Impact of hypothetical congressional-district allocation of US presidential electoral votes

In recent weeks there has been considerable attention to proposals by some Republican politicians to change the allocation of presidential electoral votes from statewide winner-take-all to congressional districts–at least in states where doing so would help Republicans. If this method had been used for all electoral votes in presidential contests from 1968 to 2008, what would its impact have been?

I happen to have district-level presidential votes for each of these elections (but not, yet, for 2012*). The graph below plots both the actual and hypothetical** electoral vote percentages for each party against the popular vote. Red for Republican, blue for Democrat. The solid symbols indicate the actual percentage of electoral votes obtained, while the open symbols indicate the hypothetical allocation by congressional district. The plotted curves are local regression (lowess) curves for each party under each condition (solid for actual, dashed for hypothetical).

EV graph 2pty

The exercise shows how any discussion of shifting to this method of allocation should be talked about for what it is: a GOP-biased proposal. Note that, under the actual allocation, the two curves are close to one another, at least through the part of the graph where it really matters–the relatively close elections. There does appear to be a slight Republican bias in the actual method, as that party’s line crosses over 50% of the electoral votes at almost exactly 50% of the (two-party) popular vote, while the curve for Democrats crosses over at just over 50% of the popular vote. In other words, the data plot predicts the Democrat needs a bigger vote lead to get the electoral vote majority. But the effect appears very small, consistent with what Thomas, King, Gelman, and Katz find.

However, under the hypothetical congressional-district allocation, there is a clear Republican bias. The Republican curve crosses over 50% of the electoral vote well to the left of the 50% popular-vote line, while that for Democrats does not break over 50% of the electoral vote until the party has a clear majority of the popular (two-party) vote.

The 2012 result is shown in the graph for the actual allocation, though I did not have the district data readily available. The 2012 result closely matches the fitted curve for the actual result. If it also matched the fitted curve for district allocation, the electoral-college result would have been very close indeed.***

Only in the case of landslides in the popular vote does the congressional-district method result in greater “proportionality”, as indicated by the flatter curve for congressional-district allocation. Otherwise, there is no sense in which the Republican proposal is “proportional“; rather, it is a partisan power grab. It is a power grab especially when employed only in states where the Republican candidate tends to have a better geographical spread of the votes in the state; it is a power grab even if employed for all electors, as assumed in the hypothetical allocations shown here.

Let’s turn to individual elections. Below is the change for the Republican candidate in electoral votes if the congressional-district method had been used instead of the actual statewide winner-take-all:

1968: -8
1972: -46
1976: 28
1980: -97
1984: -57
1988: -49
1992: 48
1996: 34
2000: 15
2004: 31
2008: 64

As we already saw from the graph, in landslide years, the Republican wins fewer electors via congressional districts. The only electoral-college landslides we have had during this time have been by the Republican candidate: 1972, 1980, 1984, 1988. All of these but 1980 were also huge wins in the popular vote. In every election since 1992, the Republican gains electors regardless of whether he wins or loses the actual election. He also gains in 1976. (In 1968, both major-party candidates lose electoral votes, as George Wallace obtains 56 under the congressional-district allocation, against the 46 he actually won.)

In 2004, despite its being a very close election, Bush would have won 317 electoral votes with a district plan, against the actual 286. His total also would have been better in 2000: 286 vs. the “actual” 271. Had Florida’s electors been awarded properly in 2000 under statewide allocation, Bush’s total would have been only 246, meaning that the congressional-district plan would have netted him 40 extra electors, despite losing the popular vote. That’s even more than his 31-elector gain in 2004, when he actually won the popular vote. (It might have been a slightly bigger change in 2000, as I am missing three districts that year: AR3, IN10 and LA2. For two of these, congressional votes are also missing; IN10, is a very safe Dem district in 2000 House race.)

Of course, an objection to any simulation such as this is that we do not know how campaign strategy might have changed under different rules. That is certainly true; if each House district actually would have awarded an electoral vote, campaigns would have targeted the marginal districts, some of which would have swung the other way. In other words, the votes themselves could have been different.

We can get a broad understanding of the opportunities for potentially swinging electoral votes by considering how often a district is marginal in the presidential contest.

There are 4,782 observations.**** There have been 730 the entire time that were decided by less than 5 percentage points (15.26%).

Of course, this varies a great deal by year, as shown below (number in parentheses indicates winner’s margin under a congressional-district allocation):

1968, 71 (104)
1972, 28 (410)
1976, 102 (2)
1980, 77 (247)
1984, 36 (398)
1988, 61 (216)
1992, 103 (106)
1996, 82 (152)
2000, 62 (37)
2004, 44 (96)
2008, 64 (64)

Obviously, 1976 could have been swung by district-focused campaigning: there were many more close districts than the margin (two electors!) that Carter would have won by under a district-based allocation. Not surprisingly, 2000 is another year when districts within the margin of 5% outnumbered the overall electoral-vote margin under the hypothetical allocation. In 2008 there are as many close districts as the electoral-vote margin, and in 1992 the two figures are within a few districts of one another. Looking only at these four elections, we can see which party had the greater number of marginal district wins.

year Rep Dem
1976 62 40
1992 46 57
2000 27 35
2008 37 27

This suggests that Bush’s district-based win in 2000 would have been relatively secure, as he had fewer close races to defend against the Gore campaign’s (hypothetical) district-swing efforts. And there would have been little risk of the Republican swinging the 1992 or 2008 outcome, though the Republican could have made the race closer. But 1976 really would have been a complete toss-up, depending on how various individual district contests turned out.

We might think that the candidate who trails in the popular vote would have more marginal districts to defend, but this is not true in either 1992 or 2000.

All in all, it is clear that congressional-district allocation of electors benefits one party more than the other, and that in a close election, the Republican candidate would be likely to have an advantage. The Republican might even be able to win with less than 49% of the two-party vote.

It is easy to see why Republicans might like a district-based electoral college. It is much harder to see why anyone would think it was a democratic (small or large d) improvement over the current method, bad though that may be.

I am actually somewhat happy that some Republicans have opened the issue of electoral-vote allocation. The country needs this conversation. However, what it needs is not one party pushing a plan that would be blatantly distorting in its favor. It needs the Democrats to engage the conversation, and come out in favor of the National Popular Vote plan, which would remove partisan bias from presidential elections.


___________________
* See sixth comment, below.

** As is standard for such proposals, I assume that the winner of the statewide plurality of the popular vote would be awarded two electors, in addition to a number corresponding to the number of individual House districts won. Two small states, Maine and Nebraska, are the only two states to have used such an allocation in at least some of the years analyzed.

*** Andrew Gelman suggests that Romney might have won, given the “huge” distortion of congressional-district allocation.

**** Would be 435*11=4,785, if not for the missing districts.

11 thoughts on “Impact of hypothetical congressional-district allocation of US presidential electoral votes

  1. Yes, I have previously expressed my dislike of the American politics specialists’ use of the mythical “two-party vote”. Yet I use that here. For presidential elections, the years with significant third candidates are such outliers that it is easier to visualize if we just ignore the third candidate. Even though we should never do such a thing.

    For purists (like me!) I also prepared a version with the actual vote percentages. The overall pattern remains the same.

  2. There actually have been some arguments for allocating the two statewide electors not to the state’s popular vote winner, but to the candidate who won the most districts. That really takes chutzpa.

  3. I was under the impression that idea of giving a bonus to the candidate taking the most districts was only voiced in Virginia. And that even the Virginian GOP leadership found that one a little too far beyond the pale.

  4. Here’s one example of how piecemeal implementation of the Electoral College “GOPmander” could go horribly wrong for the Republicans.

    Under the existing system, Romney could have won the election had he prevailed in the three most closely fought battleground states – Florida, Ohio and Virginia, with a total of 60 electoral votes – plus New Hampshire, whose four electoral votes would have put Romney at the bare minimum majority of 270-268. This could have happened had less than 250,000 voters in the aforementioned states switched from Obama to Romney…which nevertheless would have left Obama with a sizable popular vote lead over Romney.

    Now, in addition to the described changes, let’s suppose that Virginia had also adopted the “GOPmander” plan for the allocation of its 13 presidential electors. As it happens, even if Obama lost that state, he would still prevail by large margins in three of its eleven CDs…and in the process deprive Romney of an Electoral College majority: Obama would then retain the White House by 271-267, with the Virginia “GOPmander” backfiring disastrously on the Republican Party.

    Incidentally, this is probably not the only scenario of its kind, but I haven’t come across full CD-level 2012 presidential election results to confirm another example I have in mind.

  5. Following up, the district-level results show 226 districts for Romney and 209 for Obama. Combined with 24 states won (and hence 48 more electoral votes), a district-based plan would have reversed the outcome of the presidential election, although it would have been close enough (274 electoral votes for Romney, 270 needed) that campaign effects could have prevented the reversal.

    House districts with margins of less than 5%, by party:
    R 27
    D 21

    Of less than 2%:
    R 12
    D 10

    (Partially transplanted from a previous thread.)

  6. This is also from Daily Kos Elections. One diarist, Stephen Wolf, took the trouble of re-drawing the maps of all the states that had partisan gerrymanders into what they would look like if they were drawn by a California-type commission.

    The result is impressive, the lines chosen look reasonable (I’ve not had time to go carefully through the whole diary), and the categorization of the process by which the lines were drawn for each state alone is pretty helpful.

    The diary is here: http://www.dailykos.com/story/2013/04/22/1201551/-Did-Gerrymandering-Cost-Dems-the-House-A-33-State-Look-at-Alternative-Non-Partisan-Maps

    The conclusion is that if the lines were drawn by a California-style commission, there would have been a Democratic House majority in 2012.

    I have to admit that I embarked on a similar project, but though my maps are finished, and I’ve downloaded the relevant demographic and political data into a series of spreadsheets, I haven’t gone through the spreadsheets yet to see what the political effects were (I have a day job).

    However, my project was different. The neutral redistricting model I used was the British boundary commission, with the exception that I used the American rules for deviations from the mean for the size of populations in each district. But I decided to ignore the Voting Rights Act and instead strictly avoid crossing local government boundaries (mainly county boundaries), as much as one could given the low American tolerance for population deviation, and otherwise use communities of interest. One of the reasons I took this approach was that with the entire VRA up for review by the Supreme Court, I was curious to see how many majority African-American and majority Latino districts could be drawn using purely race-neutral criteria. Though racial criteria comes back in to some extent with communities of interest, the provision of avoiding crossing local government boundaries makes it really difficult to get to 50% +1 voting age populations for various racial groups.

    There is right now about an equal likelihood of the US adopting the British or Canadian methods as taking the California approach, the likelihood of both happening in the near future being zero.

    However, the “respect local boundaries” approach itself leads to some oddities, since in the US local government boundaries too are often gerrymandered.

    I do have some other issues with the diarist’s approach. First, the actual California process was not as neutral as he seems to think, the Democrats were more effective in lobbying/ manipulating the commission (this has been documented) so the resulting map should be characterized as a Democratic-friendly commission map, similar to Arizona. This means that a less Democratic-friendly commission, say on the model of Washington, would not necessarily have produced a map resulting in a Democratic majority.

    My other issue is with the assumptions that the maps drawn by judges are both politically neutral and would duplicate the maps drawn by a commission, both of which are highly questionable for all sorts of reasons. I redrew the judge-made maps as well.

    Even with these big differences in approach, the diarist’s maps came out quite close to mine. The whole diary and comments are worth taking the time to go through, and also with redistricting software now available to the general public, there really should be more interest by political scientists in map-making.

  7. Under the current Electoral method and a National Popular Vote method there is a tyranny of the majority. It is currently in each state but, under NPV method it would be countrywide making the problem dramatticly worse. Only the Congressional District method does the little guy get a chance and the small town get a voice. Currently smaller populations have no voice in elections. Currently only large urban centers have a voice in elections.

    • Doesn’t everyone get an equal voice under NPV? Whichever candidate gets the most votes wins. Doesn’t matter where the votes come from. Why should someone get elected president because lots of small towns support him but 55% of the national aggregate population does not?

  8. Pingback: So, what about districts? | Fruits and Votes

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s