(Cross-posted at Congress Matters)
The Cook Political Report's Partisan Voting Index (PVI) is a measurement of how strongly an American congressional district leans toward one political party compared to the nation as a whole. The index value for each congressional district is derived by averaging its results from the prior two presidential elections and comparing them to national results. The resulting PVI indicates which party's candidate was more successful in that district as well as the number of percentage points by which its results exceeded the national average. By using two elections this method smooths out bumps in election results that may occur because particular candidates played anomalously well or poorly in particular regions.
By trying to calculate the new PVIs after the 2008 elections it turns out that Cook's method has serious flaws and there is currently a discussion going on about this over at the Swing State Project. In my diary I want to showcase what seems to be the best approach to solve the problem.
For the years 2005-2008 the formula for the Cook PVI was defined as follows:
Cook PVI = [(Gore's 2000 two-party vote share in congressional district - Kerry's 2004 two-party vote share nationally) + (Kerry's 2004 two-party vote share in congressional district - Kerry's 2004 two-party vote share nationally)] / 2
Postive numbers result in Democratic PVIs and negative numbers in Republican PVIs.
The problem with this formula is that it compares 2000 district results with 2004 national results. Therefore it will systematically discriminate the party which improves over the previous cycle. If you have succeding elections where the two-party vote didn't deviate all that much (like 2000 and 2004) then the problem is minimal. But serious problems arise in case of an election like 2008 where one party makes a huge improvement over the previous election. This can be easily revealed with a district like OH-12 which voting habits are akin to those of the nation as a whole (Note: PVI values are based upon the two-party vote, not the actual vote share of the candidates):
In 2004 the two-party vote in the district was Bush 50.9% and Kerry 49.1% while the national values were Bush 51.2% and Kerry 48.8%.
In 2008 the two-party vote in the district was McCain 46.2% and Obama 53.8% while the national values were McCain 46.3% and Obama 53.7%.
So in both years the democratic candidate very slighty outperformed his national numbers in OH-12 (by 0.3% and 0.1% respectively) which should consequentially lead to a slightly democratic PVI for this district. However, by comparing 2004 congressional results with 2008 national results, Cook's formula turns things upside down and results in OH-12 having a new PVI of R+2.3. This is of course totally absurd and contradicts the intended purpose of the PVI!
Fortunately this error can be easily corrected by comparing 2004 district performances with 2004 nationwide performance, not 2008 nationwide performance. The adjusted formula for ths improved index (which I like to call SVI - the "Swing State Project Voting Index") is:
SVI = [(Kerry's 2004 two-party vote share in particular congressional district - Kerry's 2004 two-party vote share nationally) + (Obama's 2008 two-party vote share in congressional district - Obama's 2008 two-party vote share nationally)] / 2
Postive numbers result in Democratic PVIs and negative numbers in Republican PVIs.
To depict the effect of this formula here are the SVIs of the 25 most Democratic districts represented by a Republican based on the election data from SSP's Pres-by-CD project (Note: To learn how the unadjusted new Cook PVIs would look like just add R+2.5 to the values below):
LA-02 D+24.2
DE-AL D+7.0
IL-10 D+6.2
PA-06 D+3.8
WA-08 D+3.2
PA-15 D+2.5
NJ-02 D+1 (est.)
IL-06 D+0.4
OH-12 D+0.2
IA-04 D+0.1
MI-11 R+0 (est.)
MN-03 R+0.4
MI-06 R+0.6
FL-10 R+0.8
NY-03 R+1 (est.)
NY-23 R+1.0
IL-13 R+1.4
MI-08 R+1.7
NJ-07 R+1.9
IL-16 R+2.0
WI-01 R+2.1
VA-10 R+2.3
OH-14 R+2.8
FL-18 R+2.9
CA-50 R+3.1