A study of subjective self-identity clearly does not purport to be a census of religious affiliation or denominational membership. It is different from the normative view provided by the religious institutions themselves. The data will show that in 1990 the American public favored denominational and church names more than theological terms such as fundamentalists. In itself, this is an important finding for students of religious life and of American society. It will also be seen that the general public is somewhat confused by religious “brand names” and the complex world of church labeling. Clearly, many people do not understand the precision of theologians, nor do they always subscribe to the blandishments of denominational officialdom. Categorizing religious groups is a complex and arcane art. The NSRI makes a significant contribution by presenting a large sample size reflecting the true richness of the country’s religious tapestry, and the national coverage necessary to provide highquality statistics on what is actually happening around us.
Perhaps our most important innovation as an academically based project was our decision to use a commercial market-research firm and its omnibus market-research survey for our data collection. In recent years the commercial polling organizations have made rapid strides and large-scale investment in both computerized equipment and quality-control interviewing techniques. We chose an industry-leader, the ICR Survey Research Group of Media, Pennsylvania, which operates the twice weekly EXCEL national representative sample survey directed at 1,000 households. In effect, this involved purchasing time for a question to be included in the survey alongside a variety of other constantly changing questions on topics as varied as cable TV use, preference for consumer items, and social or political issues. Our question was included in 113 rounds over 56 weeks in the period April 1989-April 1990.
In any type of research it is important that the topic and the questions be introduced appropriately and do not appear too intrusive or threatening. We chose the multipurpose questionnaire because we believed we would gain better results than if we merely called up the general public and asked them to participate in a purely religious survey. Our question in the multipurpose context often appeared appropriate when it was positioned alongside others on life-style, attitudes, food preferences, or public policy issues. As a result, once a respondent agreed to be interviewed, the actual item refusal rate to our religious question was a very low 2.3%.
In addition to easing religious answers from the respondent, the multi-question, census-like format supplied us with detailed socio-demographic information alongside religious identification. For example, the respondent’s age sex, race, marital status, education, and political-party affiliation allowed us to see the relationship among religion and ethnicity, politics and life-style. Household characteristics such as income, family size, and geographical location provided the opportunity to analyze religious identification in light of social status, residency and economics. The unique value of the NSRI was appreciated by the authors of the study Churches and Church Membership in the United States 1990, who used this data source to obtain the first estimates of black Baptists by county throughout the nation.
One of the negatives of using commercial market research was that the response rate was lower than most social surveys would usually receive. Even after four attempts to contact respondents, the cooperation rate was only 50%. However, this did not pose a problem because we intended to carry out a massive number of interviews. We also had the benefit of access to the latest weighting and other statistical techniques, which allowed our statistician, Dale Kulp, the president of the Marketing Systems Group in Philadelphia, to weight, or recalculate, our findings in proportion to the current profile of the known composition of U.S. households and the total population. This kind of adjustment is necessary because, despite all our best efforts to obtain a representative sample, the experience of interviewing the American public shows that certain types of people are more reluctant than others to participate in surveys. Elderly women, especially those living alone, and inner-city residents are particularly hard to interview, whereas young families, especially in small towns in the Midwest, are highly cooperative. Therefore, it is necessary to adjust the raw figures accordingly. The weighting also took into account the probabilities of disproportionate household selection due to the number of separate telephone lines in the home. The most important technical innovation that made this mammoth task possible was the use of ICR’s Computer-Assisted Telephone Interviewing (CATI) system. The computer randomly chose telephone numbers across the United States in proportion to the number of residential telephone lines. This technique, known as Random Digit Dialing (RDD), creates a geographically representative sample. We used the GENYSYS system to create 113 individual national samples. Basically, GENYSYS chooses an area code, then a telephone exchange, and then the first two digits of a four-digit suffix. Then the computer randomly dials the last two numbers. In this manner a sample frame with very fine geographical stratification is created. The respondent within the household was a randomly chosen adult picked according to who had the last birthday.
The NSRI is much more accurate and detailed as to geographical distribution than any previous survey because of the number of interviews that took place within each state. For instance, there were 5,168 completed interviews in Michigan, and even small or sparsely populated states such as Vermont and Idaho had 297 and 531 interviews respectively within a carefully structured scientific sampling design. The best way to test and corroborate the validity of the results is in a small state with a stable population. Rhode Island is an obvious choice since its population totals almost exactly 1 million, which makes translation of percentages into real numbers very easy. The NSRI reported that Rhode Island was 61.7% Catholic; that’s about 617,000 people. The number used by well-informed diocesan researchers, using a comparison of parish and state demographic data, was 624,000 in 1989. This is a remarkable level of precision, considering that there were only 810 NSRI interviews in the state.
Nationally, the sheer number of interviews and careful research design resulted in a high level of precision. Nevertheless, an error factor was created. Standard error estimates for our overall national sample show that we can be 95% confident that the figures we obtained have an error margin, plus or minus, of less than 0.2%. This means, for example, that we are more than 95% certain that the figure for Catholics is in the range of 26.0% to 26.4% for the U.S. population. We must admit, however, that for a small group such as Unitarians, whom we found constitute 0.3% of the American population, the actual figure could be subject to greater proportionate fluctuation, even though the standard error for Unitarians is much less than 0.2%.
Sources of Possible Inaccuracy
Nearly every data collection concerning religion in America has had to face enormous problems of defining its terms and categories. This is due to the dynamic nature of religion in America and constant schisms and unions within Protestant denominations.
New religions arise and others fade away. Navigating the complex byways of the ever changing religious scene is fraught with difficulties. For instance, at the time of the Korean War, the Department of Defense listed 40 denominations under the general heading of Protestant chaplaincy, while today 260 groups are authorized.
Being asked to state their religion in a few words over a telephone is a problem for many people since they may find it difficult to classify their local church according to recognized denominational labels. In what denomination should we place “Christian Free-will Baptist Church of Christ?” How should we expect someone who attends the “First Church of Christ (Congregational)” to know that he is really a member of the United Church of Christ? In addition, some local congregations are members of more than one denomination (for instance, of both the main Presbyterian and Methodist bodies).
The medium of communication, the telephone, has its restrictions. It is very likely that our Mennonite total does not include the estimated 90,000 Old Order Amish who do not possess telephones. It is also possible that our methodology tended to undercount groups that live in communal settings. Such a lapse was suggested to us by the Baha’i religious organization, which operates several collective settlements in the Carolinas and claims 110,000 adherents nationwide, as a partial reason for the undercount in the NSRI (we found only 28,000). It is likely that many recent immigrants from Iran, with their expertise of persecution, were reluctant to reveal their Baha’i faith in a telephone interview. As we explained earlier, the range of miscount for small groups of 200,000 or fewer adherents could be proportionately very large. Quite possibly, the NSRI underestimated many small groups and overinflated others.
Our weighting system, which adjusted for race and inner-city dwelling, could not compensate for characteristics that were neither geographic nor demographic in nature. We had no ability to contact the homeless or the institutionalized population. Our reliance on interviewing in English may have led to underestimates of religious groups containing high proportions of recent immigrants. It is also possible, as mentioned above, that recent immigrants from societies with deep religious groups probably went into the “miscellaneous” category under “Other Religions” since they gave unclear or complicated answers. For instance, “Sunni” or “Brahmin” might not always have been correctly categorized by our interviewers under the Muslim or Hindu designation.
However, there are other sources of information that allow us to test and adjust our findings. When our Muslim figure appeared much lower than some estimates ranging from 2 million to 6 million people; we looked to the U.S. Census as well as other sources to check our accuracy. Without much evidence or objective scientific measurement, some scholars, media people, and Muslim community representatives claimed that adherents to Islam had grown enormously in the last few years as a result of massive immigration from the Middle East. The 1990 census found that only 1,067,000 people (0.4% of the U.S. population) reported their ancestry as “Arab” or from an Arab nation. Even before the census figures were available, the Princeton Religious Research Center, utilizing accumulated Gallup polls, attempted to verify our findings. “Where Are the Muslims in the United States?” an article in the September 1991 issue of the center’s publication, Emerging Trends, reported that their findings almost corroborated our own – only 0.2% nationally – and concluded that enthusiastic supporters have in the past few years greatly exaggerated the number of Muslims. Nevertheless, we believe that there are specific research design factors that resulted in an undercount; the Muslim population is probably around 0.5% of the U.S. population, and not the initial NSRI figure 0.3%.