Praying for a Recession: The Business Cycle and Protestant Church Growth

in the United States

David Beckworth
Assistant Professor of Economics
Department of Finance and Economics
Texas State University

Abstract: Some observers believe the business cycle influences religiosity. This possibility is
empirically explored in this paper by examining the relationship between macroeconomic
conditions and Protestant church growth in the United States. The findings of this paper suggest
there is a strong countercyclical component to growth of membership for evangelical Protestant
denominations while for mainline Protestants there is both a weak countercyclical component
and a strong procyclical component.

Keywords: Religiosity, Church Attendance, Business Cycle, Economic Conditions, Recession,

Yield Curve, Protestant Church Growth.

Corresponding Information: David Beckworth, Department of Finance and Economics, 550

McCoy Hall, Texas State University, San Marcos, TX, 78666. Phone: (512) 245-6067. Fax
(512) 245-3089. Email: db52@txstate.edu

I would like to thank Monte Sahlin for helpful comments on this paper. Of course, all remaining
errors are mine.
"Adrian – you're not trying to tell me that you have personally organized a slide towards
recession just to get Christianity back on its feet?” Adrian grinned [and said] "Not quite, no. But
I've prayed mighty hard for it, and – well, I'm pretty impressed with the results so far.”
--A conversation reported in The Independent during the 2001 recession.

I. Introduction

Can religiosity be influenced by the business cycle? A common view, as seen in the above

quote, is that changing economic conditions do influence religiosity. Economic theory offers

several reasons why there may in fact be a cyclical component to religiosity. The first reason is

that the opportunity cost of religiosity may change over the business cycle. During an economic

expansion, individuals may find increased opportunities for higher earnings. If their labor supply

decision is by dominated by the substitution effect, then the potential for higher earnings will

make time-intensive religiosity like church attendance increasingly costly for these individuals

and vice versa. This understanding implies there would be, ceteris paribus, a countercyclical

component to religiosity. Conversely, if the income effect dominated and religiosity is a normal

good, then the potential for higher earnings may actually lead to more time-intensive religiosity

for these individuals and vice versa. A similar response would occur if the economic expansion

increased their non-labor income or generated for them a positive wealth effect from higher asset

prices. Here, there would be, ceteris paribus, a procyclical component to religiosity. A sizable

literature going back to Azzi and Ehrenberg (1975) finds that the substitution effect generally

dominates the income effect in terms of higher earnings leading to less time-intensive

religiosity.1 This finding suggests that there should be a countercyclical component to

religiosity, although it does not rule out a procyclical component arising from a non-labor

income effect or from a wealth effect.

1
Iannaccone (1988) provides a good summary of this literature. More recently, Gruber (2004) finds that religious
financial contributions act a substitute for time-intensive religiosity, a result consistent with the opportunity cost
view of religiosity.

2
 The second reason that there may be a cyclical component to religiosity is that it may

serve to smooth consumption over the business cycle. During an economic downturn individuals

may become unemployed or find their earnings fall. In order to prevent these developments from

disrupting their consumption flows, they may turn to their faith community for both real

consumption needs such as food and clothing as well as intangible consumption needs such as a

sense of certainty and divine guidance in a job search. To the extent that faith communities

provide such consumption smoothing over the business cycle, there should be a countercyclical

component to religiosity. Several recent studies report findings consistent with this

understanding. Chen (2005) finds the Indonesian economic crisis of 1997-1998 spurred

increased study of the Koran and increased attendance at Islamic school. He also finds that those

individuals who become more religious had less of a drop in their consumption.2 Chen

concludes from these results that religion provided for them an ex-post form of social insurance

that helped to smooth their consumption. Dehejia et al. (2005) analyze data from the Consumer

Expenditure Survey and the National Survey of Families and Households and similarly find that

religious participation provides partial insurance for real consumption and happiness against

income shocks. Ellison (1991) and Smith et al. (2003) similarly find that religiosity can serve as

a buffer against stressful life events, one of which is the severe psychological stress that Di Tella

et al. (2003) show comes from recessions. Although these latter studies do not show economic

downturns lead directly to increased religiosity, their finding that religion provides a buffer

against economic shocks suggests that individuals have an incentive to increase their religiosity

during economic downturns and vice versa.

2
Chen also notes that the economic distress did not increase the demand for other leisure activities or for secular
school. He, therefore, concludes that the changing opportunity cost story outlined above does not explain the
increase in religiosity following the 1997-1998 economic crisis. Rather, the increased religiosity was a pure
consumption smoothing response.

3
 One caveat to the above analysis is that not all faith communities are capable to same

degree of providing intangible consumption flows. Some mainline Protestant denominations, for

example, do not offer absolute truths that can create a sense of certainty or an all powerful God

who is able to provide jobs. As a result, individuals who find their need for intangible

consumption smoothing dominates their need for real consumption smoothing may fail to join a

mainline Protestant denomination during an economic downturn. Conversely, these same

individuals may find a mainline Protestant denomination more appealing during an economic

upturn when the need for certainty and employment are less pressing concerns. Consequently,

the consumption smoothing view also implies that there could be a procyclical component to

religiosity for some faith communities.

To date there has been very little empirical work done exploring the relationship between

the business cycle and religiosity. In addition to the Chen (2005) and Dehejia et al. (2005)

studies discussed above, the only other study that has examined this relationship is Sales (1972)

who looks at the effect of ‘economic threat’ on annual conversion rates in eight denominations

over the years 1920-1939. Using per capita disposable personal income as a measure of

economic conditions, Sales finds the conversion rates for more conservative denominations to be

countercyclical while for more liberal denominations they are procyclical. He explains these

patterns by explaining that the ‘authoritarian’ or more conservative denominations are

psychologically appealing during economic downturns for the sense of security they bring. On

the other hand, the ‘non-authoritarian’ or more liberal denominations resonate better during

economic upturns for the flexibility they provide.3 Although grounded in psychology, Sales’

explanation is similar to the consumption smoothing ability of religion outlined above. Sales

3
One example Sales (1972) provides of these differences is that authoritarian denominations “stress the
unimaginable might (and potential punitiveness) of God, while others view the Divine more as a friend and helper”
(p. 422).

4
also acknowledges, in passing, the possibility that socioeconomic status may be a determinant in

explaining these patterns. Since the more conservative denominations on average are

characterized by a lower socioeconomic status than the more liberal denominations in his study,

Sales concedes that there may be a self-selection bias to these denominations based on the stage

of the business cycle. While interesting, Sales’ study is limited to an early 20th century period of

economic extremes. On the other hand, the periods covered in the more recent studies of Chen

(2005) and Dehejia et al. (2005) fail to span an entire business cycle.4 There simply is a dearth

of research in this area.

This paper, then, makes a major contribution to this literature by providing a

comprehensive look at the business cycle – religiosity relationship in the United States. This

paper specifically examines whether there is a cyclical component in the growth of Protestant

denominations in the United States using a variety of approaches.5 First, data from a Pew

Research Center survey administered during the last U.S. recession in 2001 is used in a probit

model to assess the determinants of weekly attendance for both evangelical and mainline

Protestants. Second, annual membership data spanning 1968 to 2004 for 14 evangelical

Protestant denominations and 11 mainline Protestant denominations are examined in a number of

time series regressions to see if they are systematically related to the NBER business cycles

dates, the unemployment rate, real GDP, real personal consumption expenditures, oil prices, real

stock prices, and the yield curve spread. Finally, using quarterly conversion data for the period

1950:Q1 through 2006:Q4 for the Seventh-day Adventist (SDA) church—one of the 14

evangelical Protestant denominations—a series of vector autoregressions are run to determine

the dynamic relationships between SDA converts and the above macroeconomic variables.

4
To be fair, Dehejia et al. (2005) never set out to explicitly study the business cycle – religiosity relationship.
5
In some instances the data sources include a trivial number of Canadian churches in their measure of American
Protestant denominations.

5
 As mentioned above, a distinction is made throughout this paper between evangelical and

mainline Protestants. This distinction is made because of the well-documented fact that

evangelicals Protestants typically fall into a lower socioeconomic grouping than do mainline

Protestants (Pyle, 2006). As a result, the substitution and income effects, the non-labor income

effect, the wealth effect, and the consumption smoothing view could have different cyclical

implications for these two groups. Presaging the conclusions, this paper does in fact find

different cyclical implications for the memberships of these two groups: evangelical Protestant

denominations have a strong countercyclical component while mainline Protestant

denominations have both a weak countercyclical and strong procyclical component. To the

extent, then, increased membership is an objective of these denominations, only leaders of

evangelical denominations should be praying for a recession.

II. Weekly Attendance at Religious Services during the 2001 Recession

The Pew Research Center conducted a survey in mid-November 2001 in the United States. The

survey had a nationwide sample of 1500 adults and covered a number of issues, including the

public’s attitude toward religion. As can be seen in Figure 1, this survey followed a steep

increase in the unemployment rate and occurred near the end of the NBER-dated 2001 recession.

The timing of the survey, then, makes it ideal for studying the relationship between economic

conditions and religiosity.

[Insert Figure 1 around here]

Since this paper focuses on church growth as a measure of religiosity, the survey question

that asks how regularly the respondents attend religious services is used here as the variable of

interest. Specifically, a dummy variable equal to one is created for all those respondents who

indicated they attend religious services at least once a week. All other respondents receive a

6
zero. This attendance variable is then regressed against a similarly-constructed unemployment

dummy variable and some other control variables in a probit regression. This approach allows

the probability of weekly attendance to be estimated conditioned upon the employment status of

the respondent. To the extent that weekly attendance at religious services has a bearing on

denominational growth and that respondent’s employment status reflects broader economic

conditions this approach should provide some insight into whether business cycles affect

denominational growth. A confounding factor in this analysis, however, is the 9-11 terrorist

attacks which took place just a few months before the survey. This traumatic event, independent

of the 2001 recession, probably caused an increase in religiosity and, if so, could create

misleading inferences about the relationship between economic conditions and religiosity.

Fortunately, this effect can be teased out in the regression by including a dummy variable for all

respondents who answered the following survey question in the affirmative: “As a result of the

terrorist attacks, are you attending religious services more or not?”

[Insert Table 1 around here]

Table 1 reports the probit regression results with p-values reported in the brackets. In

column (1), weekly attendance is regressed against the unemployment status controlling for the

9-11 effect. Columns (2) through (3) rerun this same regression but now include dummy

variables for those respondents who indicated they were Protestant, evangelical or born again

Protestant, or other (i.e. mainline) Protestant. In all cases unemployment is positively related to

weekly attendance at religious services and is highly significant. Since this is a probit model, it

is hard to interpret the coefficients directly so Figure 2 provides a probability of weekly

attendance flow chart constructed from the estimates in columns (1) through (3). In calculating

the probabilities, the 9-11 effect dummy is set equal to zero so that estimated probability of

7
weekly attendance does not reflect the impact of this event. 95 percent confidence bands for the

probability estimates are provided in the brackets. The starting point in the probability flow

chart is the rounded rectangle in the center where the overall sample probability of weekly

attendance is 43.12 percent. Moving up the flow chart, the probability for weekly attendance

jumps to 48.44 percent for unemployed respondents but drops to 35.05 percent for employed

respondents, a difference of about 13 percent. Moving down the flow chart, the probability of

weekly attendance is conditioned on employment status and type of Protestant. While

evangelical or born again Protestants have a higher overall probability of weekly attendance,

both they and the other Protestants experience about an 11 percent increase in probability if

unemployed. These increases in probability due to unemployment are non-trivial in size given

the probabilities for attendance if employed are 55.72 percent for evangelical or born again

Protestants and 29.29 percent for other Protestants. The 95 percent confidence bands indicate

these differences are significant.

To check the robustness of these results, the probit regressions were rerun with standard

control variables, all in dummy variable form except for age, taken from the survey. Columns

(4) through (6) report the results of these regressions. The unemployment dummy again remains

significant in all cases. Employment status continues to matter as a determinant for attendance at

weekly religious services. One limit of this analysis, though, is that employment status may not

fully reflect for all individuals the stresses the economic downturn. For example, some

individuals may have become more religious during the 2001 recession not because they had lost

their jobs, but because they were fearful they would lose their jobs. Others, such as individuals

working in sales, may not have had their employment status threatened but may have

experienced a financially destabilizing reduction in earnings. The impact of any recession, then,

8
may be broader than what the unemployment rate alone would indicate. Nonetheless, these

results by themselves suggest that there is a countercyclical component to weekly attendance at

religious services for both evangelical and mainline Protestants.

III. Membership Growth in Protestant Denominations

The possibility that economic conditions may influence religiosity beyond what employment

status alone may indicate is explored in this section of the paper by examining the relationship

between a number of macroeconomic variables and the growth of Protestant denominations. The

assumption here is that increased religiosity should lead to higher church attendance and, in turn,

to faster growth of denominational membership, ceteris paribus, and vice versa. A change in

denominational membership is defined as,

∆ membershipt = membershipt – membershipt-1 = convertst – dropped memberst , (1)

where t is the time period subscript and dropped members are individuals removed from

membership because of death or defection. Membership changes, then, may also reflect

influences other than religiosity, but since it is the only widely available measure across many

Protestant denominations it is adopted here as a proxy for religiosity. The macroeconomic

variables used in the analysis are a NBER recession indicator, the unemployment rate, real gross

domestic product (GDP), real personal consumption expenditures (PCE), oil prices, the real

S&P500 index, and the yield curve spread. The NBER recession indicator is a dummy variable

set equal to one for every NBER-defined recession year. The next three variables provide

concurrent measures of economic conditions while the last three variables lead economic

activity. Oil prices have spiked prior to most of the post-World War II U.S. recessions while

9
stock prices often declined prior to the economy weakening.6 In this context, though, stock

prices may also be more than a leading economic indicator, since they can reflect what is

happening to an individual’s wealth. The yield curve spread—the difference between long-term

interest rates and short-term interest rates—in particular has been found to be a good predictor of

economic activity. Whenever this measure has turned negative economic activity has either

slowed down or outright contracted (Estrella, 2005). Here two standard measures of yield curve

spread are used: the 10-year treasury minus the 1-year treasury and the 10-year treasury minus

the 3-month treasury. All macroeconomic data except for the real stock prices has been taken

from the FRED database at the St. Louis Federal Reserve bank. The real stock prices come from

Robert Shiller’s home page.7

[Insert Table 2 around here]

Data for membership in Protestant denominations comes from The State of Church

Giving through 2004. This annual volume collects membership data on a number of Protestant

denominations, but only has a consistent annual series across 25 Protestant denominations that

runs from 1968 through 2004. This volume classifies the 25 Protestant denominations into 11

mainline Protestant denominations and 14 evangelical Protestant denominations.8 Table 2 shows

the list of denominations making up the mainline and evangelical Protestant denominations. It is

interesting to note that although this list of Protestant denominations is far from exhaustive, the

decline in membership for the mainline Protestant denominations is almost fully offset by the

membership gains in the evangelical Protestant denominations over the 1968 to 2004 period.

6
See Hamilton (2005) for a discussion on oil and the macroeconomy and Stock and Watson (2003) for a review of
the forecasting ability of stock prices.
7
http://www.econ.yale.edu/~shiller/.
8
One of the denominations in the evangelical grouping, the Seventh-day Adventist church, was not listed in the
volume as an evangelical denomination but is counted here as one. The Seventh-day Adventist church considers
itself to be an evangelical Protestant denomination, a view shared by the many other observers such as the Hartford
Institute for Religion Research and the U.S. Congregational Life Survey organization.

10
During this time, mainline Protestant membership falls from 26.1 million to 18.6 million while

evangelical Protestant membership grows from 15.4 million to 23.2 million. Overall Protestant

membership in the sample, therefore, is almost unchanged going from 41.5 million in 1968 to

41.8 million in 2004. These changes are portrayed in Figure 3.

[Insert Figure 3 around here]

These three groupings of Protestant denominations and all the macroeconomic variables

are at an annual frequency for the period 1968-2004. Except for the unemployment rate and

yield curve spread, these variables are turned into growth rates to avoid unit root problems in the

analysis that follows. In the case of the unemployment rate, its first difference is also

constructed since this measure was found to be useful below. Table 3 reports the descriptive

statistics for these variables. A series of univariate regressions were run to determine if there are

any systematic relationships between the macroeconomic variables, in both current and lagged

(i.e. t-1) form, and the three groupings of Protestant denomination. Newey-West standard errors

were estimated that are robust to serial correlation. The results from these regressions are

reported in table 4.

[Insert Table 3 around here]

In the first regression, the NBER recession dummy variable is significant only for the

evangelical Protestants. The estimates show that during recessions Protestant membership grew

1.52 percent a year, a sizeable increase over the 0.98 percent a year growth in the non-recession

years. The R2 also shows that almost 19 percent of the variability in Protestant membership

growth can be explained by the NBER Recession dummy. The unemployment rate and its first

difference are similarly found to be significant for the evangelical Protestants but not the

mainline Protestants. The estimates imply a positive one-standard deviation shock to either the

11
unemployment rate or its first difference pushes up the membership growth rate 0.17 percent to

0.19 percent, a modest increase given the sample average membership growth rate is 1.1 percent.

The unemployment rate is also significant for the total Protestant denomination measure, but

again the overall effect size is modest. Real GDP and real PCE coefficient estimates are found to

be negatively related to membership growth for evangelical and total Protestants both

contemporaneously and lagged one year. For mainline Protestants, real GDP and real PCE are

found to be negatively related to membership growth contemporaneously, but are not statistically

significant.

[Insert Table 4 around here]

Real stock prices are found to be negatively related to membership growth for

evangelical Protestants, with only the one-year lag being barely significant. A one-standard

deviation real stock price shock increases the membership growth rate of evangelical Protestants

by about 0.17 percent. This result is consistent with the findings so far and indicates that a

booming stock market should reduce the religiosity of evangelical Protestants and vice versa.

For mainline Protestants, however, the estimates show the relationship is reversed: a booming

stock market should increase religiosity. Here, a one standard deviation real to the one-year

lagged stock price pushes up the current mainline protestant membership growth rate by 0.21

percent, a non-trivial amount since the sample average membership growth rate for mainline

Protestants is -0.94 percent. Moreover, the relationship is statistically significant and R2 on

lagged real stock prices is 23.25 percent.

One way to make sense of these differing responses to the stock market is the fact that

mainline Protestants typically are in a higher socioeconomic status than evangelical Protestants

(Pyle, 2006). This fact has several implications. First, mainline Protestants are more likely than

12
evangelical Protestants to have a larger percent of their wealth invested in the stock market.

They are, therefore, more likely to have a strong wealth effect, the tendency for individuals to

supply less labor and spend more if their wealth increases. Given that religiosity is normal good,

a booming stock market should lead, ceteris paribus, to more religiosity for mainline Protestants.

Second, since evangelical Protestants should have comparatively less of their wealth in stocks,

the stock market for them is a leading economic indictor. Thus, a booming stock market for

them foretells of the increased opportunity costs of religiosity.

Spot oil prices are found to be positively related to membership growth rates for all

Protestant groupings, but are only significant with the evangelical and total Protestant measures.

For these groups, then, an increase in oil prices leads to an increase in membership growth. For

evangelicals, a one standard deviation shock to oil prices modestly pushes up the membership

growth rate by 0.19 percent. By far, though, the best explainer of variability in membership

growth for the evangelicals and total Protestant is the yield curve spread. For the evangelical

Protestants, the yield curve spread explains about 22 percent of the variability. For them, a one

standard deviation shock to the spread increases the membership growth rate as much as 0.33

percent. The yield curve spread is also significant for the mainline Protestants though it explains

less variability. The signs of the yield curve spread coefficients are negative for all Protestant

groupings. This means a large positive spread, which usually indicates a growing economy, is

associated with lower religiosity. The yield curve spread, then, points to a statistically significant

countercyclical component for both evangelicals and mainline Protestant denominations.

Collectively, the results from table 4 can be summarized as follows. The evangelical

Protestants appear to be particularly sensitive in a countercyclical manner to economic

conditions as indicated both by the NBER recession dummy and the broad number of

13
macroeconomic variables that were found to be significant. The mainline Protestants, however,

were not significantly related to any of the macroeconomic variables other than stock prices and

the yield curve spread. The yield curve spread showed them to have a countercyclical

component while the stock prices pointed to have a procyclical component. The R2 for these two

variables indicate only the procyclical component to be strong. A key reason for these varying

responses is the different socioeconomic status of these two groups. As mentioned above, the

higher socioeconomic status of the mainline Protestants implies a stronger wealth effect. It also

implies, however, a stronger income and non-labor income effect. An expansionary economy for

them, then, makes religiosity more, not less attractive and this appears to be borne out in the

data. The results for the evangelical Protestant, on the other hand, suggest a strong substitution

effect for them over the business cycle. This countercyclical tendency is further reinforced by

the likelihood that their comparatively lower socioeconomic status means their faith

communities play a more important consumption smoothing role.

To get a better sense of the overall amount of variability in evangelical and mainline

Protestant denominational growth that can be attributed to economic conditions, a number of

multivariate regressions were run using these same macroeconomic variables. Once again,

Newey-West serial correlation-robust standard errors were estimated; p-values are reported in

the parenthesis. Table 5 reports the best fitting model for each Protestant grouping to emerge

from this exercise. These results indicate a sizable amount of variability in the membership

growth rate can be explained by macroeconomic variables: the R2 for evangelical Protestants is

38.77 percent, for mainline Protestants it is 45.58 percent, and for total Protestants it is 45.36

percent. As with the univariate regressions, these results again show only the evangelical

Protestants have a purely countercyclical component.

14
 [Insert Table 5 and Table 6 around here]

Another way to measure how important economic conditions are to the growth of

Protestant denominations is to evaluate whether lagged macroeconomic variables can improve

upon an autoregressive (AR) forecast of membership growth. Formally, this approach requires

adding lagged macroeconomic variables to an autoregressive regression and doing an F test for

joint significance on the lagged macroeconomic variables. If the lagged macroeconomic

variables are found to be significant, then they are said to Granger cause membership growth.

Table 6 reports the results adding a lag of both real stock prices and the 10-year treasury minus

the 1-year treasury spread to an AR(2) membership growth regression. In each Protestant

grouping, the null hypothesis that the macroeconomic variables do not granger cause

membership growth can be rejected at the 5 percent level. There are also non-trivial increases in

the R2s and reductions in the root mean squared error. Macroeconomic variables, therefore,

increase in a meaningful way the forecasting accuracy of membership growth.

IV. Seventh-day Adventist Converts

The previous section showed that evangelical Protestants have relative to mainline Protestants a

strong countercyclical component to their membership growth rate. One question that was not

addressed in this analysis is exactly how the countercyclical component plays out over time. A

challenge in tackling this question is that most denominations publish membership data at an

annual frequency. Such low-frequency data does not lend itself to the type of dynamic analysis

that could answer how the countercyclical component plays out over time. Fortunately, one of

the evangelical Protestant denominations, the Seventh-day Adventist (SDA) church, does collect

15
quarterly data for conversions.9 The Secretariat of the North American Division of the SDA

church collects and disseminates the conversion data for the United States in quarterly statistical

reports.10 Data for the period 1950:Q1 through 2006:Q4 were collected and seasonally adjusted

to create the time series seen in Figure 4.

[Insert Figure 4 around here]

A vector autoregression (VAR) was then used to estimate the dynamic relationship

between the macroeconomic variables and SDA converts. In a VAR there are n variables, each

treated as a dependent variable in an equation with an equal number of its own lags and the lags

of all the other variables entering its equation as explanatory variables. Each variable, then, in

the VAR is considered endogenous or determined within the system of equations. Unexplained

variation or shocks to variables, however, can arise from outside the system of equations and are

captured by the residuals in each estimated equation. The VAR is useful in this context because

it can show how a shock to a macroeconomic variable can affect SDA conversions over time.

Formally, this approach starts with an autoregressive structural model of the form

A0 yt = A1 yt −1 + ... + Ap y t − p + ut , (2)

where yt is a n x 1 vector of endogenous variables, A0 ,..., Ap are n x n structural parameters

matrices and ut is a n x 1 vector of uncorrelated structural shocks that are assume to be

multivariate normal with mean zero and unit variance. The vector of endogenous variables can

′
be partitioned as follows: y = ( xt , ct ) , where xt are macroeconomic variables and ct is SDA

conversions. The structural autoregressive model can be transformed into a structural moving

9
For the SDA church, conversions include individuals who join the church through baptism or through profession of
faith.
10
The data also includes Canadian converts, but they are only 4.5% of the total converts on average since 1950.
The quarterly statistical reports can be found at http://www.nadadventist.org/sec/.

16
average form so that the relationship between the endogenous variables and the structural shocks

can be defined. The structural moving average model can be shown to be

y t = ( D0 + D1 L + D 2 L2 + ...)u t

= D ( L )u t , (3)

where D0 = A0−1 , Di = ( A0−1 Ai ) i A0−1 and L denotes the lag operator.11 The coefficient matrices in

D (L) represent the dynamic multipliers of the structural shocks. As it stands, (3) is still a

structural model and cannot be estimated directly. Rather, a reduced form version must be

estimated and then identifying restrictions imposed to recover the structural model. The reduced

form moving average can be expressed as follows:

y t = ( I + C1 L + C 2 L2 + ...)ε t

y t = C ( L)ε t . (4)

There is a mapping between the reduced-form parameters in (4) and the structural

parameters in (3) since ε t = D 0 u t , C ( L) = D ( L) D 0−1 and Eε t ε t' = Σ = D 0 D 0' . However, this

mapping is not unique as an infinite number of values of D0 can satisfy Σ = D0 D0' .

Consequently, even though the reduced form parameters C (L) and Σ are directly estimable,

identifying restrictions need to be imposed to recover the structural shocks. The identification

scheme adopted here is to use the standard Choleski decomposition of Σ which restricts D0 to be

lower triangular and thus orthogonalizes the shocks into an ordering that follows the variable

ordering in the yt vector.12

The macroeconomic variables that were used in Section III are used here in the VAR and

11
Lk y t = y t −k
12
For more on VARs see Enders (2004).

17
come from the same sources. Now, however, the data are at a quarterly frequency. Also as

before, all the macroeconomic variables, except for the yield curve spread and the

unemployment rate, and the SDA conversions are transformed into growth rates for the VAR.

Standard unit root test indicate these transformations are sufficient for stationarity in the time

series. Also as before, the unemployment and its first difference are used in the VAR. Table 7

reports the summary statistics for these quarterly variables.

[Insert Table 7 around here]

As a first run, a series of two-variable VARs were estimated, where each VAR consisted

of one of the macroeconomic variables and SDA conversions. The number of lags used in each

VAR was determined by a likelihood ratio test. This lag length test indicated that 8 lags were

appropriate for all VARs except for the one using the 10 year treasury minus the 3 year treasury

spread where it suggested a 7 lag VAR. Figure 5 shows the cumulative response of the SDA

conversion growth rate to a positive one standard deviation shock to the growth rate of each

macroeconomic variable. These accumulated impulse response functions (AIRFs) are

accompanied by Monte Carlo-generated two standard error bands to provide a measure of the

precision of the estimates. The standard error bands are indicated by the dashed lines.

[Inset Figure 5 around here]

The top row of Figure 5 shows five quarters out that the number of SDA conversions

slows down about 1.9 percent in response to a real GDP growth rate shock. This response is

significantly different than zero as can be seen by the standard error bands. A shock to the real

PCE growth rate similarly creates a significant response 5 quarters out of 1.5 percent. Both of

these responses stay significant through quarter 6. Given the average quarterly conversion

growth rate is 0.9 percent over the sample, these responses are large. A shock to the

18
unemployment rate creates a significant response 4 quarters out, while a shock to the change in

the unemployment rate creates a significant response 2 quarters out. In the former case, the peak

increase in conversions is 1.5 percent and in the latter case there is a peak increase of 2.2 percent.

Like the response to real GDP and real PCE, this response is not significantly different from zero

by quarter 7. On the other hand, though, the response to both yield curve shocks does not

become significant until quarter 7. Both yield curve spread shocks lower the number of

conversions by just under 2 percent. The timing of these responses indicates the number of

conversions will begin dropping about 1.5 years after the shock. In the case of the 10 year

treasury minus the 1 year treasury spread, the response also indicates the effect of the yield curve

shock on conversions will not be significantly different than zero another 1.5 years after quarter

7. Together these responses weave an interesting story: a positive yield curve shock is a leading

indicator of future economic activity by 1.5 years at which time real GDP and real PCE increase

and unemployment decreases for another 1.5 years; these improved economic conditions then

drive down the number of conversions. Since this is a linear model the results also hold vice

versa. Finally, Figure 5 shows that shocks to real stock prices and oil prices do not respond in a

manner significantly different than zero, although the direction of their responses does indicate a

negative relationship between economic conditions and conversions.

While these impulse response functions indicate any number of macroeconomic

shocks can create a relatively large decline in the number of conversions, they do not tell what

portion of total variation can be explained by the macroeconomic shocks. This information can

be gleamed from the VARs, but as they stand each VAR only contains information on the one

macroeconomic variable in it and may miss the effect of the broader economy on conversions.

Consequently, a bigger VAR that contains multiple macroeconomic variables needs to be

19
estimated. One problem in estimating a bigger VAR is that some of the macroeconomic
13
variables may be capturing similar information about the economy and be redundant.

Therefore, a joint test of granger causality was used to determine what macroeconomic variables

should be included in the VAR. The following variables were found to be jointly significant at

the 5 percent level: yield curve spread, real stock prices, oil prices, change in the unemployment

rate, and conversions. Although both yield curve spreads were found to be significant, for

convenience the 10 year treasury minus the 1 year treasury was chosen for the analysis. The

VAR was estimated using 6 lags as was suggested by the likelihood ratio test.

Figure 6 reports the AIRFs from this bigger VAR. Once again, a standard deviation

shock to the yield curve spread lowers the conversion growth rate 7 quarters out while a standard

deviation shock to the change in unemployment increases the conversion 4 quarters out. Both

responses are different than zero with non-trivial magnitudes. However, shocks to the growth

rate of real stock prices and oil prices are not significant. Nonetheless, these latter two

macroeconomic variables are kept in the model since they along with the yield curve and the

change in unemployment were found to be jointly significant.

A decomposition of forecast variance was next created from this VAR to see how

important macroeconomic shocks were historically in explaining the variation of the conversion

growth rate. The results form this historical decomposition for the period 1960:Q1 through

2006:Q4 are reported in Figure 7.

[Insert Figure 6 around here]

This figure shows the conversion growth rate to be historically decomposed into an actual path

series, a baseline forecast series, and a baseline forecast plus macroeconomic shocks series. The

13
Moreover, it is possible that some of macroeconomic shocks may disappear in VAR that includes multiple
macroeconomic variables. Simply, the VAR may be able to account for more of the unexplained variation in some
of the macroeconomic variables by including other macroeconomic variables.

20
second series is a projection of the SDA conversion growth rate that does not include any of the

macroeconomic or other shocks in the period being forecasted. The third series makes the same

projection but incorporates the accumulated effect of past and current macroeconomic shocks.

The closer the third series is to the actual path series the more macroeconomic shocks account

for the non-forecasted variation in the growth rate of SDA conversions. The difference between

the third series and the actual path is variation attributed to non-macroeconomic shocks. Figure

6 indicates that the macroeconomic shocks make up a non-trivial portion of the forecast

variation. To make better sense of these results Table 8 reports the relative importance of the

macroeconomic shocks. Defining the accumulated effect of the macroeconomic shocks as

effectmacro and the accumulated effect of non-macroeconomic shock as effectnon-macro, the relative

importance of the macroeconomic shocks can be stated as

effect macro
Relative importance = . (5)
effect macro + effect non−macro

Table 8 shows the relative importance of macroeconomic shocks for the entire period was 35.39

percent. This amount varies by decade and reaches a low of 25.07 percent in the 1960s and a

high of 40.28 percent in the 1970s. The relative importance of macroeconomic shocks also

varies by whether or not there is a NBER-defined recession: during recession it is 45.62 percent

while in non-recession years it is 33.73 percent. This difference between recession and non-

recession years was found to be statistically significant at the 1 percent level.

[Insert Table 8 around here]

The key insight from the historical decomposition is that just over a third of the forecast

error can be explained by macroeconomic shocks. This amount is in line with the results from

Section III that similarly showed just over a third of the variation in the membership growth of

21
evangelical Protestants cab be explained by macroeconomic variables. That so much of the

VAR forecast error can be explained by macroeconomic shocks in a countercyclical manner

means the business cycle is consequential to the number of converts to the SDA church.

V. Conclusions

This paper has found evidence that indicates there is a cyclical component to the growth of

Protestant denominations in the United States. The 2001 Pew Survey showed that employment

status was an important determinant of weekly attendance during the 2001 recession for both

evangelical and mainline Protestants. Evangelical Protestants, however, were found to be

particularly responsive to downturns in the business cycle. During recessions their membership

grew on average 1.52 percent annually, a significant pick up from the 0.98 percent growth rate in

non-recession years. Evangelical Protestant membership growth was also found to be

systematically related to the unemployment rate, oil prices, real stock prices, and the yield curve

spread. Just over a third of the variation in their membership growth rate could be attributed to

changes in these macroeconomic indicators. The results for the analysis on the SDA converts,

one of the evangelical Protestant denominations, indicate that the countercyclical component of

conversions lasts for about 1.5 years after a macroeconomic shock and that the yield curve

spread does a good job predicting this change. To the extent, then, leaders of evangelical

denominations want to increase their membership they should be closely following the yield

curve spread and praying for a recession.

The growth of mainline Protestant denominations was also found to be responsive to

economic conditions. Just under half of the variation in the growth of mainline Protestant

denominations could be attributed to changes in economic conditions. The response of mainline

Protestant denominations to the business cycle, however, was less countercyclical than the

22
evangelical Protestants and even had a procyclical component to it. Stock market booms, in

particular, were found to be associated with a non-trivial pickup in the membership growth rates

of mainline Protestant denominations. Praying for a recession by mainline Protestants, then,

might bring mixed blessings for their membership growth.

One reason given for the differing cyclicality of the evangelical and mainline Protestant

membership is their dissimilar socioeconomic status. Mainline Protestant denominations are

characterized by a higher socioeconomic standing. This fact implies for mainline Protestants a

relatively stronger income effect, non-labor income effect, and wealth effect for. It also implies

less of a need for consumption smoothing by their faith communities. In short, the higher

socioeconomic status of mainline Protestants implies a procyclical component is just as or more

likely than a countercyclical component in the growth of their denominations. On the other

hand, the comparatively lower socioeconomic status of evangelical Protestants points to a strong

substitution effect and need for consumption smoothing. Hence, it implies a strong

countercyclical component for the growth of the evangelical Protestant denominations.

In sum, the findings of this paper indicate that Adrian from The Independent quote was

correct in his assessment: business cycles do influence religiosity.

23
References

Azzi, Corry and Ronald Ehrenber. (1975). “Household Allocation of Time and Church
Attendance.” Journal of Political Economy. 83(1): 27-56.

Chen, Daniel. (2005). “Club Goods and Group Identity: Evidence from Islamic Resurgence
During the Indonesian Financial Cirsis.” Working Paper, University of Chicago.

Dehejia, Rajeev, Thomas DeLeire, and Erzo Luttner. (2007). “Insuring Consumption and
Happiness through Religious Organizations.” Journal of Public Economics. 91(1-2): 259-279.

Di Tella, Rafael, Robert MacCulloch, and Andrew Oswald. (2003). “The Macroeconomics of
Happiness.” The Review of Economics and Statistics. 85(4): 809-827.

Empty Tomb, Inc. (2004). “The State of Church Giving through 2004.”

Enders, Walt. (2004). Applied Economic Time Series. John Wiley and Sons, Inc. Hoboken,
New Jersey.

Ellison, Christopher. (1991). “Religious Involvement and Subjective Well-Being.” Journal of

Health and Social Behavior. 32(March): 80-99.

Gruber, Jonathon. (2004). “Pay or Pray? The Impact of Charitable Subsidies on Religious
Attendance.” Journal of Public Economics. 88(12): 2636-2655.

Hamilton, James D. (2005). “Oil and the Macroeconomy.” New Palgrave Dictionary of
Economics, 2nd edition.

Kington, Miles. (2001). “How the Church Secretly Planned this Recession.” The Independent
(London). September 21.

Iannaccone, Laurence. (1998). “Introduction to the Economics of Religion.” Journal of

Economic Literature. 36(3): 1465-1494.

Pew Research Center for the People and the Press. (2001). Mid-November Survey.

Pyle, Ralph. (2006). “Trends in Religious Stratification: Have Religious Group Socioeconomic
Distinctions Declined in Recent Decades?” Sociology of Religion. 67(1): 61-79.

Sales, Stephen. (1972). “Economic Threat as a Determinant of Conversion Rates in Authoritarian

and Nonauthoritarian Churches.” 23(3): 420-428.

Smith, Timothy, Michael McCullough, and Jutin Poll. (2003). “Religiousness and Depression:
Evidence for a Main Effect and the Moderating Influences of Stressful Life Events.”
Psychological Bulletin. 129(4): 614-636.

24
Stock, James, and Mark Watson. (2003). “Forecasting Output and Inflation: The Role of Asset
Prices.” Journal of Economic Literature. 41(3): 788-829.

25
 Table 1

Weekly Attendance
(1) (2) (3) (4) (5) (6)
Constant -0.384 -0.542 -0.474 -0.966 -1.058 -1.103
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Unemployed 0.345 0.298 0.310 0.190 0.169 0.195
(0.000) (0.000) (0.000) (0.017) (0.035) (0.022)
Attend More Because of Sept. 11 0.685 0.687 0.502 0.690 0.691 0.528
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Protestant -- 0.332 -- -- 0.293 --
(0.000) (0.000)
Evangelical/Born Again Protestant -- -- 0.618 -- -- 0.622
(0.000) (0.000)
Other Protestant -- -- -0.071 -- -- -0.107
(0.348) (0.166)
College -- -- -- 0.133 0.155 0.321
(0.067) (0.035) (0.000)
Female -- -- -- 0.178 0.171 0.110
(0.010) (0.013) (0.132)
Married -- -- -- 0.376 0.367 0.294
(0.000) (0.000) (0.000)
Age -- -- -- 0.009 0.008 0.008
(0.000) (0.000) (0.000)
White -- -- -- -0.179 -0.189 -0.078
(0.050) (0.040) (0.421)

Number of Observations 1500 1500 1364 1488 1488 1352

LR pvalue 0.000 0.000 0.000 0.000 0.000 0.000
Psuedo R2 0.036 0.048 0.0786 0.0687 0.0795 0.1082

26
 Table 2
Mainline Protestant Denominations
American Baptist Churches in the USA
Christian Church (Disciples of Christ)
Church of the Brethren
The Episcopal Church
Evangelical Lutheran Church in America
Friend United Meeting
Moravian Church in American, Northern Prov.
Presbyterian Church (USA)
Reformed Church in America
United Church of Christ
The United Methodist Church
Evangelical Protestant Denominations
Assemblies of God
Baptist General Conference
Brethren in Christ Church
Christian and Missionary Alliance
Church of God (Cleveland, TN)
Church of the Nazarene
Conservative Congregational Christian Conference
Evangelical Congregational Church
Free Methodist Church of North America
General Association of General Baptists
Lutheran Church-Missouri Synod
Salvation Army
Seventh-day Adventist Church
Southern Baptist Convention

Table 3

Variables (Annual Frequency) Mean Std. Dev. Min Max

Mainline Protestants -0.0094 0.0047 -0.0202 0.0006
Evangelical Protestants 0.0114 0.0059 -0.0024 0.0224
Total Protestants 0.0005 0.0034 -0.0079 0.0078
Unemployment Rate 6.1611 1.4285 3.5000 9.7000
2 Year Avgerage
Unemployment Rate 6.2086 1.3226 4.1000 9.6500
Change in the
Unemployment Rate 0.0528 0.9614 -2.1000 2.9000
Real GDP Growth Rate 0.0298 0.0197 -0.0195 0.0694
Real S&P500 Growth Rate 0.0343 0.1649 -0.3247 0.2856
Real Personal Consumption
Expenditure Growth Rate 0.0330 0.0194 -0.0144 0.0680
Oil Price Growth Rate 0.1037 0.3154 -0.4709 1.3674
Yield Curve Spread (10y-3m) 0.9837 1.0939 -1.2225 3.1200
Yield Curve Spread (10y-1y) 1.4248 1.2509 -1.0301 3.4983

27
 Table 4
Univariate Regressions
Total Protestants Mainline Protestants Evangelical Protestants
Variables Coeff. (Pvalue) Coeff. (Pvalue) Coeff. (Pvalue)
NBER Recession Dummy 0.0013 (0.265) -0.0007 (0.684) 0.0054 (0.001)
Intercept 0.0001 (0.917) -0.0092 (0.000) 0.0098 (0.000)
R2 3.44% 0.44% 18.76%
Unemployment Ratet 0.0007 (0.066 ) 0.0003 (0.674) 0.0012 (0.060)
Intercept -0.0040 (0.150) -0.0114 (0.053) 0.0039 (0.418)
2
R 9.68% 1.02% 8.88%
∆ Unemployment Ratet 0.0006 (0.238) 0.0000 (0.954) 0.0018 (0.010)
Intercept 0.0004 (0.720) -0.0094 (0.000) 0.0114 (0.000)
R2 3.26% 0.00% 8.33%
∆ Unemployment Ratet-1 0.0001 (0.768) -0.0008 (0.209) 0.002 (0.032)
Intercept 0.0006 (0.253) -0.0090 (0.000) 0.0112 (0.000)
R2 0.16% 3.49% 8.34%
Real GDP Growth Ratet -0.0303 (0.260) -0.0153 (0.612) -0.0505 (0.190)
Intercept 0.0014 (0.163) -0.0089 (0.000) 0.0128 (0.000)
R2 3.16% 0.42% 2.92%
Real GDP Growth Ratet-1 -0.0093 (0.619) 0.0270 (0.336) -0.0505 (0.190)
Intercept 0.0009 (0.329) -0.0099 (0.000) 0.0128 (0.000)
2
R 0.33% 1.55% 2.92%
Real PCE Growth Ratet -0.0215 (0.358) -0.0100 (0.747) -0.0225 (0.694)
Intercept 0.0012 (0.184) -0.0090 (0.000) 0.0122 (0.000)
R2 1.53% 0.17% 0.55%
Real PCE Growth Ratet-1 -0.0121 (0.539) 0.0167 (0.547) -0.0334 (0.457)
Intercept 0.0011 (0.277) -0.0096 (0.000) 0.0124 (0.000)
R2 0.54% 0.57% 1.24%
Real S&P500 Growth Ratet 0.0027 (0.325) 0.0072 (0.075) -0.0059 (0.310)
Intercept 0.0004 (0.545) -0.0096 (0.000) 0.0116 (0.000)
R2 1.72% 6.51% 2.72%
Real S&P500 Growth Ratet1-1 0.0036 (0.333) 0.0127 (0.007) -0.0100 (0.107)
Intercept 0.0005 (0.368) -0.0094 (0.000) 0.0116 (0.000)
R2 3.31% 23.25% 7.85%
Spot Oil Price Growth Ratet 0.0029 (0.004) 0.0011 (0.374) 0.0059 (0.002)
Intercept 0.0002 (0.767) -0.0095 (0.000) 0.0108 (0.000)
2
R 7.39% 0.53% 10.01%
Spot Oil Price Growth Ratet-1 0.0018 (0.045) 0.0006 (0.609) 0.0046 (0.017)
Intercept 0.0005 0.432 -0.0091 (0.000) 0.0109 (0.000)
R2 3.28% 0.20% 6.14%
Yield Curve Spread: 10yr-3mt -0.0003 (0.492) -0.0001 (0.828) -0.0012 (0.090)
Intercept 0.0009 (0.320) -0.0092 (0.000) 0.0132 (0.000)
R2 1.26% 0.12% 6.80%
Yield Curve Spread: 10yr-3mt-1 -0.0011 (0.003) -0.0009 (0.077) -0.0020 (0.000)
Intercept 0.0022 (0.001) -0.0078 (0.000) 0.0141 (0.000)
R2 19.17% 7.04% 17.23%
Yield Curve Spread: 10yr-1yrt -0.0004 (0.438) -0.0001 0.86 -0.0018 (0.023)
Intercept 0.0009 (0.302) -0.0092 (0.000) 0.0132 (0.000)
2
R 1.65% 0.08% 10.84%
Yield Curve Spread: 10yr-yrt-1 -0.0014 (0.001) -0.0011 (0.053) -0.0026 (0.000)
Intercept 0.0020 (0.001) -0.0080 (0.000) 0.0138 (0.000)
R2 22.35% 8.20% 22.66%

28
 Table 5
Explaining Membership Variability
Variables Total Protestants Mainline Protestants Evangelical Protestants
Intercept -0.004 -0.009 0.0063
(0.038) (0.000) (0.204)
Unemployment Ratet 0.001
(0.001)
Unemployment Ratet-1 0.0013
(0.065)
Real S&P500 Growth Ratet 0.007 0.008
(0.027) (0.023)
Real S&P500 Growth Ratet-1 0.006 0.015 -0.0076
(0.010) (0.000) (0.037)
Spot Oil Price Growth Ratet 0.002 0.0022
(0.059) (0.305)
Spot Oil Price Growth Ratet-1 0.003
(0.029)
Yield Curve Spread: 10yr-1yrt-1 -0.001 -0.001 -0.0028
(0.000) (0.001) (0.001)

Number of Observations 35 35 35
R2 0.4536 0.4558 0.3877
Adj. R2 0.3594 0.3832 0.3061
F pvalue 0.0000 0.0000 0.0043

Table 6
Forecasting Models
Variables Total Protestants Mainline Protestants Evangelical Protestants
Intercept 0.0006 0.0022 -0.005 -0.005 0.0027 0.0065
(0.335) (0.000) (0.046) (0.011) (0.133) (0.005)
Dependent Variable Lagged 1 Period 0.0553 -0.0769 0.1769 -0.0162 0.4401 0.3589
(0.799) (0.668) (0.395) (0.929) (0.024) (0.048)
Dependent Variable Lagged 2 Periods 0.0333 0.1054 0.2996 0.3416 0.2919 0.2233
(0.855) (0.460) (0.019) (0.000) (0.138) (0.202)
Real S&P500 Growth Ratet-1 0.0040 0.012 -0.006
(0.160) (0.004) (0.031)
Yield Curve Spread: 10yr-1yrt-1 -0.0016 -0.002 -0.0018
(0.000) (0.000) (0.002)

Number of Observations 34 34 34 34 34 34
R2 0.0040 0.3117 0.1468 0.453 0.3976 0.5225
Adj. R2 -0.0603 0.2167 0.0917 0.3775 0.3587 0.4566
F Test (p value) 0.9536 0.0001 0.0357 0.0000 0.0004 0.0000
RMSE 0.00338 0.0029 0.00418 0.00346 0.00479 0.00441
Granger Causality Test (p value) -- 0.0047 -- 0.0016 -- 0.0344

29
 Table 7
Summary Statistics
Variables (Quarterly Frequency) Mean Std. Dev. Min Max
Conversions Growth Rate(SA) 0.0091 0.0906 -0.2061 0.3752
Unemployment Rate 5.6278 1.5081 2.6000 10.4000
Unemployment Rate Change -0.0072 0.3769 -0.9333 1.5000
Real GDP Growth Rate 0.0084 0.0096 -0.0275 0.0387
Real PCE Growth Rate 0.0087 0.0084 -0.0299 0.0502
Real S&P 500 Growth Rate 0.0100 0.0723 -0.3083 0.2072
Oil Price Growth Rate 0.0222 0.1490 -0.4401 1.3457
Yield Curve Spread (10y-3m) 1.2100 1.2109 -2.1430 3.6140
Yield Curve Spread (10y-1y) 0.8024 1.0250 -1.9400 3.2000

Table 8
Importance of Economic Shocks in SDA Conversions
% of All Shocks Creating Forecast Errors
Periods (Using Absolute Values of Shocks)
1960s 25.07%
1970s 40.28%
1980s 38.23%
1990s 35.69%
2000s 38.69%

All Decades 35.39%

Recession 45.62%
Non Recession 33.37%

30
 Figure 1

The Unemployment Rate, the 2001 Recession, and the 2001 Pew Survey
(Shaded Area = NBER Recession, Vertical Line = Pew Survey)
6.5

6.0
Unemployment Rate

5.5

5.0

4.5

4.0

3.5
2000 2001 2002 2003

Figure 2

P(Weekly Attendance|Unemployed, No P(Weekly Attendance|Employed, No

9/11 Effect) 9/11 Effect)
48.44% (43.97%,52.90%) 35.05% (31.90%,38.20%)

P(Weekly Attendance)
43.12% (40.52%, 45.71%)

P(Weekly Attendance|Unemployed P(Weekly Attendance|Employed

Protestant, No 9/11 Effect) Protestant, No 9/11 Effect)
53.55% (48.65%, 58.46%) 41.71% (37.44%, 45.98%)

P(Weekly Attendance|Unemployed P(Weekly Attendance|Employed

P(Weekly Attendance|Unemployed P(Weekly Attendance|Employed Other
Evangelical/Born Again Protestant, No Evangelical/Born Again Protestant, No
Other Protestant, No 9/11 Effect) Protestant, No 9/11 Effect)
9/11 Effect) 9/11 Effect)
40.70% (34.88%, 46.51%) 29.29% (24.65%, 33.95%)
67.49% (62.18%, 72.80%) 55.72% (50.27%,61.17%)

31
 Figure 3
45

40
Membership (Millions)

35

30

25

20

15
1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004

Total Prostestant Sample Evangelical Protestants Mainline Protestants

Figure 4
16000
14000
Conversions

12000
10000
8000
6000
4000
2000
1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004
Non S easonally Adjusted Conver sions

12000
11000
10000
Conversions

9000
8000
7000
6000
5000
4000
3000
1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004
S easonally Adjusted Conver sions

32
 Figure 5

Conversion Rate Cumulative Response to One Standard Deviation Shocks

(Dotted Line = 95% Confidence Bands)
Response to a Shock in Real GDP Growth Rate Response to a Shock in Real PCE Growth Rate
2 1. 5
1. 0
1
0. 5
0 0. 0
P ercen t

P ercen t
- 0. 5
-1
- 1. 0
-2 - 1. 5
- 2. 0
-3
- 2. 5
-4 - 3. 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Qu a rte rs Afte r Sh o c k Qu a rte rs Afte r Sh o c k

Response to a Shock in the Yield Curve Spread (10y-3m) Response to a Shock in the Yield Curve Spread (10y-1y)
2 2

1 1

0 0
P ercen t

P ercen t
-1 -1

-2 -2

-3 -3

-4 -4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Qu a rte rs Afte r Sh o c k Qu a rte rs Afte r Sh o c k

Response to a Shock in the Change in the Unemployment Rate Response to a Shock to the Unemployment Rate
4 3

3
2

2
1
P ercen t

P ercen t

1
0
0

-1
-1

-2 -2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Qu a rte rs Afte r Sh o c k Qu a rte rs Afte r Sh o c k

Response to a Shock in the Real Stock Price Response to a Shock in Oil Prices
2. 0 2. 5
1. 5 2. 0
1. 0 1. 5
0. 5 1. 0
P ercen t

P ercen t

0. 0 0. 5
- 0. 5 0. 0
- 1. 0 - 0. 5
- 1. 5 - 1. 0
- 2. 0 - 1. 5
- 2. 5 - 2. 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Qu a rte rs Afte r Sh o c k Qu a rte rs Afte r Sh o c k

33
 Figure 6

Conversion Rate Cumulative Response to One Standard Deviation Shocks

(Dotted Line = 95% Confidence Bands)
Re s pons e to a Shoc k in Yie ld Curv e Sprea d (1 0 y -1 y ) Re s pons e to a Shock in the Re a l Stock Pric e Growth Rate
1. 0 2. 0

0. 5 1. 5

0. 0 1. 0

-0. 5 0. 5
Percent

Percent
-1. 0 0. 0

-1. 5 -0. 5

-2. 0 -1. 0

-2. 5 -1. 5

-3. 0 -2. 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Qua rte rs Afte r Shoc k Qua rte rs Afte r Shoc k

Re s pons e to a Shoc k in the Cha nge in the Une m ploy m e nt Rate Re s pons e to a in Oil Price Growth Ra te Shoc k
3. 0 2. 0

2. 5 1. 5

2. 0 1. 0

1. 5 0. 5
Perc ent

Perc ent
1. 0 0. 0

0. 5 -0. 5

0. 0 -1. 0

-0. 5 -1. 5

-1. 0 -2. 0

-1. 5 -2. 5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Qua rte rs Afte r Shoc k Qua rte rs Afte r Shoc k

34
 Figure 7

Historical Decomposition
0.4

0.3
Conversion Growth Rate

0.2

0.1

0.0

-0.1

-0.2

-0.3
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Actual Baseline Forecast Baseline Forecast + Economic Shocks

35