church Board for Homeland Ministries Penny Long Marler Mark Chaves Samford University University of Notre Dame Characterizations of religious life in the United States typically reference poll data on church attendance. Consistently high levels of participation reported in these data suggest an exceptionally religious population, little affected by secularizing trends. This picture of vitality, however, contradicts other empirical evidence indicating declining strength among many religious institutions. Using a variety of data sources and data collection procedures, we estimate that church attendance rates for Protestants and Catholics are, in fact, approximately one-half the generally accepted levels. With respect to religion, the United States is an anomaly. Although thoroughly secular in many ways, religious participation and affiliation remain much higher in America than in other Western industrialized nations (Wald 1987; Bruce 1990, p. 178). Rates of religious activity also exhibit remarkable stability. Current self-reports of church attendance among Protestants do not differ significantly from levels recorded in the 1940s, and attendance among Catholics stabilized almost two decades ago after declining in the 1960s and early 1970s (Hout and Greeley 1987).1 Approximately 40 percent of the population of the United States is said to attend church * Direct all correspondence to C. Kirk Hadaway, Research Office, United Church Board for Homeland Ministries, 700 Prospect Avenue, Cleveland, OH 44115; Internet: HADAWAYK@UCC.ORG. This research was supported by grants from the Lilly Endowment, Inc. (#900754) and the University of Notre Dame. Additional financial support was provided by the United Church Board for Homeland Ministries. We are grateful to the diocesan researchers who supplied information on Catholic attendance. We also thank Lara Boucher, James Cavendish, Min Liu, and Ami Nagle for their assistance in data collection and proofing. Robert K. Shelly and three anonymous ASR reviewers gave helpful comments on an earlier version of this paper. 1 Glenn (1987) noted that, during the 1930s and 1940s, Gallup's samples were designed to represent voters rather than the entire adult population. Women, Southerners, blacks, and people at low socioeconomic levels were underrepresented. The ex- weekly. This statistic is based on extraordinarily stable results from social surveys in which respondents report church attendance. Gallup polls, for example, ask respondents, "Did you, yourself, happen to attend church or synagogue in the last seven days?" In 1991, 42 percent of adult Americans (45 percent of Protestants and 51 percent of Catholics) responded affirmatively to this question (Princeton Religion Research Center 1992, p. 4).2 In the sociological literature, this high participation rate is prominently and widely cited to bolster attacks against the secularization hypothesis (Hout and Greeley 1987; Greeley 1989; Warner 1993). The rate undergirds Warner's (1993) "new paradigm" of religious adaptability—a proposed replacement for the older secularization paradigm (also see Finke and Stark 1992). Moreover, the attendance rate elusion of several high-attendance groups may have lowered the percentage of attenders in the 1930s and 1940s. Our primary concern, however, is with stability in the polls during the last 30 years—a period when this sampling issue is not relevant. 2 The overall attendance figure is lower than the attendance figures for Protestants and Catholics because of the inclusion in the former of persons who express no religious preference (the "nones"). Smith (1991) reported similar rates from the General Social Survey. He converted NORC categories into probabilities of attending church in a given week. For Protestants, attendance estimates ranged from 41 percent to 46 percent, but there was no trend from 1972 to 1989. The converted NORC rate for Catholics was 50 percent in 1989. American Sociological Review, 1993, Vol. 58 (December:741-752) 741 742 AMERICAN SOCIOLOGICAL REVIEW is appropriated freely by historians and the mass media. The following excerpt from an award-winning book on American religious history illustrates a typical use of this "social fact" outside sociology: On any given Sunday morning, over 40 percent of the population of the United States attends religious services. In Canada and Australia this number tails off to about 25 percent; in England to about 10 percent; and in Scandinavia to around 4 percent___Statistically, at least, the United States is God's country. (Hatch 1989, pp. 210-11) Similarly, a June 15, 1991 New York Times "Religion Notes" column began: "Nearly all surveys of American churchgoing habits show that roughly 40 percent of Americans attend church once a week" (Goldman 1991). Perhaps the best indicator of the "hardness" of this social fact is that it is a standard feature in the religion chapters of introductory sociology textbooks (Johnson 1992, p. 548; Kornblum 1991, p. 514; Thio 1992, p. 393). It also appears in the most widely used research methods text (Babbie 1992, p. 398). Still, the characterization of American religious participation as strong and stable is not uniformly accepted. Many social scientists, as well as church leaders, are skeptical about consistently high rates of church attendance. Membership losses among "old-line" Protestant denominations and slowing growth rates among large conservative denominations raise serious questions about such claims (Marler and Hadaway 1992). If Americans are going to church at the rate they report, the churches would be full on Sunday mornings and denominations would be growing. Yet they are not. Is it any wonder that "regular reports . . . that 70% claim church membership and 40% attend weekly" are met with "incredulity" (Warner 1993, p. 1046)? We present evidence that church attendance rates based on respondents' self-reports substantially overstate actual church attendance in the United States. Our empirical strategy is to compare church attendance rates based on counts of actual attenders to rates based on random samples of respondents who are asked to report their own attendance. Using a variety of data sources and data collection strategies, we estimate count-based church attendance rates among Protestants in a rural Ohio county and among Catholics in 18 dioceses. The results are dramatic: Church attendance rates for Protestants and Catholics are approximately one-half the generally accepted levels. CONTRADICTIONS AND IMPLICATIONS The contradiction between poll-based reports of church participation and denominational reality prompted this research. Consistently high levels of church attendance and a growing U.S. population suggest that most major denominations should be thriving and growing (Glenn 1987, pp. S116-17). Yet most are not. Claims that losses in old-line denominations are more than offset by gains in evangelical denominations and nondenominational churches do not suffice. In addition to the fact that evangelical gains simply are not numerically large enough, Americans in declining denominations still claim high levels of membership and attendance. The Episcopal Church illustrates the contradiction. This denomination should have grown in membership by more than 13 percent from 1967 to 1990, given the percentage of Americans who claim to be Episcopalians (and church members) on social surveys.3 Instead, membership in the Episcopal Church declined by 28 percent. Moreover, attendance figures from Episcopal parishes are far below what would be expected if self-defined Episcopalians attended church in the numbers they claim. Based on Gallup surveys and other poll data, about 35 percent of Episcopalians say they attended church during the last seven days. If 2.5 percent of Americans claim to be Episcopalian and 35 percent of Episcopalians attend worship, total attendance during an average week should exceed 2 million (Johnson 1993, p. 5). Instead, average weekly attendance 3 These figures are based on aggregated General Social Survey data (NORC), Gallup (AIPO) data, our 1992 Marginal Member poll, and U.S. population estimates (U.S. Bureau of the Census 1992). Actual membership figures are from the Yearbook of American and Canadian Churches (Bedell and Jones 1992; also earlier volumes) and from information supplied to the authors by the Episcopal Church. The Episcopal Church baptizes infants, and all baptized persons are counted as members. These definitions have not changed during the period of study. Thus, our estimates of self-identified Episcopalians pertain to the entire population rather than adults. WHAT POLLS DON'T SHOW ABOUT U.S. CHURCH ATTENDANCE 743 was less than 900,000 in 1991.4 Rather than 35 percent, it seems that approximately 16 percent of self-defined Episcopalians attended worship at an Episcopal church during a typical week.5 These data suggest that many Episcopalians claim to have attended church when in fact they did not. And if Episcopalians overreport their church attendance, perhaps persons who identify with other denominations also over-report their church attendance. This assumption led to our hypothesis that the percentage of Americans who attend church worship during an average week is considerably lower than the 40 percent level that is accepted as a "social fact" in the United States. We rely on survey data and church statistics. We recognize the tendency in sociology to ignore church statistics because they are presumed to be inflated (Demerath 1974; Haďa-way 1989). This is a valid point, but we argue that inflation errors are more serious for membership statistics than for attendance counts. Membership statistics are inflated because many denominations have no accepted procedure for deciding when to purge inactive members from church rolls. Denominations also differ in the meaning of membership. Accurate, consistent attendance data, on the other hand, require only that someone periodically count the persons attending church worship. Some inflation may occur here as well, but the bias is minimal and is not a problem for this research. Our purpose is to establish the difference between self-reported behavior and actual behavior, and because such a bias in the count would decrease that difference, the fact that the count-based rates are probably overestimates of actual attendance does not threaten our argument. DATA AND RESULTS Protestants The first step in our research strategy was to compare actual counts of church attendance to self-reported church attendance. We collected 4 The attendance count from Episcopal parishes includes all persons attending (even non-Episcopalians). Of course, Episcopalians may also attend non-Episcopal churches. 5 The attendance rate estimate for self-defined Episcopalians varies according to the proportion of Episcopalians in the population. Depending on the three types of data in a circumscribed area: (1) poll-based estimates of religious preferences for residents of the area; (2) poll-based estimates of church attendance for Protestants; and (3) actual counts of church attendance for all Protestant churches in the area. We selected Ashtabula County, located in extreme northeastern Ohio, because of its manageable-sized population (100,000 persons) and the location of its population centers. The two largest towns in the county are situated near its center, and there are no large towns near the county line. Thus, the number of persons from Ashtabula County who attend church in other counties should be offset by persons from other counties attending church in Ashtabula. The Protestant population of Ashtabula County was determined through a telephone poll of 602 county residents in the Spring and Fall of 1992. Phone numbers were randomly generated from among all active telephone exchanges. Thus, the sample included listed and unlisted numbers.6 Survey results indicated that 66.4 percent of the respondents claimed a Protestant identity. Catholics made up 24.8 percent of the population, "nones" 8.0 percent and "others" .8 percent. These percentages were comparable to other data for the East North Central census region, the state of Ohio, and Ashtabula County (Davis and Smith 1990; Bradley, Green, Jones, Lynn, and McNeil 1992). Applying the proportion Protestant to the 1990 census population count for Ashtabula County, we estimated the Protestant population of Ashtabula County at 66,565. The Ashtabula survey also asked respondents if they attended church or synagogue in the last seven days. We found that 35.8 percent of Protestant respondents and 37.2 percent of all respondents claimed to have attended church. This figure is considerably lower than Gallup estimates of 45 percent for Protestants nationally and 45 percent for all respondents in the Midwest. However, a 1991 telephone sample-drawn from all Ohio residents and the 1992 Greater Cincinnati Survey (Bishop 1992) yielded results similar to our Ashtabula poll: survey used, the attendance rate varies from 14 percent to 18 percent. 6 The response rate for persons contacted was relatively high for a telephone survey (71 percent). Working phone numbers were discarded after eight unsuccessful attempts to contact someone. 744 AMERICAN SOCIOLOGICAL REVIEW 36 percent of Protestants in Ohio and 37 percent of Protestants in the Cincinnati area said they attended church in the last seven days. Based on these data, we estimated the number of Protestants in Ashtabula County claiming church attendance at 22,830 in an average week. The next step in the research process was to obtain average attendance counts from all Protestant churches in Ashtabula County. No countywide church list existed, so through exhaustive procedures we located every church in the county. We began with the countywide telephone directory, denominational yearbooks, and newspaper advertisements. Then we examined tax-exempt property listings in township plat books. One of the authors visited every church-owned site that contained a structure that was not already identified as a church. The final step was to drive the length of every paved road (and many unpaved roads) in the county searching for churches and posted signs for churches. This process identified 159 Protestant churches located in Ashtabula County. We also located 13 Catholic churches and two spiritualist congregations. Our total of 172 Christian churches can be contrasted to the 128 total for Ashtabula County in Churches and Church Membership in the United States 1990 (Bradley et al. 1992). We found 44 churches that were not included in this compilation of county-by-county reports from 133 religious denominations or fellowships. Average attendance figures were obtained through denominational yearbooks, telephone interviews, letters, and church visits. We requested membership totals, definitions of membership, and average attendance estimates (including young children who were not in the worship service) from each church in the county. Average attendance counts were received from 137 Protestant churches. For the remaining 22 (mostly small) Protestant churches, attendance was estimated using the number of cars in the parking lots or actually counting persons attending Sunday services in February and March 1992. Estimates were based on persons-per-car ratios taken from similar churches whose attendance was known.7 The attendance total for all Protestant 7 The maximum ratio from a church whose attendance was known was three persons per car. This persons-per-car ratio was used to estimate atten- churches was 13,080. This total probably overestimates attendance in Ashtabula County. Based on our estimate of Protestants and our attendance count, we calculate that only 19.6 percent of Protestants attend church during an average week in Ashtabula County. Poll-based attendance estimates for Protestants from the Gallup Organization were 130 percent higher; estimates from our Ashtabula survey were 83 percent higher. Objections. Are the people polled in the Ashtabula sample counting something other than Saturday or Sunday worship as church attendance? To deal with this possibility, respondents were asked if their attendance referred to a worship service or some other type of meeting. Only four out of 143 Protestant attenders said they went to some other kind of meeting: Three attended committee meetings or special events at a church, while a fourth attended a nonreligious church-sponsored outing. Similarly, in the Greater Cincinnati Survey, only 5.9 percent of Protestant attenders said they attended "some other type of meeting," rather than a worship service. Respondents in Ashtabula County who had attended a worship service were asked what day they attended church. Only six attended worship on a day other than Saturday or Sunday. So, if we subtract persons who attended some other type of meeting and persons who attended worship on a weekday from the total attending, the percentage who claim to have attended church worship on Saturday or Sunday only drops to 33.2 percent—a figure that remains much higher than our 19.6 percent attendance count. A second objection is that we did not count Bible studies or very small "house churches" that would not have appeared in telephone directories, newspapers, or tax-exempt lists. Although this was a possibility, none of our Protestant respondents indicated belonging to or attending such a group. Respondents who attended church were asked where they attended. No one listed a small group Bible study or prayer group. No one listed anything other than a traditional church. Thus, we assume that persons in Ashtabula County either do not count attendance at such groups as church atten- dance for all nonreporting churches, even though it was clearly too high in some cases. WHAT POLLS DON'T SHOW ABOUT U.S. CHURCH ATTENDANCE 745 Table 1. Count of Total Church Attendance and Percentage Attending Church: Selected Catholic Dioceses in the United States, 1990 Total Church Attendance Diocese (1) Corrected Church Attendance (2) Total Population (3) Proportion Catholic (4) Estimated Catholic Population (5) Percent Attending Church (6) Baltimore 173,609 182,289 2,722,904 .259 705,232 25.8 Boston 455,837 455,837 a 3,754,239 .530 1,989,747 22.9 Chicago 600,350 600,350 a 5,621,485 .392 2,203,622 27.2 Cincinnati 234,562 246,290 2,805,557 .227 636,861 38.7 Fort Worth 65,618 70,042 a 2,060,943 .129 265,862 26.3 Harrisburg 115,844 121,636 1,867,124 .159 296,873 41.0 Indianapolis 99,069 b 108,471 a 2,201,503 .169 372,054 29.2 La Crosse 132,030 b 136,894 a 781,763 .408 318,959 42.9 Milwaukee 310,933 b 326,480 2,080,883 .408 849,000 38.5 Newark 306,374 307,651 a 2,650,504 .454 1,203,329 25.6 New York 453,861 453,861 a 5,096,274 .437 2,227,072 20.4 Omaha 122,719 128,855 775,037 .351 272,038 47.4 Philadelphia 409,946 427,593 a 3,728,909 .379 1,413,257 30.3 Pittsburgh 295,662 303,629 a 2,014,935 .415 836,198 36.3 Rockford 95,553 95,553 a 1,089,576 .288 313,798 30.5 Saginaw 70,896 70,896 a 685,082 .329 225,392 31.5 San Francisco 115,312 120,414 a 1,603,678 .341 546,854 22.0 Seattle 131,596 133,905 a 3,776,852 .177 668,503 20.0 Total — 4,290,646 — — 15,344,651 28.0 a Count inflated by the known proportion of congregations not reporting. b Count for Indianapolis is for 1989; for LaCrosse, 1991; for Milwaukee, 1992. dance, or that too few persons attend such groups to show up in a sample of 602 county residents. Catholics The Catholic churches in Ashtabula County did not collect church attendance data nor could they give reliable estimates of Mass attendance. Thus, we were unable to investigate the possibility that Catholics in this rural county also overstated their attendance (53 percent of Catholic respondents in our survey reported church attendance in the last seven days). However, many Catholic dioceses in the United States conduct a diocese-wide count of all individuals who attend a religious service on a given Fall weekend. Table 1 lists the dioceses for which we obtained attendance data.8 Although not a random 8 These 18 dioceses represent all dioceses that the Catholic Research Forum knew to be conducting sample of dioceses, this list includes dioceses from every region of the country except the Southeast, where Catholics constitute only between five and eight percent of the population. The list also includes counts from two of the three largest dioceses in the country, New York and Chicago. In most of these dioceses, Mass attendance is counted each weekend in October. These counts are then averaged to arrive at a single estimate of weekly attendance. In our judgment, the overall quality of these counts is quite high, even though there are differences in quality control.9 Column 1 of Table 1 pretne attendance count and for which we were able to obtain reliable data. 9 In most of these dioceses, the counting and reporting procedure is highly institutionalized with standard forms and computer programs for aggregating the data. Some dioceses have even instituted quality-control systems in which a representative of the diocese attends Mass on an appointed weekend to confirm participation in the count. All of the dio- 746 AMERICAN SOCIOLOGICAL REVIEW sents these counts for 1990 or, in three cases, a year near 1990.10 Column 2 presents a "corrected" attendance count that represents our effort to give the counts every possible chance to match survey-based estimates of attendance. First, for the four dioceses that interpolated in order to estimate counts for parishes that did not report a count in 1990, the "corrected attendance" is identical to the number in column 1. Second, for dioceses that did not interpolate, but for which we knew the proportion of parishes that did not report in 1990, we inflated the count in column 1 by this proportion. Third, for dioceses that did not interpolate and did not provide parish-level data, we inflated the count by 5 percent, which is slightly higher than the average percent missing data we observed (4 percent). The attendance data provide the numerator of a count-based attendance rate. The denominator is more difficult to obtain because the U.S. Census does not ask religious affiliation.11 Fortunately, there was a 1990 nationally representative survey of religious affiliation with a sample size large enough (N = 113,000) to reliably estimate, for each diocese, the propor- ceses ask parish priests to count rather than estimate attendance. The majority appear to comply with this request, particularly in dioceses where this has been a standard procedure for years. As was true for Protestants, however, the institutional politics of the attendance counts is such that a parish pastor has every incentive to inflate rather than deflate the actual count. 10 The three exceptions are Indianapolis, La Crosse, and Milwaukee, for which we report 1989, 1991, 1992 counts, respectively. Most of the diocesan counts include parish attendance only. However, when dioceses counted attendance in nonparish settings (e.g., hospitals, campuses), we included those numbers in the count. 11 The Official Catholic Directory provides an estimate of the number of Catholics in each diocese. For some dioceses, the reported number is based on a demographic method that compares the number of births and deaths in the relevant counties with the number of baptisms and funerals conducted by the church. For many dioceses, however, the number in the Directory represents the registered Catholics in the diocese, which usually is considerably smaller than the number of individuals who self-identify as Catholic (Celio 1993). The demographic method produces a better estimate, but still underestimates the population of self-identified Catholics. tion of the total adult population that is Catholic (Kosmin 1991). Because these data contained county codes, and because dioceses are usually coextensive with a set of counties, we were able to calculate the proportion Catholic for each diocese (column 4).12 Calculating count-based church attendance rates then is straightforward. Total population data for each diocese were taken from county-level 1990 U.S. Census reports (column 3). Next, we estimated the total number of Catholics in each diocese by multiplying the total population by the proportion Catholic (column 5).13 Finally, we calculated a count-based church attendance percentage by dividing the "corrected attendance" figures by the number of Catholics in each diocese (column 6). The results are clear: With the exception of Omaha, in every diocese for which we have good data the church attendance rate based on actual counts is significantly below the 51 percent rate reported by Gallup and a similar figure computed from General Social Survey data (Smith 1991).14 Most dramatic, when the data are aggregated the weekly church attendance rate for Catholics is only 28 percent. Recalling the various upward biases operating on the count data (e.g., the tendency for priests to overreport attendance, our overinflation of reported counts, etc.), it is difficult to avoid the conclusion that the actual church attendance rate for Catholics is substantially below 51 percent. The use of survey data to estimate the proportion of each diocese's population that is 12 These proportions are based on data weighted so that county samples matched census distributions by age, sex, and education. 13 The census figures refer to total population while the surveys are samples of adults. However, applying the proportion of Catholic adults to the total population will overestimate the Catholic population only if Catholic fertility is lower than non-Catholic fertility. Because only an overestimate of the Catholic population would bias the attendance rate in favor of our argument (by driving the rate down), this discrepancy from ideal demographic calculations is not a cause for concern. 14 The national survey rates pertain to adults, while the rates in column 6 pertain to the total population. This difference would increase the gap between the survey-based rate and the count-based rate only in the unlikely event that the attendance rate among those under age 18 is substantially lower than that among those 18 and over. WHAT POLLS DON'T SHOW ABOUT U.S. CHURCH ATTENDANCE 747 Baltimore Boston Chicago Cincinnati Fort Worth Harrisburg Indianapolis La Crosse Milwaukee Newark New York Omaha Philadelphia Pittsburgh Rockford Saginaw San Francisco Seattle 10 51% (Gallup estimate of weekly church attendance) i 95% Confidence Interval 20 30 40 Percent Attending Mass Weekly 50 60 Figure 1. Estimated Range in Percent Attending Mass Weekly: U.S. Catholic Dioceses, 1990 Catholic introduces a sampling error component into the count-based attendance rates in column 6. Figure 1 takes this component into account by presenting a range for each rate rather than a point estimate. These ranges were obtained by: (1) calculating the 95 percent confidence interval for the proportion Catholic in each diocese; (2) using the upper and lower limits of these intervals to calculate estimates of the Catholic population in each diocese; and (3) calculating two different attendance rates, one using the lower estimate of the Catholic population and one using the upper estimate. The intervals expressed in Figure 1 can be interpreted as the ranges, for each diocese, within which we are 95 percent confident that the count-based attendance rate falls. The dashed vertical line in Figure 1 indicates the rate from the Gallup poll. For only one diocese (Omaha) does the range include the 51 percent rate of weekly attendance; for the other 17 dioceses, the range is significantly below 51 percent. Table 2 presents, for selected dioceses, direct comparisons between self-reported attendance rates and attendance rates based on actual counts. For four dioceses, we located surveys that permit a more direct comparison between survey-based and count-based rates of attendance.15 In three cases, the self-reported attendance rate for Catholics is similar to or higher than the national rate (44.8 percent in New York, 48.5 percent in Chicago, and 59.3 percent in Cincinnati), while the count-based rate is substantially lower. These results increase our confidence that the observed gap between self-reported church attendance and actual church attendance is not an artifact of local variation. Only in San Francisco do the survey-based and count-based attendance rates overlap. However, the survey-based rates for San Francisco are from 20 years ago. This suggests that the gap between self-reported and actual church attendance may have increased over time. Our conclusion, then, is straightforward: Weekly church attendance by Catholics in the United States is approximately one-half what conventional wisdom takes it to be; the true rate is closer to 25 percent rather than 51 percent. Objections. Why is the attendance rate for Omaha so high? Many parishes in the Archdiocese of Omaha grossly overstate attendance 15 The authors commissioned the inclusion of worship attendance and church membership items on the 1992 Greater Cincinnati Survey. 748 AMERICAN SOCIOLOGICAL REVIEW Table 2. Percent Attending Mass Weekly, Based on Self-Reports and Actual Counts of Attendance: Selected Dioceses, 1990 Percent Attending Mass Weekly Self-Reporta Actual Countb Diocese (95% Confidence Interval) (Range from Figure 1) Chicago 45.7-----51.3 25.7-----28.9 New York 37.8-----51.8 19.4-----21.5 Cincinnati 54.0-----64.6 35.5-----42.4 San Francisco 33.2-----44.6 29.9-----36.5 a Sources: The Chicago Survey (Taylor 1991); The New York Daily News Easter Season Poll (New York Daily News 1986); The Greater Cincinnati Survey (Bishop 1992); and The Bay Area Survey II (Shanks 1972). b The ranges for Chicago, New York, and Cincinnati are from Figure 1, as the surveys are close enough to 1990 to make such a comparison valid. The Bay Area Survey, however, is for 1972. Fortunately, the archdiocese of San Francisco has conducted attendance counts since 1961. The ranges for San Francisco draw on population data from the 1970 Census, estimates of the Catholic population based on the proportion Catholic in the 1972 survey, and the 1972 count data in this diocese to calculate the rates. at Mass in their annual reports. We attended Saturday and Sunday Masses in five parishes where overcounting was suspected and, in each case, found attendance to be significantly lower than that reported to the diocese.16 One urban parish reported 750 persons attending on average, whereas we counted only 280 in all Masses on a sunny weekend five days before Christmas. A rural parish reported 1,150 persons attending—we counted only 595. A suburban parish reported attendance of almost 4,700—we counted 2,900.17 The attendance 16 we visited parishes where the ratio of attendees to the number of registered members was high or where attendance was reported in large even numbers (e.g., 500, 750, 2500). 17 The other two parishes overestimated attendance by 20 to 25 percent. Both were unusual in that many persons attending Mass were from other parishes in the city (according to priests and Archdiocese staff). One parish was known for its "quick" Mass (no more than 30 minutes), whereas the other had a heavily attended and lively "gospel Mass." We found no parishes that undercounted attendance. We met with four members of the Archdiocese staff to discuss problems of overestimation. They acknowledged that many priests probably es- rate for Catholics in Omaha may be above the 28 percent norm presented in Table 1, but it is much lower than 47 percent. What about members of the various Orthodox Catholic churches? Our attendance counts do not include such persons, even though many probably identify their religious preference as "Catholic" rather than "Orthodox" on social surveys. Data from the Yearbook of American and Canadian Churches (Bedell and Jones 1992) indicate that approximately 93 percent of American Catholics are Roman Catholics, and the figure may be even higher owing to possible overlap in reports from various Orthodox dioceses. Adjusting for Orthodox attenders increases our estimate of the percent of Catholics who attended church weekly by only a few percentage points. DISCUSSION Our results suggest that Protestant and Catholic church attendance is roughly one-half the levels reported by Gallup. What accounts for this gap? One interpretation pertains to the problem of nonresponse in survey sampling. Since the mid-1960s, changing lifestyles and other demographic factors have resulted in increasing nonresponse rates on telephone surveys (Frankel and Frankel 1987). Hard-to-reach populations include single persons and two-career, married-couple households—persons who are also less likely to attend church (Marler forthcoming). If these less religiously active persons are underrepresented in survey results, then self-reported church attendance would be artificially high. While nonresponse bias may contribute to high attendance rates, it is unlikely that sampling problems alone explain the levels of disparity found in this study. A second possibility is that survey respondents really do overreport their church attendance. Research frequently attributes overreporting (or underreporting) to "social desirability" factors. For example, a 1988 study found that youths underreported deviant behavior like substance abuse (Mensch and Kendel timated attendance rather than counting attenders. No verification procedures were used in the Archdiocese. There was some financial incentive for accurate reporting of registered members because of per member assessments by the diocese, but there were no incentives for restricting attendance estimates. WHAT POLLS DON'T SHOW ABOUT U.S. CHURCH ATTENDANCE 749 1988). Numerous studies of voting behavior have found that more people say they voted than actually voted (Parry and Crossley 1950; Traugott and Katosh 1979, 1981; Silver, Anderson, and Abramson 1986; Presser and Traugott 1992). In a New York sample, for instance, the gap between those who claimed to have voted and those who actually voted was 16 percentage points (Presser 1990). The interpretation is that people like to see themselves (or present themselves) as "better" than they are, based on a traditionally accepted social or moral norm. Indeed, in the case of voting behavior, nearly all the error is in the socially desirable direction (Presser and Traugott 1992, p.78). If survey respondents view regular church attendance as normative or view infrequent church attendance as deviant, they may be inclined to overreport their attendance. This tendency may be greater for persons who think they "ought" to attend church (Presser and Traugott 1992, p. 85). Even with the pressures of social desirability it could be argued that Gallup's time-specific question about church attendance, "in the last seven days" provides a corrective. After all, respondents are not asked to reveal annual patterns of church attendance—only last week's participation. Still, errors in recall are possible because people often remember events and behaviors as occurring more recently than they really did, a response problem called "telescoping" (Sudman and Bradburn 1982; Bradburn, Rips, and Shevell 1987, p. 160). Furthermore, evidence from cognitive studies of survey responses indicates that much more is involved in answering questions about the frequency of particular behaviors than simple recall of events (Blair and Burton 1987, p. 280). Burton and Blair (1991, p. 51) found that survey respondents answer such questions using information stored in long-term memory. Relevant information includes data about specific episodes or events (e.g., when they occurred, how enjoyable or salient they were, their frequency) as well as nonepisodic information like rules (e.g., "I go to church every Sunday"). Such information colors survey responses whether the episode in question occurred last week or last year. What is added to the social desirability thesis is personal desirability based on the "quality" of past episodes. In the case of church attendance, a combina- tion of positive episodes and strong internalized rules about church attendance could result in inflated reports. Whatever the reasons for overreporting, overreporting is a fact. If the gap between reported attendance and actual attendance is real, the implications for religion in America are profound. For instance, is America still uniquely "God's country" among industrialized nations? Although a 20 to 25 percent attendance figure puts the United States at about the same reported levels as Australians, Canadians, Belgians, and the Dutch (Kaldor 1987; Dobbelaere 1988; Lechner 1989; Bibby 1993), survey-based rates in other countries may also be inflated. Indeed, data from the United Kingdom suggest that a small gap exists between poll estimates and attendance counts. A 1989 census of church attendance in England revealed that 10 percent of the adult population actually attended church on "Census Sunday." This percentage is lower than a 1989 poll-based estimate of 14 percent (using General Social Survey-style response categories), but is identical to findings from a 1963 survey that asked about church attendance in the last week (Hastings and Hastings 1989; Brierly 1991).18 If individuals in other countries are less likely to overreport their church attendance than are Americans, then a different sort of "American exceptionalism" is at work. Americans may not differ much in terms of behavior, but rather in how they report that behavior.19 A striking feature of survey-based church attendance rates is their stability since World War II (Hout and Greeley 1987). Our research raises an intriguing question: Are behavioral patterns truly stable over the last 50 years, or has there been a decline in actual rates of church attendance? That is, has the gap between self-reported and actual attendance remained constant, or has it increased in recent decades? Our results cast considerable doubt on the cross-sectional accuracy of the survey rates, but changes in that accuracy are another 18 The 14 percent estimate was computed from the eight response categories using a formula similar to Smith's (1991) translation of General Social Survey data on church attendance. 19 Kaldor (1987, p. 20) also reported a gap between poll-based and count-based measures of church attendance for several denominations in Australia. Unfortunately, he gave no estimate of the size of the gap. 750 AMERICAN SOCIOLOGICAL REVIEW matter. We suspect that the actual attendance rate has declined since World War II, despite the fact that the survey rate remained basically stable. Trend data for various denominations suggest a decline in the attendance rate. Although old-line Protestant denominations have recorded declines in membership and church attendance, the number of Americans who call themselves Methodist, Lutheran, Episcopalian, Disciple, or Presbyterian has increased over the last 30 years.20 Declining attendance, coupled with an increasing constituency, results in a decline in the percentage of self-identified Protestants who are counted as attending Protestant churches. For instance, we estimate that the percentage of self-identified Presbyterians who attended Presbyterian churches declined from approximately 26 percent in the mid-1960s to around 16 percent in the late 1980s. Yet throughout this period, poll data indicated that Presbyterian church attendance averaged around 34 percent (Gallup Organization 1981). If self-reported church attendance was stable and actual church attendance declined, it would appear that the gap between what people do and what they say they do increased.21 These findings have important ramifications for the interpretation of poll data—especially poll data that attempt to measure behavior rather than attitudes. Although some gap between behavior and perceptions of behavior has always been assumed, we suggest that the gap, at least for church attendance, is substantial. Further, we suggest that the gap may not 20 Attendance as a percentage of membership among old-line denominations and the Southern Baptist Convention is a remarkably constant 35 to 40 percent. Smaller evangelical, Pentecostal, and Fundamentalist groups typically exhibit higher average attendance rates. Time-series data for a variety of denominations (from old-line to Pentecostal) indicate substantial stability in church attendance as a percent of membership from the early 1970s to the present. 21 This increasing gap is paralleled by changes in another traditional institution—the family. A large gap exists between what Americans say about their families and what they actually do—a gap that has widened in recent decades (Glenn 1992). In surveys over the last 30 years, traditional family relationships are consistently identified as the most important aspect of life. Yet during this period, the proportion of Americans who were satisfied with their marriages declined and the divorce rate rose. be constant—that perceptions of behavior are subject to change. If this is true, then poll data, particularly time-series poll data, should not be taken at face value. Clearly, social surveys measure perceptions of behavior; they do not measure actual behavior. To understand behavior and the norms that interpret behavior, attention should focus on the relationship between perception and reality.22 CONCLUSION We have shown that the church attendance rate is probably one-half what everyone thinks it is. But the practical difficulties involved with this research limited our data collection efforts. Although the evidence is compelling because it is so uniform, the fact remains that our data pertain to fewer than 20 Catholic dioceses and to Protestants in only one Ohio county. To confirm the existence of this "gap" and to determine if it has widened in recent decades, researchers should examine existing time-series data from local churches, denominations, regional polls, and religious censuses—any data that permit a comparison of poll-based and count-based measures of religious activity. In addition, the meaning of the gap should be explored at three levels. At the individual level, the roles of social norms and personal "rules" in answers to value-laden questions about behaviors like church attendance require additional investigation. At the group or congregation level, research should explore the processes of regular assembling for corporate worship. Church attendance occurs in group settings, and such settings resemble "regularly organized crowds." We need to know more about the intentions of people who frequent such gatherings and how characteristics of the group affect the behavior of individuals (e.g., see Schelling 1968; McPhail and Tucker 1990). Finally, at the societal level, comparisons of poll-based and count-based measures of church attendance from other nations would be valu- 22 A possible interpretation for such gaps is provided by Swidler's (1986, p. 280) description of cultural periods during which norms and values achieve a taken-for-granted quality and people live with "great discontinuity between talk and action." Norms remain the same, but behavior changes. As the gap between norms and behavior widens, tensions to bring values and behavior back in line increase. WHAT POLLS DONT SHOW ABOUT U.S. CHURCH ATTENDANCE 751 able. If changing cultural norms and social contexts affect the gap, we would expect substantial cross-national variation. These findings undoubtedly will stimulate the ongoing debate over secularization, particularly whether the cultural norm for church-going has persisted while the behavior has diminished. Strong religious expression continues in the United States, to be sure, but these data show that it is not as strong as previously thought. Meanwhile, perhaps there should be a moratorium on claims about the singularity of the United States in terms of church attendance. C. Kirk Hadaway is Secretary for Research and Evaluation at the United Church Board for Homeland Ministries, and Adjunct Faculty at Hartford Seminary. His research interests include religious identity and cultural change. He is co-director of a study of religious marginality. Recent publications include Church and Denominational Growth (Abingdon Press, 1993) and a longitudinal study of denominational mobility. Penny Long Marler is Assistant Professor of Religion at Samford University and holds a joint appointment in the Department of Sociology. She recently completed a three-year ethnographic study, funded by the Pew Charitable Trust, of a liberal Protestant seminary. She is currently co-director of a Lilly-funded study of religious marginality and identity. Mark Chaves is Assistant Professor of Sociology at the University of Notre Dame. In addition to interests in religious organizations and secularization theories, his recent research includes cross-national analyses of the relationship between church-state institutional arrangements and religious participation. Currently, he is working on a comparative study of conflicts over the ordination of women. REFERENCES Babbie, Earl. 1992. Practicing Social Research. 6th ed. Belmont, CA: Wadsworth. Bedell, Kenneth B. and Alice M. Jones, eds. 1992. Yearbook of American and Canadian Churches 1992. Nashville, TN: Abingdon Press. Bibby, Reginald W. 1993. "Religion in the Canadian 90s: The Paradox of Poverty and Potential." Pp. 278-92 in Church and Denominational Growth, edited by D. A. Roozen and C. K. Hadaway. Nashville, TN: Abingdon Press. Bishop, George F. 1992. The Greater Cincinnati Survey [MRDF]. Cincinnati, OH: Institute for Policy Research, University of Cincinnati [producer, distributor]. Blair, Edward and Scot Burton. 1987. "Cognitive Processes Used by Survey Respondents to Answer Behavioral Frequency Questions." Journal of Consumer Research 14: 280-88. Bradburn, Norman M, Lance J. Rips, and Steven K. Shevell. 1987. "Answering Autobiographical Questions: The Impact of Memory and Inference on Surveys." Science 236: 157-61. Bradley, Martin, Norman M. Green, Jr., Dale E. Jones, Mac Lynn, and Lou McNeil. 1992. Churches and Church Membership in the United States 1990. Atlanta, GA: Glenmary Research Center. Brierly, Peter. 1991. 'Christian' England. London, England: MARC Europe. Bruce, Steve. 1990. A House Divided: Protestantism, Schism,and Secularization. London, England: Routledge. Burton, Scot and Edward Blair. 1991. "Task Conditions, Response Formulation Processes, and Response Accuracy for Behavioral Frequency Questions in Surveys." Public Opinion Quarterly 55: 50-79. Celio, Mary Beth. 1993. "Catholics: Who, How Many and Where?" America 9 Jan., pp. 10-14. Davis, James A. and Tom W. Smith. 1990. General Social Surveys, 1972-1990 [MRDF]. Principal Investigator, James A. Davis; Director and Co-Principal Investigator, Tom W. Smith. NORC ed. Chicago: National Opinion Research Center [producer]. Storrs, CT: The Roper Center for Public Opinion Research, University of Connecticut [distributor]. Demerath, N. J., III. 1974. A Tottering Transcendence: Civil vs. Cultic Aspects of the Sacred. Indianapolis: Bobbs-Merrill Company. Dobbelaere, Karel. 1988. "Secularization, Polarization, Religious Involvement, and Religious Change in the Low Countries." Pp. 80-115 in World Catholicism in Transition, edited by T. M. Gannon. New York: MacMillan. Finke, Roger and Rodney Stark. 1992. The Churching of America, 1776-1990. New Brunswick, NJ: Rutgers University Press. Frankel, Martin R. and Lester R. Frankel. 1987. "Fifty Years of Survey Sampling in the United States." Public Opinion Quarterly 51: S127-S138. Gallup Organization. 1981. Religion in America. Princeton, N J: Princeton Religion Research Center. Glenn, Norval D. 1987. "Social Trends in the U.S.: Evidence from Sample Surveys." Public Opinion Quarterly 51: S109-S126. ------------. 1992. "What Does Family Mean?" American Demographics 14(6): 30-37. Greeley, Andrew M. 1989. Religious Change in America. Cambridge, MA: Harvard University Press. Goldman, Ari L. 1991. "Religion Notes." New York Times, 15 June, p. L10. 752 AMERICAN SOCIOLOGICAL REVIEW Hadaway, C. Kirk. 1989. "Will the Real Southern Baptist Please Stand Up: Methodological Problems in Surveying Southern Baptist Congregations and Members." Review of Religious Research 31: 149-61. Hastings, Elizabeth and Phillip Hastings. 1989. Index to International Public Opinion, 1988-1989. New York: Greenwood Press. Hatch, Nathan O. 1989. The Democratization of American Christianity. New Haven, CT: Yale University Press. Hout, Michael and Andrew M. Greeley. 1987. "The Center Doesn't Hold: Church Attendance in the United States, 1940-1984." American Sociological Review 52:325-45. Johnson, Allan G 1992. Human Arrangements: An Introduction to Sociology. 3d ed. Fort Worth, TX: Harcourt Brace Jovanovich. Johnson, Benton. 1993. "The Denominations: The Changing Map of Religious America." The Public Perspective 4(3): 3-6. Kaldor, Peter. 1987. Who Goes Where? Who Doesn't Care: Going to Church in Australia. Homebush West, NSW Australia: Lancer Books. Kornblum, William. 1991. Sociology In A Changing World. 2d ed. Fort Worth, TX: Holt, Rinehart and Winston. Kosmin, Barry A. 1991. The National Survey of Religious Identification: 1989-90 [MRDF]. New York: The Graduate School and University Center of the City University of New York [producer, distributor]. Lechner, Frank J. 1989. "Catholicism and Social Change in the Netherlands: A Case of Radical Secularization?" Journal for the Scientific Study of Religion 28: 136-47. Marler, Penny Long and C. Kirk Hadaway. 1992. "New Church Development and Denominational Growth (1950-1988): Symptom or Cause?" Pp. 29-72 in Research in the Social Scientific Study of Religion, vol. 4, edited by M. L. Lynn and D. O. Moberg. Greenwich, CT: J AI Press. Marler, Penny Long. Forthcoming. "Lost in the Fifties: The Changing Family and the Nostalgic Church." In Work, Family and Faith: New Patterns Among Old Institutions, edited by N. T. Ammerman and W. C. Roof. McPhail, Clark and Charles W. Tucker. 1990. "Purposive Collective Action." American Behavioral Scientist 34: 81-94. Mensch, Barbara S. and Denise B. Kandel. 1988. "Underreporting of Substance Use in a National Longitudinal Youth Cohort: Individual and Interviewer Effects." Public Opinion Quarterly 52: 100-24. New York Daily News. 1986. New York Daily News Easter Season Poll, 1986 [MRDF #8584]. New York: New York Daily News [producer]. Ann Arbor, MI: Inter-University Consortium for Political and Social Research [distributor]. Parry, Hugh and Helen Crossley. 1950. "Validity of Responses to Survey Questions." Public Opinion Quarterly 14: 61-80. Presser, Stanley. 1990. "Can Changes in Context Reduce Overreporting in Surveys?" Public Opinion Quarterly 54: 586-93. Presser, Stanley and Michael Traugott. 1992. "Little White Lies and Social Science Models." Public Opinion Quarterly 56: 77-86. Princeton Religion Research Center. 1992. "Church Attendance Constant." Emerging Trends 14(3): 4. Schelling, Thomas C. 1968. Micromotives and Macrobehavior. New York: W. W. Norton. Shanks, J. Merrill. 1972. Social Indicators: Bay Area Survey II [MRDF]. Ann Arbor, MI: Inter-University Consortium for Political and Social Research #8540 [distributor]. Silver, Brian D., Barbara A. Anderson, and Paul R. Abramson. 1986. "Who Overreports Voting?" American Political Science Review 80: 613-24. Smith, Tom. 1991. "Counting Flocks and Lost Sheep: Trends in Religious Preference Since World War II." GSS Social Change Report No. 26. (Revised Jan. 1991). Chicago, IL: National Opinion Research Center. Sudman, Seymour and Norman M. Bradburn. 1982. Asking Questions. San Francisco, CA: Jossey-Bass. Swidler, Ann. 1986. "Culture in Action: Symbols and Strategies." American Sociological Review 51:273-86. Taylor, D. Garth. 1991. The Chicago Survey [MRDF]. Chicago, IL: Metropolitan Chicago Information Center [distributor]. Thio, Alex. 1992. Sociology: An Introduction. 3d ed. New York: Harper Collins. Traugott, Michael and John P. Katosh. 1979. "Response Validity in Surveys of Voting Behavior." Public Opinion Quarterly 43: 359-77. ------------. 1981. "The Consequences of Validated and Self-Reported Voting Measures." Public Opinion Quarterly 45: 519-35. Wald, Kenneth D. 1987. Religion and Politics in the United States. New York: St. Martin's Press. Warner,R. Stephen. 1993. "Work in Progress Toward a New Paradigm for the Sociological Study of Religion in the United States." American Journal of Sociology 98: 1044-93.