Wednesday, March 05, 2008

Primary vs. General Elections

Al Gore won the 2000 New Hampshire Democratic Primary. George W. Bush lost the 2000 New Hampshire Republican Primary. One would assume that if primaries are good predictors of general election results that Al Gore would have won New Hampshire in the 2000 General Election. That is not the case.

Primaries are a non-random subsample of the entire electorate and making an argument that they are good predictors of future strength in a general election campaign is a very weak argument. However Jerome Armstrong is making that argument about Ohio and Obama:

Obama has a huge electability problem in the state. He took a total of 5 counties, and lost in 82 counties. Even though he's able to rack up a large number of urban black voters he did terrible among white voters, winning just 34 percent....

You don't win a general election in Ohio if you can only win in 5 counties.

There are a couple of problems with this argument.

The first is that most Democrats will vote for a Democrat in the general election over any Republican (and vice versa) so most Clinton voters in the Democratic Primary will vote for Obama in the general, and most Obama voters in the Democratic Primary will vote for Clinton in the general election.

Secondly, the universe of voters is much larger and more diverse in the general election. Independent and non-aligned voters who are less invested in the Democratic Party are more likely to vote in the general election, and they are a bit more unpredictable than independent but functionally Democrats in voting behavior independents who are more likely to vote in an open primary.

The third problem is a strategic decision making problem. The goal of the Democratic Primary process is to achieve as many delegates as possible. We know that the delegate allocation rules gives more weight to a vote in Cleveland than a vote in Cadiz. Winning as many votes as possible is one strategy maximize delegate counts as an 85/15 split will give the winner 100% of the delegates, but it is not an optimal strategy when there are two viable candidates. Each candidate has a delegate gathering strategy, and they interact with each other in a non-cooperative manner. And these strategies are not necessarily vote maximizing strategies for the primary.

However the goal of a general election is to achieve a plurality of all votes cast within an electoral college voting group. A vote in Tuscarawas County, OH is now just as valuable as a vote in Cleveland. Furthermore, McCain's base coalition is a very different base coalition than the one to be expected by either Democratic candidate in the general election and very different than the base coalition that either candidate assembled last night. In a primary between two popular figures, there is a decent chance of stealing significant elements of the other's base coalition (Clinton seems to have done this with the service unions last night) while there is a minimal chance of Clinton/Obama stealing significant support from McCain in the 'bomb everyone to the stone age' crowd. Two very different objectives with two very different sets of initial moves and therefore countering moves.

Now if you want to argue based on a combination of demographics, policy, persona and organization, generic issue trends etc... that Clinton has a higher probability of beating John McCain in Ohio than Barack Obama, then that is a legitimate and interesting argument. But a linear extrapolation from a primary to the general election is an extraordinarily weak argument.

No comments: