Prof.
Bryan Caplan
bcaplan@gmu.edu
http://www.bcaplan.com
Econ
854
Week 5: Voter Motivation, II:
Ideological Voting
I.
Factor Analysis
A.
One statistical technique social scientists outside of economics use a
great deal is factor analysis.
B.
The main idea of factor analysis: reducing a lot of variables to a
smaller number of "summary" variables, aka "factors" or
"dimensions."
C.
The classic example: intelligence testing. A test has 100 items. Is it possible to extract a smaller number of
summary variables?
D.
Yes. In fact, factor analysis on
variables related to cognitive ability normally finds ONE overriding factor
(called g for "general intelligence"). Cognitive ability is essentially
"onedimensional."
E.
Performance on individual test items can be seen as a function of g
plus noise. The greater the predictive
power of g, the higher we say the item's gloading is.
1.
Ex: Analogies have a higher gloading than pure memory tasks.
F.
Factor analysis in no way guarantees the existence of a single
overriding factor. For example, on
personality tests, factor analysis normally extracts FIVE unrelated
factors.
G.
Factors do not label themselves.
Ordinary language terms are convenient, though occasionally misleading.
1.
Ex: OCEAN
H.
On purely random data, no factors would emerge.
II.
The Dimensionality of
A.
There are many different ways to analyze political beliefs.
1.
Libertarianstatist spectrum
2.
Christiansecular spectrum
B.
What can factor analysis tell us about the dimensionality of
C.
Strong result: As with intelligence, empirical tests typically find
that political opinion is roughly one dimensional.
D.
What is the dimension?
Empirically,
B.
On a deep level, this spectrum may not make a great deal of sense. Libertarians, for example, often argue that
there are really two dimensions  personal freedom and economic freedom:
1.
Libertarians  propersonal, proeconomic
2.
Populists  antipersonal, antieconomic
3.
Liberals  propersonal, antieconomic
4.
Conservatives  antipersonal, proeconomic
E.
But empirically, most people line up on the diagonal, and the other two
boxes are sparsely inhabited.
F.
G.
A second dimension (related to race) occasionally pops up, but is no
longer important. P&R's story:
During the 50's, otherwise liberal Southern Democrats often opposed civil
rights measures, and otherwise conservative Republicans often favored them. Once the Southern Democrats left the party,
and debate shifted from "equality of opportunity" to "equality
of result," position on further civil rights measures began to correlate
well with the rest of the liberalconservative dimension.
H.
Similarly, Levitt and earlier researchers have found that
onedimensional ideological measures of lc like
I.
Less work has been done on the dimensionality of individual citizens'
opinions, but once again, a strong liberalconservative dimension pops out of
the data.
J.
Remarkably, voting in the U.N. is also onedimensional, in spite of the
extreme heterogeneity of the participants.
The dimension is something like "attitudes towards the
U.S./Israel."
III.
Ideological Voting
A.
As mentioned earlier, the main problem with the simple sociotropic
voting model is that it has trouble explaining disagreement.
B.
The empirical evidence on ideology suggests a more sophisticated
interpretation of sociotropic voting.
C.
Motivation is indeed sociotropic: People support the policies they
think are in the public interest.
D.
But: There are large ideological disagreements about the public
interest. Ideology determines beliefs
about what policies "work" and what counts as
"working."
E.
Ex: Affirmative action.
Conservatives and liberals argue about whether it works (are blacks
betteroff because of it?), but also disagree about what it means to
"work" (a "level playing field" versus a "fair
outcome"?).
F.
Important theoretical point: If ideology is onedimensional, and people
largely vote ideologically, then the simple MVT's seemingly strong assumptions
are satisfied. Perhaps the issuespace
only looks multidimensional.
V.
Ideology and Reduction
A.
Main objection to ideological voting model: Can't ideology be reduced
to personal interests?
B.
Ex: Isn't conservatism just the "ideology of the rich," and
liberalism the "ideology of the poor"?
C.
No. The correlation between
income and professed ideology is very low.
In the GSS, for example, the correlation between real income and
POLVIEWS (a 17 measure of leftright ideology) is .06.
D.
So what does determine ideology?
Is it education?
E.
Once again, no. Education and
ideology are close to unrelated (r=.03)
when you look at a random sample of Americans from the GSS (as opposed
to, say, a 50/50 sample of random Americans and university faculty!).
F.
In a multiple regression framework, there is a tendency for income to
make people more conservative and education to make people more liberal. [Table 2]
G.
Both are clearly statistically significant, but the actual effect is
small. On a 6point scale:
1.
Raising log of real income by 1 – a huge change  makes people .096
units more conservative.
2.
Going from a high school degree to a BA makes people .084 units more
liberal.
H.
What then is ideology? As far as
anyone can show, ideology is an independent causal force. Ideology explains a great deal about people's
beliefs, but no standard social science variable does much to explain ideology.
I.
Maybe someone will one day show that ideology reduces to something
else, but given the failure of all the obvious candidates, I doubt it. (But stay tuned for the genetics of politics
next week!)
VI.
Case Study: The Determinants of Party Identification, II
A.
Question: Returning to last
week's linear probability model of party identification, what happens in the
GSS if you also control for stated ideology?
N≈41,000, so focus on
magnitudes, not tstats.
B.
[Tables 3a&3b]
C.
Answer: Ideology matters even
more than race. Moreover, the slight
change in the other coefficients shows that ideology is far from a "mere
proxy for selfinterest."
D.
Consider two examples for 2010.
1.
Ex. #1: Black female with $1M
annual income in 1986 dollars, 30 years old, college graduate.
2.
Ex. #2: White male with $10k
annual income, 30 years old, high school education, conservative ideology.
E.
Ex. #1: [Since we don't know
ideology, use Tables 1a and 1b]
Estimated probability of being a Democrat: 56.4%; estimated probability
of being a Republican: 26.6%.
F.
Ex. #2: [Using Tables 3a and 3b] Estimated probability of being a Democrat:
6.8%; estimated probability of being a Republican: 59.1%. (Age coefficient to one more decimal
place=.0005).
VII.
Income, Education, Ideology, and Opinion
A.
For specific opinions (as opposed to party identification), income
empirically often seems to make a large difference.
1.
Ex: High income people seem much more in favor of immigration than low
income people.
B.
But
the effect of income almost always disappears once you control for education. Ph.D.s who drive cabs think like other
Ph.D.s, not other cab drivers.
C.
How does education affect opinion?
More educated people tend to be both more tolerant and more appreciative
of free markets.
D.
Even though voting is onedimensional, opinion looks
twodimensional.
E.
Moreover, the two dimensions more or less fit the twodimensional
personal freedom/economic freedom diagram.
Education shifts the diagonal up and to the right.
F.
This fact suggests that politicians might really compete over two
dimensions rather than one, again raising doubts about the median voter model.
G.
In practice, however, the liberalconservative dimension appears to be
far more electorally salient. Education
affects issue beliefs, but appears to be independent of party
identification.
H.
Why? How come liberals ally, but
not high school dropouts?
VIII.
Case Study: Economic Beliefs
A.
Now let us go through two illustrations from the SAEE: tendency to
blame economic difficulties on:
1.
Immigration
2.
"Excessive profits"
B.
If we do not control for education, income appears to have a large effect
on these beliefs. [Table 4a, 4b]
C.
Controlling for education, though, makes the apparent effect of income
almost disappear. [Table 5a, 5b]
D.
Immigration.
1.
Opposition shrinks as education rises.
2.
Opposition grows as conservatism rises.
E.
"Excessive profits."
1.
Assigning blame falls as education rises.
2.
Assigning blame falls as conservatism rises.
IX.
The Ideology*Education Interaction
A.
Ideology and education interact in an interesting way. Despite their slight correlation,
ideology*education has more predictive power than ideology alone.
B.
Simple explanation: The higher your education level, the more likely
you are to know what your ideology says about a given topic. For someone with a gradeschool education,
"liberal" is just a word; for a Ph.D., it is an integrated worldview.
C.
This works for party identification: The tstat on ideology*education is
higher than the tstat on ideology alone, rising from 44 and 61 to 48 and 67. [Tables
3a&3b vs. Tables 6a&6b]
D.
It also works on individual issues.
For immigration, the tstat rises from 3.9 to 4.3 [Table 4a versus 7a];
for excessive profits, from 4.6 to 4.9 [Table 4b versus 7b].
E.
Returning to the twodimensional diagram, education
"stretches" the liberalconservative spectrum.
Table
2: The Determinants of Ideology (POLVIEWS rescaled to go from 3 to +3)
Table
3a: Conditional Probability of Being a Democrat, with Ideology
Table
3b: Conditional Probability of Being a Republican, with Ideology
Table
4a: Effect of Income on Beliefs About Immigration, No Education Control
Dependent Variable: IMMIG 


Method: Least Squares 


Date: 10/23/01 Time: 13:02 


Sample(adjusted): 1 1510 IF ECON<1 


Included observations: 1362 after
adjusting endpoints 


Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 

C 
1.581155 
0.176059 
8.980843 
0.0000 

BLACK 
0.141790 
0.076408 
1.855686 
0.0637 

ASIAN 
0.002224 
0.092337 
0.024084 
0.9808 

OTHRACE 
0.004465 
0.090074 
0.049576 
0.9605 

AGE 
0.009174 
0.007457 
1.230223 
0.2188 

AGE^2 
0.000139 
7.59E05 
1.832582 
0.0671 

MALE 
0.130501 
0.042039 
3.104298 
0.0019 

IDEOLOGY*(1OTHIDEOL) 
0.106427 
0.023119 
4.603419 
0.0000 

OTHIDEOL 
0.242322 
0.150883 
1.606028 
0.1085 

JOBWORRY 
0.049389 
0.019877 
2.484734 
0.0131 

YOURFAM5 
0.018488 
0.033123 
0.558180 
0.5768 

YOURNEXT5 
0.037205 
0.033983 
1.094799 
0.2738 

INCOME 
0.041745 
0.010383 
4.020541 
0.0001 

Rsquared 
0.069468 
Mean dependent var 
1.218796 


Adjusted Rsquared 
0.061191 
S.D. dependent var 
0.779419 


S.E. of regression 
0.755196 
Akaike info criterion 
2.285819 


Sum squared resid 
769.3625 
Schwarz criterion 
2.335612 

Log likelihood 
1543.643 
Fstatistic 
8.392399 

DurbinWatson stat 
2.049180 
Prob(Fstatistic) 
0.000000 
Table
4b: Effect of Income on Beliefs About “Excessive Profits,” No Education Control
Dependent Variable: PROFHIGH 

Method: Least Squares 

Date: 10/23/01 Time: 13:02 

Sample(adjusted): 1 1510 IF ECON<1 

Included observations: 1355 after
adjusting endpoints 

Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 
C 
1.346526 
0.164472 
8.186947 
0.0000 
BLACK 
0.078105 
0.071559 
1.091486 
0.2753 
ASIAN 
0.011367 
0.087285 
0.130229 
0.8964 
OTHRACE 
0.160538 
0.085611 
1.875199 
0.0610 
AGE 
0.010419 
0.006962 
1.496472 
0.1348 
AGE^2 
7.23E05 
7.09E05 
1.020087 
0.3079 
MALE 
0.202624 
0.039320 
5.153159 
0.0000 
IDEOLOGY*(1OTHIDEOL) 
0.090241 
0.021657 
4.166787 
0.0000 
OTHIDEOL 
0.180299 
0.140710 
1.281355 
0.2003 
JOBWORRY 
0.037830 
0.018623 
2.031381 
0.0424 
YOURFAM5 
0.056647 
0.030934 
1.831217 
0.0673 
YOURNEXT5 
0.104313 
0.031768 
3.283568 
0.0011 
INCOME 
0.036220 
0.009713 
3.729038 
0.0002 
Rsquared 
0.108802 
Mean dependent var 
1.272325 

Adjusted Rsquared 
0.100833 
S.D. dependent var 
0.742522 

S.E. of regression 
0.704092 
Akaike info criterion 
2.145732 

Sum squared resid 
665.2902 
Schwarz criterion 
2.195732 

Log likelihood 
1440.733 
Fstatistic 
13.65318 

DurbinWatson stat 
2.008430 
Prob(Fstatistic) 
0.000000 
Table
5a: Effect of Income on Beliefs About Immigration, Education Control
Dependent Variable: IMMIG 

Method: Least Squares 

Date: 10/23/01 Time: 12:49 

Sample(adjusted): 1 1510 IF ECON<1 

Included observations: 1362 after
adjusting endpoints 

Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 
C 
1.883690 
0.174664 
10.78466 
0.0000 
BLACK 
0.174951 
0.074420 
2.350864 
0.0189 
ASIAN 
0.035971 
0.089924 
0.400013 
0.6892 
OTHRACE 
0.032613 
0.087676 
0.371975 
0.7100 
AGE 
0.004571 
0.007273 
0.628464 
0.5298 
AGE^2 
8.37E05 
7.41E05 
1.129602 
0.2588 
MALE 
0.115403 
0.040928 
2.819625 
0.0049 
IDEOLOGY*(1OTHIDEOL) 
0.088741 
0.022578 
3.930411 
0.0001 
OTHIDEOL 
0.253523 
0.146774 
1.727304 
0.0843 
JOBWORRY 
0.036076 
0.019394 
1.860182 
0.0631 
YOURFAM5 
0.004961 
0.032329 
0.153453 
0.8781 
YOURNEXT5 
0.025312 
0.033084 
0.765072 
0.4444 
INCOME 
0.011501 
0.010667 
1.078253 
0.2811 
EDUCATION 
0.121877 
0.013828 
8.814086 
0.0000 
Rsquared 
0.120175 
Mean dependent var 
1.218796 

Adjusted Rsquared 
0.111690 
S.D. dependent var 
0.779419 

S.E. of regression 
0.734604 
Akaike info criterion 
2.231255 

Sum squared resid 
727.4387 
Schwarz criterion 
2.284878 

Log likelihood 
1505.485 
Fstatistic 
14.16323 

DurbinWatson stat 
2.020208 
Prob(Fstatistic) 
0.000000 
Table
5b: Effect of Income on Beliefs About “Excessive Profits,” Education Control
Dependent Variable: PROFHIGH 

Method: Least Squares 

Date: 10/23/01 Time: 12:49 

Sample(adjusted): 1 1510 IF ECON<1 

Included observations: 1355 after
adjusting endpoints 

Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 
C 
1.509230 
0.166386 
9.070651 
0.0000 
BLACK 
0.060476 
0.071038 
0.851317 
0.3947 
ASIAN 
0.008011 
0.086629 
0.092480 
0.9263 
OTHRACE 
0.144138 
0.084945 
1.696828 
0.0900 
AGE 
0.012815 
0.006920 
1.851821 
0.0643 
AGE^2 
0.000101 
7.05E05 
1.432204 
0.1523 
MALE 
0.194440 
0.039020 
4.983073 
0.0000 
IDEOLOGY*(1OTHIDEOL) 
0.099322 
0.021551 
4.608611 
0.0000 
OTHIDEOL 
0.185962 
0.139513 
1.332934 
0.1828 
JOBWORRY 
0.030960 
0.018516 
1.672024 
0.0948 
YOURFAM5 
0.044179 
0.030774 
1.435562 
0.1514 
YOURNEXT5 
0.097896 
0.031524 
3.105476 
0.0019 
INCOME 
0.020394 
0.010153 
2.008678 
0.0448 
EDUCATION 
0.064849 
0.013178 
4.920943 
0.0000 
Rsquared 
0.124610 
Mean dependent var 
1.272325 

Adjusted Rsquared 
0.116123 
S.D. dependent var 
0.742522 

S.E. of regression 
0.698080 
Akaike info criterion 
2.129311 

Sum squared resid 
653.4895 
Schwarz criterion 
2.183157 

Log likelihood 
1428.608 
Fstatistic 
14.68370 

DurbinWatson stat 
2.000165 
Prob(Fstatistic) 
0.000000 
Table
6a: Conditional Probability of Being a Democrat, with Ideology*Educ
Table
6b: Conditional Probability of Being a Republican, with Ideology*Educ
Table
7a: Effect of Income on Beliefs About Immigration, Ideology*Educ Interaction
Dependent Variable: IMMIG 

Method: Least Squares 

Date: 10/23/01 Time: 12:54 

Sample(adjusted): 1 1510 IF ECON<1 

Included observations: 1362 after
adjusting endpoints 

Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 
C 
1.901975 
0.174285 
10.91305 
0.0000 
BLACK 
0.167020 
0.074339 
2.246730 
0.0248 
ASIAN 
0.038885 
0.089935 
0.432370 
0.6655 
OTHRACE 
0.032774 
0.087630 
0.374001 
0.7085 
AGE 
0.004735 
0.007263 
0.651871 
0.5146 
AGE^2 
8.50E05 
7.40E05 
1.148399 
0.2510 
MALE 
0.116930 
0.040887 
2.859876 
0.0043 
IDEOLOGY*(1OTHIDEOL)*EDUCATION 
0.020108 
0.004718 
4.261634 
0.0000 
OTHIDEOL*EDUCATION 
0.062666 
0.032606 
1.921896 
0.0548 
JOBWORRY 
0.036512 
0.019397 
1.882333 
0.0600 
YOURFAM5 
0.005987 
0.032285 
0.185437 
0.8529 
YOURNEXT5 
0.025103 
0.033047 
0.759605 
0.4476 
INCOME 
0.011887 
0.010661 
1.114952 
0.2651 
EDUCATION 
0.124634 
0.013817 
9.020040 
0.0000 
Rsquared 
0.122481 
Mean dependent var 
1.218796 

Adjusted Rsquared 
0.114018 
S.D. dependent var 
0.779419 

S.E. of regression 
0.733641 
Akaike info criterion 
2.228631 

Sum squared resid 
725.5321 
Schwarz criterion 
2.282253 

Log likelihood 
1503.698 
Fstatistic 
14.47294 

DurbinWatson stat 
2.021390 
Prob(Fstatistic) 
0.000000 
Table
7b: Effect of Income on Beliefs About “Excessive Profits,” Ideology*Educ
Interaction
Dependent Variable: PROFHIGH 

Method: Least Squares 

Date: 10/23/01 Time: 12:54 

Sample(adjusted): 1 1510 IF ECON<1 

Included observations: 1355 after
adjusting endpoints 

Variable 
Coefficient 
Std. Error 
tStatistic 
Prob. 
C 
1.504577 
0.166017 
9.062768 
0.0000 
BLACK 
0.048497 
0.070958 
0.683463 
0.4944 
ASIAN 
0.004565 
0.086640 
0.052685 
0.9580 
OTHRACE 
0.135789 
0.084901 
1.599375 
0.1100 
AGE 
0.013030 
0.006910 
1.885486 
0.0596 
AGE^2 
0.000104 
7.04E05 
1.471935 
0.1413 
MALE 
0.192414 
0.038981 
4.936041 
0.0000 
IDEOLOGY*(1OTHIDEOL)*EDUCATION 
0.022115 
0.004490 
4.924966 
0.0000 
OTHIDEOL*EDUCATION 
0.049965 
0.030994 
1.612075 
0.1072 
JOBWORRY 
0.028983 
0.018517 
1.565193 
0.1178 
YOURFAM5 
0.046264 
0.030730 
1.505501 
0.1324 
YOURNEXT5 
0.096444 
0.031489 
3.062790 
0.0022 
INCOME 
0.019645 
0.010146 
1.936233 
0.0530 
EDUCATION 
0.065029 
0.013171 
4.937448 
0.0000 
Rsquared 
0.126969 
Mean dependent var 
1.272325 

Adjusted Rsquared 
0.118506 
S.D. dependent var 
0.742522 

S.E. of regression 
0.697138 
Akaike info criterion 
2.126612 

Sum squared resid 
651.7280 
Schwarz criterion 
2.180458 

Log likelihood 
1426.780 
Fstatistic 
15.00220 

DurbinWatson stat 
2.001392 
Prob(Fstatistic) 
0.000000 