Assessing Voter Partisanship: A New Model for North Carolina

How Does It Work?

How Well Does It Work?

In 2006, which was an unusual election year by most standards, the outcome in state House races was more highly correlated with the NCPI than with voter registration. In fact, holding party registration constant, on average a 1 percentage point increase in the NCPI (i.e. from D+1 to D+2) translated to an increase of 0.6 percentage points in the vote for that party’s candidate. Of the 85 contested legislative seats in 2006, all but 16 seats went to the winner of the party indicated by the NCPI (two seats had a neutral NCPI). Among the 16 districts that were not in line with the NCPI score, seven were in the range of R+3 to D+3, and several others involved races with multi-term incumbents, well-known challengers, or significant spending differences between the candidates.

More About The Model

The NCPI is based on voter data from the 2004 general election results for governor and other council of state offices. Although president and U.S. Senate results were also available, we found that state-level races give a more accurate picture of how voters will vote in a state legislative race than do national races.

In general, North Carolinians tend to send more Republicans to Washington, D.C. and more Democrats to Raleigh. In 2004, in national races, North Carolinians voted 45.3 percent Democratic and 53.8 percent Republican. By contrast, in council of state races, North Carolinians voted 53.4 percent Democratic and 46.3 percent Republican. However, many council of state seats have been unevenly contested in the past, making it difficult to predict results “down ticket” looking at raw numbers alone. (In contrast, in the three most hotly contested Council of State seats – seats which were either open or resulted in an upset – North Carolinians voted 49.9 percent Democratic and 50.1 percent Republican, a virtual dead heat.) By looking only at deviations from the state average, the Civitas model is able to mitigate the effects of incumbency and uneven contests.

The NCPI model also adjusts for outliers due to the popularity of a specific candidate in a specific area (for example, the governor’s popularity in his home town). Other data assumptions are also applied.