Did Obama’s ambitious agenda—specifically, health care reform, the economic stimulus, cap and trade, and his vigorous support for TARP (even though he wasn’t president at the time)—contribute to the Democratic seat losses? This has been discussed here, here and here, and Obama addressed the issue himself in his first post-election news conference.
The simplest approach to this question is to compare Democratic incumbents on the ballot who cast votes for these bills to those who didn’t. (The roll call vote data are here, although they are gated.) For each Democratic incumbent, I counted how many of these four bills they voted for, and then tabulated the average Democratic vote for each group:
Support for these bills appears to have helped rather than hurt. But of course, Democrats who opposed these bills were more likely to represent competitive or even Republican districts, meaning they did worse for other reasons (see Brendan Nyhan for a similar point). So I modeled Democratic vote share in contested House districts using this count of “yes” votes, plus campaign money in 2010 (from here and here) and each district’s House and presidential vote in 2008 as controls (here] The model also estimates whether the effect of roll call votes depended on the partisanship of the district, as captured by the 2008 presidential vote. This model predicts a Republican majority of 242 seats, compared to the 244 it currently looks like they are going to win. So the model is pretty good, understating Republican performance only a little bit.
What does this model tell us about roll call votes on these four bills? Simple answer: they mattered. A lot. A Democratic incumbent in the average district represented by Democratic incumbents actually lost about 2/3 of a percentage point for every yes vote. That doesn’t sound like a lot, but that’s for incumbents in districts that voted 63% for Obama.
For Democrats in the least Democratic districts (Chet Edwards of TX or Gene Taylor of MS), the model suggests a loss of about 4% for every yes vote. Does that mean poor Chet lost 16 points on roll call votes alone? No, because he wasn’t a big supporter of Obama’s agenda. But he did vote for both TARP and the stimulus. In fact, virtually every Democratic incumbent on the ballot yesterday supported at least one of these four bills. That support was costly.
What might have happened if vulnerable Democrats hadn’t voted for any of the four bills? I’ll define “vulnerable” as any Democratic incumbent who lost. The graph below shows the balance of power as predicted by the regression, and then what it might have been if everything else was the same but these vulnerable Democrats had voted “no” on everything. The result is stunning:
The Democrats gain back 32 seats, enough to retain control of the House. The margin of error around that prediction crosses the majority control line, so we can’t be fully confident that Democrats would have maintained their majority, even with these predictions. But the difference between the actual result and the counterfactual is itself outside the margin of error, so the effect is large no matter how you slice it.
On one level, this makes sense. You take a tough vote, you pay for it. Otherwise it wouldn’t be a tough vote. Indeed, political science research has shown that citizens do hold members accountable for their votes and members who are ideologically out of step with their district are more likely to find themselves out of office. See this paper (gated) by Stephen Ansolabehere and Philip Edward Jones and this paper (gated) by Brandice Canes-Wrone, David Brady, and John Cogan.
On another level, it’s puzzling. The outcome of the election was reasonably well predicted by fundamentals like the state of the economy and presidential approval. So why the strong effect for roll calls?
I can address some possibilities right away. First, it doesn’t look like it was the money. The effect of the roll calls is the same whether money is in the model or not, and the effect of roll call votes does not depend on the partisan balance in fundraising. Second, the effect of these roll call votes is not simply capturing the effect of the incumbents’ ideology or overall records. I tried adding a general measure of ideology (DW-Nominate scores) along the lines that Brendan Nyhan suggested, and it’s these big votes that matter. In short, these votes seem an independent force all their own.
So do I think the roll call votes actually cost the Democrats the House? Not necessarily. I would suggest four caveats:
- The effect of roll calls on the vote is large on the seats where it matters, but this is largely because those seats were lost by small margins (all but one of them under 10 points) so a meaningful but not gargantuan roll call effect ends up producing gargantuan changes in seats. (The effect of this simulation on the overall House vote is less than one percentage point.) That’s why we can’t say for sure that the Democrats would have retained control under the “no” vote scenario.
- Even small indiosyncracies that we’re not capturing could prevent the roll call effect from being realized in any particular race. These aren’t causal estimates so much as comparisons of incumbents who were similar on many other factors but nonetheless voted differently on these roll calls.
- As the table above makes clear, there were almost no Democratic incumbents running for reelection who voted against all four bills. So the result is really an extrapolation, and it’s important to be careful about extrapolations. If we just make the vulnerable Democrats vote no on one of the four bills, the gain for Democrats is only 12 seats, not 32.
- Finally, the roll calls may still just be capturing some other aspect of the political climate. In other words, if it hadn’t been the roll calls it might have been something else.
On balance, I don’t think we can say these four big items definitively cost Democrats the majority. But it seems safe to say that they had a big negative effect on Democratic performance, and they certainly didn’t help.
fn1. Brendan Nyhan pointed out that controlling for money is problematic because money itself is endogenous to the roll call votes. He’s absolutely right, but it didn’t seem to make any difference whether the money was included or not, so I erred on the side of controlling for as much as possible.
{ 13 comments }
This is an interesting analysis. But according to many economists, if TARP and the stimulus hadn’t been passed, the economy would have been even worse – it’s hard to imagine the dems winning under that scenario. I’d like to see an analysis with that extra detail considered.
The congressional elections literature suggests that challenger quality plays an important role in determining election outcome/vote shares. Why not include that?
Eric, I conducted a similar analysis and (not surprisingly) came up with similar results: among Democratically held districts, a vote for health care reform cost the Democratic candidate between 3.8 and 4.8 percentage points of the two-party vote, controlling for 2008 and 2004 presidential vote in the districts along with district demographics and redistricting institutions. Another finding of note: Among Democratically held districts, legislature-controlled redistricting process cost the Democratic candidate about 2 percentage points compared to those Democrats running in states where a commission or non-partisan redistricting controlled the process.
My point is similar to the comment above but perhaps more general: what about general equilibrium effects? That is, does the actual passage of any of these pieces of legislation influence the intercepts? Could the Democrats have faced a collective action problem, in that they had an overall incentive to pass legislation even as individual legislators would lose support?
>>He’s absolutely right, but it didn’t seem to make any difference whether the money was included or not, so I erred on the side of controlling for as much as possible.
Tyson and Dan: I didn’t mean to suggest the Dems had an alternative, just that these votes appear to have hurt them. It might be that they were damned if they did, damned if they didn’t. In fact, that could help explain how these roll calls could matter even as the economy predicted the outcome pretty well. No matter what they did they were going to lose seats.
Jack: We left out experience for a simple reason–we weren’t able to get the data in time.
dsquared: What I meant to say was it didn’t make any difference to the coefficient estimate for roll calls whether money was included or not.
This seems to attribute way too much issue voting to real world voters.
How does controlling for the partisanship of the district control for the partisanship of individual voters making individual vote decisions? Is it really all issue voting in U.S. House races and no party effect at all? I have conducted hundreds of focus groups among swing voters over the years and I have to tell you I find that very hard to believe.
Wouldn’t you need a separate dummy variable for each vote? Otherwise using a simple count on those 4 votes would be indistinguishable from any other set of votes where they voted in a similarly patterned way. Thus you may assume you have modeled 4 Votes A, B, C, and D but you could just as well be modeling Votes E, F, G, and H where the candidates exhibited a similar pattern of voting. That might happen if the votes reflected a pattern of ideological voting. You say you controlled for the ideology of the incumbent but aren’t you committing the ecological fallacy by inferring individual level vote decision factors from aggregate results?
I wonder if you had individual level data on voters in a given district and if you included party and/or ideology as control variables do you still really think that those variables would explain so little of the variance?
Another possible problem: presumably it is not candidates’ actual voting record but perceived voting record that counts. It seems likely that many, probably most, voters do not know the candidates’ actual voting records. They may assume that there is a Democratic position on those 4 votes and just attribute the Democratic positions to the Democratic candidate. Maybe your 4-valued vote variable isn’t measuring how the candidates voted on those 4 specific votes but is instead measuring perceived “Democraticness” of the candidates?
“if TARP and the stimulus hadn’t been passed, the economy would have been even worse”
The following counterfactual is not something I necessarily wish had happened (to the extent that I do, it’s for reasons other than the electoral success of the Dems), but:
(1) Assume no TARP. There is a possibility that the economy would have been MUCH worse. But if it was, (a) those effects would likely have occurred before the Obama inauguration, (b) the damage that did occur anyway would have been accelerated and even more clearly linked to Bush, (c) with (say) 15% unemployment, there would have been massive support for more aggressive action to stimulate the economy, and (d) with all of the above, there is a very good chance that the trajectory of unemployment would have looked very different – say, 15% when Obama took office, 9 or 10% now.
In that scenario, the Dems keep the House, probably adding to their majority, and likely get to 60 in the senate.
Point taken about the stimulus, though.
(Parenthetically, the reason I almost wish that this had happened is a moral hazard issue. At some point we need to stop bailing out the banks, and the longer we wait the uglier it will be when the day finally comes.)
“The outcome of the election was reasonably well predicted by fundamentals like the state of the economy and presidential approval. So why the strong effect for roll calls?”
Surely you can’t separate out the poor presidential approval rating from roll calls on four unpopular laws. If all the vulnerable Democrats had voted against those bills (really if only some had), then those laws wouldn’t have passed. If those laws hadn’t passed, then perhaps the presidential approval wouldn’t have been as bad.
“On another level, it’s puzzling. The outcome of the election was reasonably well predicted by fundamentals like the state of the economy and presidential approval. So why the strong effect for roll calls?”
Because you are conditioning on contested house elections. In a sense, this is like further delving into the residuals from a model based on “fundamentals.” The fundamentals made some leaning districts solidly GOP, and made some leaning Dem districts tossups. The influence of the fundamentals is already present when you look at the districts that are contested/competitive. But there is still variation in outcomes that the fundamentals cannot explain.
This analysis doesn’t say that the roll calls mattered more than the fundamentals or that the fundamentals story line is wrong. It says that given the fundamentals, a secondary effect of roll calls is detectable in those districts that the fundamentals had put in play.
Why take a raw count of these votes, rather than include each of them separately as indicator variables? Using just the count only makes sense if the effect of each vote is identical. But my analysis, and other independent analyses find very different magnitudes from different votes. Consensus seems to be ~7-8 points lost for voting for Affordable Care, while the effect of voting for stimulus was negligible (possibly offset by the effects of pork brought home) as was the effect of a vote for the climate bill. I haven’t looked at the TARP vote.
LarryM:
I’m assuming that economists such as Blinder and Zandi are close to the mark when they estimate that without TARP, unemployment would currently be between 12 and 13% this year, not 9 or 10% as you say. I know people have a beef with some of Blinder and Zandi’s methods; maybe you have a better model than they have.
http://www.economy.com/mark-zandi/documents/End-of-Great-Recession.pdf
Maybe you are right that more voters would have blamed Bush if TARP didn’t pass and the recession was even worse, but a majority of the population already blames Bush for the recession and they still voted Republican on Tuesday. http://dailycaller.com/2010/11/02/most-voters-still-blame-bush-for-bad-economy/
So,
Eric:
I think you are right that the Democrats were damned if they do and damned if they don’t. But as a group they would have been MORE damned if they didn’t, is the way I see it. This isn’t what you said in your piece above. You said:
“What might have happened if vulnerable Democrats hadn’t voted for any of the four bills? I’ll define “vulnerable” as any Democratic incumbent who lost. The graph below shows the balance of power as predicted by the regression, and then what it might have been if everything else was the same but these vulnerable Democrats had voted “no” on everything. The result is stunning: The Democrats gain back 32 seats, enough to retain control of the House.”
So you are saying that if all of the vulnerable Dems had voted against stimulus and TARP, they might have retained control of the House. I don’t think TARP or the stimulus would have passed without those votes (maybe I’m wrong, I didn’t look this up), and so unemployment would currently be over 12%, and so I think the Dems would have still lost control of the House. You qualify your finding with “if every else was the same but these vulnerable Dems had voted “no”…” But if they voted “no” everything wouldn’t be the same, the economy would have been worse, so in my opinion your counterfactual doesn’t work.
I think it is important to make these caveats clear, otherwise naive readers might think that it is a good strategy to vote against unpopular but effective policies.
I think Dan is exactly right. As a group it is better for the Dems to pass TARP and simulus. As individuals, it is better to vote against, but still enjoy the benefits of those bills passing. Measuring the latter effect is the contribution of this post. But I think it is mislieading to then project this finding and say that if all vulnerable Dems voted against, they would have retained control (again, I’m basing this all on my guess that the bills wouldn’t have passed without those votes, I didn’t look that up).
Josh: When we estimate the effects of the stimulus, cap and trade, and health care reform separately, we find that all three are statistically significant. Furthermore, we cannot reject the null that their effects are actually equal to each other.
This is from a model estimate on all Democratic incumbents in contested general election races, and including the other variables discussed above.
Comments on this entry are closed.