I’m afraid I have to take exception to this statement by Goldy:
Oh… and the fact that polls generally show Edwards as being the toughest Democrat to beat… that doesn’t hurt him in my book either.
I suppose Goldy is relying on national head-to-head polls like these. The problem with such national polls is that they don’t reflect the way we elect our Presidents.
Rather than looking at the national head-to-head polls, we should be examining state head-to-head polls and take into consideration the number of votes each state gets in the Electoral College.
In fact, I have been doing just that for a number of months. Essentially, I’ve collected the state head-to-head polls taken in 2007 and have been analyzing the polls as a way of evaluating the relative strength of candidates.
Now I am going to switch into statistical wonk mode and explain my analyses. If you just want to see the results, skip over the Methods section and pick up from the Results.
Methods
To analyze the poll data I take the last month of polls for each state as a way to increase the certainty and (hopefully) minimize biases inherent in individual polls. If there is no polls taken in the last month, I use the most recent poll available in 2007. The analysis could stop at this stage after simply tallying the number of Electoral College votes each candidate would receive for each state based on the poll data.
The one problem with this approach is that it doesn’t account for the uncertainty in the polls. For example, suppose a poll in Pennsylvania of 500 individuals gives Clinton 51% and Giuliani 49% of the vote. Clinton’s lead comes from only five individuals who went for Clinton instead of Giuliani. In fact, statisticians would tell us that there is substantial sampling error because of the small sample size and the very close percentages. The statistician would do some calculations (or simulations) and tell us that the poll indicates that Clinton has only a 69.9% chance of winning, and Giuliani has a 30.1% chance of winning.
In simulating a national election, I do this same evaluation over all states. Here is how it works. I simulate elections using only information from state head-to-head polls (with one exception discussed below). Each single election proceeds state by state, pooling polls from the last month (or the most recent poll if no polls were taken in the last month). For each person polled in the state, I randomly draw votes according to the observed probabilities found by the state’s poll(s).
After conducting such elections in all fifty states (plus Washington D.C.), the electoral vote is totaled and a winner determined from the electoral vote count.
This process is repeated 10,000 times. The result is a distribution of electoral votes for the pair of candidates that fully accounts for the sampling error in the polls used. For example, here is the distribution of electoral votes for a Clinton—McCain match-up from a few days ago:
In this example Clinton won 9,167 simulated elections and McCain won 779 simulated elections. (There were also 54 ties that would go to the House of Representatives and almost certainly result in a Clinton victory.) Thus, the poll data suggests that, if the election were held today, Clinton would have a 92.2% chance of beating McCain.
Oh…about that exception I mentioned above. Some states have had no polls taken at all. In that case, I always assign the electoral votes for the state according to the 2004 presidential election outcome. For the most part, states that have had no polls taken are not likely to hold any surprises. In any case, this procedure slightly favors the Republican candidate (since Bush won in 2004).
Results
Here are the results after simulating a variety of match-ups. (Additionally, I provide a link to my most recent analysis. In most cases the published analysis is slightly older than the analysis from today given in the table below, but the numbers are close.)
Republican | Democrat | Probability the Democrat wins | Average electoral votes for Democrat | Link |
Giuliani | Clinton | 100% | 342 | Analysis |
Huckabee | Clinton | 100% | 335 | Analysis |
McCain | Clinton | 92.1% | 293 | Analysis |
Romney | Clinton | 100% | 385 | Analysis |
Thompson | Clinton | 100% | 354 | Analysis |
Giuliani | Edwards | 4.90% | 237 | Analysis |
McCain | Edwards | 99.4 | 303 | — |
Romney | Edwards | 100% | 388 | — |
Thompson | Edwards | 100% | 358 | — |
Giuliani | Obama | 27.7% | 258 | Analysis |
Huckabee | Obama | 88.7% | 277 | — |
McCain | Obama | 4.4% | 237 | — |
Romney | Obama | 100% | 376 | Analysis |
Thompson | Obama | 100% | 329 | — |
Right now Clinton does better against Republican challengers—she beats every one of them with a high degree of certainty. Edwards does very poorly against Giuliani, although he does a little bit better than Clinton against McCain. Obama doesn’t do well against either Giuliani or McCain right now.
Keep in mind that the analysis only suggests what would happen if the election were held right now. (Interpret this the way you might the speedometer on a long trip—it gives you some idea of your progress even though you know your speed is going to change along the way.)
Things will certainly change in the next ten months, but what we can say now is that Clinton has some advantage over both Obama and Edwards in a general election. Is Clinton’s advantage right now important in the long run? It’s hard to say. It’s not even clear to me that her advantage should be considered over more fundamental characteristics like political philosophy and policy positions. Perhaps some readers will use this information as a tie-breaker.
As for me? I still have no idea who I will support at tonight’s straw caucus. Maybe I’ll pretend to be a Republican….