In the previous planting on presidential approval and impact on House votes and seats for the president’s party, I noted there was no evidence that the translation of votes to seats had become less responsive to changes in votes totals (as might be expected from gerrymandering or other trends making House districts more homogeneous).
However, when we look at the scatterplots of each election since 1946, we might draw a different conclusion. This first graph shows Republican votes percentages and seats won in the House.
(click on the image for a larger version in a new window)
The horizontal line shows 218 seats, a majority. The diagonal is a best fit. There is not a lot of scatter around the best fit–consistent with my statement in the earlier post that the relationship between the popular votes and House seats is more predictable than I might have imagined.
Now, for the sake of comparison, here is the same graph for the Democratic party.
Less here stands out as anomalous, at least to me. Not even the recent elections. (The one outlier at 260+ seats on 48% of the vote, which I forgot to label, is 1948.) However, the best-fit line is a bit less steep: Democrats, over time, have gotten a bit less payoff (or decline) in seats from any given gain (or loss) in votes. (Partly this may be a diminishing returns phenomenon, given how dominant the party’s votes majorities were for many years.)
Now, let us return to the Republican graph. Despite an overall tendency for seats to follow from votes in a fairly predictable way, it is clear that recent elections are deviating from the pattern. Every election since 1996 is above the trend. More to the point, in each House election since 1996, the Republican party has won a majority of seats on less than 50% of the vote. The only time that had happened before was in 1952. In 1996, as well as 1952, the Republican party did not even win a plurality of the House vote.*
With all due caveats about the small number of data points (5), if we eyeball a best fit to the 1996-2004 data, the line is clearly much less steep than the overall trend, suggesting that today’s Republican party can afford to lose vote share and yet retain the House majority. It could fall to 47% (in 2004 it was 49.3%) and possibly still have at least 218 seats. Nonetheless, such an outcome is not very likely. Part of what allows the Republicans to retain a majority on less than 50% of the vote is the growth in recent elections in third-party voting. If 2006 is going to be a change year, and if the Republican vote falls to 47% (or lower), it is highly unlikely that the third-party vote will rise (it was 4.2% in 2004) enough to keep the Democratic party from enjoying a plurality of a few percentage points. (Compare 1994, when the third-party vote share was 3.7% after having been 5.3% in 1992.) Is the House so gerrymandered or would the regional vote swings be so skewed as to allow a Republican majority if the votes were something like 50-47 in favor of the Democrats? I don’t think so, though it is certainly possible.
* In 1952, Democrats actually had a majority of the vote, yet Republicans won a majority of seats.
UPDATE: Readers interested in these themes will find Jack’s post at the Fair Vote Blog, linked in the comments below, to be of interest. I think the hypotheses that Jack is referring to can be tested only at the district level and over time. I hope to look into this, although the district-level data I have contains no third-party votes, and as I allude to above, it is clear that growing third-party voting is at least part of the picture of “skew” in the sense of regional and partisan biases. It would be quite an effort to assemble district-by-district data on all candidates and not just those with R or D by their names. The country specialists who study the US House simply throw out anything that is not D or R. We comparative electoral systems specialists know that doing so is likely to miss something important–or at least prevent the opportunity to find out if other party voters are making a difference.
If anyone knows of machine-readable data that includes all candidates, please tell!