Analyzing the House vote on health care


The House voted on its version of the health care bill on November 7, 2009. The vote was largely along party lines, with 219 of 258 Democrats voting “Yes” and 176 of 177 Republicans voting “No.” You can find the full vote results along with a nifty interactive map at the New York Times.

Conventional wisdom holds that the lone yes-voting Republican did so because he was elected in a district that voted heavily for Obama in the presidential election. Similarly, many have noted that several of the no-voting Democrats represent districts that voted for McCain in the presidential election. This observation suggests that members of Congress may sometimes be subject to cross pressures. Their party leaders want them to vote in support of the party’s program, but the members may also feel compelled to vote in accordance with the majority sentiment in their district. Sometimes these pressures push a member in different directions. Below the fold I present some statistical analysis that confirms the conventional wisdom with respect to this vote and briefly discuss some implications of the analysis for the 2010 congressional elections.

Political scientists like to predict political behavior. Interestingly, we are usually trying to predict the past, rather than the future because, as Edward Banfield suggests, this is difficult enough.

Anyway, what we want to do is develop a (hopefully simple) mathematical model that allows us to predict how each member of Congress voted on the bill.

One very simple model is to predict that all Democrats vote “Yes” and that all Republicans vote “No.” Because of the partisan nature of this particular vote, even this very simple model performs pretty well. We can demonstrate the performance in a classification table:

——– Actual ——–
Predicted YES NO Total
YES 190 20 210
NO 30 195 225
Total 220 215 435

The table above shows that this model correctly predicted 190 of the 220 “Yes” votes and 195 of the 215 “No” votes. In other words, it correctly predicted (190+195)/435 = 395/435 = 90.8% of the votes. This model, however, predicts that 30 members will vote “No” when they actually voted “Yes” and that 20 members will vote “Yes” when they actually voted “No.” We have additional information about each member that might help us improve our model.

I suggested earlier that some members might be cross-pressured, that the people in their districts might want them to vote against their party’s program. Congress members might use the presidential vote in their district as one indicator of constituent sentiment toward health care reform. If many voters voted for John McCain, then maybe many voters oppose the Democratic plan for health care reform. On the other hand, if many voters voted for Barack Obama, then maybe many of them support the Democratic plan.

We have information about the presidential vote in each congressional district, so we can add this information to our model. So, now our model predicts each members vote using their party affiliation and the percentage of voters in their district who voted for John McCain as predictors. (The type of model I’m using to do these predictions is called a logistic or logit regression. I’ll be glad to share the specific results with anyone who emails me; in this post I’m focusing on the predictions.) So how does our new model perform. The classification table is shown below:

——– Actual ——–
Predicted YES NO Total
YES 212 18 230
NO 8 197 205
Total 220 215 435

The table above shows that we have improved our predictions by adding information about the presidential vote to our model. We now correctly predict (212+197)/435 = 409/435 = 94% of the votes. This model predicts that one Republican – Cao of Louisiana’s 2nd district – casts a “Yes” vote; in other words, it perfectly predicts Republican votes. All prediction errors are related to Democratic votes. The increase in predictive power provides statistical evidence to support the idea that there were a number of cross-pressured Democrats who voted “No” in response to constituent sentiment in opposition to the Democratic health care reform plan.

These findings don’t mean, however, that members of congress necessarily took their voting cues directly from the presidential vote. More likely, they sensed constituent sentiment from turnout at town hall meetings, letters and phone calls from constituents, and other common means of communicating with constituents. But it is more likely that these measures of sentiment were more heavily negative toward health care reform in districts that voted for McCain for president.

Now, let’s turn to some more detailed analysis. There were 39 Democrats that voted “No” on HR 3962. Our model correctly predicts the votes of 21 of these members. The table below displays these members, the district they represent, and the percentage of the district presidential vote cast for John McCain.

Democratic NO votes predicted by model

Member State CD McCain%
Bright AL 2 63
Griffith AL 5 61
Ross AR 4 58
Boyd FL 2 54
Marshall GA 8 56
Minnick ID 1 62
Chandler KY 6 55
Melancon LA 3 61
Kratovil MD 1 58
Skelton MO 4 61
Childers MS 1 62
Taylor MS 4 68
Boren OK 2 66
Altmire PA 4 55
Herseth SD AL 53
Davis TN 4 64
Gordon TN 6 62
Tanner TN 8 56
Edwards TX 17 67
Matheson UT 2 57
Boucher VA 9 59

These prediction results indicate that we can expect a Democratic member to have cast a “No” vote if McCain received more than 52% of the vote in his or her district. Those are the predicted “No” votes. What about the Democrats who voted “No,” but that our model predicted a “Yes” vote? These 18 members are displayed below.

Democratic NO votes not predicted by model

Member State CD McCain%
Davis AL 7 27
Markey CO 4 50
Kosmas FL 24 51
Barrow GA 12 45
Peterson MN 7 50
McIntyre NC 7 52
Kissell NC 8 47
Shuler NC 11 52
Adler NJ 3 47
Teague NM 2 50
McMahon NY 13 51
Murphy NY 20 48
Massa NY 29 51
Kucinich OH 10 39
Boccieri OH 16 50
Holden PA 17 51
Nye VA 2 49
Baird WA 3 46

Despite the predictions of our model, Democratic leaders might be willing to give some of these members a pass for their “No” vote. Several of them represent districts that voted for McCain, if only narrowly. Even in most of the districts which Obama won, the victory was fairly narrow. Two of these Democratic “No” votes really stand out.

Artur Davis of Alabama’s 7th district voted “No” despite the fact that Obama won the district by a huge margin. One explanation for Davis’ vote is that he is running for governor and is, therefore, responsive to all voters in Alabama and not just his current constituents. As the only member of the Congressional Black Caucus to vote against HR 3962, he was harshly criticized by Jesse Jackson, although Jackson has since softened his criticism.

Dennis Kucinich is another notable “No” vote. Voters in his district strongly supported Obama for president; McCain received less than 40 percent of the vote. Consequently, his “No” vote is a protest against a health care bill that he feels does not go far enough in reforming the health care system.

Based on these results, I believe that the calls by progressives for retribution against many of these no-voting Democrats represent a misguided political strategy. Given the districts that many of these members represent any successful attempt to defeat them in the primary with a more liberal nominee, risks losing the general election to a Republican candidate.

A group that has gotten less attention is the group of Democrats who voted “Yes” despite McCain’s strong showing in their district. The table below displays the members who voted “Yes” while our model predicted a “No” vote.

Democratic YES votes not predicted by model

Member State CD McCain%
Berry AR 1 59
Snyder AR 2 54
Kirkpatrick AZ 1 54
Pomeroy ND AL 53
Carney PA 10 54
Spratt SC 5 53
Mollohan WV 1 57
Rahall WV 3 56

Berry and Mollohan faced no opposition in the previous election. Most of the others won their own elections pretty handily with more than 60% of the vote, so these “Yes” votes may reflect the strength of incumbency. It is worth pointing out, however, that without some of these votes, the bill would have failed. Look for these representatives to get support from the Democratic leadership and the president in their reelection efforts, if they think it will be helpful.

-chiptaylor

Advertisements

3 Responses

  1. Nice work, very interesting! To bad I never had the chance to take a class with you! Keep up the good work!

  2. Thanks Alex.

  3. […] states in the left half, where uninsurance rates are low. Recall that red state congress members tended to oppose the house health care bill. Does this mean that GOP congress members from red states are voting […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: