Editore"s Note
Tilting at Windmills

Email Newsletter icon, E-mail Newsletter icon, Email List icon, E-mail List icon Sign up for Free News & Updates

February 11, 2008
By: Kevin Drum

CONVENTIONAL WISDOM WATCH....Brendan Nyhan crunches some numbers to test the conventional wisdom about Barack Obama's support and concludes that some of it is true and some isn't. He looks at five pieces of CW: (1) Obama does better in caucus states, (2) he does better in states with either few blacks or lots of blacks, (3) he does worse in states with lots of Hispanics, (4) he does worse in big states, and (5) he does worse in heavily Democratic states. He concludes that only (1) and (2) are true:

It's hard to separate the associations between these variables because larger states are (on average) more black and Hispanic, more Democratic, and less likely to have caucuses. But when we put all these factors together in a linear regression (including both black population and black population squared), we find that the U-shaped quadratic relationship for black population and the positive relationship for caucuses are statistically significant, while the other factors are not. In other words, the evidence so far is consistent with the conventional wisdom that Obama does best in heavily black and heavily white states and in caucuses and he does less well in moderately black states and primaries.

I'd add a caveat to this. Brendan actually finds that all five pieces of CW are true, but that the last three aren't statistically significant. In other words, there's at least a 5% possibility that they might be the result of chance.

But this is a one shot deal, and I wonder if the results are significant at, say, a 90% level? In an academic setting this wouldn't be good enough, but in a real-life setting where this is the only data you have (no followup studies, folks!), most people would probably think that 90% certainty was fairly convincing. For better or worse, it looks to me like the CW is likely true on all five counts.

UPDATE: In a demonstration of the blogosphere at work, Brendan responds almost instantly:

To answer the question, the other variables aren't close to being significant. However, I wouldn't put too much stock in the results of any of these hypothesis tests because (a) hypothesis testing is riddled with epistemological problems and (b) it's difficult to achieve significance in small samples.

So don't pay any attention to any of this stuff. But at least there are some nifty charts for you to go look at.

Kevin Drum 11:59 AM Permalink | Trackbacks | Comments (40)

Bookmark and Share
 
Comments

Brendan Nyhan? The grad student that built his reputation on "the democrats are even worse?"

Back to work....

Posted by: jerry on February 11, 2008 at 12:07 PM | PERMALINK

Obama does better

Isn't this the wrong verb tense? Shouldn't it be, "Obama did better"? It seems to me, looking at this weekend especially, that over time these CW points have lessened or disappeared. Further, certainly, Brendan and Kevin are not suggesting they can predict the future?

Posted by: Patrick on February 11, 2008 at 12:08 PM | PERMALINK

McCain: Like hope, only different

http://talkingpointsmemo.com/archives/177927.php

Posted by: Margaret on February 11, 2008 at 12:10 PM | PERMALINK

The regression almost certainly assumes that his popularity is unchanging with time. There may be a significant effect based upon time of primary/caucus. I think you are right about a 95% confidence level being too high, the other results could be considered suggestive, but unproven.

I suspect exit polls, or district by district results would do a better job of isolating the factors.

Posted by: bigTom on February 11, 2008 at 12:13 PM | PERMALINK

He's already updated his blog. The results aren't even close to significant.

Posted by: Dave Munger on February 11, 2008 at 12:15 PM | PERMALINK

Let's avoid the formal statistics stuff in this blog please Kevin. It's been 26 years since I took my two semesters of stat and I don't remember any of it. So let's keep it qualitative, shall we?

Posted by: Tom in Houston on February 11, 2008 at 12:17 PM | PERMALINK

Actually, Obama does better across the board, and all 5 are wrong.

Obama does better in caucus states,

The how come he won South Carolina, Alabama, Connecticut, Delaware, Illinois, Missouri, Utah, and Georgia even though they are all primaries?

he does better in states with either few blacks or lots of blacks

What do you mean by few or lots? That sounds pretty racially charged to me. You're beginning to sound like Bill Clinton talking about Jesse Jackson. And you call yourself an Obama supporter?

Posted by: Al on February 11, 2008 at 12:17 PM | PERMALINK

Ok Kevin, if you actually read your blog, take a look at this -http://fish.blogs.nytimes.com/

Are you flexible enough to take a 180 degree look at Hillary bashing?

Posted by: james b on February 11, 2008 at 12:20 PM | PERMALINK

The notion of a CW in such a fluid campaign is flawed in and of itself. Any type of direct analysis like this is backwards-looking. What are the trend lines? Also, what if Obama wins VA by 10-15 points tomorrow? Will there be a CW (6) stating that Obama does well in states that have former black governors?

Posted by: MikeO on February 11, 2008 at 12:23 PM | PERMALINK

Sounds as if nobody really has a clue why Obama does better in some states than in others.

Sounds as if the whole study does nothing but confirm my private theory that at some point, once you get past basic things like fund raising, name recognition, and the general state of the economy, whoever becomes President is mostly determined by random factors.

We're just reading patterns into cloud formations. Yonder cloud, methinks it is backed like a camel.

Posted by: nemo on February 11, 2008 at 12:23 PM | PERMALINK

As Mr. Drum says, this is a one-shot deal. Thus a piece of conventional wisdom, discovered post facto, may be true on average but still have no value -- I might as well observe that Mr. Obama wins in most states beginning with the letter "M". Thus Mr. Nyhan (and commenter "nemo", above) are right to reject insignificant explanations as useless.

Posted by: sammler on February 11, 2008 at 12:27 PM | PERMALINK

@nemo: You will never make any money in political consulting. You have to keep the candidates thinking there is meaning in the numbers, so that they hire you to process the data and make sense of it.

Posted by: Christopher on February 11, 2008 at 12:31 PM | PERMALINK

so... the point is all those things seem to be true, may in fact be true, but can't be proven to be true at the moment - think tobacco industry 30 years ago re cigarettes and cancer. Thanks.

Posted by: guff on February 11, 2008 at 12:31 PM | PERMALINK

I am bored with the Barack Obama conventional wisdom watch. I would much rather talk about Maine.

I think the Maine caucuses yesterday are more telling than most observers seem to believe. Maine is a North Eastern state, just the kind of state where Hillary has done well in the past few weeks. Both Obama and Hillary spent time in Maine. She drew reasonably large crowds. You would think that Hillary might do well in Maine. Still Obama blew her away. Why? Any ideas?

Is her poor showing the result of the caucus factor? Is her poor showing the result Maine's population being primarily white men?

What about the possibility that are moving from Clinton to Obama because they like Obama better? Just a thought.

Posted by: corpus juris on February 11, 2008 at 12:34 PM | PERMALINK

off topic, but not really -

it really turns my stomach when I hear one of the wingnuts refer to Obama as "Barack Hussein".

I'm wondering if Barack might should come right out and refer to himself that way occasionally - and nip the wingnuts whistle call to the muslim haters right in the bud... snip their balls off at the same time -

might work -

Posted by: christAlmighty on February 11, 2008 at 12:34 PM | PERMALINK

"of" between "result" and "Maine's" and "voters" between "that" and "are." Man, I should never post on Monday when I am working.

Posted by: corpus juris on February 11, 2008 at 12:40 PM | PERMALINK

"Is her poor showing the result Maine's population being primarily white men?"

The white women of Maine might object to their exclsuion from your rationale.


Male
645,154
48.8
49.2%
+/-1,259


Female
676,420
51.2
50.8%
+/-1,259

White
1,265,541
95.8
73.9%
+/-3,360
rank

Black or African American
13,669
1.0
12.4%
+/-2,692

Posted by: solar on February 11, 2008 at 12:41 PM | PERMALINK

It just gets sillier and sillier.

Posted by: Chris Brown on February 11, 2008 at 12:48 PM | PERMALINK

Kevin, Just a thought, but I think that the states that have *low* AA populations that are going for Obama says more about the composition of the *white* population there. I think the white baptist voters will not vote for him and you don't have as many baptists out west. I suggest the stats guys look at *religious composition* of the white voters and you might see something useful there.

Posted by: Doc at the Radar Station on February 11, 2008 at 12:49 PM | PERMALINK

I wonder if the results are significant at, say, a 90% level? In an academic setting this wouldn't be good enough...

Just for the record, the U.S. Census Bureau tests for significance at the 90% level in its demographic surveys. Not just here and there, either, but across the board.

Posted by: low-tech cyclist on February 11, 2008 at 12:56 PM | PERMALINK

Kevin -

It might be worth taking a look at Poblano's demographic analysis. He finds some interesting ways to analyze results (e.g. looking at percentage of Southern Baptists in the population of a given state). He finds that Hillary's advantage lies with non-college educated voters (regardless of income) and naturalized citizens (regardless of ethnicity), which is an important distinction. It's not that Latinos are voting for Hillary more, it's that people without college education and recent immigrants vote for her more.

Posted by: handsofaten on February 11, 2008 at 12:56 PM | PERMALINK

I haven't seen any discussion on the possibility of voting machine errors, whether by incompetence, design, or just because of the nature of the beast.

Posted by: Sandy on February 11, 2008 at 1:26 PM | PERMALINK

Re previous comments. The primaries tend to have higher turn-outs and the use of machines for voting are more prominent.

Posted by: Sandy on February 11, 2008 at 1:30 PM | PERMALINK

In general, it's very, very hard to find statistical significance when the numbers are so very small -- even very critical trends can fail statistical significance when there are so very few samples.

And this makes anything that is statistically significant all the more impressive, because the underlying trend it reports must be very gross in nature. So if Obama does much better in caucuses than in elections to the point of statistical significance, that's got to be a very key trend.

Of course, Nyhan's approach to the other questions appears also to be vitiated by the fact that he gloms together elections and primaries when looks at those other questions. The same weirdnesses that make caucuses particularly favorable to Obama may account for unobserved regularities when they are mashed together in a larger reference group, but which might pop out if they were separated into subgroups (of course the numbers would be even smaller in that case, making statistical significance harder to achieve).

Posted by: frankly0 on February 11, 2008 at 1:31 PM | PERMALINK

Well, that's something, data and nifty charts. It is statistical, Obama does better in caucus states and in states with either many or few African Americans. But still no explanation.

Maybe we should reframe the question. Why is Clinton doing so badly in caucus states? Or why do people without a college education and recent immigrants vote for Clinton more? What about Clinton appeals to that demographic? It isn't the details of her health care proposal--presumably mandates v mandates is not going to be a big issue for them. Is it race? Name recognition?


Posted by: PTate in MN on February 11, 2008 at 1:31 PM | PERMALINK

Is her poor showing the result of the caucus factor? Is her poor showing the result Maine's population being primarily white men? What about the possibility that are moving from Clinton to Obama because they like Obama better?

Yeah, I'm sure that the big difference was that, suddenly, right after Super Tuesday, which was a draw, they just liked Obama so much more up in Maine than one would expect; somehow, magically, everything changed.

I thought I was the one who couldn't get past some bias?

Look, the caucus effect is obviously very, very real. No other explanation is necessary, I should think.

Why not draw your conclusions rationally instead of emotionally?

Posted by: frankly0 on February 11, 2008 at 1:53 PM | PERMALINK

Can we stop chalking everything up to demographics?

Obama has done well where he has been able to campaign effectively. Small states are places where a few stops and having an organization on the ground are both feasible and can move the whole result significantly. Big states like California, that's certainly not a possibility for a challenger.

Clearly Obama has done well in places like Washington state with many asian voters and in New Mexico where he must have done well among Hispanics to pull a draw.

Not everything is demographics.

I'm not accusing you Kevin, I'm criticizing the Conventional Wisdom about these things.

Posted by: david in norcal on February 11, 2008 at 1:57 PM | PERMALINK

It's pretty hard to look at the caucus effect and not conclude that Obama is riding high mainly because of what amounts, for all practical purposes, to a voter suppression technique called caucuses.

Everybody knows that the gold standard for democracy in elections is the secret vote in standard polling booths. Obama is systematically profiting from scenarios where that is NOT what's going on, but rather some process highly skewed toward his voters. Again, if this were to happen in an ordinary general election, everyone would be screaming "voter suppression".

Sorry, Obama supporters, but it's the simple truth. Democracy in its purest form has not been so kind to Obama.

Posted by: frankly0 on February 11, 2008 at 2:00 PM | PERMALINK

You know, Nyhan *could* have done a multivariate regression, which would have dealt with frankly0's point that the other results may be due to the caucus effect...

One thing that you can do is eyeball the plots that Nyhan made, looking to see if the orange dots (caucuses) and green dots (primaries) show the other effects independently. If you do this, it looks pretty clear that Obama's good performance in states with low population of African-Americans is mostly due to caucuses; he only won UT among the primaries in states that are less than 10% black. If you just look at primary states, it looks more like a simple trend, rather than a "U".

With a proper regression I suspect that only the caucus result (and possibly a simple trend for proportion of black population) would prove to be statistically significant, even with a fairly modest confidence level.

Posted by: Alex R on February 11, 2008 at 2:08 PM | PERMALINK

Frankly0

There is a difference between drawing conclusions and asking questions. I just asked questions. For example I have learned that there are more white women than white men in Maine and that there aren't many blacks. Those white men for Obama must really be manly men to suppress the pro-Clinton white woman vote so effectively. I guess what I am saying is I am not so sure the caucus effect is really an effect. It is more likely a coincidence.

Of course, it could be that lots of folks go to caucuses planning to vote for Clinton but when the Obama man in black wearing dark Raybans blinks that funny light they just have to caucus for Obama.

It just gets sillier and sillier.

I am currently happy with either Obama or Clinton. I just want to beat John McCain. I think that Obama might be the better candidate to do that. Give us reasons why Clinton is a better McCain beater.

Posted by: corpus juris on February 11, 2008 at 2:52 PM | PERMALINK

Let me add one more observation about the so called caucus effect. It is based on the assumption that Clinton people are either too stupid or lack motivation to find the neighborhood caucus location.

How do you explain the fact that voter turnout to both Democratic caucuses and Democratic primaries are up big every where? The reports I saw indicate that turn out to the Kansas caucuses were stunning. I think that Missouri primary turnout was equally high.

Posted by: corpus juris on February 11, 2008 at 3:03 PM | PERMALINK

(2) he does better in states with either few blacks or lots of blacks,

Matt Yglesias had the most ridiculous comment the other night. He wrote that Obama doesn't do well in states where racial competition (or something like that) is a big part of local politics, and he held up NJ as one of his examples. As if in all states where people are really integrated, we're all somehow at each other's throats or something!

I'm from NJ, and I can tell you, I have no clue what he was talking about. Does he think the came thing about NY? I think the better explanation is that some states that have a notorious history of slavery and racial prejudice, as well as especially large black populations (South Carolina, Georgia) the black people there are really excited about electing a black president, so the black Democrats turn out in large numbers. In very racist, mostly-white states, on the other hand (Iowa, Idaho), lots of white raiders turn out to vote for Barack or encourage regular Dems to come out for Barack because they want him instead of Hillary. Barack may already lead in those states butthe raiders inflate his totals noticably, since people through their scruples about politics out the window ever since a few years ago.

Posted by: Swan on February 11, 2008 at 3:26 PM | PERMALINK

(1) Obama does better in caucus states, (2) he does better in states with either few blacks or lots of blacks

The problem is that two of these three groups (caucus states, few blacks, lots of blacks) are for the most part the same thing.

Obama does better in caucus states, which *are* the states that have few blacks.

The problem is that the caucus states aren't delivering the voice of all voters. If the caucus states ran primaries, there's little reason to believe that Obama would hold such commanding leads.

So the correct assessment is 1) Obama does well in caucus states; 2) Obama does well in states with lots of black voters.

There's no reason to think he will do well in states that have few black voters, as we can't consider the caucuses meaningful indicators.

Posted by: Cal on February 11, 2008 at 4:40 PM | PERMALINK

I'm waiting to see what happens after March. If Sen. Obama posts wins with decided margins (5% to 10%) then the caucus theory will be disproven. Wins by a smaller margin may simply be attributed to "momentum" and increased campaigning by the candidate. Whatever occurs, it will be interesting.

Posted by: Doug on February 11, 2008 at 5:39 PM | PERMALINK

I agree. If Obama posts big wins in Texas, Ohio,and Pennsylvania, then he's made his case. I'm not at all convinced that happens, though.

Posted by: Cal on February 11, 2008 at 5:44 PM | PERMALINK

I recommend Poblano's work as well. We link to it at Economists for Obama

Posted by: lerxst on February 11, 2008 at 5:59 PM | PERMALINK

Wow...Economists for Obama must not have any econometricians! And there are so many GOOD economists for Obama out there!

I'm pretty sure that Poblano would NOT get a good grade in any stats class. This reminds me of the overfitted regressions that you always see for the freshman intro stats class: "I can perfectly predict the stock market!"

Posted by: economists? on February 11, 2008 at 7:40 PM | PERMALINK

It's pretty hard to look at the caucus effect and not conclude that Obama is riding high mainly because of what amounts, for all practical purposes, to a voter suppression technique called caucuses.

And it's also pretty hard not to conclude that Clinton supporters are grasping at whatever straws are within reach to explain why their candidate has blown Obama away.

Voter suppression is what the Republicans did in Florida and Ohio (and other places).

The caucases were not set up to give Obama an advantage. They've been in place for a long time and all the candidates knew the rules when they entered the race. With Obama's background in community organizing, it makes sense that he can put together a very effective organization for those states.

If Clinton chose not to contest those states, or did not put together an equally effective organization in them, then how does that turn into voter suppression?

And the fact that these caucases have consistently had record turnouts, I'd say that Obama has been effective at democratizing the caucases, not voter suppression.

Posted by: tomeck on February 11, 2008 at 9:10 PM | PERMALINK

"The caucuses were not set up to give Obama an advantage."

This is obviously true. Caucuses were setup however to favor candidates similar to Obama. Caucuses were initially set up to allow challenges to the candidate(s) favored by the party establishment. Caucuses amplify the influence of activists since they feature a small window in which to arrive at the site but then may take a considerable amount of time before the "voting" is completed. The conventional wisdom is that caucuses gauge the intensity of support rather than the depth. Caucuses also generally have much lower turnout than primaries increasing the significance of caucuses goers over primary voters. Finally, employers are not required to allow time off for caucusing unlike voting.

It is unfortunate that the only reason anyone is discussing caucuses is because one side is crushing the other in caucus states. Caucuses are certainly more fair than party insiders selecting delegates but hopefully once the election is over, the DNC and state parties will revisit the system.
I would prefer to have caucuses in all of the early states and then switch to primaries for the remainder. This would allow candidates to establish themselves early without breaking the bank before switching to contests that mirror the general election later.

Posted by: activists & organization on February 12, 2008 at 4:15 AM | PERMALINK

There are too many explanatory variables here relative to the number of observations. This regression is "overfitted," in a way that often leads to apparently strong, but spurious, statistical results. No one should have much confidence in it.

Posted by: Matt on February 12, 2008 at 8:50 AM | PERMALINK




 

 

Read Jonathan Rowe remembrance and articles
Email Newsletter icon, E-mail Newsletter icon, Email List icon, E-mail List icon Sign up for Free News & Updates

Advertise in WM



buy from Amazon and
support the Monthly