Sunday, 15 February 2004

Knowledge, ideology, and party identification

Somehow I missed Eugene Volokh’s post on the whole “conservatives are stupid” kerfuffle. There are a few caveats in order with this use of ANES data:

  1. Partisanship is not ideology. While the two concepts are correlated highly in the United States, a “strong Republican” is not necessarily a conservative. (The correlation has improved over time, however, as the South has dealigned.) The ANES does include an ideology item.
  2. The partisanship scale used by the ANES has a weird inflection point noted by Larry Bartels several years ago: “weak identifiers” are generally more independent in their behavior than “independent leaners.”
  3. Political knowledge is not general knowledge (as Volokh’s correspondent notes). Again, while general knowledge is correlated with political knowledge, the latter is only one component of the former.
  4. There are numerous disputes over whether “quiz-type” political knowledge questions properly tap overall respondent knowledge about politics. Read chapter 2 of my dissertation if you need all the morbid details. Read the appendix of chapter 4 if you want to see how various types of questions performed in a 1998 Dutch election survey that uses a larger battery of questions than the contemporary ANES (I suppose the Dutch are more willing to subject themselves to quizzes than Americans, although Delli Carpini and Keeter disagree).

The GSS data cited in this InstaPundit post is more dispositive, although I am not as familiar with the GSS—and there may well be caveats that Prof. Lindgren does not mention or is unfamiliar with.

Viewer mail on ideology and knowledge; substantive and statistical significance

Prof. Jim Lindgren of Northwestern dropped me an email responding to this post*. Lindgren writes:

I appreciate your thoughtful comments on Somin's note posted on the Volokh Conspiracy.

As you know but your readers might not, the two leading academic cross-sectional surveys are the American National Election Studies (ANES) from the Univ. of Michigan (which Somin used) and the General Social Survey (GSS) from the University of Chicago (which I used in my note to Instapundit). Political scientists naturally tend to use the ANES, while sociologists tend to use the GSS, with the rest of the social sciences using both to a substantial extent.

Each have their advantages and disadvantages. In my opinion, because the ANES is taken around the time of national elections, it is better for understanding elections and voting. Because the GSS is not taken around election time, it is better for understanding how large political groups, including Republicans and conservatives, tend to think at times other than the few months on either side of a national election. For this reason, the GSS trends in political orientation tend to be far more stable than the ANES data on this question, which are unreliable in their high variability from election to election. [Greg Caldeira (political science, Ohio State) and I are doing a paper on this phenomenon.]

Your observation about independent leaners behaving more like party adherents than weak identifiers as a Republican or a Democrat is true of voting as revealed in the ANES. It is not true as a generalization for issues across the board (it varies by issue).

For example, in the 1994–2002 GSS, independents who lean Democratic are like Republicans in their high performance on vocabulary and analogical reasoning tests. Leaners to either party tend to fall between Republicans and Democrats in their educational level. Independents who do not lean either way usually score down with the Democrats, either below them or between strong and not too strong Democrats.

That is why my analyses usually discuss types of people, rather than treating liberalism/conservatism and party identification as left/right ordinal or interval variables.

First, I’d like to thank Prof. Lindgren for his correspondence.

Second, I’d like to clarify that I advisedly used the word behavior; opinionation (such as issue positions) and attitude-holding are not behavior, and it is true that the relationship Bartels notes between party identification as measured by the “standard” 7-point scale and voting behavior may not apply to opinionation or attitude-holding.

I think the observations about Democrats are interesting, because I suspect they reflect a bifurcation in “strong Democratic” identifiers: on the one hand, you have groups who are identified with the Democrats on the basis of social affiliation, such as the poor, most minority groups, the working class, and organized labor; on the other, you have highly-educated people with “social consciences” who have a more psychological attachment to the Democratic party on the basis of ideology. This certainly doesn’t seem like an original observation, although I’m not sure anyone has shown it empirically rather than impressionistically (citations welcome).

In other viewer mail, another Chicago correspondent—Debian developer/economist Dirk Eddelbuettel—notes an article in a recent week’s Economist ($) on statistical significance versus substantive significance, a distinction social scientists probably need to pay more attention to. (Also see this week’s letters page.)

* Hopefully it was pasted in here properly—I’m borrowing my mom’s laptop for this post, and my email access is via a teletype-style SSH session rather than the Mozilla Thunderbird setup I’m using at home. Any formatting problems are mine and mine alone. (If you’re expecting an email from me in response to something you sent, I should get back to you sometime Monday.)