All of this takes huge time away from his academic work, and the vast majority of top academics do little but work huge hours on their academic work. Thus, he has seriously risked his chance of winning a Nobel prize, and of being recognized as one of the great economists in history. This huge amount of time could also have been spent on his personal life, which, with all he does, I doubt he has much time for. So he's truly a caring man who has sacrificed much to help others.

O.k., a moment to dry our eyes...Now to my comments on Paul's recent post:

Some key things to keep in mind:

Define the parameters a1 through a7 as follows;

-a2 = Decrease in earnings greater than 10%

-a3 = Decrease in earnings between 5% and 10%

-a4 = Decrease in earnings between 1% and 5%

a5 = Increase in earnings between 1% and 5%

Then, come up with numbers for the parameters a1 through a7 by calibrating to the historic data before 2008 (finding the values for the a1 through a7 that will make this model best fit the historic data).

So, you might find a1 = 5, a2 = 3.7, a3 = 1.4, etc.

Then, plug in 2008's numbers and see what comes out for the probability of the incumbent party losing.

The idea behind catagorizing is that it stops false linearity. What do I mean by that? The way the model is now, if a person's income increases by 50%, it's saying it's 10 times more likely that he will vote for the incumbent party than if his income increased by only 5% (this is approximately true, technically due to a non-zero intercept -- sorry to the laypeople). This is probably very unrealistic. In reality it's probably something like there's typically a 60% chance of a person voting for the incumbent party if he has a 5% increase in income and an 80% chance if he has a 50% increase in income. So the probability increase from 60% to 80%, a 33% increase, not a 10 fold, or 1,000%, increase.