The importance of brand and competition in defining U.S. religious markets.

1. Introduction

Over the past 40 years, economists have vastly expanded their study of behavior in markets where there is no explicit trade of goods or services. One of the most fundamental and ubiquitous of these markets is that of religious behavior, where economists and other social scientists have explored a variety of influences on religious participation. At the outset of this research program, economic models of religious choice borrowed from long-used choice-theoretic concepts such as the law of demand or the cost of time. More recent contributions have included household production models taking account of human capital or the role of market-level forces, particularly religious competition. Prior empirical research has resulted in the identification of a variety of determinants of religious participation. These include income, education, race, and the degree of monopolization or competitiveness in the religious market.

As the empirical application and testing of these models has proceeded, researchers have been aware of problems in measuring various aspects of religious choice such as religious participation or "religiosity." At the most general level, religiosity may imply a difference in a secular versus religious worldview. At the level of actual observation, it often measures how frequently one attends church. While some of the basic measurement issues related to religious choice and behavior may be beyond the power of researchers to investigate, one important aspect is not but has largely been overlooked, that is, the type or brand of religious choice. By and large, empirical research has skirted this important definitional issue and treated different types or brands of religious choice much like one would treat different brands of carbonated soft drinks. Using such a broad, homogeneous definition of the religious product implicitly assumes that economic and demographic characteristics exert similar influences on different types or "brands" of religious participation and, thereby, assumes that such branding is not of empirical significance in defining religious markets and behavior within them. (1) Moreover, some tension has existed in prior empirical work as to whether religious competition is an exogenous factor--the common assumption--or whether it helps to define the choice variable. (2)

Our central question is whether the same processes are driving religiosity across brands. We ask, do consumers of religion respond differently to key economic variables such as income, religious competition, or other factors based on the brand of religion being consumed? Likewise, do some economic variables exert their influence independently of brand? An important part of our investigation centers on how to measure brand. We first rely on two indicators of brand: broad denominational categories and intensity of participation in religious activity. The denominational categories are mainline Protestant, evangelical Protestant, and Catholic. The intensity breakdown is based on geographic regions of low, moderate, and high religious participation. In addition, we also examine whether the degree of religious competition itself helps to define brand.

In the next section, we develop theoretical explanations for variation in overall religious participation as well as by brand and discuss prior empirical work in this area. The third section contains the data description and our empirical framework for explaining religious participation. In the fourth section, we estimate the empirical model using both the denominational and intensity brand indicators. In the final section, we condense and summarize the key findings.

2. Theoretical and Empirical Background

Analytical study of religious choice from an economic viewpoint dates at least to Adam Smith, who drew attention to the importance of competition and monopolization on religious choice. (3) In more recent times, household production models have become the theoretical standard. Azzi and Ehrenberg (1975) link age with religious participation using a household production model with a lifetime budget constraint. Iannocone (1984, 1990) and Neuman (1986) have developed consumption capital models of religion, building on the work of Stigler and Becker (1977). Iannocone (1998) summarizes these theoretical models. An adapted version of this is given in Equation 1:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (1)

Here, utility for consumer j is determined by consumption of i different generic, secular Z-goods and afterlife consumption, [A.sub.j], and is subject to the traditional time and budget constraints, T and Y. The Z-goods are produced by consumers combining time, [T.sub.i], in the production process with market goods, [X.sub.i]. Afterlife consumption for consumer j is produced by k different religious activities, [R.sub.k], and these activities are produced by combining time; market goods; human capital specific to the activity, [S.sub.k]; and quality of associations in a particular religious group noted by [Q.sub.m]. Religious human capital is a function of both an initial endowment of capital, [S.sub.k0], and investment into human capital, [DELTA][S.sub.kt].

In this framework, human capital and investments in it become important influences on choice. What economists refer to as human capital is similar to what many other social scientists would call cultural influences--the set of values, beliefs, rituals, and related matters making up the environment in a society. The approach described in Equation 1 provides economists with theoretic models in which these "cultural" forces become more than merely exogenous forces. (4) For example, endogenous life cycle changes in depreciation rates or time horizons can influence choice. Exogenous endowments of inputs such as doctrine, practice, or networks of relationships inherited from family and other external sources matter. Capital stocks can also be built by endogenous investments in doctrines, practice, and relationships. Along with capital stocks themselves, goods, services (both traded and nontraded), and institutions that are complementary to the utilization of the stocks will also influence choices.

At a general functional level, Equation 2 summarizes the reduced-form model for participation ([R.sub.k]) that lies behind most of the econometric studies of religious choice:

Participation = f(Income - Wealth; Religious Competition; Human Capital). (2)

Barro and McCleary (2003) found cross-country evidence that religious participation increases with income levels. Iannacone (1992, 1995) finds income and wealth negatively related to "fundamentalism"-sectarianism and positively related to mainline or less strict portfolio diversification. Lipford and Tollison (2003) find income and participation inversely related in a simultaneous model. (5)

Finke and Stark (1988); Iannacone (1991); Lipford, McCormick, and Tollison (1993); Zaleski and Zech (1995); and others have found supporting evidence that higher levels of competition promote more religious participation. (6) Hull and Bold (1998) find a negative relationship between religious product variety and church membership. Ekelund et al. (1996) and Ekelund, Hebert, and Tollison (2002) explored the importance of supply-side, institutional features in a wide-ranging study of medieval Christianity. Barro and McCleary (2005) find a positive association between monopolization (nationalizing) of religious choice and religiosity, but the direction of causation in their study runs from religiosity to monopolization.

The central issue that we address in this article is whether the broad categories of variables described previously (income, market competition, and human capital) would have different effects on different brands of religion. While this issue surfaces in the literature, it has not been systematically explored. Additionally, the relationship between brand choice and religiosity is analogous to the more general relationship between brand and intensity (quantity) of purchase of any product. This relationship has received extensive study from theoretical and econometric perspectives in economics journals as well as by economists in marketing journals, starting with McFadden (1980, 1986). (7) This literature has demonstrated that empirical models that estimate a single equation for intensity or quantity of purchase are misspecified. Instead, either intensity equations for each brand should be estimated separately or a multiequation system for brand and intensity, if identifiable, should be estimated.

The implications of the brand choice-theoretic literature applied to the previously mentioned religious choice framework leads to an extension of the model in Equation 1 as described in Equation 3:

[Q.sub.m] = f([Q.sup.a.sub.m]; [C.sup.a.sub.j])

[R.sub.k] = R([T.sub.kt], [X.sub.kt], [S.sub.kt], [Prob[[Q.sub.m]]), (3)

where [Q.sup.a.sub.m] is a vector of attributes for group m and [C.sup.a.sub.j] is a vector of attributes for consumer j. Religious association (brand) is no longer an exogenous influence on religious activity. Instead, brand choice is determined by attributes of the brand and characteristics of the consumer, and the likelihood of choice of brand m enters the religious activity production function.

While this choice problem can be characterized in broad theoretical terms, at this point the issue of what group and consumer attributes drive brand choice is largely empirical. Even in the relatively well-developed literature on brand choice and quantity, theory does not provide guidance as to the direction or magnitude of effects of different brand types. Instead, the studies have centered on brand variations of empirical price elasticities.

Narrowing brand choice research to the religious context, in particular, is purely empirical and still relatively young, at least as far as systematic quantitative and qualitative estimates are concerned. Market-level competition may interact with group attributes in ways that influence choice that are, at least, suggestive of direction of the effect. If one of the attributes of evangelical Protestantism is the evangelizing of members of other Christian faiths as well as non-Christian individuals, then the impact of higher degrees of (interbrand) religious competition will likely be larger on evangelical Protestant participation than the reverse.

For consumer attributes influencing choice, theory points toward income and religious capital as two important influences. If certain brands, such as evangelical, are more time intensive and if higher income suggests higher opportunity cost for participating in your chosen religion, then income effects on participation should be different (lower) for evangelicals than for brands that are less time intensive. In empirical work, even this income relationship may not be so clear-cut because of differential impacts of current income, income changes, or permanent income or wealth. Although religious capital specific to a group may be important and identifiable at a general theoretical level, linking empirical measures or indicators of such capital to specific theoretical predictions is not possible.

3. Data and Empirical Model

The measurement of religiosity or religious participation and identification of brand are central to our study. One of the broadest measures is simply the share of religious adherents in a county. This offers a measure of choice between (organized) religious participation versus either a more secular orientation or, at least, a more individualized form of religious practice. Using county-level church data from Religious Congregations and Membership in the United States 2000 (Bradley et al. 2002), we measure the share of the general religious adherents in the county and divide adherents into three subcategories or "brands": mainline Protestant, evangelical Protestant, and Catholic. These subcategories represent 14%, 28%, and 14% of population across counties and 27%, 44%, and 25% of the counties' religious participants belonging to these three categories, respectively.

Finke and Scheitle (2005) have demonstrated that these published attendance rates underreport attendance, primarily for traditionally black congregations, and propose a correction to the reported numbers. We utilize the authors' state-level correction factor for our county-level study by first regressing the state factor against the percent of the black population in each state. Thus, we are attempting to factor out the degree to which the state's black population results are underreported. Because our study is done at the county level, we then multiply the coefficient on percent black from the state-level equation by each county's percent of black population. Finally, we add this product, our county-level adjustment, to the reported adherence rate for the county. As a result, counties with larger black populations receive a larger adjustment than those with smaller black populations. For the remainder of the article, we refer to this as simply the adherence rate. (8)

Our use of cross-sectional county-level data as the unit of analysis strikes a balance between many of the foregoing empirical studies that have used either church-level data or more aggregated state- or national-level data. Although market size is uncertain and varies across locations, county designations are correlated with the geographic market for church competition. The exact geographic market relevant for church competition in any given location depends on the population density and infrastructure within that church's location. In some cases, county may be broader than the relevant geographic market. For example, an individual church may be competing with other churches in its immediate neighborhood, over several city blocks, or in an entire city. In other settings, religious participants may be drawn from an entire county, or in a few cases with relatively small county sizes, they may be drawn from multiple counties. We include controls for population density to adjust for these differences. Beyond its correlation to the relevant market, county-level data also offers two additional advantages. It is the lowest level of aggregation for several variables important for religious participation, including religious competition. Further, while individual choice models are the focus of theory, Becker (1973), who draws from Cramer (1964), noted the value of grouped data in empirical settings because grouped data randomize idiosyncratic influences and measurement errors. This is especially important for a variable such as religious participation, where matters such as theological views are difficult to measure or model in choice-theoretic ways.

Our data suggest that, at the low end, only 55% of a county's residents were adherents, while at the upper end, a few counties exhibit greater than 100% adherence rates because of definitional handling of in-migration of adherents. Mean adherence is close to 60% with about an 18% standard deviation.

Equation 4 presents our basic equation used to estimate religious choice drawn from prior theoretical and empirical studies:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

Appendix A summarizes descriptive statistics for all the variables. Appendix B provides sources. We quickly summarize the variables next.

The specific income and wealth variables that we use are median household income in dollars (MHI), the percent change in MHI from 1989 to 1997 (Change MHI), and per capita bank assets (Bank Assets). (9)

We use three direct measures of religious competition. The first is a religious Herfindahl index based on overall adherence (Religious Heft) calculated using the 149 Judeo-Christian denominations found in the Religious Congregations and Membership in the United States 2000 data described previously. The second is a Herfindahl index (Mainline Herf) for competition among denominations categorized as mainline Protestant. The third is a Herfindahl index (Evang Herf) for competition among denominations categorized as evangelical. The overall Herfindahl accounts for interbrand competition, while the denomination-specific Herfindahls account for intrabrand competition for members.

We also use Population Density and (Population Density) (2) to control for differences in availability of substitutes for religious association. Because religion offers an alternative form of entertainment or leisure, urban dwellers may have more substitutes for their leisure time and exhibit lower religious participation rates as a result. In this way, location is another means of measuring competition for religious consumption. Also, given population for an area, larger areas imply less direct interaction and, therefore, lower complementarity to religious consumption.

Stocks of human capital related to religious choice are multidimensional and difficult to measure directly. At best, observable variables are indicators of human capital or are measurements of goods or services complementary to the use of such capital. Educational level attained and racial variables are used as indicators of an individual's stock of human capital in religious and political models. Following these practices, we use percent of high school graduates in a county (High School) as our measure of educational attainment and percent Asian (Asian), percent black (Black), percent Hispanic (Hispanic), and percent native American (Native) as the racial variables. (10) In addition, measures of age or death rates are often used to identify differences in the stock of human capital and investments in that stock over time. The basis for the inclusion of a measure of age in our human capital discussion is driven from two very distinct theories: the traditional life cycle model (Azzi and Ehrenberg 1975) and a "formative years" framework that suggests the importance of a cohort effect in religious participation (Walrath 1987; Chaves 1989; Zech 2000). We use death rates (Death Rate) in each county to capture this effect, regardless of the theoretical premise.

General religious preferences and preferences for specific religious brands have been dispersed unevenly through geographic regions of the United States since its inception. This dispersion creates differences in religious heritage--a major influence on religious human capital. It also creates social costs to nonparticipation or differential branding. While region-based binary variables capture some of the cross-sectional differences in human capital and culture, we use two measures that provide even more detailed, quantitative indications of underlying and, often, long-standing differences--percent farm population (Farm) and an indicator of linguistic differences. (11) The linguistic measure is based on county-level data regarding the percent of the population using a particular term for carbonated soft drinks. We use the percent of the county population using "Coke" (Coke) as the preferred term for carbonated soft drinks.

4. Results

Overall and Brand-Specific Participation

Table 1 presents regression results for all religious adherents as well as the three broad subgroups described previously. While the marketing-economics literature on brand choice uses a two-stage estimation approach (brand choice and quantity choice) at the individual level, we do not observe actual brand choice in our county-level data. Instead, we observe only aggregated shares of brands. Thus, we choose to use a single-equation approach where we examine the influence of brand through estimation of brand-specific models.

The overall adherence equation in Table 1 explains 47% of the overall religious adherence rate with all the specific regressors significant at or below the 5% level. All the income and wealth measures have positive effects. Consistent with Hull and Bold's (1998) study of church membership in U.S. counties, we find that greater monopolization in the religious markets (higher interbrand Herfindahl levels) leads to higher participation levels. The sign on the intrabrand Herfindahl mainline and evangelical variables shows the opposite result. Greater intrabrand competition increases participation. Population density has an increasing but diminishing effect over the relevant range. Among the human capital-related variables, death rate, high school education, and farm population are all positive. Among the ethnicity controls, Asian is negative, while all the other categories are positive. The linguistic-human capital indicator, Coke, is positive.

The importance of brand categories can be seen across the denominational subgroup regressions in Table 1, where influences on religious choice vary widely based on the specific denominational brand. Overall, the model's ability to predict Catholic shares is far less than half its performance in predicting evangelical participation. The coefficients on the income-wealth variables are positive with some variation in magnitude. Death rate and farm population have positive coefficients in all three groups, but the magnitudes differ substantially. For example, death rate has twice the effect on evangelical participation than the other two, and farm population has three times the effect on mainline than evangelical participation. High school education affects Catholic participation only. The ethnicity variables differ across groups. The Coke-linguistic indicator is large and positive for evangelicals, large and negative for mainline, and zero for Catholic.

The results for the competition variables are of special interest in view of prior results in the literature using pooled denomination regressions. Greater interbrand competition (lower Religious Herf levels) raises mainline and evangelical participation, a result in line with the Lancaster-type models described in Hull and Bold (1998), but the evangelical coefficient is five times higher. The larger impact on evangelicals is consistent with the notion of an evangelizing denomination that attracts members from other Christian and non-Christian competitors. In contrast and consistent with Barro and McCleary (2005), greater interbrand monopolization raises Catholic participation. It is also interesting to note that while controlling for interbrand competition, higher intrabrand competition for evangelical Protestants lowers adherence. This result is consistent with the Protestant-only estimations of Hull and Bold (1998).

So far, we have employed broad denominational categories (mainline Protestant, evangelical Protestant, and Catholic) to identify possible brand differences and differences in underlying processes driving religious participation. The intensity of religious participation offers an alternative indicator of religious brand. In terms of overall adherence, low intensity suggests a relatively secular geographic region, while high intensity suggests a relatively religious region. There is sizable variation in overall religious participation across U.S. counties. We break down U.S. counties into intensity subgroups based on low (lowest 25% of adherence for given category), mid (middle 50% of adherence of counties), and high (highest 25% of adherence).

Table 2 presents regression results for overall religious adherence by counties of similar intensity of participation (low, mid, and high). The income-wealth variables and the (interbrand) competition variable (Religion Herf), along with death rates and farm population, exhibit consistent signs and significance across all three intensity levels but with sizable differences in magnitude. More interbrand competition (lower Religion Hero reduces participation across all groups but is five times larger in its effect in the mid-intensity counties than in the high-intensity counties. County size, death rate, percent black, and percent Hispanic are consistent across the mid- and high-participation counties.

Table 3 presents regression results for overall religious adherence with county subgroupings based on areas where a given denomination is dominant. As in Table 2, the income-wealth variables, death rate, and farm population exhibit consistent signs and significance levels but different magnitudes. As in the intensity breakdown, more interbrand competition reduces participation across each subgroup. The effect in counties where evangelicals are dominant is only about half the size of the effect in mainline-dominant counties and only about three-fifths of the size of the effect in Catholic-dominant counties.

We perform Chow tests on these results to formally test whether the regression coefficients from the different brand groupings, whether by intensity or by area of denominational dominance, are equal and, therefore, can be estimated as a single equation. Table 4 presents the results. In all cases, the reported F-statistic rejects the null that the subgroup coefficients are equal.

Religious Competition and Brand Identification

To this point in our analysis, we have treated religious competition as a variable to explain religious participation rather than as a way to help identify brand of choice. Like denominational categories or intensity of consumption, however, the degree of competition may help define markets. This may be especially important at the extremes of competition. For example, in Barro and McCleary (2005), the dependent variable is determined on the basis of whether religion is nationalized. In the United States, state sponsorship or monopolization of religion does not occur in a legal sense. Nonetheless, historical settlement and environmental conditions in certain areas of the country have led to near de facto state religions, at least around some localities over some periods of time. For example, the Mormon migration to and settlement of Utah created many communities where individuals of this faith dominated religious activity in ways very similar to state-sponsored religions.

In addition, the interaction of participation rates and competition may signal important market differences. For example, where a market is dominated by a single religious brand but participation rates are low, religious consumers may more easily have the option of not consuming. Where participation rates are very high and competition is very low, monopolization of the market by a particular brand creates a very different market in which consumers choose.

The extreme values show the dominance of Mormonism as the religious brand in sections of the intermountain West. Of the 20 least competitive markets as measured by the Herfindahl, Utah contains 11 of the top 20 Herfindahl counties and 20 of the top 50. Idaho accounts for seven more of the top 50. Among the 20 counties where the product of the Herfindahl and religious participation are the highest, Utah accounts for 10 of the top 20 and 19 of the top 50, while Idaho accounts for five more of the top 50.

Table 5 presents estimates of the explanatory model for two samples of counties with limited competition including the top 10% highest Herfindahl counties and counties with the top 10% of products between the Herfindhal and adherence. In these very low competition counties, wealth, death rates, farming populations, and percent black continue to be important influences with a positive relationship. It is also interesting to note that, even while focusing on the most religiously monopolized counties in the country, our measure of interbrand competition suggests that the more monopolized the religious market, the greater the adherence rate, while greater intrabrand competition raises adherence.

5. Concluding Remarks

We have explored the structure of religious markets in the United States by looking at the influences on religious participation with a large number of estimates. In condensing these results, our results support the following conclusions:

(i) The brand of religious consumption matters for the modeling of religious behavior. Aggregate-level denominational categories and intensity of religious participation indicate important branding differences.

(ii) The direction of effect of some economic influences in markets work independently of brand influence. Specifically, wealth is a consistently positive contributor to religious participation regardless of brand, and higher death rates increase participation regardless of brand.

(iii) Several influences on religious participation are highly sensitive to brand choice. Specifically, the magnitude of effects of wealth and the degree of religious competition depend heavily on brand.

(iv) The relationship between the level of competition and religious consumption is complex. As a purely explanatory variable, greater religious competition lowered participation except in the case of mainline and evangelical Protestant brands. It is also interesting to note that the effects of intrabrand and interbrand competition are often in opposition.

These results suggest that results from previous research using a more general definition of religious participation may not have a complete picture of religious markets and that empirical examination of religious activity should be pursued on a more disaggregated level.

Appendix A
Descriptive Statistics

                                                             Standard
Variable               Minimum       Maximum         Mean    Deviation

Religious adherence        0.1         1.66 (a)        0.58       0.18
Mainline share             0           0.84            0.14       0.11
Evangelical share          0           1.12 (a)        0.27       0.20
Catholic share             0           0.95            0.14       0.15
MHI                   14,178      77,513          32,604      7,868
Change MHI                 0.8        81.3            37.6        9.7
Bank assets                0         117              11.5        6.6
Pop density                0.1    66,835             227       1689
Death rate                 1.2        25.4            10.3        2.8
High school               32.0        96.0            69.6       10.3
Farm                       0          66.0             6.3        7.0
Asian                      0          46.0             0.8        1.9
Black                      0          86.5             8.8       14.5
Hispanic                   0.1        97.5             6.1       11.9
Native                     0          94.0             1.9        7.6
Coke                       0           1               0.32       0.34
Religious Herf             0.07        1.0             0.32       0.16
Mainline Herf              0.0         1.0             0.05       0.06
Evangelical Herf           0.0         0.94            0.15       0.18

(a) For a few areas, the share is more than 1.0 because the variable
includes individuals whose county of religious participation is
different than county of residence.

Appendix B

Data Sources

Religious adherence and computation of data for the Herfindahl indices of religious groups in counties are from Religious Congregations and Membership in the United States (Bradley et al. 2000).

The following variables are from the U.S. Bureau of the Census, Count), and City Data Book (2000), accessed through Geospatial and Statistical Data Center, University of Virginia (specific years in parentheses): median household income (MHI) (1997), percent change in MHI (1989-1997), bank assets (1999), death rate (1997), population density (2000), percent high school graduates (1990), percent of county population that is Asian American (2000), percent of county population that is Hispanic (2000), percent of county population that is African American (2000), percent of county population that is Native American (2000), and farm population (1990).

Data on linguistic terms for carbonated soft drinks (Coke) are from Spatial Graphics and Analysis Lab, East Central University (Oklahoma), and compiled from data available at http://www.popvssoda.com/countystats/total-county.html.

The authors thank Bob Tollison, colleagues participating in the Economics+ Workshop at WKU, and two anonymous reviewers for comments. Any remaining flaws are the authors' responsibility.

Received August 2006; accepted May 2007.

References

Anderson, Gary M. 1988. Mr. Smith and the preachers: The economics of religion in the wealth of nations. Journal of Political Economy 96:106-88.

Azzi, Corry, and Ronald G. Ehrenberg. 1975. Household allocation of time and church attendance. Journal of Political Economy 83:23-56.

Barro, Robert E., and Rachel M. McCleary. 2003. Religion and political economy in an international panel. American Sociology Review 68:760-81.

Barro, Robert E., and Rachel M. McCleary. 2005. Which countries have state religions? Quarterly Journal of Economies 120:1331 70.

Becker, Gary S. 1973. On the new theory of consumer behavior. Swedish Journal of Economies 75:378 95.

Bradley, Martin B., Norman M. Green, Jr., Dale E. Jones, Mac Lynn, and Lou McNeil. 2002. Religious congregations and membership in the United States, 2000. Nashville: Glenmary Research Center.

Chaves, Mark. 1989. Secularization and religious revival: Evidence from U.S. church attendance rates, 1972 1986. Journal for the Scientific Study of Religion 28:464-77.

Chintagunta, P. K., and Alok R. Prasad. 1998. An empirical investigation of the "dynamic McFadden" model of purchase timing and brand choice. Journal of Business and Economic Research 16:2 12.

Cramer, J. S. 1964. Efficient grouping, regression, and correlation in Engel curve analysis. Journal of the American Statistical Association 59:233-50.

Ekelund, Robert B., Robert F. Hebert, and Robert D. Tollison. 2002. An economic analysis of the Protestant reformation. Journal of Political Economy 110:646-71.

Ekelund, Robert B., Robert F. Hebert, Robert D. Tollison, Gary M. Anderson, and Audrey B. Davidson. 1996. Sacred trust: The medieval church as an economic firm. New York: Oxford University Press.

Finke, Roger, and Christopher P. Scheitle. 2005. Accounting for the uncounted: Computing correctives for the 2000 RCMS data. Review of Religious Research 47:5-22.

Finke, Roger, and Rodney Stark. 1988. Religious economies and sacred canopies: Religious mobilization in American cities, 1906. American Sociological Review 53:41-9.

Hamberg, Eva M., and Thorleif Pettersson. 1994. The religious market: Denominational competition and religious participation in contemporary Sweden. Journal for the Scientific Study of Religion 33:205-16.

Hanemann, W. Michael. 1984. Discrete-continuous models of consumer demand. Econometrica 52:541-62.

Hull, Brooks B., and Frederick Bold. 1998. Product variety in religious markets. Review of Social Economy 56:1 19.

Iannacone, Laurence R. 1984. Consumption Capital and Habit Formation with an Application to Religious Participation. PhD dissertation, University of Chicago.

Iannacone, Laurence R. 1990. Religious practice: A human capital approach. Journal of the Scientific Study of Religion 29:297-314.

Iannacone, Laurence R. 1991. The consequences of religious market structure. Rationality and Society 3:156-77.

Iannacone, Laurence R. 1992. Sacrifice and stigma: Reducing free-riding in cults, communes, and other collectives. Journal of Political Economy 100:271-91.

Iannacone, Laurence R. 1995. Risk, rationality, and religious portfolios. Economic Inquiry, 33:285-95.

Iannacone, Laurence R. 1998. Introduction to the economics of religion. Journal of Economic Literature 36:1465-96.

Knack, Stephen, and Philip Keefer. 1997. Does social capital have an economic payoff?. A cross-country investigation. Quarterly Journal of Economics 112:1251-88.

Krihnamurthi, Lakshan, and S. P. Raj. 1988. A model of brand choice and purchase quantity sensitivities. Marketing Science 7:1-20.

Lipford, Jody W., Robert E. McCormick, and Robert D. Tollison. 1993. Preaching matters. Journal of Economic Behavior and Organization 21:235-50.

Lipford, Jody W., and Robert D. Tollison. 2003. Religious participation and income. Journal of Economic Behavior and Organization 51:249-50.

McFadden, Daniel. 1980. Econometric models for probabilistic choice among products. Journal of Business 53:13-29.

McFadden, Daniel. 1986. The choice theory approach to market research. Marketing Science 5:275-97.

Neuman, Shoshana. 1986. Religious observance within a human capital framework: Theory and application. Applied Economics 18:1193-202.

Stigler, George J., and Gary S. Becker. 1977. De gustibus non est disputandum. American Economic Review 67:76-90.

Walrath, Douglas. 1987. Frameworks: Patterns of living and believing today. New York: Pilgrim Press.

Zaleski, Peter, and Charles Zech. 1995. The effect of religious market competition on church giving. Review of Social Economy 53:350-67.

Zech, Charles. 2000. Generational differences in the determinants of religious giving. Review of Religious Research 41:545-59.

(1) In a theoretical study, Iannacone (1990) notes that even the use of more narrow term such as "fundamentalism" suffers from definitional problems.

(2) See Barro and McCleary (2005) as an example.

(3) See Anderson (1988) for a survey of Smith's contributions on this subject.

(4) Becker (1973) first observed and Stigler and Becker (1977) later developed the framework for examining human capital as more than merely an exogenous influence on a model. In reduced-form empirical models, the distinction between human capital and cultural factors is not as critical in that measures of human capital or complementary variables are largely treated as exogenous.

(5) This contribution, as well as Ekelund, Hebert, and Tollison (2002), also indicates that the income-religiosity relationship may be more complex than presented in many studies.

(6) Hamberg and Pettersson (1994) find this same relationship using Swedish data.

(7) Some other contributions following on McFadden are Hanemann (1984), Krihnamurthi and Raj (1988), and Chintagunta and Prasad (1998).

(8) The average county-level adherence rate increased from 0.53 to 0.58 with the adjustment. Most of the denominations affected by underreporting are evangelical Protestant. Thus, the adjustment factor was also used for that denomination.

(9) We also examined percent of county population below poverty threshold, but its correlation with MHI is very high (-0.77), so it was not used in the final estimates.

(10) The correlation between percent high school and percent college graduate rates is very high (0.69), so we use only high school. None of the other explanatory variables are highly correlated, at least in the full sample.

(11) See Knack and Keefer (1997) for an example of the use of linguistic-based identifiers of group-based social capital.

Brian Goff * and Michelle W. Trawick ([dagger])

* Department of Economics, 1906 College Heights Boulevard, Western Kentucky University, Bowling Green, KY 42101, USA; E-mail [e-mail]; corresponding author.

([dagger]) Department of Economics, 1906 College Heights Boulevard, Western Kentucky University, Bowling Green, KY 42101, USA; E-mail [e-mail].

Table 1. Regression Results for Religious Adherence by Affiliation

                                 Dependent Variable

                             All              Mainline

MHI                          5.2e-6 **        1.90E-06
Change MHI                   0.003 **         0.002 **
Bank assets                  0.008 **         0.002 **
Pop density                  1.0e-5 **        4.2e-6*
[(Pop density).sup.2]       -2.7e-10 **      -9.6e-11 **
Death rate                   0.021  **        0.005 **
High school                  0.001 **         0.001
Farm                         0.007 **         0.003 **
Asian                       -0.004 **        -0.002
Black                        0.004*           0.001  **
Hispanic                     0.003 **         0.001  **
Native                       0.002 **         0.001*
Coke                         0.079 **        -0.041 **
Religious Herf               0.442 *         -0.052 **
Mainline Herf               -0.042 *          0.670 **
Evangelical Herf            -0.032 *          0.001
Constant                     0.365 *         -0.120 **
[R.sup.2]                    0.47             0.48
F-statistic                162 **           169 **

                                 Dependent Variable

                             Evangelical      Catholic

MHI                          8.8e-7 **        1.3e-6 **
Change MHI                   0.001 **         0.001
Bank assets                  0.003 **         0.001 **
Pop density                  2.5e-06          2.1e-06
[(Pop density).sup.2]       -9.9e-11 **      -1.8e-11
Death rate                   0.010 **         0.006 **
High school                 -2.5e-05          0.001  **
Farm                         0.001 **         0.002 **
Asian                       -0.001            0.001
Black                        0.003 **        -0.001
Hispanic                     0.001 **         0.001
Native                       0.001*           0.001 **
Coke                         0.078 **         0.008
Religious Herf              -0.269 **         0.342 **
Mainline Herf               -0.225 **        -0.315 **
Evangelical Herf             0.826 **        -0.484 **
Constant                    -0.010 **        -0.101
[R.sup.2]                    0.83             0.32
F-statistic                914 **            88 **

* Significant at 0.05 level or below.

** Significant at 0.01 level or below.

Table 2. Regression Results for Religious Adherence by Regions of
Similar Intensity

                                   Adherence Intensity Level

                               Low           Mid           High

MHI                         1.1e-6 *      1.1e-6 **      2.9e-6 **
Change MHI                 -8.40e-06      0.001 **       0.001 *
Bank assets                 0.002 **      0.002 **       0.004 **
Pop density                 6.2e-5 **     1.5e-5 **      1.0e-06
[(Pop density).sup.2]       9.0e-9 **    -7.6e-10 *     -1.2e-10 *
Death rate                  0.006 **      0.007 **       0.007 **
High school                 4.9e-05       0.001 *       -0.001
Farm                        0.001 **      0.001 **       0.004 *
Asian                      -0.002 *      -0.002 *        0.013 **
Black                       0.002 **      0.002 **       0.001
Hispanic                    0.001         0.001 **       0.002 *
Native                      0.001         0.001          0.002 **
Coke                        0.01          0.021 *        0.041
Religious Herf              0.057 *       0.102 **       0.017 **
Mainline Herf              -0.076        -0.033          0.003
Evangelical Herf           -0.016         0.023         -0.007
Constant                    0.179 **      0.245 **       0.393 **
[R.sup.2]                   0.13          0.17           0.21
F-statistic                 4.7 **       19.3 **        16.1 **

* Significant at 0.05 level or below.

** Significant at 0.01 level or below.

Table 3. Regression Results for Religious Adherence by Regions of
Denominational Dominance

                                         County Sample

                           Mainline      Evangelical     Catholic
                           Dominant      Dominant        Dominant

MHI                        8.3e-6 **     4.1e-6 **       4.6e-6 *
Change MHI                 0.004 **      0.001 **        0.004 **
Bank assets                0.008 **      0.010 **        0.005 **
Pop density                8.8e-06       2.0e-05         5.8e-06
[(Pop density).sup.2]     -2.3e-09      -1.9e-10        -1.2e-10
Death rate                 0.020 **      0.022 **        0.014 **
High school               -0.002 *       0.003 *        -0.002 **
Farm                       0.008 **      0.005 **        0.007
Asian                     -0.002        -0.008          -0.001
Black                      0.004 *       0.004 **        0.004 *
Hispanic                   0.003 **      0.005 **        0.001
Native                     0.002 **      0.002 **        0.001
Coke                      -0.165 **      0.086 **        0.024
Religious Herf             0.659 **      0.333 **        0.577 **
Mainline Herf             -0.285        -0.046           0.069
Evangelical Herf          -0.488         0.084 *         0.109
Constant                   0.237 **      0.426 **        0.058
[R.sup.2]                  0.56          0.50            0.51
F-statistic               45.1 **      101.6 **         47.6 **

* Significant at 0.05 level or below.

** Significant at 0.01 level or below.

Table 4. Chow Tests for Pooling Coefficients by Intensity Level

Dependent Variable (Test Samples) (a)                 F-Statistic

Overall adherence (pooled sample vs. subsamples
  by intensity level)                                    345 *
Mainline adherence (pooled sample vs. subsamples
  by intensity level)                                    261
Evangelical adherence (pooled sample vs.
  subsamples by intensity level)                         210 *
Catholic adherence (pooled sample vs. subsamples
  by intensity level)                                    322 *
Overall adherence (pooled sample vs. subsamples
  by dominant denomination)                               13.5 *

(a) The null hypothesis is that the coefficients for the subgroup
regressions equal the pooled regressions coefficients. The first
four rows are based on comparing column 1 of Table 1 with subsample
results from Table 2 as well as subsamples for denominations broken
down by intensity. The last row is based on comparing column 1 of
Table 1 to the subgroup regressions in Table 3.

* Significant at the 1% level or below.

Table 5. Adherence Rate Regressions for Counties with Low Competition

                                     Sample

                            Top 10%          Top 10%
                           Herfindahl     High Ad * Herf

MHI                        6.2e-07           -3.8e-06
Change MHI                 0.001              0.005 *
Bank assets                0.011 **           0.008 *
Pop density                1.5e-06            3.2e-05
[(Pop density).sup.2]      5.3e-11           -5.2e-09
Death rate                 0.011              0.010 *
High school                0.001              0.003 **
Farm                       0.010 **           0.006 **
Asian                     -0.010              0.045 *
Black                      0.003 **           0.003 **
Hispanic                   2.92e-05           0.001
Native                     0.002              0.002 *
Coke                       0.80               0.055
Religious Herf             0.506 **          -0.162
Mainline Herf             -0.428 **          -0.202 *
Evangelical Herf          -0.251 **          -0.079
Constant                   0.039              0.391
[R.sup.2]                  0.34               0.50
F-statistic                5.4 **             8.7 *

* Significant at 0.05 level or below.

** Significant at 0.01 level or below.