December 5, 2004

Correlation of Presidential Votes by State

An extraordinary change in politics: I've discovered something I am almost flabbergasted by concerning how much Presidential politics has changed since the 1950s.

I've mentioned before how stable the election results by state and by demographic group were from 2000 to 2004. Bush's share of the vote in 2004 by state correlated at the 0.98 level with his performance in 2000. What that means is that if you spent November in a cave and just surfaced today and asked "What happened in the election?" you could be 96% (that's 0.98 squared) accurate in guessing Bush's share in each state with just three kinds of information: his 2000 performance, his new intercept (start Bush off 3.9 percentage points higher), and his new slope (for each 10 percentage points his 2000 share goes up per state, his 2004 share goes up 9.77 percentage points). For example, if he earned a 50% share in a particular state last time, you would expect him to earn 52.7 points this time (3.9 + (5 * 9.77).

So, how does the stability from 2000 to 2004 compare to elections in the past? The impact of third party candidates makes it somewhat difficult to compare seemingly similar pairs of elections, such as the President's father's campaigns in 1988 and 1992. The correlation of Bush41's share in 1988 to 1992 on a state-by-state basis was only .83 (71%), but, perhaps, the intervention of Ross Perot, who captured 19% of the vote, had something to do with that.

The cleanest comparison to 2000 and 2004 is the 1952 and 1956 elections, which twice in a row matched up Dwight Eisenhower and Adlai Stevenson. You would think that the results would have been almost identical from 1952 to 1956, but they correlated only at the 0.78 level, meaning you could only be 61% accurate at plotting Eisenhower's 1956 results knowing his 1952 results and Eisenhower's intercept and slope for 1956. In other words, there was hugely more shifting at the state level between 1952 and 1956 than between 2000 and 2004.

Eisenhower's overall share grew 2.3 points from 1952 to 1956, only a little less than Bush's improvement from 2000 to 2004, but Ike's share fell in 19 of 48 states. In contrast, Bush lost share in only 2 of 51 states (although this may change slightly as final vote counts come in).

Were voters in 1956 much more sensitive to the actual policies advocated by the candidates, and how they would affect their states, and thus more likely to change their votes as both candidates altered their stance on issues to try to appeal more broadly? In contrast, did voters in 2004 vote not so much on the issues as on which (relatively unchanging) part of society they wished to affirm their membership in?

By the way, the correlation between Eisenhower's share by states in 1952 and Bush's in 2004 is -0.01, or utterly random.

Here are the r-squareds for state-by-state correlations for the last eight elections. For 1992 and 1996, I've laid out the correlations both with the GOP candidate by himself and with the GOP candidate plus Perot (i.e., the non-Democratic share of the vote). There seems to be an upward trend over time for elections to become more stable, although 1984 to 1988 was 88%, which is low only compared to 2000 to 2004 (96%). The 1992 and 1996 elections were somewhat perturbed by Perot and by Clinton, who had a certain amount of Southern appeal.

R-Sqd 1984 1988 1992 1992 1996 1996 2000

Reagan Bush Bush Bush Dole Dole Bush

+ Perot
1988 Bush 88%

1992 Bush 59% 71%

1992 Bush+Perot 84% 68% 53%

1996 Dole 68% 68% 75% 67%

1996 Dole+Perot 77% 70% 66% 83% 93%

2000 Bush 70% 64% 66% 68% 89% 93%
2004 Bush 72% 66% 72% 69% 88% 91% 96%


My published articles are archived at -- Steve Sailer

No comments: