Not Gerrymandering, but Districting: More Evidence on How Democrats Won the Popular Vote but Lost the Congress

by John Sides on November 15, 2012 · 29 comments

in Campaigns and elections

This is a guest post by political scientist Nicholas Goedert, who is a Postdoctoral Fellow at Washington University.

*****

Expanding on recent posts by Dan Hopkins and Eric McGhee, there appears to be evidence at a state-by-state level that the disparity between the popular vote in the House and the distribution of seats is not just due to Republican gerrymanders, but due to a skewed geographic distribution of population putting the Democrats at an inherent disadvantage, along the lines of Chen and Rodden’s recent work.  That is, the Democrats’ loss in the House was caused largely not by gerrymandering, but districting itself.

McGhee’s post compares the results of the 2012 elections to what the election might have looked like using the 2002-2010 maps, and finds that the most recent round of redistricting had relatively minimal effects.   An alternate way of measure districting effects is to compare the 2012 results with historical patterns from recent congressional elections for seats won for a given popular vote share.  Using this technique, I find slightly greater effects of partisan gerrymandering, but also a persistent bias in favor of the Republicans.

To measure the distortion caused by districting, I calculated each party’s mean vote share in 2012 across each state’s congressional districts, then estimated the share of seats a party could normally expect to win given this vote share under a historically average seats/votes probit curve; for example, over the past 40 years, winning 55% of the popular vote has generally led to winning about 60% of seats.  I then compared this to the actual share of seats won.  The tables below depict the results for all states with at least seven districts, excluding a few Deep South states (where districting is dominated by Voting Rights Act considerations) and Washington (where only about 80% of the vote had been counted at the time of writing).

Table 1 shows nine states with maps drawn by Republicans:

In every state districted by Republicans, Democrats won fewer seats than their historical expectation, and in six cases they underperformed by 20% or more (as a percentage of total seats up for election).   So it appears that Republicans gained benefits across the board from controlling the redistricting process.

By contrast, Democrats exceeded their expected seat share only slightly in the three states where they controlled the process.  As shown in Table 2, Democrats gained just a fractional seat above expectation in each such state.  For instance, Illinois Democrats won a smaller majority in their delegation than Pennsylvania or Ohio Republicans won in theirs, despite winning a much larger vote share.

But partisan control of redistricting cannot completely explain each party’s performance relative to the hypothetical unbiased map.  Instead, we still observe bias even where we should expect none in the redistricting process.  Table 3 shows that Democrats also fell short in several states with bipartisan or court-drawn maps, winning on average 7% fewer seats than expected.

So how many seats did this underlying disadvantage cost the Democrats?  If we were to imagine that these bipartisan or court maps were unbiased, and Democrats received the same benefit from their own maps that Republicans received from theirs (let’s say a 13% advantage as an average), this would have yielded 14 additional seats, likely getting the Democrats within 3 or 4 seats of the majority.

If there is any area of the country where the geographic distribution of partisans has not led to an underrepresentation of, we might expect to observe it where Democratic voting strength does not hue as closely to the black/white or urban/rural divide.  In particular, we find this pattern interrupted in areas with very strong Hispanic populations.  Table 4 shows the three large states with the highest proportion of Hispanics, revealing that Democrats won a seat share very similar to their expected share in each of these states, despite not controlling the process in any of them.  It is possible that nonpartisan commissions may have contributed to greater fairness, but the ease of drawing geographically large, majority Hispanic districts in these states, (e.g. AZ-2, CA-16, CA-51, and TX-23) might have also mitigated the natural advantage Republicans have in other regions in the distribution of the their vote.

In direct support of the Chen and Rodden argument, states that are heavily urbanized (such as New Jersey and Pennsylvania) are more distorted against Democrats than more rural states (such as Minnesota and Wisconsin).   Indeed, urbanization has a negative and significant effect on the difference between seats won by Democrats and expected seats, even after controlling for the party in control of redistricting.

Of course, this analysis does not imply that Democrats are doomed to the minority for the foreseeable future, or even the next decade.   The Pennsylvania map includes five Republican seats won by Obama in 2008, suggesting that a wave of sufficient strength could reverse the delegation’s majority.  But because of unequal concentrations of vote share in most states, not just those with Republican gerrymanders, a Democratic majority will be more difficult than it should be.   And this difficulty persists even when both parties agree to the maps.

Changing our redistricting institutions alone will not assure national proportionality.

{ 29 comments }

John Griffin November 15, 2012 at 1:23 pm

Ironic that Paul Ryan is (erroneously it would appear) claiming urban areas cost Republicans the White House when they, along with districting, appear to have aided the party’s retention of the House . . .

Rob Richie November 15, 2012 at 2:04 pm

We’re really glad this analysis is getting aired. We’ve been making this point for some time and suggesting modest, American forms of proportional representation that can be established by statute. See, for example,congressional plans with real data for every state in the country accessible through online reports and our interactive map at http://www.FairVoting.US

ClearEye November 15, 2012 at 5:22 pm

I completely agree that from the perspective of governing nationally multi-member districts (at least for large states) would give us a more representative mix for each STATE in Congress. However, I worry about how that may limit access (e.g. in the sense of constituent services) for those that don’t hold majority views and are not represented by someone holding their view.

I live in a small state where the county is the only form of local government, and our county politics operate a twist of a multimember district.

Each of nine seats have a defined area in which the council candidates must live in order to run, but we vote county wide for all 9 seats. The difficulty here is that 2 of those areas are geographically isolated from the others and too small to be an independent district in themselves. Specifically because they are isolated they NEED their own representative on the council.

You could say that we are like an urban state in that we have 4 “urban” districts, 2 of which are adjacent and constitute what you would consider the “urban core”. The other 2 urban areas are distant from the core, and 2 other districts are primarily rural.

The essence of our local elections is that the two “core” districts essentially elect all of the council because they constitute 1/3 of the vote in the county. Whether distant urban districts have true access and representation or not is HIGHLY dependent on the office holder. At the county level they can almost ignore significant majorities that our State House members (in roughly corresponding districts) must pay close attention to. Furthermore at the council level many widely held views that don’t align with the majority view in the urban core are rarely are discussed or when they are, they are presented as extreme oppositional views.

I realize our local political situation is probably not of interest to you but it does raise a concern I have about multi-member districts. I’d prefer to see a majority of Democrats in Congress now, but as I think back to the purpose of the House of representatives – which I think was not only 1 man 1 vote, but also so that voter who are far from DC might have a chance of knowing and being heard by their representatives. I see a failure of the 2nd piece here and for those with views held by a significant minority, it effectively quashes their voices and perspectives.

Your thoughts?

Andre Kenji November 16, 2012 at 12:10 am

There is another point: the fact that an US Representative is really the Representative of the people that lives there is enviable all over the world. That´s something much more difficult to be dealt if you have three or four representatives in your district.

Arizona Eagletarian November 16, 2012 at 5:25 am

And yet, it is (already) NOT a requirement that one’s Congressional Representative live in the district which he or she represents.

John November 16, 2012 at 1:37 pm

There’s plenty of other countries with single-member districts – the United Kingdom and Canada come to mind. Do they envy us too?

Keith Agaran November 15, 2012 at 2:11 pm

Great that it’s “out”, but it needs a wider audience… Maddow could get it into the thick heads of media pundits…

Andre Kenji November 16, 2012 at 12:05 am

There is another point: the large size of the Congress Districts is a big thing that works against the Democrats. It would be more difficult to gerrymander smaller districts, by the way. I also imagine that the problem of wasted votes would be smaller.

Arizona Eagletarian November 16, 2012 at 5:19 am

The redistricting process should belong to the PEOPLE, not to the legislators. The people deserve the right to choose their representatives, NOT vice versa.

Even if the author is correct, and I certainly cannot dispute his findings, there MUST be reform to redistricting throughout the country.

My blog contains the most detailed record — available on any news or blog on the internet — of Arizona’s INDEPENDENT redistricting as it played out for the 2011 cycle.

Demo Rep November 16, 2012 at 2:37 pm

Nonstop ANTI-Democracy gerrymanders in the U.S.A. since 4 July 1776 — i.e. Day ONE — carried over from the gerrymanders in England in 1200s-1700s.

Democracy NOW – to END the EVIL oligarchs in control since 1776.

Total Votes / Total Seats = EQUAL votes needed for each seat winner.
Excess votes down. Loser votes up.

At stake – the survival of the human race.

Policyguy November 16, 2012 at 4:04 pm

One problem with comparing the 2012 results with what might have happened if the 2002 maps were still in effect is that that presumes the 2002 maps were also fair and reasonable. In 2002 in Pennsylvania, the Republicans controlled the process as well and one would argue that those maps were equally skewed. That said, the state election in Pennsylvania might also be illustrative. The courts ordered the state to use the 2002 maps, rejecting the plan developed for the state House and Senate. The Senate Democrats won 3 seats this election. The general consensus is that had the newer maps been used, they would have lost 2 seats. The salvation here is that maybe demographic changes and realignment of party affiliation, which have been occurring, eventually catch up to a gerrymadered districts.

Nicholas Goedert November 16, 2012 at 4:34 pm

Just to clarify, my analysis in this post does not compare the 2012 results to what would have happened under the 2002 maps. Rather, the benchmark is a nationwide historical average over the last 40 years of congressional election results, with the fit estimated using a probit curve.

Couldyou Repeatthat November 17, 2012 at 6:59 am

My first question is, who is this article for? Who is your audience? If it’s for people like the people who wrote it? If so, well, isn’t that a bit like preaching to the chior? My next question is, who would you like your audience to be? I’m going to go out on a limb here and guess that the people you are trying to enlighten are the people who could most benefit from this information. People like me, your averarge Joeanne, who votes. Well, here’s some advice from people like me. Your languang is too heady. Yes, the difference between the right world and the wrong word used can be as important as the difference between a lightening strike and static cling. Yeah, I get that. However, well, you aready know what I’m going to say. Rewrite the article. Make the language less “specialized”. Yes, it’s hard, but you’re smart. Steven Hawkens understood this. Look at what he accomplished. Smiles

Sidney18511 November 17, 2012 at 1:15 pm

I agree with the poster couldyourepeatthat. Please rewrite in simpler language. I had, and am still having a hard time understanding what your message is.

John Sides November 17, 2012 at 1:34 pm

Simply put, the reason that Democrats win more votes for the House of Representatives than they do actual seats in the House is partly due to redistricting: Republicans have drawn district lines that favor the GOP. But it is partly due to other factors as well, and in particular the fact that Democrats are more geographically concentrated in urban areas and so even “neutral” district boundaries might still favor the GOP somewhat.

Glen Tomkins November 18, 2012 at 2:02 am

The underlying problem for the Ds is that their voters are more geographically concentrated than R voters. This makes them difficult to disperse into districts that closely match the overall D:R ratio of a state, instead “wasting” large numbers of D voters overkilling in districts that often exceed 90% D performance.

Take, for example, NY’s geography, and give it a notional D:R ratio of 60:40. Were it possible to create districts that were all close to that statewide ratio, the Rs would never send anyone to Congress from NY. But because so many of those D voters are concentrated in NYC, it is impossible to create these uniformly 60:40 districts without accepting not just noncontiguity, but noncontiguity with great distances between the pieces parts of which a district is contructed.

I don’t think that this inherent disadvantage can be entirely compensated for, at least not in states with the largest urban areas. But it is only insurmountable in those states with really big urban areas, and in which those big urban areas are far from evenly distributed geographically. But there are many states in which the D concentration disadvantage creates a vulnerability that has to be surmounted, where it can be used by R district-drawers to devastating effect.

The Ds can never match the Rs in gerrymandering because of the concentration problem. As long as gerrymandering in D-controlled states is the only means available to compensate for R gerrymandering, the Ds have to do that. But that’s second-best as a strategy, because the Rs can always profit more from any given level of ruthlessness in gerrymandering. The best D strategy would be to get national control of districting, and give the task to an algorithm that prioritizes districts that have a partisan split close to that of their state. The algorithm would have to also give enough weight to compactness of district that we don’t end up with geographic monstrosities, and so there would be some anti-D bias. But that concession to an inherent disadvantage would be more than be made up for by the safety play of preventing intentional exploitation of that weakness.

Couldyou Repeatthat November 18, 2012 at 5:07 am

Thanks Glen! Your explanation of Dr. Sides’ article was very helpful. To Dr. Sides I would just like to say, thank you for your article. It’s people like you who take the time to write these kinds of articles to help people like me by exposing subtle unknowns and, thereby, helping everyone to see the entire picture . Even if we, the average American, can’t understand everything in the articles there are people, like Glen, who do, and will act as translators and even offer solutions to the problems. Thanks, again Glen and Dr. Sides.

Couldyou Repeatthat November 18, 2012 at 5:30 am

Opps, Sorry Dr. Goedert

Nichol November 18, 2012 at 8:15 am

Check out the german system. They have a mixed system with district representatives, that is subsequently corrected for state-wide proportional votes .. by adding a certain fraction of seats from party lists. This guarantees that the seats in their parliament divided proportional to the popular vote in the state. It also gives a chance for spread-out minorities to get somebody to represent their voice.

C. November 18, 2012 at 3:57 pm

Stand your ground Rep., find a solution for the fiscal cliff, but make it as difficult as possible for Mr. Socialist I mean Mr. The entitlement, oops mr. obama

Gillian November 20, 2012 at 12:35 am

Journo Annabel Crabb on US elections and Australia’s Electoral Commission….

Dear Australian Electoral Commission,

This is just a little note to thank you for being the way you are. And an apology for taking you for granted all this time. Like many millions of Australians, I’m guilty of thinking you a little fusty, with your anally retentive rules and your little cardboard demountable booths.

On occasion, I have rolled my eyes at you.

Like that time in, I think, 2004 when one of your electoral officials told Tony Abbott he had to remove his Liberal Party T-shirt before voting (no campaign material within six metres of a polling place). I’m all for sticking to the rules, but surely the broader cost to the sighted community on that occasion could have been considered?

Or that time you reprimanded me for not voting in the New South Wales local government elections, when I was living overseas and couldn’t care less who was the mayor of Queanbeyan. But I take it all back, dear old Australian Electoral Commission. The scales have fallen from my eyes. After a recent close inspection of the American voting system, I returned to Australia with a strong urge to kiss your feet.

The United States of America will vote on Tuesday to elect a president. They will also vote to elect members of Congress, senators, sheriffs, judges, members of the county soil and water boards, local lollipop guys and anyone else that county administrations with way too much time on their hands think ought to be elected. Including election officials.

Read more: http://www.smh.com.au/opinion/politics/votes-are-in-and-our-plodding-way-of-having-a-say-wins-in-a-landslide-20121103-28r17.html#ixzz2CjYalX8e

Australian politicians realised a few decades ago that gerrymandering was bad for everyone, and they handed the responsibility over to the placid cardigans in the AEC (Australian Electoral Commission) – comprised of public servants hired on merit, not appointed by politicians or, God forbid, voted in!

Thanks for your interesting article… the machinations of the US political system reveal the hollowness of American-style so called democracy.

Sam Wang December 11, 2012 at 2:07 am

There are some potential weaknesses in this analysis. In my view, the headline of this article is likely to be incorrect.

For example, in Tables 3 and 4, the tendency for the last column to show negative percentages could be explained by a bad seats/votes curve. This is the case because most Democratic vote share percentages in Table 3 are >50% (57%, 57%, 67%), whereas 2 out of 3 in Table 4 are <50% (47%, 43%).

It would be better to sample from all current districts to get a general relationship, along with confidence bands. It should be possible to quantify the extent to which asymmetric R vs. D districting strategies account for most of the antidemocratic outcome. I'll analyze that in the near future.

Nick Goedert December 11, 2012 at 4:32 am

I’m not sure I understand a few of your points. What do you mean by a “bad” seats/votes curve? Are you saying that the curve used to generate the “expected seats” is too steep? If you flattened this curve, it would indeed reduce the appearance of bias. The curve used here is just the slope of the best-fit for the national seats/votes relationship for congressional elections from 1972-2010. The intercept is assumed to be unbiased toward either party (i.e. 50% of the vote should win 50% of seats), as allowing the curve to be biased would pretty much defeat the point of the exercise.

Also, I don’t think I am claiming that the parties have asymmetric districting strategies. I think their strategies are pretty much the same….it’s just that the same strategy is much more effective when applied by Republicans because the population distribution is asymmetric.

But I’m eager to see your results. Certainly this is far from the most sophisticated way to weigh these effects; it is barely more than back-of-the-envelope math written up a few days after the election.

But I think it is strongly suggestive….it’s difficult to see how Democrats could have gerrymandered any state as effectively as Republicans gerrymandered PA or NC (at least under circumstances of a tied national election). And even ostensibly “nonpartisan” maps in NJ and FL appear as though they pretty much applied the strategy of a Republican map.

DavidT December 11, 2012 at 2:23 am

“McGhee’s post compares the results of the 2012 elections to what the election might have looked like using the 2002-2010 maps, and finds that the most recent round of redistricting had relatively minimal effects. ”

That particular approach proves very little, because the 2002-2010 districts were themselves often gerrymandered, far more often by Republicans (Ohio, Pennsylvania, Michigan, Florida, Texas) than by Democrats (even in California Pelosi decided on a “protect all incumbents” gerrymander instead of one that could expand the Democratic edge). Admittedly some of the gerrymanders did not work in Democratic “wave” years–e.g., 2006 and 2008 in PA, 2008 in OH–but on the whole they held up pretty well.

Joe Noonan December 11, 2012 at 7:20 am

Gee, I didn’t see the Democrats whining about this when they controlled Congress for 40 years. As stated, it’s simple. Dems live in high concentrations in a smaller number of districts and Rs live in lower concentrations in a larger number of districts. When the Dems retook Congress in 2006, it was with the help of conservative Blue Dog Democrats which the party ostracized because well, they were conservatives. Perhaps if the party wasn’t so hostile to right leaners in their own party, they’d be in control of congress….

Patrick B December 13, 2012 at 2:03 pm

This is some great work. It also explains some conflicts I saw in my own data analysis and in some that I read. There may be a long term tilt towards gerrymandering, but it’s primarily districting overall. There are a lot of ways to draw equally populated districts in and around cities, at this point districts in states like PA and OH are drawn to severely minimize them. That is why on average it take 2.5 Dem votes to overcome a single GOP vote for the house.

http://www.motherjones.com/politics/2012/11/republicans-gerrymandering-house-representatives-election-chart

http://ballotlines.com/2012/12/04/its-not-what-you-think-the-structural-causes-of-our-partisan-politics-part-2/

MOPstr February 7, 2013 at 7:14 am

This article seems be premised upon a matter of semantics. It provides no definition for gerrymandering but assumes it is different than redistricting. Any district drawn to create partisan majorities that do not reflect the average division between the parties in a state is a gerrymander that can skew representation. Sometimes such districts, such as in urban areas, are geographically compact. In others they look like the ungainly shape that gave gerrymander its name. An algorithm that would produce the most geographically compact district that would also divide partisan voters according to the statewide average is what is required to avoid the effects of deliberately increasing the margin of victory in some districts for the purpose of partisan skewing of representation.

YM March 21, 2013 at 3:13 pm

Using an algorithm to produce “the most geographically compact district that would also divide partisan voters according to the statewide average” is just another form of gerrymandering, IMHO. Perhaps the real problem is that in many areas, too many people vote Democrat; also if you divide the national vote between liberal and conservative, then some of the Democratic majority falls away, because the Democratic candidates were actually conservatives, not liberals.

DigitalDao February 21, 2013 at 2:00 pm

Dr. Goedert,

How much variation is there around the predictions generated by your probit model?

Comments on this entry are closed.

Previous post:

Next post: