Bell Curve The Law Talking Guy Raised by Republicans U.S. West
Well, he's kind of had it in for me ever since I accidentally ran over his dog. Actually, replace "accidentally" with "repeatedly," and replace "dog" with "son."

Saturday, October 10, 2009

Why the Census and Redistricting Really Matter

There's an excellent post by Tom Schaller on that contains a number of nuggets of neat information. The post is about the "generic congressional ballot" (GCB) which is a nationwide aggregate survey of party preference in congressional elections. The question is simply "will you vote D or R?" We are so far away from the 2010 elections that only the broadest trends of red/blue have any vague predictive value now. My personal take on election polling is that polls more than six months out are as useful as predictions of the weather - on a particular day - six months out. Right now the GCB is trending more Republican than it has since 2005. But, as I like to point out, unemployment is at 10% and the Democrats are slogging through very hard agenda items in Congress. Next summer, with a year of legislative accomplishments behind them and a recovering economy, I would expect the GCB to tilt Democratic again.

Most politicos are waking up to the fact that the conventional wisdom that the president's party loses seats in the midterm elections didn't hold true in 1998 or 2002. In fact, other than big-change elections in 2006 and 1994, there has been no such "natural" drift in midterms for almost 20 years. Many things affect what may be a change, but here's the basic analysis: we no longer have coattail driven presidential politics, where the president's personal popularity sweeps in a set of reps who would never have won otherwise, and who therefore lose in midterms. We have presidents and parties much more politically aligned and polarized than in previous years. Thus the presidential popularity and vote share more closely match congressional vote shares. The 1994 and 2006 elections were examples of the president's unpopularity being reflected in Congress.

What is fascinating is the comment by a noted pollster that, in this decade, he simply subtracts 2 from the Democratic column in the GCB because Democrats are concentrated in fewer districts. Because of the GOP advantage in statehouses in 2000-2001 redistricting period (and the unbelievable mid-term redistricting in Texas later), Democratic-voting districts are more democratic than Republican districts. So much is the tilt that pollsters simply discount some Democratic strength in polling. Remember that the main use of gerrymandering is to arrange a situation where your party has 51% majorities in most districts, leaving the opposition to have supermajorities in a few districts.

Now realize that - despite this - the Democrats have surged to big electoral majorities in the House in 2006 and 2008.

So in 2010, we will see another round of redistricting. Odds are that, unlike 2000, the Democrats will control the process in Ohio, Iowa, California (somewhat), Colorado, and New York, and will have acquired some veto power over the process (e.g., control of one legislative house or governor's seat) in Virginia, Montana, and Nevada. This will make the House trend more Democratic in the next decade. The big question is what happens in Texas. Texas will get up to 4 new seats depending on how the math is figured. 3 is likely. Right now, the Texas lower house is split almost 50/50 for the first time in years. Dems have been reorganizing and gaining strength there as internal migration and demographic change in Texas has brought in non-Texans from around the USA and minority groups, including the Katrina refugees. If the Democrats can take over a legislative chamber in Texas in 2010, this will be very huge.

Note that the Senate is not gerrymandered. So with advances in gerrymandering technology (data and computers) the Senate has become almost more responsive to changes in national politics than the House.


Raised By Republicans said...

I think you are right about the ideological (and geographical) polarization in the USA now. But I disagree (mildly) with your assertion that there are no more coattail effects. A reasonable person could argue that the 2006 results were related to Bush's tremendous unpopularity. Similarly, Obama was both enormously popular and black. Both tend to increase turnout for Democrats and both will not be in play in 2010 because Obama won't be on the ballot. While he is still popular, I doubt that will increase turnout in marginal districts.

I don't have the numbers at my finger tips but I'm guessing that a lot seat gains in 2006 and 2008 came in the South, the West and/or rural areas usually represented by Republicans for the last 20 years or so. If I'm right in that guess, then these Democrats may be at risk in a low turnout election.

The Law Talking Guy said...

The term "coattail" means the winning presidential candidate sweeping in additional Congressional seats in a presidential election year particularly in swing districts. It doesn't really refer to the president's unpopularity spreading to the rest of his party in midterm elections. So 2006 is not an example of coattails.

The argument you are making, I think, is that Obama did have coattails in 2008 because of high turnout, particularly African-American turnout. I'm not sure we saw a whole lot of that, actually. In many Senate races, the Democrat was elected with a smaller, often much smaller, margin that Obama. Oregon and Minnesota are examples. North Carolina (Hagan), by contrast is likely an example of coattails. I don't know if anyone has stats on this.

Apparently in house races, something like a coattail effect may be evident. At least according to we see that Obama got 53% of the two-party vote versus Democratic congressinoal candidates getting 56% of the two-party vote. And in a good many states Dems outperformed Obama.

But do we call it coattails if Democrats in Congress won races with the votes of people who wouldn't vote for Obama for president? The typical coattail effect would show the President shifting districts from the opposite party.

Unbelievably, the suggestion from much of the 2008 data is that it was a such a huge anti-Bush, anti-Republican voting year where Obama's race was a bit of a handicap from an even bigger victory.

If that is true, and I think it may be, the lack of Obama on the ticket is not that material to 2010. It's the lack of Bush/Cheney/McCain on the ticket that will make the difference.

Raised By Republicans said...

I don't think comparing the winning margin of the Democrat in question Obama's winning margin is the right comparison to identify coattails for the purposes of this discussion (i.e. whether 2010 will see a decline in Democratic voters - particularly in marginal districts). I think what we should be looking at is whether the Democrat in question did unusually well in 2008 for a Democrat in that district and whether we can see any evidence of a surge in black turnout being the cause of it.

If a spike in African American turnout in marginal areas wasn't decisive then I'll concede the coattail issue.

I certainly hope you are right that Republicans are still screwed in 2010. I've seen reports that they appear completely screwed in California for 2010 at least. But how representative is California of the rest of the country?

The Law Talking Guy said...

California is certainly not representative of the country. For one thing, white people are barely a majority. For another, unemployment is 12%, well above the national average.

But that's not the question. The question is whether there will be an anti-incumbent mood in 2010. I seriously, seriously doubt it. Now, if there were some Dems who won with a big black turnout, they may lose, but those will be marginal losses. I suspect, however, that this analysis fails too because the African-American turnout was not in the Blue Dog disticts, who are the really vulnerable candidates. I suspect the black vote pumped up urban candidates mostly, and - in the racially geerrymandered rural south - did not have a big electoral effect.

Let me put it this way. The only two times in the since 1990 when the incumbent president's party lost in midterm elections were 1994 and 2006 when the president was massively, monstrously unpopular as was his party. This is unlikely to be the case in 2010.

Raised By Republicans said...

I agree with you about the possible effects of African American turnout. If it was distributed as you suggest, it would have the kinds of effects you suggest.

But why should we believe that this phenomenon has radically changed in 1990? Why do you emphasize the "since 1990" part of your argument? Why not look as far back as we have data for? If we do that the midterm losses for the President's party looks a lot more solid.

If I'm looking at your analysis from a neutral data analysis perspective, my first question would be why cut it off at 1990 when we have data for all the elections?

The Law Talking Guy said...

I emphasize "since 1990" because the data suggest to me that something has changed in American politics in the last 20 years.

Let's take my premise as true: My premise is that while we've been saying that the president's party loses seats in midterm elections since I was in college in before, it hasn't really proven true for 20 years. Put another way, I am saying that it seems that we aren't seeing the same temporary pro-president coattails that disappear at midterms that we used to. Rather, we are seeing more consistent ideological voting for a president and his party, with the president's popularity and congressional voteshare moving in tandem.

Is it possible that something has changed about the electorate? Simply adding in past electoral data won't help answer the question.

Here are three suggestions. First, studies of party ideologies show a significant increase in ideological polarization in Congress during this same period. Second, the science of gerrymandering has improved quite a bit. Three, information and media has changed dramatically over the past 20 years. In 1990, there were three major TV networks that provided news, and CNN was a newcomer. Today, the share of TV news has shrunk dramatically, and the biggest TV news source is Fox News, which has a stalwart ideological bent and didn't exist before.

All of these could be explanations.

Another possiblity is that coattails are an artifact of lopsided presidential wins that aren't likely to happen anymore now that both parties are acting more rationally in nationwide positioning, so we are coming closer to the 50%+1 coalitions.

Raised By Republicans said...

Based on your suggestions we should not simply truncate the analysis at 1990 but rather look at all midterm elections going back as far as we can and compare them to your hypotheses: ideological polarization, gerrymandering, and the telecommunications revolution. Some of those are clearly restricted to recent years but ideological polarization probably isn't and the accuracy of gerrymandering might have been very high when districts were smaller, had few inhabitants, and had demographic make ups that were less complex.

What were the midterm election results in Gilded Age (post reconstruction)? What about the 1840s and 1850s?

The way to test a proposition is to confront it in an atmosphere of potential variation not to assume that the proposition is correct and truncate the scope of the analysis accordingly.

The Law Talking Guy said...

I think you're getting lost here and have reverted to a basic explanation of how you isolate variables. Okay, but it's really not on topic. I'm not sure why you think that I intend to discard previous elections in my analysis. You are absolutely correct that we have to compare present conditions with prior conditions in prior elections. That's what you have to do. And I was doing that whe I suggested that ideological polarization has increased (I left out the phrase "relative to pre-1990 period" but that was implied bythe comparative term "increase").

I was pointing out that the conventional wisdom (based on 150 years or so of elections) is that the party in power loses in midterms. Although if I remember correctly, that analysis is largely restricted to the post-civil war period and even to the post-17th-amendment period. Anyway, that's the basic fact: if you exclude post-1990 elections, you get that conclusion about losing in midterms. So we kind of know that. At least we know it well enough to start looking for causes. If you include the past 20 years, the situation looks less clear.

Is this a change? Well the likely answer is yes. Thousands of variables affect elections, and studies usually focus on a few and ignore the rest. And new variables arise that you didn't know to account for. For example, the "existence of internet" variable wouldn't have been imagined before 1996. And the "existence of ultrascientific gerrymandering" would not necessarily have been a variable beforehand. One of the reasons why the analyses don't go back before the Civil War is that there were so many differences in context that it just becomes too hinky to deal with it.

I think you'll find, by the way, that Americanists agree that ideological polarization is at its highest point today in a very, very long time.

I also think you'll find that gerrymandering was relatively crude, ad hoc, and based largely on vote counts and anecdotal evidence until late, late in the 20th century. I mean, that's how state legislatures used to do everything. The ability, funds, and desire to pick through census tracts and use complicated mathematical analysis to define districts is relatively new phenomenon, I will bet.

These are all ways of suggesting why a model that worked well in the past may no longer be predictive of the future.

I suppose I also want to suggest that electoral outcomes of this kind are a different animal than the kind of work you more often do with analyzing systems of government or institutions. You can probably compare the internal workings of the Polish Diet of the 1600s with the Japanese Diet of the 1890s and the European Parliament and make cogent observations. But comparing elections in the 1730s with elections in the 1990s is hazardous work if you are trying to build in variables about public opinion and sentiment. I'm afraid the only conclusions you will be able to draw with any certainty will be so broad and boring (e.g., the party in power loses when masssively unpopular) that they aren't worth it.

Raised By Republicans said...

My point wasn't just that you needed to change wording (that would be silly and nit picky). It was that it is possible that we could look back at other times when there was a lot of ideological polarization and see if there were midterm losses for the President's party then or not. If there were then it would undermine one of your hypothesized mechanisms for why things appear different now.

You're totally right about the ad hoc nature of gerrymandering in the 19th century. But let's imagine what it would have been like when districts were much smaller and less complex. Those crude, ad hoc measures might have been more effective than we assume now. I'm sure though that there was a long period of time - after districts got big and complex and before technological advances in gerrymandering - when gerrymandering was particularly ineffective.

My broader point here is that the hand full of midterm elections that we have seen since 1990 might be a fluke just as easily as be evidence that you're right about this. We can try to tease it out by looking at the specific hypotheses you proposed across a broader scope of elections.

As for the myriad of things that contribute to elections, that just argues in favor of looking at as many observations as possible, not truncating the analysis to the post 1990 era.

Raised By Republicans said...

By the way, I'm not convinced you're not correct about all of this. I'm just not convinced you are either.

The Law Talking Guy said...

I wasn't suggesting truncating the analysis. I was suggesting trying to figure out if the last 20 years are a "fluke" or not by comparing current conditions with previous conditions.

The Law Talking Guy said...

I wonder if there has been work done on 19th century gerrymandering. Bet there has been. It's an interesting topic in a totally geeky wonky way.

Here's my thought. All you really had for data in the 19th century was the list of registered voters in each precinct, their party registration, and whether they voted, IF you had that. No other demographic data (except basically all were white men, of course...) They also had anecdotal evidence about the Irish side of town, the Polish side of town, etc.

I am not sure that having fewer people in a district makes this gerrymandering game easier or harder.

Raised By Republicans said...

I think I understand your point now.

RE: 19th century gerrymandering... A smaller district would mean that anecdotal evidence would be more likely to be representative. It would also mean that politicians might have a very thorough knowledge of the districts in which they were running. Add to this the effect of machine politics or other forms of patron-client politics and I can imagine that gerrymandering could be very effective despite the lack of technology.