In the run-up to the 2006 elections there was great fear among Democrats, and hope among Republicans, that gerrymandering would effectively maintain GOP dominance despite the will of the voters. The fear was that, just as Gore won the popular vote but lost in the electoral college, the Democrats might in fact win the "generic congressional ballot" but still fail to take the House. Even after the election, there was much discussion of how changing a few thousand votes hear and there might have (as Karl Rove said) kept the Republicans in power or (as has been noted here) given the Democrats a huge majority. And so I decided to look at what effect the drawing of congressional boundaries has on the representation in congress. In particular I was concerned with the questions: which states are the most gerrymandered to help one party? and How well does the composition of the next congress represent the actual proportion of votes won by the parties? My results show that (for our next congress) the votes are represented remarkably well.
First, a brief discussion of the electoral system and what I am not saying. I am not saying that the results reflect those that would come from a proportional representation (PR) voting system. In a PR system, there would be different incentives for party formation and for voters to select parties. How many parties received representation and what share of the votes they got would depend on a variety of factors (size of PR districts, minimum percentage of votes as a cutoff, etc). Undoubtedly many people on this site who did in fact vote for a Democrat would have voted for the Green list, and probably a few would have gone Libertarian, Socialist-Worker, or something else. The GOP also likely would lose votes to other parties (Libertarians, perhaps the fundamentalists would form their own "Party of God"). While it can be fun to speculate which parties might form and perhaps someday I will diary my thoughts on the subject, there is no way to tell what the results of a PR election would be based on the results of elections under our plurality system.
Second, I should state that I do not really attempt to determine where redistricting crosses into gerrymandering. A plurality system can lead to results that differ greatly from the overall public will without any gerrymandering. As an example, imagine that the voters of a state preferred Democrats to Republicans 51-49. Now imagine that these preferences were evenly distributed around the state. No matter how you drew the boundaries, the Democrats would win all of the seats, despite having only a slim majority. This is one of the common complaints about plurality systems. Obviously, voter preferences are not evenly distributed. This can mean that outcomes will more accurately reflect the distribution of voters preferences. It also gives those who draw the boundaries (generally state legislatures) the chance to affect how many seats are likely to go to each party. Still, it is entirely possible that a states representation in the House will fail to reflect the overall voting spread for a variety of reasons without intentional gerrymandering.
Finally, I should note that unlike a presidential election, in a congressional election people are voting for different candidates in different localities. Thus incumbancy advantages and the failure of both parties to contest every seat may distort the vote totals of the parties. With that I will finally get to the data.
I compiled the results of congressional contests from the 2006 elections and looked at the vote totals of the two major parties, as well as what share went to other candidates. A more detailed description of the data and problems therein appears at the end of this diary. First I compiled the data based on states. The resulting voteshare by state is in the chart below (apologies in advance for not spending more time making the charts readable).
Based on this data, I came up with a rather simplified version of how many seats each party should (proportionally) have won in each state. To do so I took each parties share of the two-party vote in each state (throwing out independents for the present), multiplied this by the number of seats in the state, and rounded to the nearist integer. By subtracting these results from the actual number of seats one by each party, I founnd the amount of over- or under- representation in each state. The results for the Democratic party are shown below (the GOP would obviously be the reverse of this).
The big offenders are not a surprise. The Democrats lost out on the most seats in Texas and Florida (assuming that FL-13 stays Republican), the Republicans lost out biggest in California and New York. It is not coincidental that these are also the four largest states. What was more surprising was the symmetry. Adding all seats shows that the Democrats had a net loss of one seat over what they "should" have had based on each states share of the vote, coming out with 232 instead of 233. A change in any of the disputed races could fix this. I would be interested to see whether previous elections worked out this closely.
While the overall national picture looked relatively good, there were some states that showed particular problems. The great fear is that gerrymandering will allow for a minority to rule over a majority. This was famously seen in the city of Londonderry/Derry in Northern Ireland during much of the 20th century. Despite the fact that the city had a large Catholic majority, the city council lines were drawn such that most Catholics were grouped into a single district where they had a vast majority, while the other two districts each maintained small Protestant majorities. While this has not been successfuly done in the United States as a whole, in six states a minority of voters elected representatives to a majority of seats. Two of these went for the Democrats. In Iowa they won 3 of five seats with about 47.7% of the vote (the Republicans had 50.6%). In Indiana, 48.6% of voters elected 5 Democrats while 50.5% got 4 Republicans. Republicans won the majority in four states: Nevada (2-1, 45.2-50.1), New Mexico (2-1, 44.2-55.8), Michigan (9-6, 44.6-52.7) and Ohio (11-7, 47.4-52.4). New Mexico saw the largest majority to lose out, while Ohio and Michigan each saw the minority win two extra seats. Iowa obviously has non-partisan redistricting, so the idea of a gerrymander there seems slim. I know little of the others (although I am particularly suspiscious of Ohio), but perhaps others will comment on what has happened in these states.
I also looked briefly at what would happen if the independents and third parties were left in. Generously, if all independents and third party voters were grouped together, they might have accounted for as many as five seats (2 in California, one each in Arizona, Texas, and Virginia). Individually, it is doubtful that any party would have one a single seat. As I mentioned before, this is not representative of what would have happened in a PR election, that merely represents the vote total not going to either Democrats or Republicans.
Finally, the question of distortions caused by federalism arrises. For reasons that will be discussed below, I am more confident in my data within states than I am across states. Still, running the total percentages I calculated (53.6% Democrats, 44.8% Republican), one would get a total of 236 seats for the Democrats to 199 for the Republicans based on a two party vote or 231D, 195R and 9 other if independents are included. Thus on a national scale, the Democrats do lose out slightly (about four seats, if no disputed seats change) based on the federal nature of the system and the drawing of districts. Still, the overall totals seem to reflect the proportions of votes cast remarkably well, and far better than I expected. I would be interested to see how this has played out in other elections, although I have no immediate plans to do so.
Some notes on the data (boring stuff you can skip): I compiled my data on vote totals from a variety of sources, making the compatibility across states somewhat suspect. Wherever possible, I used official websites of the Secretaries of State, Board of Elections, etc of different states. Some states do not post this data until after the results are finalized (often in December). For these states I was forced to rely on news websites, generally CNN. While much of the data does not include provisional ballots, I made an effort to only include data where all or nearly all of precincts had reported (those without all precincts were only included recently, as I wanted to stop wasting time looking for the data and finish this report). A more serious problem came from missing data. Six states had a total of 23 seats where I could find no data on incumbants who were running unopposed. The breakdown was AL(3), FL(5), MA(5), NY(4), TX(5), and WI(1). As 19 of these seats were held by Democrats, this would seem to have a strong distortion on the vote totals. Thus I decided to impute this data (through rather unsophisticated means). I looked at all states in which I had data on uncontested races, and compared the vote totals of winners in these races to those of other winners within the same state. Comparing across states, I found that on average winners of uncontested races recieved slightly fewer (about 97%) votes than did those who faced opponents within the same state. I used this number, multiplied by the average number of votes received by winners in the relevant state, to come up with my expected vote total for winners of uncontested races. As I imagine that, at the least, these numbers would be closer to the true total than 0, I used this modified vote total for all analysis above.