Prof. Bryan Caplan
bcaplan@gmu.edu
http://www.bcaplan.com
Econ 849
Fall, 2001
Week 10: Ideology
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 over-riding factor (called g for "general intelligence"). Cognitive ability is essentially "one-dimensional."
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 g-loading is.
1. Ex: Analogies have a higher g-loading than pure memory tasks.
F. Factor analysis in no way guarantees the existence of a single over-riding 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 U.S. Political Opinion
A. There are many different ways to analyze political beliefs.
1. Libertarian-statist spectrum
2. Christian-secular spectrum
B. What can factor analysis tell us about the dimensionality of U.S. political opinion?
C. Strong result: As with intelligence, empirical tests typically find that political opinion is roughly one dimensional.
D. What is the dimension? Empirically, U.S. political opinion "fits" well on the liberal-conservative or left-right spectrum.
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 - pro-personal, pro-economic
2. Populists - anti-personal, anti-economic
3. Liberals - pro-personal, anti-economic
4. Conservatives - anti-personal, pro-economic
E. But empirically, most people line up on the diagonal, and the other two boxes are sparsely inhabited.
F. Poole and Rosenthal's long-term study of the U.S. Congress finds that a one-dimensional l-c model works very well.
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 were often opposed to civil rights measures, and otherwise conservative Republicans were often in favor. 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 liberal-conservative dimension.
H. Similarly, Levitt and earlier researchers have found that one-dimensional ideological measures of l-c like ADA scores give better predictions of politicians' behavior than measures of constituent interests. Marginal predictive value of alternative scores is limited.
I. Less work has been done on the dimensionality of individual citizens' opinions, but once again, a strong liberal-conservative dimension pops out of the data.
J. Remarkably, voting in the U.N. is also one-dimensional, 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 disagreeing.
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 better-off 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 one-dimensional, and people largely vote ideologically, then the simple MVT's seemingly strong assumptions are satisfied. Perhaps the issue-space only looks multi-dimensional.
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 (r=.08). [Table 2a]
D. So what does determine ideology? Is it education?
E. Once again, no. Education and ideology are close to unrelated (r=.05) when you look at a random sample of Americans (as opposed to university faculty!). [Table 2b]
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 2c]
G. Both are clearly statistically significant, but the actual effect is small. On a 4-point scale:
1. The gap between the bottom and top of the income distribution is .39.
2. The gap between the bottom and top of the education distribution is .32.
H. There is evidence of inter-generational transmission of ideology, though these studies rarely have enough control variables.
I. So 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 nothing we know of does much to explain ideology.
J. Maybe someone will one day show that ideology reduces to something else, but given the failure of all the obvious candidates, I doubt it.
VI. Case Study: The Determinants of Party Identification, II
A. Question: Returning to last week's linear probability model of party identification, what happens if you also control for stated ideology (very liberal, liberal, moderate, conservative, very conservative)?
B. [Table 1b]
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 self-interest."
D. Consider two examples.
1. Ex. #1: Black female millionaire Ph.D., with maximal job security and income growth, 30 years old.
2. Ex. #2: White male minimum wage worker, minimal job security and income growth, 30 years old, high school education, very conservative ideology.
E. Ex. #1: [Since we don't know ideology, use Table 1a] Estimated probability of being a Democrat: 58.8%; estimated probability of being a Republican: 16.9%.
F. Ex. #2: [Using Table 1b] Estimated probability of being a Democrat: 10.9%; estimated probability of being a Republican: 49%.
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 one-dimensional, opinion looks two-dimensional.
E. Moreover, the two dimensions more or less fit the two-dimensional personal freedom/economic freedom diagram. Education probably 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 liberal-conservative 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 drop-outs?
VIII. Case Study: Economic Beliefs
A. Now let us go through two illustrations: 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 3a, 3b]
C. Controlling for education, though, makes the apparent effect of income almost disappear. [Table 4a, 4b]
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 grade-school 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 8.4 and 11.1 to 10.0 and 12.5. [Table 1c]
D. It also works on individual issues. For immigration, the tstat rises from 3.9 to 4.3 [Table 4a, 5a]; for excessive profits, from 4.6 to 4.9 [Table 4b, 5b].
E. Returning to the two-dimensional diagram, education "stretches" the liberal-conservative spectrum.
X. Mainstream and Polarization Effects
A. There are interesting empirical connections between political awareness and opinion. Political scientists call these the "mainstream" and "polarization" effects. (Zaller 1992)
B. The "mainstream" effect: When elite opinion is united, agreement with elite opinion is an increasing function of political awareness.
C. Interpretation: For non-partisan issues, the more aware you are, the more likely you are to know what everyone is "supposed to believe."
D. The "polarization" effect: When elite opinion is divided along ideological lines, agreement with "your" ideological leaders is an increasing function of political awareness.
E. Interpretation: For partisan issues, the more aware you are, the more likely you are to know what people on your side of the fence are "supposed to believe."
1. Ex: Nixon on price controls.
F. Contrast: The Vietnam War in 1964 versus 1970.
G. Limits of the mainstream effect: at least under censorship, susceptibility to propaganda peaks at around the 67th percentile of awareness, then declines. It takes some sophistication even to be brain-washed!
XI. Does Policy Match Public Opinion? What Are the Unpopular Policies?
A. Now that we have a better grip on voter motivation, let us return to the earlier question: Is the median voter model correct?
B. It is hard to do formal empirical tests on the federal government (without internationally comparable data, N=1). But there is a lot of informal evidence that the median voter gets what he wants.
C. What does the federal government do? Is this what the median voter wants?
1. Spending:
|
Source |
Share |
|
Social Security |
23% |
|
Non-Defense Discretionary |
19% |
|
Defense Discretionary |
16% |
|
Medicare |
12% |
|
Net Interest |
11% |
|
Medicaid |
7% |
|
Other Means-Tested |
6% |
|
Other |
6% |
2. Taxes:
|
Source |
Share |
|
Individual Income Taxes |
48% |
|
Social Insurance Receipts |
34% |
|
Corporate Income Taxes |
10% |
|
Excise Taxes |
4% |
|
Other |
4% |
3. A lot of regulation: Environmental, worker safety, drug safety, anti-competitive behavior, labor...
D. Starting with the budget: Social Security and Medicare remain extremely popular programs; the military is also usually well-regarded. The remaining items are more contentious.
E. Broadly defining "welfare" as Medicaid and "Other Means-Tested" spending, we get 13% of the budget. But:
1. Few people want to actually abolish these programs
2. Medicaid also pays for middle-class nursing home residents who have run through their personal savings.
F. The national debt is unpopular, but repudiating it would be even less popular. So "net interest" ultimately has voter support.
G. That leaves 25% of the budget for "non-defense discretionary" and "other" spending. Some of this spending is "waste. Waste is unpopular. But outside of isolated examples of $500 toilet seats, what spending do a majority of Americans agree is wasteful?
H. Turning to spending: It is surprising that income and SS taxes are such a large percentage of the budget. But insofar as business "passes on" corporate and other taxes, do a majority of Americans really want significant changes here?
I. Regulation is more complicated. Are there majorities in favor of weaker (or stronger) environmental regulation? Worker safety? Drugs?
J. Challenge: What policies exist that a majority of American voters oppose? Consider all the clichés of politics. Do any hold water?
1. Relatively weak gun control?
2. Foreign aid?
3. NAFTA?
XII. Application: State-Level Policy
A. There have been a number of empirical studies of state-level policy.
B. Main findings: Variations in degree of liberalism are strong predictors of variation in state policy. When public opinion is liberal (as in NY), policy is liberal; when public opinion is conservative (as in Colorado), so is policy.
Table 1b: Conditional Probability of Being a Democrat/Republican
|
|
Probability (Democrat) |
Probability (Republican) |
||
|
Independent Variables |
Coefficient |
tstat |
Coefficient |
tstat |
|
Constant |
0.148 |
1.391 |
0.266** |
2.618 |
|
Ideology |
-0.117*** |
-8.431 |
0.147*** |
11.153 |
|
Other Ideology |
-0.109 |
-1.234 |
-0.102 |
-1.216 |
|
Income |
-0.008 |
-1.292 |
0.008 |
1.330 |
|
Job security |
-0.022 |
-1.865 |
0.023* |
2.049 |
|
Recent Income Growth |
0.022 |
1.091 |
-0.003 |
-0.171 |
|
Expected Income Growth |
-0.022 |
-1.070 |
0.046* |
2.392 |
|
Education |
0.003 |
0.315 |
0.008 |
0.997 |
|
Male |
-0.043 |
-1.733 |
0.003 |
0.107 |
|
Age |
0.010* |
2.165 |
-0.005 |
-1.263 |
|
Age2/100 |
-0.007 |
-1.587 |
0.005 |
1.271 |
|
Black |
0.333*** |
7.313 |
-0.272*** |
-6.277 |
|
Asian |
0.023 |
0.420 |
-0.073 |
-1.378 |
|
Other race |
0.165** |
3.067 |
-0.129* |
-2.515 |
|
Mean Dependent Var |
.337 |
.299 |
||
|
SD Dep. Variable |
.473 |
.458 |
||
|
R-squared |
.108 |
.136 |
||
|
N |
1360 |
1360 |
||
|
*=p<.05 **=p<.01 ***=p<.001 |
||||
Table 1c: Conditional Probability of Being a Democrat/Republican
|
|
Probability (Democrat) |
Probability (Republican) |
||
|
Independent Variables |
Coefficient |
tstat |
Coefficient |
tstat |
|
Constant |
0.066 |
0.624 |
0.346*** |
3.438 |
|
Ideology*Education |
-0.028*** |
-9.963 |
0.034*** |
12.514 |
|
Other Ideology*Educ. |
-0.029 |
-1.503 |
-0.022 |
-1.195 |
|
Income |
-0.007 |
-1.120 |
0.007 |
1.134 |
|
Job security |
-0.021 |
-1.745 |
0.021 |
1.848 |
|
Recent Income Growth |
0.020 |
1.014 |
-0.001 |
-0.031 |
|
Expected Income Growth |
-0.021 |
-1.063 |
0.045* |
2.336 |
|
Education |
0.004 |
0.520 |
0.007 |
0.894 |
|
Male |
-0.040 |
-1.595 |
-0.001 |
-0.037 |
|
Age |
0.010* |
2.223 |
-0.006 |
-1.320 |
|
Age2/100 |
0.000 |
-1.631 |
0.000 |
1.335 |
|
Black |
0.320*** |
7.118 |
-0.256*** |
-5.964 |
|
Asian |
0.017 |
0.310 |
-0.061 |
-1.159 |
|
Other race |
0.163** |
3.064 |
-0.123* |
-2.430 |
|
Mean Dependent Var |
.337 |
.299 |
||
|
SD Dep. Variable |
.473 |
.458 |
||
|
R-squared |
0.126 |
0.154 |
||
|
N |
1360 |
1360 |
||
|
*=p<.05 **=p<.01 ***=p<.001 |
||||