Monday, September 15, 2008

Aye, Iowa

Yesterday I noted that there had been no post-convention poll in Iowa and that, in its absence, my model was showing McCain ahead in the state. Well, the pollsters heard of my dilemna and rustled up real quick an Iowa poll to fill that gaping gap. According to the Des Moines Register's poll, Obama is ahead in Iowa by a 52% to 40% margin. As this margin is similar to that shown in at least one pre-convention poll in the state, it is highly likely that Obama is leading in Iowa and quite likely that his margin is fairly substantial. To compensate for what appears to be reality, I have tacked five percentage points on Obama's share for Iowa (and correspondingly have taken five percentage points off of McCain's share).

The map below shows how unusual Obama's performance in Iowa is:

The figures in each state are Obama's estimated vote percentage shares. Obama appears to be struggling to stay on top in Minnesota and Wisconsin two upper mid-western states once reliably Democratic but which have now become swing states. Note the overwhelming Republican support in the tier of agricultural states running north to south from North Dakota to Kansas. Obama's performance in Iowa is just a shade behind that of his home state of Illinois, a state dominated by the Chicago metropolitan region. In contrast, Iowa's economy revolves around agriculture and Des Moines is the only significant city of any size in the state.

So why is Obama's support in Iowa so strong? Corn might well be the reason. Corn is the state's most important crop and virtually the entire state is located within America's prime corn growing area. There's this little thing called ethanol, the production of which is heavily subsidized by the US Government and which uses corn as its primary input. Iowa has benefited handsomely from this generous US Government program. So has Illinois, which is also a major corn producer. Now it just so happens that one Barack Obama, Senator from Illinois, is a strong supporter of corn ethanol. Looks like the good folks of Iowa just may be showing their gratitude for good old fashioned pork.

I mentioned yesterday that I would discuss a bit more the election forecasting model. My model has been tracking quite closely the performance of their model both at the national and state levels. It appears we are using somewhat the same approach, that is, we are using results from the national polls, other states polls and the in-state polls in order to estimate the current condition in a given state. When I compared the results of my model yesterday with the model on a state by state basis, I noticed a very curious pattern. In spite of the fact that the overall percentage shares at the national level were virtually the same, Obama's percentage shares in the model were higher than those in my model in 31 of 50 states. When I checked these 31 states where Obama's percentage shares were higher than in my model, my model was showing McCain ahead in 23 of these 31 states (McCain was ahead in 21 of 31 states according to the model). In constrast, for those 19 states in which Obama's percentage shares were lower than those in my model, Obama was ahead in 11 these states while McCain was leading in 8 in both of our models.

In other words, the model appears (compared with my model) much more likely to estimate higher percentage shares for Obama when McCain is ahead in a given state and is conversely more likely to estimate lower percentage shares for Obama when Obama is ahead in a given state. I've tested this relationship using a simple chi-squared test and found that these differences were significant at about the 2% level. That is, it is pretty unlikely that the observed relationship has occured by chance. Instead, either I am biasing the results in favor of McCain by tilting swing states in McCain's direction or is biasing the results in favor of Obama by tilting swing states in Obama's direction. Since I am without bias or prejudice, I can only point the finger (tongue-in-cheek) in their direction since the folks appear to be quite highly partisan in favor of Obama. Jokes aside, this does point out that the human factor plays an important role in these models. Studies suggest that the more mechanical the model (that is, the less reliant on human judgement) the better the model's performance.

Other than the Iowa poll, two Minnesota polls came out today suggesting that the race there is very close wnile a New Jersey poll reinforces the belief that Obama has a modest, but real, lead in that state. Oh, yes, there was another poll from Utah shows McCain far, far ahead. If someone hadn't gone through the trouble to take a poll in that state, how could we possibly know who was ahead? The updated tracking polls show a breath of movement towards Obama, now trailing McCain by 2.0 percentage points, and, after the above-noted Iowa adjustment, McCain leads in the Electoral College by 283 to 255 votes.

No comments: