Why More Americans Have No Religious Preference: Politics and Generations

by Michael Hout, Claude S. Fischer
Why More Americans Have No Religious Preference: Politics and Generations
Michael Hout, Claude S. Fischer
American Sociological Review
Start Page: 
End Page: 
Select license: 
Select License



University of California, Berkeley University of California, Berkeley

The proportion of Americans who reported no religious preference doubled from 7 percent to 14 percent in the 1990s. This dramatic change may have resulted from demographic shifts, increasing religious skepticism, or the mix of politics and reli- gion that characterized the 1990s. One demographic factor is the succession of generations; the percentage of adults who had been raised with no religion in- creased from 2 percent to 6 percent. Delayed marriage and parenthood also contrib- uted to the increase. Religious skepticism proved to be an unlikely explanation: Most people with no preference hold conventional religious beliefs, despite their alien- ation from organized religion. In fact, these "unchurched believers" made up most of the increase in the "no religion" preferences. Politics, too, was a significant factor: The increase in "no religion" responses was confined to political moderates and liberals; the religious preferences of political conservatives did not change. This political part of the increase in "nones" can be viewed as a symbolic statement against the Religious Right.

THE MINORITY of American adults who the 1990s were a time when religion was as- claim no religious preference doubled cendant in the United States (e.g., Kohut et from 7 percent in 1991, its level for almost al. 2000). For the preference for no religion 20 years, to an unprecedented 14 percent in to double in less than a decade is not only a 1998. This trend is likely to surprise the startlingly rapid social change in its own many researchers who have described Ameri- right but also a challenge to these widely held cans as especially religious (e.g., Caplow impressions of American culture. It may even 1985; Inglehart and Baker 2000), those who signal that century-old predictions of secu- included religiosity as part of "American larization may be (finally) coming true. They exceptionalism" (e.g., Greeley 1991; Lipset probably are not, though, as secularization 1996), and the many observers who thought proves to be inconsistent with some key evi- dence. The trend nonetheless points to im- portant changes in religion's role in the cul-

Direct all correspondence to Michael Hout,

tural milieu of fin-de-sibcle America, when

Survey Research Center, 2538 Channing Way, Berkeley, CA 94720-5 100 (mikehout@uclink4. many political controversies were about or berkeley.edu; fischerl @uclink4.berkeley.edu). entwined with religion (e.g., Williams 1997). This research is part of the "USA: A Century of We seek to explain why American adults Difference" project funded by the Russell Sage became increasingly likely to express no re- Foundation. The Survey Research Center, Uni-

ligious preference as the 1990s unfolded.

versity of California, Berkeley, and the National

Briefly summarized, we find that the increase

Science Foundation (SES-96-17727) provided

was not connected to a loss of religious pi-

additional support. We have benefited from the

ety, and that it was connected to politics. In

invaluable research assistance of Jon Stiles and

the 1990s many people who had weak attach-

from comments from Clem Brooks, Paul

ments to religion and either moderate or lib-

Burstein, James A. Davis, Otis Dudley Duncan, John Evans, Philip Gorski, Andrew Greeley, Ann eral political views found themselves at odds Swidler, and Melissa Wilde. with the conservative political agenda of the


REVIEW,2002,VOL.67 (APRIL:1 65-1 90)

Christian Right and reacted by renouncing their weak attachment to organized religion.

Our analysis proceeds in four steps: (1) We identify three theories on the doubling of "no religion" survey responses; (2) we examine the trend more closely and estab- lish that the change is a real historical change and not an artifact of survey meth- odology; (3) we assess secularization by ex- amining the beliefs, practices, and social ori- gins of people who have no religion; and (4) we quantify the contributions that demogra- phy, politics, and religious beliefs make to explaining the trend in religious preference and find that demographic changes and po- litical changes combine to account for it.


National surveys taken since the early 1990s show a sharp increase in the percentage of American adults who reported having no re- ligion.' The percentage doubled between 1990-1 99 1 and 1998-2000-from 7 percent to 14 percent-according to the General So- cial Survey (GSS), a large, nationally repre- sentative survey of American adults con- ducted annually or biennially from 1972 to 2000 (and contin~ing).~

After 17 years of no significant change in surveys, from 1974 to 199 1, this sudden increase is one of the most dramatic proportional changes in any of the

Most surveys include the word "preference" in the question about religion. Many fail to in- clude "no religion" as one of the suggested responses. Our principal data source is the General Social Survey (GSS) which asks: "What is your religious preference? Is it Protestant, Catholic, Jewish, some other religion, or no religion?"

We average two surveys together at each end of the decade to reduce the sampling error of our estimate of the change in religious preference. In this and all other calculations involving GSS data, we restrict attention to persons who are be- tween 25 and 74 years old to ensure that our co- hort comparisons are as unbiased as possible. We exclude the 1972 GSS because it did not include a question about religious origins-an important variable in the multivariate analysis to come. We also exclude persons who have data missing on their age, marital status, parenthood, or education because these are important variables in subse- quent analyses. Excluding these cases removes less than 1 percent of the cases that would other- wise have been available to estimate the trend.

variables measured by the GSS. Figure 1 shows the trend. The circles show the ob- served percentage in each survey, the thin vertical lines show the 95-percent confi- dence intervals (adjusted for survey sam- pling effect^),^ and the heavy dark line shows a spline function that smooths over the fluctuations from year to year that are at- tributable to amp ling.^ The total change be- tween 1991 and 2000 on the trend line is 8.5 percentage points. The trend through the 1990s would be clear even if we were to leave the spline function off the chart.

Other surveys confirm this increase. The National Election Study (NES) shows a rise from 8 to 13 percent from 1992 to 2000, and a 1996 study of religion and politics esti- mated that 14 percent of American adults had no religious preference (Kohut et al. 2000). Gallup is the one exception among major data sources; Gallup polls as late as the first quarter of 2001 continued to report that 8 percent of American adults claimed no reli- gi~n.~

While it is conceivable that Gallup is right and the other major surveys are wrong, we are inclined to accept the preponderance of evidence. which indicates an increase of 6

The adjustment takes account of the oversamples of African Americans included in the 1982 and 1987 GSSs and of the variations among the sampling frames (updated in 1983 and 1993) and, within sampling frames, variation among primary sampling units.

A spline function splices lines; specifically it joins together two lines with different slopes. The slopes are usually estimated using maximum- likelihood methods. We used a logistic regression of the log-odds on having no religious preference on a transformation of year that had the value of 0 for years 1973 through 1991 and (r-1991, where t is the year) for subsequent years. A par- tition of the total association between year and preferring no religion (L~ = 275.97; d.f. = 21; p < .01) shows 6 percent of the association is due to differences in the percentage with no religion from 1973 to 1991 (L2 = 16.66; d.f. = 16; p > .lo), and 94 percent of the association is due to the last 6 periods (L2 = 259.31; d.f. = 5; p < .01). A uniform association model that corresponds to the spline function in Figure 1 accounts for 92 percent of the association; its residual is not sig- nificant (L2 = 21.61; d.f. = 20: p > .lo).

The Gallup figure refers to a poll conducted February 19-21, 2001 and is reported on their website (www.gallup.com/poll).


,5~n,acm,mr~,mr~ , 8m, m m

8 8,

1972 1976 1980 1984 1988 1992 1996 2000


Figure 1. Percentage with No Religious Preference, by Year: Persons 25 to 74 Years Old, Born 1900 to 1974, GSS, 1973 to 2000

Note: Observed data are smoothed by a spline function hinged at 1991 (b = ,099). Cases missing data on religious origins, age, marital status, parenthood, or education are excluded; N = 31,678.

to 8 percentage points. One important dis- tinction between Gallup and the other sur- veys: Gallup interviewers accept "no reli- gion" as an answer but do not suggest it to their respondents; NES, Pew, and the GSS interviewers all read "or no religion" as a possible answer. Other differences between Gallup and the GSS include mode (Gallup is a telephone survey while the GSS is in-per- son) and response rate (Gallup's response rate is about 60 percent while the GSS's av- erages 77 percent).


There are at least three ways to explain the upsurge in "no religion" in the 1990s, and each has its own theoretical significance and implications. The three are not mutually ex- clusive; one or more might be true.


quently disengage from organized religion when they leave the family they grew up in and re-attach themselves about the time they start a family of their own (Glenn 1987; Greeley and Hout 1988; Roof 1993, chap. 6). The extended schooling and delayed family formation of recent cohorts may have con- tributed to increased nonpreference. Recent cohorts are more likely than those born 60 to 70 years earlier to have been raised with- out religion. As the less religious recent co- horts replace the more religious former co- horts, the religious attachment of the popu- lation will drop.

It seems unlikely that demography is the whole story, though. The religious change is more sudden than the longer-running and slower demographic trends so that it is un- likely that family events or cohort succes- sion can fully account for the sudden in- crease we seek to explain.


Demographic changes may be causing an in- The increase may reflect a suddenly acceler- crease in "no preference" responses. Reli- ated historical trend toward secularization. gion follows a family life cycle; people fre- The debate over whether modernization brings secularization is generations old in sociology. (Even the briefest bibliographies would include, in addition to the classic works of sociology's founders, sources from the 1990s such as Butler 1990; Bruce 1992; Finke and Stark 1992; Casanova 1994; Chaves 1994; Swatos and Christian0 1999.) Secularization seems to have been long de- layed in the United States, compared with other, mostly Protestant, English-speaking nations. Perhaps the 1990s upsurge heralds the coming of secularization at last.6

The term "secularization" is itself a sub- ject of debate. Some suggest that distinc- tions be made, especially between public and private religious expressions (Dobbelaere 198 1 ;Chaves 1994). Casanova (1994) cautions that the public-to-private transition may be reversible; there was as much evidence (from Spain, Brazil, Poland, and the United States) of religion moving from the private to the public sphere as there were clear indications that religion had "retreated" to the private sphere. Be- low, we look at the association between in- dividual piety and denominational identity as we attempt to assess whether the trend to no religious preference reflects seculariza- tion.


Controversies that connect politics and reli- giosity may be pushing some people away from organized religion. This is an old asso- ciation in many other nations, where to de- clare oneself religious is to take a political stance, typically a conservative one, while anticlericalism remains deeply ingrained in leftist politics (Lipset and Rokkan 1967; Casanova 1994; Gorski 2000). We have in mind the Dutch confessional parties, the Christian Democrats in Italy and Germany, and several parties in Israel including Shas and the National Religious Party. That kind of institutionalized connection between reli- giosity and party did not exist in the United States for much of the twentieth century (Dalton 1988; Lipset 1996), although reli- gion did affect voting (Manza and Brooks

Glenn (1987) treated earlier, smaller trends in "no preference" as the leading edge of secularization.

1997). With the emergence of the Religious Right as a force in Republican Party politics, a connection may have emerged (Casanova

1994). Research suggests that Americans did not become more polarized on most cultural matters in the last few decades, but it also suggests that religious identities and politi- cal party affiliations have become more closely aligned to positions on cultural mat- ters (like abortion) that touch on the public regulation of moral choices (DiMaggio, Evans, and Bryson 1996; Miller and Hoffman 1999; also see Evans 1996; Lay- man 1997; Hout 1999). Our conjecture is that the growing connection made in the press and in the Congress between Republi- cans and Christian evangelicals may have led Americans with moderate and liberal po- litical views to express their distance from the Religious Right by saying they prefer no religion.

These three explanations, demographics, secularization, and politics, guide and orga- nize our analysis. We do not rule out either complementary or overlapping effects from each in crafting our understanding of the in- crease in null religious preferences. For ex- ample, prolonged education may not only be delaying religious attachment, but it may also be increasing the likelihood of never at- taching (melding demographic and secular- izing effects). More subtly, the activism of some evangelical Christians may be simul- taneously increasing the religious vigor of fellow evangelicals who share their sympa- thy for a conservative social agenda and prompting a withdrawal from public reli- gious expression among other Protestants (and even some Catholics) who dissent from the conservative agenda.


Cohort succession implies that no individu- als changed their religious identification; it argues that the most religious cohorts sim- ply passed out of view (by death or exceed- ing our upper age limit) while less religious cohorts came into view for the first time in the 1990s. There are two parts to the cohort succession argument. One refers to the gross differences among cohorts that may be due to several factors; the second focuses on the growing fraction of recent cohorts that were raised with no religion.



Figure 2 presents GSS data on the trend in religious preference for each of five birth cohorts. The circles show observed percent- ages; the splines trace the percentages ex- pected from a logistic regression of having no religious preference on the transformed time variable described in connection with Figure 1 and four dummy variables that dis- tinguish among the five birth cohorts. The cohort succession argument implies that all the change should be evident in the contrast between the younger and older cohorts; no cohort should show a dramatic increase in the 1990s. If the life-cycle thesis is correct, then the two youngest cohorts should have high prevalence of no religion early and move downward toward the average as they age.

People from younger cohorts that entered the adult population after 1973 expressed significantly less attachment to organized re- ligion than did the cohorts they replaced. That much of the cohort succession argu- ment is correct. However, the younger co- horts also increased their preference for no religion by a wider percentage-point margin after 1991, thus widening, not narrowing, the gaps among cohorts.

Turning to the family life-cycle thesis, we see little evidence of its importance; there is a slight downturn in having no religious preference in the 1945-1959 cohort. A full decomposition of age, period, and cohort components of the patterns in Figure 2 might yield additional insight, but Figure 2 estab- lishes that the increase in the percentage of American adults with no religious prefer- ence after 1991 was not limited to people who were too young to have been inter- viewed before 1991. Cohort succession of- fers an important first step toward forming a fuller explanation of the upsurge in no reli- gious preference. The average change from 1991 to 2000 within cohorts is only 3.8 per- centage points-slightly more than half of the gross change (7.0 points). Thus genera- tional succession, by itself, increased the percentage of American adults with no reli- gion by between 3 and 3.5 percentage points.


There are two limits to the cohort succes- sion argument. (1) Cohorts replace one an- other gradually, so it would be reasonable to expect a gradual rise in the prevalence of no religious preference earlier than 199 1. Yet we have only the most tenuous evi- dence of change in the 1980s-1988 and 1989 are higher (but not significantly so) than other years in the decade. (2) The co- horts that had the highest percentage ex- pressing no preference before 1991 (the 1945-1959 and 1960-1974 cohorts) also experienced the most change between 1991 and 2000. So something else is also pushing more Americans toward having no religious preference.


About 6.5 percent of American adults in the late 1990s had been raised within no specific religious tradition, an increase from 2.5 per- cent in the early 1970s. This increase alone would be enough to raise the percentage of adults with no religious preference by that same 4 percentage points if nobody raised without religion acquired one in adulthood. In fact, many people raised without religion took up religion later in life. In cohorts born before 1945, a wide majority took up a reli- gion in adulthood despite their lack of reli- gious upbringing-72 percent of people born before 1945, raised without religion, and interviewed before 199 1 had a religious preference at the time of inter vie^.^ People from recent cohorts who were raised with- out religion were much less likely to affili- ate with a religion-half of those born be- tween 1945 and 1959 found a religion and only one-third of those born between 1960 and 1974 did. The multivariate analysis pre- sented below confirms that the increasing tendency for those raised without religious affiliation to stay that way is an important part of the explanation or part of the phe- nomenon to be explained.

Prior to the 1990s, marriage contributed to the tendency of people who were raised without religion to take up a religion in adulthood, as the religion they adopted was nearly always the religion of their spouse.

This calculation is made from among cases that we included in Figure 1.

Three trends converged to alter that pattern in the late 1990s. (1) Americans of all reli- gious origins married later (if at all), so a smaller fraction of adults raised without re- ligion had a spouse to conform (2) As their numbers grew, people who were raised without religion saw their chances of find- ing mates who likewise had no religious preference also increase, so that in more couples neither spouse has a religion for the other to conform to; the percentage of mar- ried persons raised without religion who had a spouse with no religion doubled from 16 to 32 percent from the early 1970s to the early 1990~~

(3) Finally, the pressure on people raised without religion to adopt their spouse's religion may have diminished as the proportion of married people raised with- out religion who preferred no religion at the time of interview rose from 27 percent in the 1970s to 51 percent in 1996-2000.1° This is consistent with a historical increase in the proportion of couples in which the spouses have different religious affiliations.

The foregoing evidence of growing inter- generational stability and homogamy among those raised outside a faith suggest that hav-

In the 1970s, 11 percent of adults raised with- out religion had never married; in the 1980s, 15 percent had never married; in the first half of the 1990s, the figure was 18 percent. In 1996-2000 the never-married reached 29 percent for persons with no religious upbringing. The conditional probability of having no religion given that one was raised with no religion and never married has not changed significantly over time; the chi- square tests for a table with six periods and a di- chotomy (no religion versus some religion at the time of interview) are X2 = 7.89 and L2 = 7.66

(d.f. = 5; p > .10 for each).

The GSS asked about spouse's religion repeatedly between 1974 and 1994. The chi-square tests for a table with five time periods and a di- chotomy (spouse currently prefers no religion versus spouse prefers some religion) are L2 =

10.59 and x2= 10.71 (d.f. = 4; p < .05 for each), for persons 25 to 74 years old and born 1900- 1970 who were raised with no religion.

lo We cannot restrict our attention to persons married to spouses who have a religion because the GSS contains no data on spouse's religious origin after 1994. The chi-square tests for a table with six time periods and a dichotomy (no cur- rent religion versus currently prefers some reli- gion) are L2= 21.51 and X2 = 21.39 (d.f. = 5;p < .O1 for each).


ing no religion is gaining momentum. Is the proportion of Americans with no religion likely to double again in the next genera- tion? The record of social forecasting is too humbling to give us any confidence in a pre- cise prediction at this point, but we can study the mathematical properties of the data in the hope of tendering a tentative answer. The cross-classification of religious origins by destinations obtained from the 1998-2000 GSS can be thought of as a "transition matrix" of probabilities that transform the reli- gious distribution of one generation into the distribution of the next generation. A com- mon result in linear algebra tells us that if such a transition matrix is "regular" and ap- plies for an indefinitely long time, eventu- ally the population comes into an equilib- rium, that is, the origin distribution exposed to the transition matrix yields a destination distribution that is identical to the origin dis- tribution." The United States was clearly far from religious equilibrium in 2000 because 14 percent of adults had no religious prefer- ence but 6.5 percent had no religious origin. What percentage of adults would have no re- ligious preference if the 2000 transition ma- trix were to hold sway until equilibrium is reached? We did the math and discovered that just under one-quarter of adults (24 per- cent) would ultimately have no religious preference if the most recent intergenera- tional pattern were to persist long enough to achieve equilibrium.'* Numerous caveats apply to a calculation such as this (e.g., each religion would have to have the same fertil- ity), but the main substantive implications are robust: (1) the momentum of recent growth in the percentage of adults with no religion is sufficient to raise the percentage higher even if no new changes add to the

l1 A transition matrix (T) is "regular" if it has no 0 entries in at least one of its positive integer powers including the initial matrix itself. For- mally, there exists some integer n = 1, . . . ,-for which element tij@) in Tnis not equal to 0 for all i, j (Kemeny, Snell, and Thompson 1966).

l2 We used a 6 x 6 transition matrix; the origin and destination categories were conservative Protestant, mainline Protestant, Catholic, Jewish, other religion, and no religion. The distinction between conservative and mainline Protestant is that defined by T. Smith (1990) and coded as the FUND variable in the GSS.

Protestant (N = 20,300) Catholic (N = 8,910)


Jewish (N = 693) Other (N = 570)

n: 0 0 0

20 ......................... .......... 20


'S2 0 0


0 00 .o 0 0 oo;/

-0... 10 ...................................


z = 0 O 0 vl p

0 0 OO 0



2 .F

O 0 0 0


I 0 -,s,, , ,*a,, , , , , , , , 1972 1982 1992 2002 1972 1982 1992 2002

Year Year

Figure 3. Percentage with No Religious Preference, by Year and Religious Origin: Persons 25 to 74 Years Old, Born 1900 to 1974, with a Religious Origin, GSS, 1973 to 2000

Note: Observed data smoothed by spline functions hinged at 1991. Each panel has its own best-fitting

spline function.

trend, but (2) the momentum is not sufficient to double the percentage of adults with no religious preference in the next generation (as it has in the most recent generation), let alone make no religion the largest prefer- ence. In other words, current patterns of in- tergenerational religious mobility imply that the most dramatic consequences of recent changes are already visible.

Although being raised without religion has spread and become more salient, it is not a complete explanation. Adults who were raised as Protestants or Catholics were sig- nificantly more likely to prefer no religion in 1998-2000 than in the past-up from 5 percent of people with Protestant roots in the 1970s to 11 percent in 1998-2000 and from 8 percent of Catholics in the 1970s to 11 per- cent in 1998-2000 (see Figure 3). Adults from the heterogeneous "other" origins probably increased their propensity to prefer no religion as we11.13 Jews are the only reli-

l3 The trend for "others" is not significantly different from 0, nor is it significantly different

gious group to show no sign of increased apostasy. Although the "falling-away" from childhood religions in the 1990s (except among Jews) is far more modest than the strong trends among people from a nonreli- gious background, it contributed almost as much to the overall growth in no religious preference because almost 95 percent of Americans were raised Christian or "other." Thus, a full explanation must also account for rising apostasy of all but the Jews.


If secularization accounts for the rise in no religious preference, then we should see (1) significant decreases in important beliefs such as belief in God or life after death be- fore or coincident with the trend to null pref-

from the Protestant trend. Thus we say the others "probably" increased their defections.

erences, and (2) evidence that people who have no religious preference also have no religious faith. We see neither. Survey data offer no evidence that Americans suddenly lost faith in the 1990s, or even raised new doubts. Furthermore, at most one-third of the people who prefer no religion are atheists or agnostics, and that fraction decreased slightly in the 1990s.


If secularization was the cause of the rising preference for no religion, then other reli- gious indicators would have suddenly turned against religion in the 1990s, too. That did not happen. The widely circulated Gallup poll data show no change since 1976 in its estimate that 95 percent of Americans be- lieve in "God or a universal spirit" (Bishop

1999). The six-response item developed by Glock and Stark (1965) shows a richer array of beliefs about God; among other things it allows us to distinguish atheists from agnos- tics, those who believe in a personal God from those who believe in a higher power, and believers who have doubts from believ- ers who are certain. The top panel of Table 1 shows the percentage distribution of Ameri- can adults on this item for the six years in which the GSS asked the question. The per- centage saying they do not believe in God increased from 1.5 percent to 3.0 percent; though statistically significant,14 this change is trivial compared with the increase in null religious preferences, and each increase in nonbelief is reversed, at least partially, in the next survey. Other beliefs-in heaven, hell, and religious miracles-did not change.15 Belief in life after death actually increased (Greeley and Hout 1999a), especially among people with no religious preference. Thus, the first condition for interpreting the in- crease in having no religious preference as secularization is not met.

The beliefs of people who prefer no reli- gion are, nonetheless, relevant. If people drop religion and then quit believing, for ex-

l4 The likelihood-ratio (L2) and Pearson (X2) chi-square tests are: L2 = 12.42 and X2 = 12.00

(d.f. = 5 and p < .05 for both).

l5 The data are available on the GSS website (www.icpsr.umich.edu/gss).


ample, the recent trends in religious prefer- ence might be a harbinger of future secular- ization rather than the result of ongoing secularization. The lower panel of Table 1 shows responses to the Glock and Stark (1965) item among the "nones." Belief in God among people with no religious prefer- ence appears to have increased, suggesting that most new "nones" are believers (al- though the changes over time are not statis- tically significant). Over two-thirds (68 per- cent) of adults with no religious preference expressed some belief in God or a higher power in 1998 or 2000; one-fourth said they do not doubt that God really exists. Less than one-third gave the atheist (16 percent) or agnostic (15 percent) response. While 31 percent is far more than the 4 percent atheist or agnostic among people who have a reli- gious preference,16 atheists and agnostics would have to be a strong majority-2 or 2.5 times more prevalent than they are now- before we could equate having no religious preference with being skeptical of religious beliefs.

Two other questions about belief in God in the 1998 GSS asked people to agree or disagree with the statements: "I believe that God watches over me" and "I believe in a God that concerns himself with each human being personally." These questions are more specific about what God is or does, and smaller percentages of the adults with no re- ligious preference agreed with them-59 percent and 32 percent, respectively-than indicated belief in response to the Glock and Stark question. Not surprisingly, adults with a religious preference are significantly more believing-94 percent and 80 percent, re- spectively. The question about God being concerned with each person was also asked in 1991. The percentage of people with no religion who agreed that God is concerned about people rose from 22 to 32 percent as their numbers grew. This is more evidence that the new "nones" were believers opting out of organized religion rather than people who lost faith as well as religion.

l6 Chi-square tests indicate that the differences between people with and without religious pref- erences are statistically significant at conven- tional levels: L~= 287.60 and x2= 372.84 (d.f. = 5; p < .05 for both).

Table 1. Percentage Distribution of Beliefs about God, by Year and Religious Preference: Persons 25

to 74 Years Old, GSS, 1988 to 2000


Belief 1988 1991 1993 1994 1998 2000

All Persons 25 to 74 Years Old

I don't believe in God.

I don't know whether there is a God and I 3.8 4.7 4.2 2.8 4.7

4.1 don't believe there is any way to find out.

I don't believe in a personal God, but I do 8.1 7.2 8.2 10.2 9.6

6.9 believe in a Higher Power of some kind.

I find myself believing in God some of 4.0 4.9 3.4 3.9 4.4 3.5 the time but not at others.

While I have doubts, I feel that I do 18.7 18.6 15.4 15.5 14.9 16.5 believe in God.

I know God really exists, and I have no 64.0 62.9 65.7 65.3 63.1 66.2 doubts about it.

Total percent
Number of cases

Tests of null hypothesis of no change (d.f. = 25): L2 = 44.28*; X2 = 43.73*

Persons 25 to 74 Years Old with No Religious Preference

I don't believe in God. 12.9 8.7 15.0 13.5 18.3 14.2

I don't know whether there is a God and I 21.2 31.9 17.7 14.4 14.6 15.0 don't believe there is any way to find out.

I don't believe in a personal God, but I do 25.9 18.8 23.9 29.8 24.8 18.6 believe in a Higher Power of some kind.

I find myself believing in God some of 8.2 8.7 7.1 2.9 7.3 2.7 the time but not at others.

While I have doubts, I feel that I do 12.9 18.8 14.2 14.4 15.3 20.4 believe in God.

I know God really exists, and I have no 18.8 13.0 22.1 25.0 19.7 29.2 doubts about it.

Total percent
Number of cases

Tests of null hypothesis of no change (d.f. = 25): L~ = 3 1.70; x2= 31.98

*p < .05

Table 2 compares the beliefs of those who 1974 to 1998 (Greeley and Hout 1999a); their did not have a religious preference with re- belief in heaven, hell, and miracles did not spect to beliefs in life after death, heaven, change significantly between 1991 and 1998. religious miracles, and hell in 199 1 and 1998. The vast majority of religiously identified



people believe in the each of these things. Persons with no religious preference are The most distinctive fact about the people more skeptical about these articles of reli- with no religious preference is their lack of gious faith, but over half believe in life after participation in organized religion (see Table death, and about a third believe in heaven and 3). Although two-thirds of people with a re- hell. Belief in life after death actually in- ligious preference attended church services creased among adults with no religion from several times a year or more, only 12 per- cent of persons with no religious preference attended more than once a year (a 55 per- centage-point gap). Almost two-thirds (64 percent) of those with no religious prefer- ence said that they never attend religious ser- vices. The data in Table 3 refer to the 1998 and 2000 GSS, but the same pattern is evi- dent throughout the 1990s.

Few people with no religious preference showed any sign of religious activity. Three- fourths did not read the Bible at home in the 12 months prior to their interview. Less than 3 percent belonged to church-affiliated orga- nizations.

But they do pray. On average, people with no religious preference prayed less often than others did, but 93 percent reported praying sometimes and 20 percent reported praying every day (see Table 3). Prayer among the nonaffiliated may have been more common in the late 1990s than it was in the mid-1960s. For example, a Gallup poll from 1965 asked how often people pray." The Gallup and GSS questions differ, so precise comparisons are not possible, but while only 12 percent of adults with no religious pref- erence had attended services in the prior three months, 60 percent said "yes" when asked if they "ever" prayed.'*

Our analysis of additional items about God from the 1998 GSS reveal that people who professed no religion relied on God in times of trouble in the three ways that people with a religion did.19 Most adults-with or without a religious preference-responded to trouble by thinking of themselves as part of a larger spiritual force, working together with God as partners, and looking to God for strength, support, and guidance at least some of the time. Neither the affiliated nor the nonaffiliated thought of hard times as a sign

l7 We know of no publications that analyze the Gallup respondents who had no religious prefer- ence. We obtained the original data from Univer- sity of California Data Archive and Technical Assistance (UCDATA) and made our own calcu- lations.

l8 Surveys get asymmetrical results from seemingly symmetrical comparisons, so we are reluctant to infer that 60 percent "ever" praying implies 40 percent "never" praying.

l9 Details were cut because of ASR space limi- tations. The referees and editors examined the full analysis. Contact the authors for details.

Table 2. Percentage Distribution of Those Holding Specific Religious Beliefs, by Year: Persons 25 to 74 Years Old with No Religious Preference, GSS, 1991

and 1998
Definitely or   Year
Probably Believes in: 1991 1998
Life after death 43 57
Heaven 43 42
Hell 30 36
Religious miracles 36 38
Number of cases 56 116

Note: Cases with missing data on any of the four belief items, religious origins, age, marital status, parenthood, or education are excluded. None of the changes from 1991 to 1998 is significant at the p < .05 level.

of God's punishment or that God has aban- doned them or that they try to make sense of bad situations without relying on God. The two groups differed significantly on four of these six items, but most people with no re- ligion nonetheless said that they relied at least somewhat on God in times of trouble.

The key fact, in sum, about people who express no religious preference is that most are believers of some sort, and many are quite conventional. Relatively few are secu- lar, agnostic, or atheist; most actually pray. Their most distinguishing feature is their avoidance of churches.


We might get a better sense of the "un- churched believers" we have just identified if we knew whether they were attached to other social institutions. Perhaps the increase in their numbers reflects a decline in social participation of many kinds (Putnam 2000). Conversely, people with no religious attach- ment may be more active in nonreligious pursuits.

Those with no preference are, it turns out, less socially active than are those who have a religious preference: One-third volun- teered for charity in 1997 compared with 42 percent of religiously affiliated ~mericans."

20 Not surprisingly, very few unaffiliated per-

Table 3. Percentage Distribution of Frequency of Attendance at Religious Services and Prayer, by Presence or Absence of Religious Preference: Persons 25 to 74 Years Old, GSS, 1998 and 2000

Religious Preference

Variable No Religion Religion

Frequency of Church Attendance

Never 63 13
Less than once a year 12 9
Once a year 10 12
Several times a year 6 13
Once a month 1 8
Two or 3 times a month 1 10
Nearly every week 1 6
Every week 2 20
More than once a week 1 9
Don't know 4 2
Total percent 100 100
Number of cases 656 3,989
Test of null difference   L2= 958.70
between those with and X2= 1,033.38*
without a religious (d.f. = 9)
Frequency of Prayer  
Never 7 0
Less than once a week 55 16
Once a week 5 8
Several times a week 13 15
Once a day 11 32
Several times a day 9 29
Don't know 0 0
Total percent 100 100
Number of cases 332 1,985
Test of null difference   L2 = 334.18
between those with and   X2= 399.26
without a religious   (d.f. = 5)
preference a  

Txcludes "don't know" responses

*p< .05

They belong to fewer nonreligious organiza- tions and are significantly less likely to vote than persons with a religious affiliation. People with no religious preference are not totally inactive-they are more likely to at-

sons volunteered at church-sponsored charities, but they are also significantly less likely to par- ticipate in secular charity.

tend concerts, see movies, and spend an evening with friends at a bar than are people with religious affiliations, but they are less likely to spend an evening with relatives or

neighbor^.^^ They also reported having fewer friends. In sum, people with no reli- gion are generally less attached to nonreli- gious organizations than are their religious counterparts, although perhaps they are like- lier to go out in the evenings.

We do not propose that a "bowling alone" (Putnam 2000) disengagement explains the decrease in religious preference. Among other comparative shortcomings, the two trends are out of synch. Most indicators of affiliation collected by Putnam (2000) began falling in the 1960s and 1970s; the trend in religious preference is a phenomenon of the 1990s. The pattern is, however, interesting as background. We have shown that the people with no religious affiliation are un- likely to have a compensating attachment to other social institutions but do participate in cultural consumption.


The 1998 GSS asked people whether they think of themselves as "religious" and also if they think of themselves as "spiritual." People who had a religious preference gave similar answers to both questions (Table 4). Over two-thirds described themselves as at least "moderately" religious and/or spiritual. People who had no religious preference overwhelmingly rejected the "religious" label; only 15 percent saw themselves as even moderately religious. But 40 percent de- scribed themselves as at least moderately spiritual. This difference between people who have a religion and those who do not confirms our sense that the nonreligious dis- sent from organized religion but maintain nonsecular beliefs and identities.

Why do so many believers claim no reli- gion? A few GSS items gauge attitudes to- ward organized religion, and responses to

21 Details for these activities are not reported here, but interested readers can find the relevant data on the data analysis website maintained by the UC-Berkeley Computer-Assisted Survey Methods program (csa.berkeley.edu:7502).

Table 4. Percentage Distribution for Self-Image as a Religious andlor Spiritual Person, by Presence or Absence of Religious Preference: Persons 25 to 74 Years Old, GSS, 1998

Self-Image Very religiouslspiritual Moderately religiouslspiritual Slightly religiouslspiritual Not religiouslspiritual Don't know

Total percent Number of cases

Test of null difference between those with and without a religious preference (d.f. = 3)a

a Excludes "don't know" responses. *p< .05

them show significant antipathy toward or- ganized religion among unchurched believ- ers. Two items ask about the confidence that people place in "churches and religious or- ganizations" or in "the people running orga- nized religion." We compared the answers given to these two questions by people with religious preferences, believers with no reli- gious preference, and nonbelievers (details available from the authors). Even among those people who preferred an organized re- ligion, the level of confidence in the churches and religious leaders was low (fewer than half expressed "a great deal" of confidence). But just over 10 percent of un- churched believers expressed "a great deal" confidence in religious leaders: People with no religious preference had significantly less confidence than people with religious pref- erences. But the differences between un- churched believers and nonbelievers were not statistically significant.

In 1998, the GSS also asked people whether they agreed with three statements about the effects of religion: "Looking around the world, religions bring more con- flict than peace"; "People with very strong religious beliefs are often too intolerant of others"; and "The U.S. would be a better country if religion had less influence." People who have no religious preference differ sharply from those who do on each of

Religious: Spiritual: Religious Preference Religious Preference No Religion Religion No Religion Religion

4 22 16 25 11 47 24 43 24 23 27 2 61 7 34 8

< 1 < 1 < 1 < 1

~~=259.24*; L2=78.40*; X2 = 100.01*

X2= 329.71'

these statements (see Table 5). By ratios of about 2:1, people who have no religious preference agree more with these critical statements than do other Americans. These items show that the unaffiliated are not merely uninvolved in organized religion- they have some antipathy to it.

Table 5. Percentage Distribution for Attitudes about Religions and Religious People: Persons 25 to 74 Years Old with No Religious Preference, GSS, 1998

Response Strongly agree Agree Neither agree nor disagree

Disagree Strongly disagree Total percent

Number of cases

1a 2 3

20 3 1 14 4 1 3 8 26 25 22 46

11 7 12 4 2 1

100 100 100 138 138 138

a "Looking around the world, religions bring more conflict than peace." "People with very strong religious beliefs are often too intolerant of others." "The U.S. would be a better country if religion had less influence." *This category also includes the "can't choose" response.


This analysis of the beliefs, practices, atti- tudes, and origins of persons who have no religious preferences have shown the major- ity to be "unchurched believersv-only a minority appear to be "atheist," "agnostic," or "skeptical." Many of them described themselves as spiritual but not religious. And while they did not attend religious services or read the Bible, these unchurched believers did pray and ask God's help in times of trouble. Their quarrel was not with God but with people running organized religion. They expressed little or no confidence in religious leaders and churches, and many saw them as the source of conflict and intolerance.

Our general description of nonreligious Americans is confirmed by several other sur- veys we examined. In a 1996 Gallup poll, for example, 70 percent of those who said they had no religion also said that they believed in God, and 30 percent were absolutely cer- tain of God; 54 percent prayed at least occa- sionally; 43 percent said that the Bible was inspired by or was the literal word of God. As in the GSS data, 70 percent of these re- spondents rarely if ever attended services, showing that it was this feature that most distinguished them. Again, "unchurched be- lievers" best describes this prominent fea- ture of the American religious landscape.22

Did the rapid increase in no religious pref- erence in the 1990s reflect an increase in un- believers, unchurched believers, or both? The best indicator of belief is the Glock and Stark item on belief in God (see Table l), which asks people to pick from among six statements the one that best describes their

22 Recall that people with no religious prefer- ence are a much smaller fraction of the 1996 Gallup data than the GSS reports. This may be due to question-wording-the Gallup question does not mention "no religion," but the GSS question does. If that is the only difference be- tween the two surveys, then the Gallup sample of people with no religion is probably composed of more "hard core" skeptics than the GSS sample is. Even with this bias, we find signifi- cant levels of belief among the Gallup "no reli- gion" respondents-a finding that builds confi- dence in our conclusion that "unchurched believ- ers" describes the majority of the adults with no religion.

view of God. For the purposes of this calcu- lation, we considered people to be believers if they expressed belief in God or a higher power (the third through sixth response op- tions); otherwise we consider them nonbe- lievers (the first two response options). Un- churched believers were 4.5 percent of adults in 1988-1991 and 7.9 percent in 1998-2000-a 3.4 percentage-point increase. Nonbelievers with no religious pref- erence were 3.7 percent of adults in 1988- 1991 and 5.3 percent in 1998-2000-a 1.6 percentage-point increase. Thus, two-thirds of the increase in preferring no religion was due to an increase in unchurched believers, and one-third was due to an increase in non- believers.

A longer time-series in the GSS bolsters our conclusion that a change in the religious preferences of believers in the 1990s con- tributed more to the increase in no religious preference than disbelief did. The GSS has asked about people's beliefs in an afterlife since 1973; it is a narrower belief than be- lieving in "God or a higher power," but with it we can see change over two more decades. Figure 4 shows that unchurched believers- people who prefer no religion but believe in life after death-have risen from 3 percent to 8 percent of adults, while nonbelievers have risen from 3.5 percent to 5 percent. This decomposition of the overall change is very close to the two-thirds versus one-third breakdown using the belief in "God or higher power" item. All the change occurred in the 1990s.

In sum, the secularization explanation for the growth in no religious preference is in- correct in so far as secularization means a decrease in belief and piety-the fraying of the "sacred canopy."


Few would be surprised to learn that religion has played a role in American politics throughout American history. From abolition to populism to the progressive era and on to the Civil Right Movement, religion provided a wellspring from which political movements could draw ideas and supporters. Here we are less concerned with what the trend toward not expressing a religious preference might do to future faith-based social move-

Unchurched Believers

0 Nonbelievers

1972 1982 1992 2002


Figure 4. Percentage Who Have No Religious Preference, by Belief in Life after Death and Year: Persons 25 to 74 Years Old, Born 1900 to 1974, GSS, 1973 to 2000

Note: Observed data smoothed by spline function hinged at 1991.

ments than we are with the possibility that the cause-effect relationship linking religion and politics might have become reciprocal in the 1990s. Although religion propelled some people into politics, the politicization of re- ligion might have caused people who dissent from the conservative agenda of vocal Chris- tian leaders to stop identifying with those religions.

In the 1990s, the Religious Right became a political factor for its critiques of what it saw as eroding family values. Religious leaders made pronouncements on abortion, gay rights, school prayer, and public spending on art they considered sexually explicit or anti- religious. Their power and the consequences are widely debated (e.g., DiMaggio et al. 1996; Evans 1996; Williams 1997; C. Smith 2000), and their numerical strength is easily exaggerated (T. Smith 1999; Greeley and Hout 1999b). But religious conservatives definitely received more attention in the press in the 1990s than during the earlier years covered by the GSS. Our search of ar- ticles in "major newspapers" compiled by the Lexis-Nexis service revealed that the num- ber of listings with the keywords "religious right" increased from 72 in 1980-1984 (that is 14 per year) to 1,736 in 1994-1996 (578 per year). It tapered off slightly to 1,017 articles in 1997-1999 (339 per year), and then spiked to 216 in just the first quarter of 2000 (864 for the year if the other three quarters kept pace with the first).23 In addition, con- siderable political emotion between 1992 and 2000 concerned moral issues that religious people care about-from the murder of abor- tioiproviders to President Clinton's personal life. We suggest that this religiously tinged political atmosphere not only brought some religious people out of apathy into politics but also pushed some moderate and liberal Americans with weak religious attachments away from religion.

Figure 5 presents the first evidence of the relationship between politics and increasing religious di~affiliation.~~

From 1974 to 2000,

23 The first quarter of 2000 was distinctive be- cause the presidential primaries were going on then. That link of politics with coverage is ex- actly the point we are making.

24 We classify people according to their politi- cal views as ascertained by a GSS question asked each year since 1974: "We hear a lot of talk these days about liberals and conservatives. I'm going to show you a seven-point scale on which the po- litical views that people might hold are arranged from extremely liberal-point 1-to extremely conservative-point 7. Where would you put yourself on this scale?" [The respondent is handed a card that corresponds to the wording of the question.] The "extreme" answers are rela- tively rare, so we combine responses 1 with 2 and 7 with 6 to avoid having to make inferences from

and controlling for marital status, liberals increased their preference for no religion by 11 percentage points, moderates who lean to liberal increased theirs by 7 percentage points, moderates increased theirs by 5 points, moderates who lean to conservative increased by 4 points, and conservatives in- creased by a statistically nonsignificant 1.7 points. In short, the significant increase in no religious preference was confined to liberals and moderates (even among married people), and the magnitude of the change in- creased with political distance from the right. As liberals were more likely than con- servatives to have no religious preference throughout the "stable" period from 1974 to

1991, the trend through the 1990s polarized the religious climate.

Of course, religious preference is usually thought of as a cause of political views (Manza and Brooks 1997). In this trend away from religion, we are inclined to see the usual order as being reversed by the politics of the 1990s. In defense of our in- terpretation, we cite the relative amounts of change in religious affiliations and political views. While the preference for no religion was doubling between 1991 and 2000, the liberal-conservative balance did not shift.25 If the usual causal direction were dominant in the 1990s, then the increase in Americans with no religious preference should have in- creased the prevalence of liberal political views. That did not happen. From this ob- servation, we arrive at the interpretation we favor: The disaffinity of liberals and moder- ates for the social agenda of the Religious Right led the ones who had weak religious attachments to disavow organized religion.

sparse data. We use the same missing data re- strictions as we used in making Figure 1. Further- more, we restrict attention to ever-married per- sons because singles are both more liberal and less likely to have a religious preference than other American adults are. To include singles in the calculations would exaggerate the political nature of the trend toward having no religious preference. We fitted separate spline functions to the trend for each category of political views.

25 The chi-square statistics are L2= 30.29 and x2= 30.22 (d.f. = 20; p = .06); adjusting for sam- pling design we get F = 1.31 (d.f. = 15.60, 2,854.26; p > .lo).



We have dismissed secularization and have seen how demography and politics each of- fer partial explanations for the sudden growth of no religious affiliation among Americans. For a more precise accounting, we now turn to multivariate analysis, which can weigh each potential contribution while statistically controlling for the other influ- ences. Our strategy is to start with a logistic regression of the propensity to claim no reli- gious preference. We use the spline function introduced in Figure 1 to model the sharp acceleration in having no religious prefer- ence after 1991. This approach gives us one number with which to track our success (or lack of success) in explaining the rise of the "nones7'-a logistic regression coefficient that measures the gross change over time in a bivariate regression and measures the un- explained trend once we add explanatory factors to the equation. If a factor or group of factors explains a substantial part of the increase after 1991, then the net spline coef- ficient will be noticeably smaller than its gross coefficient; if the model explains little or none of the increase, then the gross and net spline coefficients will be similar.

A conventional approach would measure the time effects with dummy variables. Al- though techniques exist for summarizing the combined effect of a list of dummy variables (e.g., Manza and Brooks 1997), our spline function gives mathematical form to our in- terest in increases through the 1990s. We use it because it is simple and because it is un- ambiguous in this way. The usual procedures for combining dummy variables' coefficients into a single index are indifferent to the rank order of the coefficients.

This logic of explanation applies to the demographic and secularization explana- tions but not to the political one. That is be- cause the demographic and secularization explanations propose that there is no trend within categories of their explanatory vari- ables-the preference for no religion grew because one cohort replaced another or be- cause people married later or quit believing in God. So the efficacy of the demographic and secularization explanations will show up as a net spline coefficient little different from 0. The political explanation is differ- ent, however; it says that political moderates and liberals changed but political conserva- tives did not. The efficacy of the political explanation will show up in the differences among the spline coefficients for political liberals, moderates, and conservatives.

The mathematical form of the logistic re- gression model (LRM) is important to keep in mind when interpreting the results. The LRM is a linear regression of the log-odds of having no religion on the explanatory vari- ables. This means that the spline coefficient measures the uniform effect of time after 1991 on the log-odds of having no religious preference, but that works out to an acceler- ating trend in the probability of having no religious preference. While each subsequent year after 1991 raises the expected log-odds by the same amount, its effect on the ex- pected percentage grows bigger over time.26 We saw this feature of the model in Figure 1, in which the gross spline coefficient of .099 implies an increase of 1.4 percentage points between 1991 and 1993, then an increase of

2.4 points between 1998 and 2000. This loglinear feature of the model becomes im- portant for our analysis because it implies a larger increase in the probability of having no religious preference for groups that had a relatively high probability of no religious preference prior to the 1990s and a smaller increase among groups that had a relatively low initial pr~bability.~~

Petersen (1985) and Long (1997:51-82) present methods making the results more interpretable.

26 The acceleration stops and begins to reverse once the expected probability reaches 50 percent, but that limit is not relevant to us because as the overall expected percentage with no religion does not reach that point (it comes to exceed 50 per- cent for those who were raised without religion).

27 We used this implication of the model to test whether it is an appropriate functional form for the data at hand. If the logistic model is appro- priate, then groups that had relatively high prob- abilities of preferring no religion in the 1970s should change more in the 1990s than groups with lower initial probabilities of preferring no religion. In six comparisons, the group with the higher initial percentage increased more in the 1990s than the group with the lower initial per-


Turning now to the actual multivariate analysis, we assess how changes in demo- graphic, social, and political variables that significantly affect respondents' odds of claiming no religion in the cross-section may have contributed to the increase over time in having no religious preference. As we saw in Figure 2, cohort differences-and any age differences they harbored-were appreciable. The cohort contrasts are not purely demographic, for cohort succession is a de- mographic process that can lead to cultural change as cohorts with one outlook are re- placed by new cohorts that view the world (or the hereafter) differently. Thus, the re- placement of more religious cohorts with less religious cohorts can potentially explain the increase in having no religious prefer- ence: The religious did not leave their churches-cohorts that were predominantly religious died or reached age 75 while less religious cohorts reached age 25. This would be the extreme form of the demographic ex- planation-nobody changes, the young en- tering the adult world are just different from the old ones they have replaced. Figure 2 showed us that the cohort replacement pro- cess accounts for some but not all of the in- crease in having no religious preferen~e.~~

centage. Specifically, the percentage preferring no religion increased more in the 1990s among men than women, among Pacific residents than southern residents, among whites than African Americans, among childless people than parents, among 20-to-29-year-olds than 50-to-59-year- olds, and among people with no religious up- bringing than people with a religious upbringing. These relative changes are captured well with the LRM that has one time effect and additive effects of each of these attributes, so we accept it as the appropriate functional form. The interaction be- tween political views and time that we use to test our political explanation is change in excess of that built into the LRM. That means we have set a high standard for our preferred explanation.

28 We initially thought to include dummy vari- ables for age groups in the multivariate analysis. But as we noted in discussing Figure 2, there is little indication that age effects matter so we avoid the complications of age-period-cohort analysis by leaving age out of the sociodemo- graphic model.

Substantively, the cohort differences reflect, we believe, the cultural experiences of com- ing of age in the 1960s.

It would be better, of course, if we could find the specific variables that make cohorts different from one another. Finding that co- hort differences "explain" a significant part of the trend in preferring no religion mean less progress in understanding religious preferences than finding specific substantive factors that account for the trend. Therefore, in building our sociodemographic model, we looked for variables that not only affected the propensity to have no religious prefer- ence but that also changed across cohorts.

To facilitate this search for specific vari- ables, we developed a spline function that expresses the cohort effects in a single coef- ficient, just as we did in Figure 1 for the pe- riod effects. The spline function we use equals 0 for cohorts born before 1935, is equal to the difference between the year of birth and 1934 for cohorts born 1935 to 1949, and stays equal to 15 for cohorts born 1950 to 1974.29

The cohort spline function implies that co- horts born prior to 1935 are more religious than those that came after, each cohort from 1935 to 1950 is increasingly less religious than the one right before it, and those born after 1950 are at the same (low) level of re- ligious attachment as the 1950 cohort. We think that this pattern of cohort differences reflects a "sixties" effect. Prior to the 1990s, the last well-documented increase in the per- centage of American adults with no religious preference occurred in the 1960s when it rose from 2 or 3 percent to 6 or 7 percent (Glenn 1987). The cohort differences in the GSS are consistent with the conjecture that people who were old enough in the 1960s to have well-established religious identities were less affected by the changes of those times than were cohorts just coming of age then. Thus, the cohorts that were over 30 years old in the 1960s less often expressed preference for no religion in the 1990s than did cohorts that were in their teens and twen-

29 We arrived at this specification after explor- atory analyses of single-year cohorts using lo- cally estimated regression techniques (Cleveland 1994) and five-year cohorts using dummy vari- ables in logistic regression analyses.


ties then. The "sixties effect" levels off but does not reverse for cohorts born after 1950 (they were less than 15 years old in 1965).

We have already documented the growing importance of religious origins. We incorpo- rate both the main effect of having been raised in a religious tradition and its recent increase in efficacy in the multivariate analysis. The main effect is a dummy vari- able equal to 1 for people raised with no re- ligion and 0 for those with a religious up- bringing. We experimented with models that treated the increase in the effect of religious origins as either a period effect or a cohort effect and found that an interaction effect that equals 0 for cohorts born prior to 1960 and increases linearly for cohorts born 1960-1974 works best.

We include family life-cycle events-mar- riage, divorce, remarriage, and parenthood- that underlie the correlation between age and having a religious preference (Greeley and Hout 1988). However, the arguments for how they affect religious preferences only apply to people who had a religious upbring- ing, so we specify the effects of family life cycle as operative for those who were raised in a religion but nil for those who were not.30 Between 1991 and 2000, each of these fac- tors except divorce changed in ways that can be expected to decrease the incidence of having no religious preference: Higher frac- tions of each cohort had been married and had become parents at the end of the 1990s than at the beginning.

Rising education has long been thought of as having a secularizing influence on those who were raised in a religious tradition. Al- though education increases certain kinds of religious beliefs and practices (e.g., belief in life after death among Catholics [Greeley and Hout 1999a]), it also contributes to the pro- pensity to claim no religion (e.g., Kohut et al. 2000). Americans were more highly edu-

30 This amounts to including an interaction ef- fect between marital status and religious origins and a three-way interaction involving marital sta- tus, parental status, and religious origins but not the main effects of marital status or parental sta- tus. Preliminary models that include all relevant effects confirm the supposition that marital sta- tus and parental status did not affect those who had no religious upbringing.

cated in 2000 than in 1991, thus rising edu- cation may have been a factor in the-1990s increase in preferring no religion. We include education in our sociodemographic model with the constraint that its effect applies only to those who had a religious upbringing.

Other sociodemographic factors-are im- portant for explaining cross-sectional varia- tion in religious preference but did not change much between 1991 and 2000. For example, men are far more likely than women to prefer no religion-18 percent for men compared with 11 percent for women in 1998-2000. Other sociodemographic fac- tors such as the racial and ethnic ancestry composition of the U.S. population changed but in ways that were unlikely to have con- tributed to the 1990s increase in having no religious preference. Both Asian Americans, a not-very-religious group (39 percent of Americans with Chinese or Japanese ances- try preferred no religion in 1998 and 2000), and Latinos, a more-religious-than-average group (11 percent preferred no religion in 1998 and 2000) increased in the population. Similarly, regional differences are quite large, but the "most" and "least" religious regions grew fastest between 1991 and 2000-washing out region as an explanatory factor. Even though we do not expect them to explain much of the trend or the cohort differences, we include dummy variables for being female, African American, Latino, Chinese-Japane~e,~'and living in the Mid- west, South, or Pacific states in the multi- variate models to assure that we are focus- ing as clearly as possible on net effects.

Table 6 presents our sociodemographic model (along with the standard errors ad- justed for sampling design). We limit atten- tion to persons born from 1900 to 1974 who were 25 to 74 years old at the time of inter- view and who were not missing data on any of the variables in the

31 These are not mutually exclusive categories; adults with Latino, Chinese, and Japanese ances- try can be of any race. Therefore, we enter each ancestry as an independent contrast. A with two of the three ancestries in his or her background would be scored on each, and his or her predicted log-odds on having no religion would be the sum of the separate effects.

32 Excluding missing data reduces the number of cases available for analysis by less than 1 percent.

I Table 6. Logistic Regression Coefficients from the Sociodemographic Model Predicting No Religious Preference: Persons 25 to 74 Years Old, GSS, 1973 to 2000

Independent Variable b A.S.E.a
1990s expansionb .038* (.009)
1960s1e&acyC ,050' (.005)
Raised with No Religion    
Main effect 2.062' (.164)
Interaction: 1960-1 974 .048* (.012)
woman -.679* (.044)


African American -.302* (.078)

Latino -.349* (.130)

Chinese or Japanese .907* (.165)


Northeast ,000

Midwest and Mountain -.lo2 (.089)

South -.425* (.091)

Pacific ,556' (.090)

Effects that Apply Only to Persons with Religious Upbringing

Education (years) .043* (.011)

Marital status:

Married once -.479* (.08 1)

Remarried -.030 (.108)

Widowed -.282 (.154)

Divorced or separated ,155 (.086)

Never married ,000 -

Parenthoodd -.407* (.062)

Intercept -2.804* (.165)

"Asymptotic standard errors, adjusted for survey effects; = 31,678.

The "1990s expansion" term is a spline func- tion based on the year the data were collected; it equals 0 for years 1973-1991 and equals t -1991 thereafter (i.e., for t = 1992, . . . ,2000).

The "1960s legacy" term is a spline function

based on the respondent's birth cohort; it equals 0 for cohorts born 1900-1934, it equals c -1934 for c = 1935, . .. , 1949 and equals 15 thereafter (i.e., for c = 1950, . . . , 1974).

The parenthood effect applies only to ever-mar- 'ied Persons.

*p< .05 (two-tailed tests)

The most important result in the socio- demographic model is the coefficient for the 1990s expansion (.038), compared with its gross effect (.099; see note to Figure 1). The difference between these two coefficients quantifies how much of the 1990s expansion can be attributed to changes in the variables in the sociodemographic model: [(.099 .038)/.099 = .62]. Thus 62 percent of the in- crease in the log-odds on having no religious preference is a result of sociodemographic changes. Translating this result into expected percentages, we find that more cohorts with a 1960s experience, more prevalent nonreli- gious origins, and delayed marriage and par- enthood together would have raised the per- centage of adults with no religion by 4 or 5 percentage points, even if there had been no period effect in the 1990s.

As noted above, the cohort effects are largely legacies of the 1960s' defections from organized religion. So, too, having no religious background reflects previous moves away from religion (on the part of the respondent's parents). The cohort most af- fected by the upsurge in nonreligious origins is the 1960-1974 cohort-the children of the cohorts most directly affected by the 1960s. This raises the prospect of a legacy for the recent changes, too-one that will be re- flected in data for the next generation. Be- fore going too far with that projection, how- ever, we note that origins are only imper- fectly related to parents' religions. The 1991 and 1998 GSSs included questions on par- ents' religions as well as the usual question about religious upbringing. To our surprise, only 68 percent of the people who said that both their mother and their father preferred no religion also said that they were brought up with no religion. Where the nearly one- third of people whose parents had no reli- gion got a religious upbringing we cannot say. Nor does it bear on our explanation of changes that have already occurred. We bring this up to show how hard it is to pre- dict the future from trends like these.


Our political hypothesis is that the actions of the Religious Right prompted political moderates and liberals to quit saying they had a religious preference. We now add po- litical effects to our sociodemographic model of having no religious preference to see if the net period and cohort effects differ for people with different political views. If, as Figure 5 indicated, more Americans had no religious preference in 2000 than in 1990 because moderates and liberals left orga- nized religion, then the trend and cohort co- efficients should be positive and statistically significant for moderates and liberals but close to 0 and not significant for conserva- tives. The results in the top panel of Table 7 affirm our political hypothesis: The 1990s expansion and 1960s legacy effects vary by political views as expected. The coefficients for the 1990s and 1960s terms are not sig- nificant for conservatives but are positive and significant for moderates and liberals. Both coefficients actually turn out to be larger for moderates than for liberals, but transforming them into expected changes in probabilities indicates more increase among liberals than moderates-just as we ob- served in Figure 5.33

We have argued that the political effects evident in Figure 5 and Table 7 operate through an a;ersion to the politics of the 1990s--a politics that made religious iden- tity seem like an endorsement of conserva- tive views. Although the results presented to this point are all consistent with our inter- pretation, they are indirect. We have yet to show that the Religious Right is the link. The divisive issues themselves, for example, abortion or gay rights, could be the link. The GSS includes good measures of these hot- button issues, but we do not include them in our analysis because we think they are more likely the consequence of changed religious identity than a cause of it. We look instead at peoples' attitudes about mixing politics and religion. The causal order is just as am- biguous with these items as with the specific issues, but getting the causal order right here is not essential. Knowing whether thoughts about religious politics are the cause or con- sequenceof affiliation on the relationship would not change our interpretation of the data. The 1991 and 1998 GSSs included three items that ask about the overlap be-

33 This is a consequence of the nonlinear rela- tionship between the expected logits and the ex- pected probabilities.

Table 7. Period and Cohort Spline Coefficients after Adding Political Views and Belief in God to Sociodemographic Model Predicting Preferring No Religion: Persons 25 to 74 Years Old and Born 1900 to 1974, GSS, 1974 to 2000

Category of Control Variable

Political views (N = 28,484)

Conservative Moderate Liberal

Belief in God (N = 6,590)

Skepticalb Has some belieF Believes without doubt

1990s Expansion Effect 1960s Legacy Effect b A.S.E.a b A.S.E.a

,014 (.025) ,015 (.012)

,058' (.010) .054* (.006)

.03 1 * (.015) .040* (.OlO)

.054* (.022) ,073' (.017)

.09 1 * (.025) ,020' (.020)

Note: The "1990s expansion" and "1960s legacy" terms are defined in Table 6. a Asymptotic standard errors, adjusted for survey effects. Skeptics chose either "I do not believe in God" or "I do not know if God exists and there is no way to find out" as the statement that comes closest to their belief about God. People with some belief are those who believe in a higher power or say that they do not believe some- times or that they have doubts about God's existence.

*p< .05 (two-tailed tests)

tween politics and religion: Question 1 asks whether church leaders should influence their followers' votes; Question 2 asks whether church leaders should attempt to in- fluence government decisions; and Question 3 asks whether churches have too much or too little power. Table 8 presents the distri- bution of responses for each year as well as the percentage with no religious preference within each response category for each year.

The first two items-about whether church leaders should influence their fol- lowers or political leaders-changed significantly between 1991 and 1998. The ex- treme positions grew while the middle shrunk. The third item did not change sig- nificantly. The third and fourth columns show that the growth in preferring no reli- gion is concentrated among the people who think that religious leaders should not influ- ence politics and among people who think that religion is too powerful. We cannot say anything about cause and effect from these tabulations, but our argument does not re- quire us to resolve that issue. Finding the link between having no religion and reject- ing clerical activism in politics supports our interpretation of the effect of political views on religious preference.


To put the secularization hypothesis on an

equal footing with our political hypothesis,

we repeat the multivariate analysis including

the six-statement belief-in-God item in the

sociodemographic model (the statements are

spelled out in Table If secularization

explains the increase in the no religious pref-

erence, then the 1990s expansion and 1960s

legacy coefficients-significant in the socio-

demographic model shown in Table 6-will

be close to 0 and statistically nonsignificant

once we control for belief in God. The sur-

prising results are presented in the lower

panel of Table 7.35The effects of the 1990s

34 We consider this to be the most generous test possible because it attributes all of the asso- ciation between belief and religious preference to the effect of disbelief on disaffiliation. Any cor- rection that purged the observed belief variable of the reciprocal effect of religious preference on belief would reduce the efficacy of the belief variable.

35 Because the question about belief in God was not asked until 1988, we lose the earliest cases. However, 1988 through 2000 still includes

I the period of the upsurge in no religious prefer-

to 74 Years Old, GSS, 1991 and 1998

Distribution of Responses Percentage With No Religion Item and Res~onses 1991 1998 1991 1998

"Religious leaders should not try to influence how people vote in elections. "

Strongly agree Agree Neither agree nor disagree Disagree Strongly disagree 3 1 35 16 15 2 36 30 17 12 4 9 3 5 8 19
Total percent Number of cases      
Test of null difference between      

survey years (d.f. = 4)

"Religious leaders should not try to influence g overnment decisions."

Strongly agree 22 30 11 Agree 3 1 29 4 Neither agree nor disagree 24 19 7 Disagree 20 17 6 Strongly disagree 3 6 6

Total percent Number of cases

Test of null difference between survey years (d.f. = 4)

"Do you think that churches and religious orga nizations in t his country have too much power or too little power?"

Far too much power 8 6 25

Too much power 16 17 10

About the right amount 57 55 5

Too little power 15 17 1

Far too little power 3 5 0

Total percent Number of cases

Test of null difference between survey years (d.f. = 4)

expansion and the 1960s legacy are not sig- significantly increased their disaffiliation. nificant only among atheists and agnostics Far from explaining the trend, controlling for (combined into a "skeptical" category for beliefs about God presents a new puzzle. The these purposes). Believers who have doubts increase in no religious preference is concen- and those that are certain of God's existence trated among those with the firmest beliefs,

ences. We considered some alternative specifica- an interaction term that allowed the effect of be- tions of how to include belief in the model. As liefs to increase in the 1990s; that term is highly the trend is limited to the 1990s, we introduced collinear with the period effect itself as only the

not among skeptics. These multivariate re- sults confirm our earlier conclusion that the rise in no religious preference responses stems from growing numbers of unchurched believers, not from a loss of religious belief. Substantively it means that, absent the de- mographic and political factors that have en- couraged disaffiliation, there would have been fewer adults with no religious prefer- ence in the late 1990s than before.


An important aspect of American religion changed dramatically in the 1990s. The mi- nority of adults who prefer no religion doubled from 7 percent to 14 percent be- tween 1991 and 2000. The very small (but growing) segment of the population raised without religion quit joining churches (for the most part), and between 5 and 7 percent of Americans raised in a Christian tradition, especially in cohorts that came of age in the 1960s and their offspring, left organized re- ligion. The Christian decline appears to have political content: Organized religion linked itself to a conservative social agenda in the 1990s, and that led some political moderates and liberals who had previously identified with the religion of their youth or their spouse's religion to declare that they have no religion. Had religion not become so politi- cized, these people would have gone on identifying as they had been and the percent- age of Americans preferring no religion would have risen only 3 or 4 percentage points.

In a country with as much emphasis on re- ligion as we see in the United States-today and throughout American history-the growing detachment of a significant portion of the adult population from organized religion is

1988 survey is not in the 1990s, so we dropped it. We also extended the analysis over a longer period by using the belief-in-life-after-death item instead of belief in God. Because belief in life after death has increased, especially among per- sons who had no religious preference (Greeley and Hout 1999a), controlling for it results in an even bigger coefficient for the period spline function than we report in Table 8. This is even stronger evidence that secularization (defined as the erosion of belief) is not responsible for the increase in no religion preference.

important. Equally important is the evidence that indicates how the new religious dissent- ers have distanced themselves from the churches, not from God. The data offer no support for conjecture that a long-expected secularization has finally asserted itself. The majority of adults who prefer no religion continue to believe in God and an afterlife. Few are atheists or agnostics. Most pray. Many reject the "religious" label, but they think of themselves as "spiritual." They sel- dom if ever attend religious services or read the Bible. In short, the critical feature of most such people is not their beliefs or personal piety but their estrangement from organized religion.

For 5 to 7 percent of American adults,36 holding no religious preference in the late 1990s was a political act, a dissent from the affinity that had emerged between conservative politics and organized religion. With- out panel data we cannot be sure, but we in- fer from the available data that people who changed from some religious preference to none rarely attended services anyway; they simply quit using the name of the denomi- nation they were raised in because the meaning of religious identification had changed for them. This account makes sense of our two key observations: (1) Po- litical conservatives did not change their re- ligious preferences, and (2) most people who prefer no religion have conventional religious beliefs, and many are even pri- vately pious or describe themselves as "spiritual."

Aside from helping to explain a major so- cial change, our analysis of politics and reli- gion underlines the point that the meanings expressed in identities only make sense in context. Even if Americans' religious beliefs and practices are stable-as we showed they basically are-the symbolic meaning of their religious identities can change. In this case, affirming religion increasingly carries the meaning of being conservative, much more so than in an earlier era. Furthermore, spe- cific historical events or shifts in the spirit of the times can change these defining con-

36 We include here some of the adults who were raised without religion because the politi- cal effects apply to them as well as to the ones who were raised in a religious tradition.

texts. (This view is in contrast to theories that describe a long-term unfolding of mo- dernity and secularization.) We have identi- fied two historical events here: the cultural turmoil of the 1960s (reflected in the cohort effects), and the politicization of religion in the 1990s (reflected in how only liberals and moderates changed). One cannot, therefore, simply extrapolate from the 1990s trend that the rise in nonpreference will continue. His- torical events, such as the attacks of Septem- ber 11, 2001, and new cultural movements may reverse these trends. Religious affilia- tion has increased before, not only in the eras of the Great Awakenings, but as recently as the 1940s and 1950s when, according to polling data from the period, both church at- tendance and religious beliefs increased (Glock and Stark 1965). On the other hand, if the identification of religious affiliation with political conservatism strengthens, then liberals' alienation from organized religion may become, as it has in many other nations, fully institutionalized.

Michael Hout is Professor of Sociology at the University of California, Berkeley. From 1997to 2001 he chaired the GSS Board of Overseers. He has published widely on demography, inequality, and, most recently, religion. With Claude S. Fischer and four other Berkeley colleagues, he co-authored Inequality by Design: Cracking the Bell Curve Myth (Princeton University Press, 1996). Currently, he and Fischer are collaborat- ing on a study of the major social and cultural trends of the twentieth century.

Claude S. Fischer is Professor of Sociology at the University of California, Berkeley and editor of the ASA's new magazine, Contexts. His most recent book before the co-authored Inequality by Design (Princeton University Press, 1996) is America Calling: A Social History of the Tele- phone to 1940 (University of California Press, 1992). In addition to his project with Hout, he is currently completing a book on American social history.


Bishop, George. 1999. "Trends: Americans' Be-

lief in God." Public Opinion Quarterly 63:

421-34. Bruce, Steve. 1992. "Pluralism and Religious Vi- tality." Pp. 170-94 in Religion and Moderniza- tion: Sociologists and Historians Debate the


Secularization Thesis, edited by S. Bruce. Oxford, England: Clarendon.

Butler, Jon. 1990. Awash in a Sea of Faith: Christianizing the American People. Cambridge, MA: Harvard University Press.

Caplow, Theodore. 1985. "Contrasting Trends in European and American Religion." Sociological Analysis 46: 101-08.

Casanova, JosC. 1994. Public Religion in the Modern World. Chicago, IL: University of Chicago Press.

Chaves, Mark. 1994. "Secularization as Declin- ing Religious Authority." Social Forces 72: 749-74.

Cleveland, William S. 1994. Visualizing Data. Summit, NJ: Hobart.

Dalton, Russell J. 1988. Citizen Politics in West- ern Democracies. Chatham, NJ: Chatham House.

DiMaggio, Paul, John H. Evans, and Bethany Bryson. 1996. "Have Americans' Social Atti- tudes Become More Polarized?" American Journal of Sociology 102:690-755.

Dobbelaere, Karol. 198 1. "Secularization: A Multi-Dimensional Concept." Current Sociol- ogy 29: 1-216.

Evans, John H. 1996. "Worldviews or Social Groups as the Source of Moral Value Atti- tudes: Implications for the Culture Wars The- sis." Sociological Forum 12:37 1-404.

Finke, Roger and Rodney Stark. 1992. The Churching of America, 1776-1990: Winners and Losers in Our Religious Economy. New Brunswick, NJ: Rutgers University Press.

Glenn, Norval D. 1987. "The Trend in 'No Reli- gion' Respondents to U.S. National Surveys, Late 1950s to Early 1980s." Public Opinion Quarterly 5 1 :293-3 14.

Glock, Charles Y. and Rodney Stark. 1965. Religion and Society in Tension. Chicago, IL: Rand McNall y.

Gorski, Philip. 2000. "The Mosaic Moment: An Early Modernist Critique of the Modernist Theory of Nationalism." American Journal of Sociology 105: 1428-68.

Greeley, Andrew M. 1991. Religious Change in the United States. Rev. ed. Cambridge, MA: Harvard University Press.

Greeley, Andrew M. and Michael Hout. 1988. "Musical Chairs: Patterns of Denominational Change in the United States, 1947-1986." Sociology and Social Research 72:75-86. . 1999a. "Americans' Increasing Belief in Life after Death." American Sociological Re- view 64:813-35. . 1999b. "Measuring the Strength of the Religious Right." Christian Century, August 25, pp. 810-15.

Hout, Michael. 1999. "Abortion Politics in the United States, 1972-94: From Single Issue to

Ideology." Gender Issues 17:3-32.

Inglehart, Ronald and Wayne E. Baker. 2000. "Modernization, Cultural Change, and the Per- sistence of Traditional Values." American So- ciological Review 65: 19-5 l.

Kemeny, John G., J. Laurie Snell, and Gerald L. Thompson. 1966. Introduction to Finite Math- ematics. Englewood Cliffs, NJ: Prentice Hall.

Kohut, Andrew, John C. Green, Scott Keeter, and Robert C. Toth. 2000. The Diminishing Divide: Religion's Changing Role in American Poli- tics. Washington, DC: Brookings Institution.

Layman, Geoffrey C. 1997. "Religion and Politi- cal Behavior in the United States." Public Opinion Quarterly 61:288-3 16.

Lipset, Seymour Martin 1996. American Excep- tionalism: A Double-Edged Sword. New York:

W.W. Norton.

Lipset, Seymour Martin and Stein Rokkan. 1967. "Cleavage Structures, Party Systems, and Voter Alignments," Pp. 1-40 in Party Systems and Voter Alignments, edited by S.M. Lipset and S. Rokkan. New York: Free Press.

Long, J. Scott. 1997. Regression Models for Cat- egorical and Limited Dependent Variables.

Thousand Oaks, CA: Sage. Manza, Jeff and Clem Brooks. 1997. "The Reli- gious Factor in U.S. Presidential Elections:

1960-1992."American Journal of Sociology 103:38-81.

Miller, Alan S. and John P. Hoffman. 1999. "The Growing Divisiveness: Culture Wars or a War of Words?" Social Forces 78:721-52.

Petersen, Trond. 1985. "A Comment on Present- ing Results from Logit and Probit Models."

American Sociological Review 50: 130-3 1. Putnam, Robert. 2000. Bowling Alone: The Col- lapse and Revival of American Community.

New York: Simon and Schuster.

Roof, Wade Clark. 1993. Generation of Seekers: Spiritual Journeys of the Baby Boom Genera- tion. San Francisco, CA: Harper Collins.

Smith, Christian. 2000. Christian America? What Evangelicals Really Want. Berkeley, CA: Uni- versity of California Press.

Smith, Tom W. 1999. "The Religious Right and Anti-Semitism." Review of Religious Research 40:244-58.

Swatos, William H., Jr. and Kevin J. Christiano. 1999. "Secularization Theory: The Course of a Concept." Sociology of Religion 60:209-28.

Williams, Rhys H. 1997. "Culture Wars, Social Movements, and Institutional Politics." Pp. 283-96 in Culture Wars in American Politics, edited by Rhys H. Williams. New York: Aldine de Gruyter.

  • Recommend Us