About two weeks ago I altered our model to reflect the mere possibility that the Bradley Effect could rear it's ugly head; the article was met with mixed reviews. Perhaps I wasn't clear enough in my original explanation so I'm going to do it again.
In the next twelve days there will be a plethora of articles relating to the Bradley Effect, but none of them will be conclusive; not because they are poorly reasoned or factually inaccurate, but rather due to the complexity of the issue. Nobody can definitively say whether the Bradley Effect will or will not occur; it's simply impossible to know. There is no historical precedent, on the national scale by which an accurate conclusion can be drawn. Given the aforementioned limitations of the problem, I'll take a different approach. Rather than trying to prove that the Bradley Effect will or will not play a role in the upcoming election, I'll try and show that it simply doesn't matter.
In order to debunk the affect of the Bradley Effect, I'll try and simulate the potential outcome of the election should the Bradley Effect occur; we'll assume a historically significant racial offset as dictated by the Princeton University graph below:
Source: Princeton Election Consortium
Judging from the graph and more specifically the data points on the far right corresponding to the year 2006, it can be seen that the Bradley Effect was distributed between +5%, (meaning the African American candidate polled 5% lower in the final result than in the public polls leading up to the election) and -5%. For the purpose of this analysis we are trying to find the maximum degree to which the Bradley Effect could occur; in order to achieve this end we will use the maximum racial offset of +5% as prescribed in the 2006 data.
Incorporating the 5% into the projection algorithm presented the following changes:
I calculated his initial projection and then subtracted 5% of it to arrive at his adjusted projection.
I started by figuring out the number of undecided voters at the end of the initial projection; this is done by subtracting 100 from McCain's and Obama's initial projection. I then added this number to McCain's projection; to account for the adjustment made to Obama's total above, 5% of Obama's initial projection is then added to McCain's running sum to arrive at McCain's final adjusted projection.
Translating the pure formula into words results in a much more succinct correlation to the core principles behind the Bradley Effect. I assumed that 5% of Obama's support was racially tinged so I added it to McCain's total while subtracting it from Obama's. I then also assumed that of all the currently undecided voters, 5% are racist and as a result they will cast their vote for McCain. The result of this application can be seen in the map below:
Using a severe over exaggeration of the Bradley Effect's potential outcome, Obama would still emerge victorious. In fact, Obama has gained three Electoral Votes during the two weeks that have elapsed since our previous experiment. The Bradley Effect would likely manifest itself differently in each state, but in assuming this large discrepancy between what would likely happen and what could happen the absolute worst possible outcome (for Obama) is presented. If the most severe Bradley Effect outcome still results in an Obama win, it's fair to say that the Bradley Effect will not alter the outcome of this election.
While altering the outcome of our model, I wondered what would happen if all currently undecided voters suddenly moved to McCain en masse:
Wow, Obama still wins. Given the outcome of this scenario McCain must win every currently undecided voter in the United States along with a handful of Democrats already committed to supporting Obama; a monumental task. Our models don't show any chance of this happening.
Update: Several comments have suggested that I also create a voter suppression model; this is a good idea. I'll work on it this weekend.
Debunking the Bradley Effect, Again
Leave a Reply:
Name: (Defaults to Anonymous)
Type the characters you see in the image below: