Another element of our forthcoming Senate race ratings involves assigning a liberal-conservative score to each candidate. Although there are resources like the National Journal -- and a whole host of interest groups -- that provide such ratings for sitting senators, there is nothing in so far as I am aware that provides them for prospective ones.
So I built my own from scratch. The process was to assign each candidate a number from 1 (conservative) to 5 (liberal) on six critical issues: Iraq/national security, health care, taxation, immigration, environment/energy, and values (e.g. abortion and gay rights). The scores were then averaged and translated to a 0 to 100 scale, with 0 indicating conservative and 100 liberal.
A whole series of sources were consulted, including the candidates' voting records, the policy positions set forth on his or her website, and liberal-conservative ratings provided by third parties. So, these ratings were not done without thought. But I'm sure there are a few that can be improved. Which of these strike you as wrong?
Tuesday, June 24, 2008
Liberal-Conservative Scores for Senate Candidates
-- Nate Silver at 8:50 AM
Labels: political spectrum, senate, site
92 comments
Lunsford is higher than Mark Udall?
That's surprising.
Not surprised about Franken.
Coleman is one of the ten most liberal Republicans in the.
I'm splitting my votes...Obama and Coleman.
Is Smith's 33 a career rating, or is it significantly biased by how he changed his position on Iraq soon after the 2006 elections?
I wonder if there is a correlation or even a collinerity between spread (difference between the candidates) and safety of the incumbant? Smaller spread = at risk (I)? I think you might be able to leverage this scaling into more info -
I gave Udall fairly conservative scores on immigration and tax policy. He's liberal in other areas like environment/energy.
Lunsford is pretty explicit about wanting a pullout from Iraq (which automatically qualifies as at least a 4 out of 5 on the rating for that category) and seems to favor a health care mandate, although he tapdances somewhat around the issue. He's not in the same group as someone like Ronnie Musgrove, who does pretty much everything in his power to conceal the fact that he's a Democrat.
@ 9:18. Although this is inherently a subjective exercise, the ratings are primarily intended to be based on the platform that the candidate is putting forward at present.
90-100: 0 Dems, 0 Reps
80-89: 10 Dems, 0 Reps
70-79: 10 Dems, 0 Reps
60-69: 10 Dems, 0 Reps
50-59: 2 Dems, 0 Reps
40-49: 1 Dems, 1 Reps
30-39: 0 Dems, 1 Reps
20-29: 0 Dems, 8 Reps
10-19: 0 Dems, 12 Reps
0-9: 0 Dems, 11 Reps
Please provide more transparency and consistency to your ratings.
Can you in detail, besides saying "a variety of sources" were consulted say how these were achieved. And what happened when their voting record didn't match their web site, etc.?
Interesting that the Republicans tend to be closer to the 0 than Democrats are to the 100. No wonder they think Obama is so liberal.
What would Clinton, McCain, and Obama be on this scale, Nate?
Hoogendyk=Troglodyte. The western side of MI is loaded with these types.
Not that any R candidate could beat Levin this year, even if McCain were able to steal the state.
Very thought provoking, but it appears there is an issue with the data sorting - the senators don't always match their state or party.
I would presume that you keep the subcategories broken out somewhere, though, since in all likelihood, different states will weight the importance of different issues differently.
So, say, in southern border states, a candidate's position on immigration might be more salient than their tax policy, whereas in California, "values" issues (esp. marriage, given recent events), might be weighted more heavily.
By and large, The Netherlands is an enlightened and liberal country.
Maybe they shipped the Dutch cavemen to western Michigan.
New PPP poll from Michigan:
Obama 48
McCain 39
@9:32. It's possibly a reflection of my bias that the scores tend to be closer to 0 than 100 on average. But for the way I'm applying these -- including them as a variable in the regression estimate -- it's only the scores relative to one another that matter; the 0-to-100 scale is just an abstraction.
FWIW, I would probably give Obama a 4.5/5 on environment/energy, immigration, and values, and a 4.0/5 in the other three categories. That would translate to score of 81/100.
I would give Clinton a 4.5 on health care and values, a 4 on immigration and taxes, and a 3.5 on Iraq and environment/energy. That would translate to 75/100.
McCain, I would probably assign a 3 on immigration, a 2.5 on environment/energy, a 2 on health care, taxes and values, and a 1.5 on Iraq/national security. That would work out to a score of 29/100.
Unless you provide breakdowns, it's really hard to critique these numbers. Of course, if you provide the breakdowns, you may get too much input from the peanut gallery.
On your Obama, Clinton, and McCain scores, I think you've overrated McCain and perhaps Obama by half a point in environment/energy. But it is really hard to say without providing the breakdowns of your other ratings - there are no comparison points. In other words, you might just as well have underrated Clinton by half a point.
First, thanks for the excellent statistical analyses on a wide range of questions : )
I'd like to offer a statistical comment on this issue.
A polychoric correlation-based factor analysis that applied your ratings of both sitting and prospective senators would estimate the importance of each of the six questions, reducing the "measurement error" if one question (e.g., values) is much more important than another.
Even better would be a graded item response model
http://edres.org/irt/baker/
Keep up the interesting work!
I suggest you do some crossvalidation before going ahead with your scores.
For incumbent Senators, as well as those with House voting records for non-incumbents with a HR record available, you should use NOMINATE scores from last two years as predictor variables, and your L-R ideology score as dependent variable and run a regression. Look for outliers and evaluate. You can also include some other conservative or liberal or labor organization scores as predictors in the same equation.
More generally, a correlation matrix of your scores, the NOMINATE scores, and other ratings would be instructive.
Are you doing any such subjective ratings for campaigning strength, or letting fundraising speak for that? I would suspect that there are some weaknesses which would not impact fundraising (party faithful) too much but would weigh with independents - the less concrete scandals.
For the sitting senators, did you compare your rankings to voteview.com?
This is a great start.
One suggestion: since many Americans' values include fair play, live and let live, and keeping our noses out of other people's bedrooms, I wonder if the "values" category might be given a more neutral name. As it is, it's almost as though people who don't care that much about these two issues aren't voting based on their values. "Marriage/fertility regulation" perhaps? "Morality issues"?