VOTING AND IMITATIVE BEHAVIOR PHILLIP NELSON* Political behavior generates private benefits by helping people fit in with desired fiiends. A voter imitates other voters, but at the same time they imitate him. An equilibrium solution requires exogenous variables: the narrow self-interest of the par- ticipants. The reduced form makes one’s vote a function of the narrow self-interest of others as well as one‘s own. In accord with the model, a person’s party ident@cation depends on his ethnic group’s current income and its income in 1909 as well as his own income. I. INTRODUCTION What determines how people vote? Most economists, such as Peltzman [1980], have emphasized self-interested motiva- tions. But Kau and Rubin [1982] and Kalt and Zupan [1984] claim that ideology, based in part on altruism, influences vot- ing patterns. Both hypotheses have the same funda- mental flaw: they focus on the conse- quences of the policies for which people vote. In the Peltzman model a person votes for those policies which maximize his real income. I call this narrow self-in- terest. In the standard altruism models of economists, a person maximizes some weighted average of the real income of himself and 0thers.l However, an individual’s vote has a minuscule impact on the outcome of any election. Therefore, his vote for a policy yields extremely small expected returns through the impact of that policy on the voter. Other returns from voting will dominate in determining behavior, if such returns exist. * Professor, Binghamton University. I acknowl- edge the help of Vivian Carlip, Ken Greene, Cliff Kern, and the editor and referees of this Journal. 1. I will use ’‘altruism‘’in that sense throughout this paper. Some sociologists,on the other hand, take a broader view of altruism. The criticism that follows is not applicable to this broader view. Economic Inquiry VoI. XXXII, January 1994, 92-102 Some economists, for example, Brennan and Buchanan [1984], have recognized this problem with the standard or ”economic” view. No one, however, has developed an alternative theory based on other returns to voting. The key point of this paper is that political behavior can generate pri- vate benefits by helping people fit in with desired friends and associates: a “herd effect” to give this phenomenon a name, Political positions are thus chosen not be- cause they produce the desired outcomes for voters, but rather because voters want to associate with certain people who have certain positions; that is, people imitate others in choosing political positions. The idea for imitation in political posi- tions is hardly new. Berelson et al. [1954] provide a prima facie case for imitation in voting behavior. They found that a person’s political position was closely re- lated to the political positions of his co- workers and friends. Of course, these re- sults could be attributable to the fact that these groups share common characteris- tics, but the Berelson data also show that people change their political position to- ward that of their associates-a phenom- enon more difficult to explain by the com- mon characteristic hypothesis. Imitation is uninteresting as a predictor of behavior by itself, however. A voter imitates other voters, but at the same time 92 @WesternEconomic Association International
Transcript
VOTING AND IMITATIVE BEHAVIOR PHILLIP NELSON*
Political behavior generates private benefits by helping people fi t in with desired fiiends. A voter imitates other voters, but at the same time they imitate him. A n equilibrium solution requires exogenous variables: the narrow self-interest of the par- ticipants. The reduced form makes one’s vote a function of the narrow self-interest of others as well as one‘s own. In accord with the model, a person’s party ident@cation depends on his ethnic group’s current income and its income in 1909 as well as his own income.
I. INTRODUCTION
What determines how people vote? Most economists, such as Peltzman [1980], have emphasized self-interested motiva- tions. But Kau and Rubin [1982] and Kalt and Zupan [1984] claim that ideology, based in part on altruism, influences vot- ing patterns.
Both hypotheses have the same funda- mental flaw: they focus on the conse- quences of the policies for which people vote. In the Peltzman model a person votes for those policies which maximize his real income. I call this narrow self-in- terest. In the standard altruism models of economists, a person maximizes some weighted average of the real income of himself and 0thers.l However, an individual’s vote has a minuscule impact on the outcome of any election. Therefore, his vote for a policy yields extremely small expected returns through the impact of that policy on the voter. Other returns from voting will dominate in determining behavior, if such returns exist.
* Professor, Binghamton University. I acknowl- edge the help of Vivian Carlip, Ken Greene, Cliff Kern, and the editor and referees of this Journal.
1. I will use ’‘altruism‘’ in that sense throughout this paper. Some sociologists, on the other hand, take a broader view of altruism. The criticism that follows is not applicable to this broader view.
Economic Inquiry VoI. XXXII, January 1994, 92-102
Some economists, for example, Brennan and Buchanan [1984], have recognized this problem with the standard or ”economic” view. No one, however, has developed an alternative theory based on other returns to voting. The key point of this paper is that political behavior can generate pri- vate benefits by helping people fit in with desired friends and associates: a “herd effect” to give this phenomenon a name, Political positions are thus chosen not be- cause they produce the desired outcomes for voters, but rather because voters want to associate with certain people who have certain positions; that is, people imitate others in choosing political positions.
The idea for imitation in political posi- tions is hardly new. Berelson et al. [1954] provide a prima facie case for imitation in voting behavior. They found that a person’s political position was closely re- lated to the political positions of his co- workers and friends. Of course, these re- sults could be attributable to the fact that these groups share common characteris- tics, but the Berelson data also show that people change their political position to- ward that of their associates-a phenom- enon more difficult to explain by the com- mon characteristic hypothesis.
Imitation is uninteresting as a predictor of behavior by itself, however. A voter imitates other voters, but at the same time
92
@Western Economic Association International
NELSON IMITATION 93
they imitate him. An equilibrium solution requires exogenous variables. I assume these to be the narrow self-interest of the participants. The resulting model differs from a simple narrow self-interest model. In the reduced form, one’s political deci- sion is a function of the narrow self-inter- est of others as well as one’s own narrow self-interest. Imitation along with the self- interest of the participants leads both to additional tests of imitation and makes the imitation model of interest in understand- ing political behavior.
Imitation could be an attempt to please others who have a primitive preference for friends who imitate themselves. Alterna- tively, imitation could signal an offer of friendship by the imitator. Without invok- ing preferences for friends who imitate, such signalling could therefore explain why people care whether somebody imi- tates their political behavior or not. Imita- tion could also arise from another process. People’s views of appropriate behavior could be determined by the behavior of those with whom they associate.2 Narrow self-interest can also generate a form of imitation: imitating the behavior of people with like interests who are more informed. Because of the low private payoff to more- informed voting, one would not expect this to be a major source of voting imita- tion. My analysis is based on the first two sources of imitation: that people want their friends to imitate themselves. Except for the section on lying, below, the impli- cations I examine do not depend upon the reasons for imitation.
I I . IMITATION
To deal systematically with political po- sitions it is necessary to quantify them. If there is but a single issue, the issue itself generates a simple metric. Fortunately,
2. In the language of Coleman [1990] the first two processes are externalizations and the third an inter- nalization of social norms.
Poole and Romer [1985] show that a single dimension does dominate political behav- ior, at least as shown by campaign contri- butions. For expository purposes I as- sume, 6 la Peltzman [1980], that the single issue is non-defense government expendi- ture with fixed proportions among its components and its financing. The politi- cal position of any person i, Pi, is mea- sured by the amount of that expenditure that he advocates. Si is defined as the amount of that expenditure that maxi- mizes his own income.
Utility should be a declining function of both the difference between one’s polit- ical position and that of someone else whose friendship is valued, and of the difference between the political position one adopts and one’s income-maximizing political position. I assume that utility for the ith person takes the following form:
2 (1) Ui = ci c wij[-(Pi - Pi)*] - hi (P i - Si)
i
where wij is the weight that i gives to im- itating j ’ s political behavior, and
c w.. = 1. ‘I i
Maximizing Ui with respect to Pi,
(1 + bJP; = c WijPi + b,S, ‘ 1
where b, = hi / c i . To get an explicit solution for the Pi,
consider a simple case. Assume that bi is the same for all i. Suppose that there are only two groups, with nl people having S , = 0 and n2 having S2 = x . Assume further that all those in a group have the same wii. Then, equation (2) becomes
94 ECONOMIC INQUIRY
(3) (1 + b)P, = (n,-~)wllP1 + n2w12P2,
(1 + b)P, = (n2-1)w2,P2 + n,w,,P, + bx.
Given that the sum of each person's weights equals 1, the solution for political positions is
P, = x(b + n,w,,)
If wii > 0 and j f i, 0 < Pl< P2<x. Pl < P, is consistent with a simple narrow self-inter- est model. But the other part of the ine- qualities-a shift of political positions to- ward the mean-is not. But this shift could be substantial. It does not require prepos- terous assumptions about the parameters of equation (4) to obtain a substantial im- pact of the political position of others on one's own political position. Suppose, for example, that nl = n2 = 5, b=.01, x = 1, and wii =lowij with j f i. (Our theory requires a low b and one observes association pat- terns by income consistent with a low wij relative to wii.3) Then Pl=.478 and P, = .522 much closer to the mean political position than to their respective self-inter- est political positions (0 and 1). Even if wii = 100wip j + i, and all other conditions remain the same, Pi = ,356 and P2 = .644
The imitation hypothesis can be tested by examining how behavior changes as the parameters of equation (4) change.
111. GROUP EFFECTS
Call a group an association group if a member of the group has a higher proba-
3. The relationship between association patterns and w is discussed below.
bility of associating with another member of the group than with a person outside the group. As one can confirm by data on marriage patterns, ethnic and religious groups are clearly association groups. There should be a close relationship be- tween these association patterns and the weights a person gives to the political positions of others for two reasons. First, actual associations are determined in part by the preferences measured by the w. Second, the very fact of association tends to produce a higher w. It pays to give greater weight to the political position of those with whom one might associate than to the position of others.
The imitation model predicts that indi- viduals in association groups will vote in terms of the income of their group as well as their own individual incomes. A mem- ber of a high-income group has more high-income people to court by imitation. Such imitation produces more advocacy of high-income behavior, holding constant one's own income. For example, suppose a person associates exclusively with a dominantly high-income group. Assume, in terms of equation (4), that n,=2, n2=8, wii = 10wij, j f i, b=.01, x = 1. A low-income person in that group has P1=.922 (P2=.942). If he were a member of the comparable low-income group (nl=8, n2=2) under the same other conditions, his P,=.058
To test this hypothesis one must have an empirical measure of political posi- tions. The Republicanism of voters serves as a measure of opposition to non-defense expenditures as shown by Peltzman [1984] and Kau and Rubin [1982]. Along with observations from 15,125 individuals on other variables over the period 1972-1986, NORC [1986] provides data with seven levels of that variable from Strong Demo- crat (0) to Strong Republican (6). I call this measure RN.
While obvious this scaling is also some- what arbitrary. Party identification is of
(P,=. 078).
NELSON: IMITATION 95
interest because it can help predict the behavior of voters. If the difference be- tween Strong Democrat and Not Very Strong Democrat has half the impact on electoral decisions of the difference be- tween Not Very Strong Democrat and In- dependent Close to Democrats, that differ- ence should be scaled by half as much. I have devised another scaling, called R, based on that pr in~iple .~
To test the hypothesis that people in high-income groups will have high P, holding constant their own income, one must separate the effect of an individual’s income from the effect of the income of his group. The coefficients ( B ) of the ethnic dummies in the Republicanism equation of column 1 in Table I measure the effects of group membership on the individual’s political position, holding constant a wide array of individual characteristics includ- ing individual income. Those group ef- fects should be positive, i.e. more Repub- lican, when the group is a high-income group. Hence, the imitation model pre- dicts that B should be positively related to a measure of the income of ethnic groups. The measure I use is the set of ethnic coefficients ( I ) in a regression of a person’s family income on ethnic dummies (in col- umn 2 of Table I) when I control for age and those variables that have big associa- tion effects of their own (region and city- size).5
4. I ran regressions with the dependent variable Ford; Ford=l if a person voted for or would have voted for Ford in 1976, Ford=O if a person voted for or would have voted for Carter in that year. The independent variables were all the independent variables of Table I and dummy variables reflecting the value taken by RN. To determine, for example, the impact of the effect on R as one moved from Strong Democrat to Not Very Strong Democrat, observations were restricted to in- dividuals with these two values of RN, and the dummy variable was set to 1 for Very Strong Democrat and 0 if not Very Strong Democrat. Analogous regres- sions were repeated for all possible adjacent values of RN. I used the coefficients of these dummy variables to scale R. This results in the following conversion of RN to R: 0 to 0, 1 to .124, 2 to .139,3 to .323,4 to ,507, 5 to .522, 6 to 546.
5. Results are not substantially changed when these controls are not used.
This prediction is confirmed for a sam- ple of thirty-one ethnic groups. In testing for this relationship I eliminate the ethnic groups that are beneficiaries of affirmative action-Blacks, Native Americans, and American Hispanics-because their be- havior can be explained by narrow self-in- terest.6 A regression of B on I for the remaining ethnic groups yields
(5)7 B = -.0098 + .173 1. (-.89) (2.53)
(t-values in parentheses)
An alternative explanation for these re- sults is the permanent income hypothesis: that ethnic group income provides a mea- sure of permanent income, even given current income, education, and unemploy- ment status. Hence, narrow self-interest could explain the role of the ethnic dum- mies in the same way that it explains the role of current income.8
A version of the altruism hypothesis also generates a relationship between group income and voting behavior. If peo-
6. The primary system insures that the party that specializes in the interests of low-income voters qua low-income voters will tend to be the advocate of their other interests. To succeed in a primary a candidate must appeal to the majority of his own party. In con- sequence, the Democratic party has pushed affirmative action and some kinds of civil rights legislation. Blacks, Native Americans and American Hispanics were ben- eficiaries of that legislation. I do not control for the differential effects of affirmative action on the ethnic groups that are hurt by it. But one expects the losing low-income groups to be harmed by affirmative action more than the losing high-income groups because the former are closer competitors on jobs, schools and res- idences with low-income affirmative action benefici- aries. This bias clearly cannot explain the observed positive relationship between group income and Re- publicanism for the ethnic groups who are losers from affirmative action.
7. This procedure is better than including the pres- ent income of ethnic groups directly into the original Republicanism equation of column 1, Table I in place of the ethnic dummies. Income does not fully explain the ethnic effects, so an individual’s disturbance term will be related to the disturbance terms of other mem- bers of his ethnic group in the alternative procedure. The alternative leads to results similar to the text’s.
8. I estimate that effect to be about 2 percent of the total effect given in equation (5). A discussion of the procedure is available on request.
96 ECONOMIC INQUIRY
TABLE I Regression Results and Related Dataa
Independent Variablesb Dependent Variables' 1909 Republican Income Wages
Self Interest Variables: Income Self Employed Full-Time Employee Government Aid
Industries: College Other Education Public Administration Hospitals Union
Regions: North East Middle Atlantic East North Central West North Central South Atlantic East South Central West South Central Mountain
City: Large SMA: Central City Large Sh4A: Suburb Other SMA: Central City Other SMA: Suburb Other Urban
Ethnic: Africans Chinese Japanese Philippine Indian Arab Greek Yugoslav Spanish Portuguese Hungarian Russian
.0240
.0273
.0024 -.0225
-.0442 -.0196 -.0129 -.0146 -.0338
.0364
.0511
.0213
.0071 -.0148 -.0320 -.0358
.0206
-.0548 -.0003 -.0371 -.0069 -.0127
-.1231 .1104 .0092 .0085
-.0054 -.0272
.0942 -.0309 -.0082 -.1102 -.0568 -.0262
(8.30)d (5.03)
(-4.63) (.58)
(-3.49) (-2.89) (-2.18) (-2.16) (-9.02)
(3.54) (7.09) (3.16)
(-82) (-2.06) (-3.41) (-4.30) (2.01)
(-6.78) (-.05)
(-5.42)
6.97) (-2.30)
(-14.60) (2.30)
(*20) (W
(-.09) (-.38)
(1.87) (797) (-.35)
(-2.72) (-2.52) (-1.40)
S669 .lo36 .0474 .0270
-.0399 -.0522
.0233 -.0075
.0751
.3698
.0586
.2569
.Of344
-.1572 .2725 .loo2 .1690 .0472
.3166
.0609
-.0727
-.0514 .1702 .1283 .2081
(5.34)d (4.73)
(1.02) (2.32)
(-1.85) (-1.85)
C93) (-.23)
(3.07) (15.71) (2.84)
(11.94) (5.04)
(-6.16)
(1.87) (.70)
(.27) (-.34)
(2.07) (.63) C73)
(1.28)
(1.38) (1.87) (3.66)
8.12 8.41
11.69 10.51 8.10
11.65 11.01
NELSON IMITATION 97
TABLE I continued Regression Results and Related Dataa
Independent Variablesb Dependent Variables' 1909 Republican Income Wages
Irish Lithuanian Rumanian Mexican German English Scot t ish Danish Finnish Italian French Belgian Austrian Czechoslovak Dutch Nonuegian Swede Pole West Indian Puerto Rican South, Central American American Indian French Canadian Other Canadian Swiss "American
A g e Age Square Age Slopef Education Sex Year Intercept R Squared
-.0062 .00006
-.0012 ,0060 .0038 .0015 .2444 ,1254
(-9.44) .0584 (30.90) (8.23) -.0007 (-34.45)
(-9.41) (8.14)
(3.46) C95)
(6.42) -.3580 (-.17) .1668
asample Size: 15,125 Source of Data: National Opinion Research Center [1986]; Higgs [1971]. bRepublican = Republican party identification scaled by the 1976 Presidential Election. (For details see
fn. 4). Range from 0 to ,646. The results are robust with respect to the scaling procedure. Income = Family income in a year divided by mean family income in that year. Quadratic, cubic and
log income functions were also tried. The squared and cubic terms were not significant and the log income terms worked no better than income.
Self Employed Dummy = 1 if person or spouse was self employed. Full-Time Employee Dummy = 1 if person was a full-time employee. Gouernmenf Aid Dummy = 1 if person or spouse received government aid in the last five years. Regions: Pacific is the region of comparison. City: Rural is the city category of comparison. Ethnic: Ethnicity unspecified is the ethnic group of comparison. Religions: Other Religions, mainly Protestant, is the religion of comparison. Attendance - Number of days attended church per year. The attendance variables are attendance times
Education = Years of school. A squared education term was also tried, but not found to be significant. Sex = 1 if Male. Year - year of interview. ‘Republican = Republicanism (R defined in fn. 4). Income = Family income defined in fn. b above. 1909 Wages = Observed past income. See fn. 10. dt-statistics in parentheses. yhese averages are not additional variables. They are the coefficients of the given religious variables
taken at the mean level of church attendance for that group compared to the omitted group (Protestants) taken at the mean level of church attendance for that group.
‘Slopes are also not additional variables. Instead, they are combinations of the appropriate independent variables that yield slopes at the mean value of that independent variable.
the appropriate religious dummy.
NELSON: IMITATION 99
ple give greater weight to the well-being of other people the more contact they have with them, they will vote in terms of group income as well as individual in- come. Fortunately, there is a version of the imitation hypothesis that more clearly distinguishes itself from the ob- vious competition.
IV. LAGS
Consider individual voting behavior as a function of group income sixty years in the past. Neither the altruistic nor the narrow self-interest models other than im- itation for information will produce that relationship with any reasonable specifica- tion. All of the alternative hypotheses- self-interest and altruism alike-require group income to measure either the self- interest position of the group or the per- manent income of the individuals in the group. Group income of sixty years ago will not serve as such a measure, espe- cially when present group income is also included in the regression.
In contrast, long lags make sense in the imitation model, though not required by it. There are reasonable specifications of the imitation model with long lags. Con- vert equation (4) to a set of simultaneous difference equations. Assume that people know the political position that maximizes their immediate self-interest, but that they discover with a lag the political position of others.
Look at the case where a person associates exclusively with a dominantly high-in- come group. Assume that within the group n,=2, n2=8, wii = loulij, j # i, b=.01,
x = l . But suppose that group was for- merly a low-income group (nl=8 and n2=2). It takes seventy-three periods for both low-income and high-income mem- bers of that group to move their political positions half-way between their equilib- rium low-group-income position and their equilibrium high-group-income po~i t ion .~ Equation (6) is a model of custom: how people’s concern with the present posi- tions of others, as they see them, generates only a slow response to changes in the self-interest of the group.
While this is just an example, equation (6) implies in general that lags will be longer the smaller is b. The term b-the relative returns to a person of the conse- quences of his actions compared to the imitative returns from those actions-will be smaller in the case of large group pro- cesses such as voter behavior than in pri- vate market decisions. If b = l rather than .01 in the previous example, it takes only one period to move half way between the two equilibria. One, therefore, expects greater lags in the political behavior case than in usual economic problems with their larger b’s. However, one does not know precisely how long a lag to expect, so the imitation model itself does not provide any precise guide to the expected period of the lag. My choice of the lag period is, therefore, dictated by other con- siderations. I want to choose a lag suffi- ciently long that it cannot be explained by the alternative hypotheses, and that pres- ent and past group income are clearly
9. Given equation (6) and the assumed values in the text, then
P1f=.0091495(.52306)f - 37285q.99014)‘ + .92166,
P 2 f r .0005614(.52306)‘ - .872918(.99014)’ + .94240.
Solve for the integer t that yields an average of Plt and P2f as close to .5 as possible. (.5 is half-way be- tween the average of the equilibrium values of PI and P2 for our low- and high-income groups.)
100 ECONOMIC INQUIRY
distinguishable, and I must work within data constraints. The voting behavior ob- served is for 1972-86. The past-year in- come chosen is for 1909. My test will be of the imitation model with lags of sixty- three to seventy-seven years against the hypotheses of non-imitative narrow self- interest, altruism, or the imitation model with much shorter lags.
I test the lag hypothesis using ethnic groups. I use as an imperfect measure of their past income the weekly wages of the foreign-born in 1909 by ethnic groups, as shown in column 3 of Table I.'O Call these wages PI. Results of adding 1909 wages (PI) as an additional explanatory variable in equation (6 ) for the twenty-seven groups for which it is available are
(7) B = -.113 + .095 I + .0091 PI. (-2.28) (1.33) (2.15)
This test helps establish the importance of imitation, but does not establish the reason for imitation. There is, however, a way of distinguishing between two of the reasons: imitation for reputation and imi- tation through information.
V. LYING
Imitation for reputation, unlike imita- tion through information, produces incen- tives for a person to lie. The lying can take two forms: (1) voting one's self-interest by secret ballot while adopting verbal politi- cal positions to please others; (2) varying one's verbal political position to please the particular audience addressed. The re- turns from the first type of lying are lim- ited because the private gains to voting one's self-interest are so minuscule. Even a small cost of lying can bring verbal and voting behavior together, and there do
10. Higgs [197l] estimates wage rates for the for- eign-born by country in 1909 for twenty-four relevant countries. In addition I estimate three others-Austria, Spain and Switzerland-by taking the unweighted av- erage of the wage rates for the countries bordering the country with the missing observation.
seem to be such costs. Not only is lying somewhat unpleasant to most of us, but our own responses to this discomfort make the act of lying sometimes observ- able to others." It is not surprising that there is not much of the first kind of lying. As a result polls are useful predictors of voting behavior.
In contrast, there can be substantial private returns to the second kind of lying: to please different associates with differ- ent political views. (The second kind of lying also has greater costs, since the prob- ability of detection is greater if the two sets of friends talk to one another.) There is substantial evidence for the second kind of lying when the probability of detection is small. For example, Reese et al. [1986] found that answers to ethnically sensitive political and social questions depended significantly on the ethnicity of interview- ers as well as the ethnicity of respondents. At least some part of imitative behavior is attributable to imitation for reputation.
VI. RELIGIOUS GROUPS
Equation (7) includes ethnic groups, but not religious groups. While the behav- ior of religious groups is explicable only in a complicated way by the imitation model,'* that behavior is utterly devasta-
11. Nietzsche says, "One can lie with the mouth, but with the accompanying grimace one nevertheless tells the truth." (Frank [1988])
12. The behavior of Catholics and Protestants is consistent with equation (5), though not in terms of present income. Catholics had lower past wages ( P o and are Democrats; Protestants, with higher PI, vote Republican. But Jews are Democrats yet have higher than average PI, at least in terms of our measures. There are two explanations: (1) Measurement error. Our measure is the income of the foreign-born, not the income of the entire ethnic group including the native-born. Jews had higher than average wages among the foreign-born in 1909, but their smaller pro- portion of native-born significantly lowers the relative wages of the whole religious group in 1909. (Native- born have higher wages than the foreign-born.) (2) Lags from their pre-U.S. past. Baron [1976] showed that Jews in Europe were decidedly left-wing. How- ever, this latter explanation only transfers the problem to an earlier period.
NELSON IMlTATION 101
ting to the narrow self-interest model. Given the higher present income of Cath- olics and Jews than Protestants, the signs of the religious coefficients in the Repub- licanism equation of Table I are clearly inconsistent with the predictions of any variant of the narrow self-interest model (except imitation for information) that produces a relationship between individ- ual behavior and group income.
Religious groups provide another test of the imitation model. Group effects should be larger the more a person asso- ciates with the group. One way in which a person associates with co-religionists is through church activities. The more a per- son attends church, the greater (in abso- lute value) the expected effect of the reli- gious dummies on Republicanism. That is, indeed, the case. The cross-product of church attendance with religion is positive for Protestants (t=4.38). Protestant, the control group in my study, has an implic- itly positive coefficient, since all the reli- gious dummies have negative coefficients. This cross-product is negative for Catho- lics and Jews with t= -3.36 and -1.50 respectively. Both Jewishness and Cathol- icism are negatively related to Republican- ism. It is hard to see how the narrow self-interest model, except, again, imita- tion for information, would lead to that prediction. (Incomes, for example, signifi- cantly increase with church attendance no matter what the religion.)
VI. CONCLUSION
The imitation model is better both the- oretically and empirically than either the standard narrow self-interest or altruism models of voter behavior. In contrast to the other theories, the imitation model appropriately focuses on private returns to private actions rather than public re- turns. Empirically, the observed impacts of present and past group income and church attendance on voting behavior are inconsistent with any version of narrow self-interest except imitation for informa- tion. That version of narrow self-interest has two problems: (1) the large-group problem endemic to any narrow self-inter- est model of voting behavior; (2) a failure to explain why people vary their political position as the position of those to whom they talk varies.
The main empirical conclusion is that associations matter profoundly in deter- mining voting patterns. Ethnic and reli- gious groups, not including the benefici- aries of affirmative action, explain 32 per- cent of the explained variance of Republi- canism, while the self-interest variables of Table I and education (partially a measure of permanent income) explain only 20 percent.13 This theory helps explain what the narrow self-interest model (except im- itation for information) cannot explain- the large proportion of conflicts generated by ethnic and religious differences.
13. In both cases these percentages are determined by R2 with all variables included, minus R2 with the relevant variables excluded as a percent of the former. This comparison exaggerates the importance of narrow self-interest in political behavior because most of the self-interest groups are also association groups. How- ever, one’s membership in a religious and ethnic group is less likely to vary over time, especially over an inter- generational time span, than is one’s membership in self-interest groups. This increases the importance of present ethnic and religious membership on political behavior compared to membership in present self-in- terest groups.
102 ECONOMIC INQUIRY
REFERENCES
Baron, Salo. The Russian Jews under Tsars and Soviets. New York Macmillan, 1976.
Berelson, B., P. Lazarsfeld, and W. McPhee. Voting. Chicago: The University of Chicago Press, 1954.
Brennan, G., and J. Buchanan. "Voter Choice and the Evaluation of Political Alternatives." American Behavioral Scientist, November 1984, 185-201.
Coleman, James. Foundations of Social Theory. Cam- bridge: Harvard University Press, 1990.
Frank, Robert. Passions Within Reason. New York Nor- ton, 1988.
Higgs, Robert. "Race, Skills, and Earnings: American Immigrants in 1909." Journal of Economic History, June 1971,420-28.
Kalt, J., and M. Zupan. "Capture and Ideology in the Economic Theory of Politics." American Economic Review, June 1984,279-300.
Kau, J., and P. Rubin. Congressmen, Constituents, and Contributors. Boston: Martinus Nijhoff, 1982.
National Opinion Research Center. General Social Sur- veys 1972-86. Chicago, 1986.
Peltzman, Sam. "The Growth of Government." Journal of Law and Economics, October 1980, 209-87.
-. "Constituent Interest and Congressional Vot- ing.'' Journal of Law and Economics, April 1984,
Poole, K., and T. Romer. "Contributions to the 1980 Campaigns for the United States House of Rep- resentatives." Public Choice 47(1), 1985, 63-111
Reese, S. et al. "Ethnicity of Interviewer Effects Among Mexican Americans and Anglos." Public Opinion Quarterly, Winter 1986, 563-72
