Saturday, August 30, 2008

The Barr/Nader Effect Revisited

Earlier this week, we looked at a prediction of how well both Bob Barr and Ralph Nader would do in November based on the Libertarian/Nader vote in 2004 and the state of polling on them both on the state and national levels thus far in 2008. A simple model, but one we can enhance. FHQ commenter and Election-Projection proprietor, Allen, spoke about the 2000 election in response to that post (...posing an altogether different question, but certainly one to look at.). And that got me thinking: What would adding in the data from 2000 do to the regression? It would do a couple of things. First, it provides a more consistent measure -- across two elections -- of the libertarian vote. This is advantageous because it eliminates the possibility that the events unique to 2004 are driving the changes we see. However, the drawback to adding in that data in is that it likely inflates to some extent the vote share Nader would be predicted to receive in November. As I said earlier in the week, though, the goal right now -- especially with the limited amount of polling we have for both third party candidates during this cycle -- is to get an idea about the relative effect each will have across the 14 states that FHQ has as toss ups at the moment in our electoral college projections.

What happens is that we don't see any monumental shake up, but there are some subtle shifts.
[Click to Enlarge]

In looking back at the Libertarian scatterplot from the previous post, there's not much difference in the predicted vote share that Bob Barr would get in November here. [Though there is a bit more dispersion here the focus should be on how high or low the point is.] There are three main groupings of states: Alaska and Indiana in the upper right, a group nine states in the middle, and Florida, Michigan and New Hampshire at the bottom left. To reiterate a point from earlier, the three closest states on the most recent Electoral College Spectrum -- Nevada, Ohio and Virginia -- are close enough that two to three points won by Barr could make a difference. However, if those states are that close, what we see here is likely to have been an exaggeration come November. Swing states across the 2000 and 2004 data typically yielded smaller vote shares to third parties than the less competitive states. Voters are willing to vote in protest if the candidate from their party has already seemingly won or lost the state.

One more thing we can add to this is how Ron Paul did this year in the Republican primaries. These are voters -- his supporters -- who are organized and perhaps inclined to vote for the Libertarian candidate. Ultimately, what this is measuring is the intensity of Paul support across states. A variable controlling for caucus states has been included to deal with contests where Paul did better on the whole than in primary states.
[Click to Enlarge]

Montana, Nevada, North Dakota and Pennsylvania all see modest jumps while the remaining states hold relatively steady when compared with the plot above. That's three caucus states (Montana held a caucus on the Republican side) and one primary state; two McCain toss ups and two Obama toss ups. Again, the same caveats as above apply in the case of a competitive state -- which all of these are. However, Nevada is in a bit of a gray area here. Yes, it did have a caucus, but Nevada was a state where the Paul forces were very well organized. They completely disrupted the state convention in the Silver state and left Nevada without a delegation to next week's convention until just hours ago -- when the state Republican Party named the delegation. In a state that is as close as Nevada, this matters. Whether Barr's numbers are inflated in the state is beside the point. If those Paul supporters turn out and if -- this is a big if -- the opt for Barr, then McCain may have issues turning the tide there.

That's the story on the Libertarian front, but what about the impact Nader is predicted to have later in the fall? As I said at the outset in explaining the inclusion of the 2000 data, Nader would be expected to gain as a result of the inclusion of an election where he outperformed the 2004 numbers we used before.
[Click to Enlarge]

This distribution is also largely similar to the original plot with just the 2004 vote data. Ohio is the only state that really makes a move. Even with that 2000 data, Nader's predicted vote share for the upcoming election is still modest, only just more than 2 points at the most.

In the end though, the message is largely the same as what we saw earlier in the week among these toss up states with regard to the Barr/Nader effect. There is the potential for influence, but the main question is whether close states follow form, not giving third party candidates as large a share of the vote as in other states.

Somewhat tangentially, there's another issue I'd like to raise in this context. Earlier this week when FiveThirtyEight ran the latest CNN state polls, they used the version with the two party vote question as opposed to the four way race data. That has since been changed, but it started something of a discussion over there, and that is a discussion that is relevant here as well. It has implications for our electoral college projections. As I've discussed in this post and in others on the subject, it is likely that the third party percentages in polls are inflated in relation to where vote choices will ultimately be. That being said, is it beneficial to proceed with the four way polls or to go for the two way race version? In one version the third party aspect is supressed and that has an impact on the accuracy of that poll. But the accuracy of the four way polls are questionable as well since those numbers may be skewed here during the late summer weeks. What are people's thoughts on this? I have, to this point, included that four way race data when available.


Recent Posts:
The Links (8/30/08): Sarah Palin/GOP Convention Edition

More Thoughts on Penalties to Prevent Frontloading

If Taking Away Delegates Won't Stop Frontloading, What Will?

11 comments:

Jack said...

My view:

You don't account for other likely distortions (e.g. convention bounces, VP bounces, Zogby Interactive craziness) so it would be arbitrary to say, "the third party candidates are getting x percentage in polls but it's unlikely that will really happen, so we'll pretend they don't exist." We have to take the numbers as they are.

SarahLawrenceScott said...

I agree with Jack. 538 is trying to forecast the election, rather than reflect how things are now. It is a more accurate reflection of the current state of the electorate to include the versions with third parties.

Allen said...

I think my comment about 2000 may have been misunderstood. My hypothesis is the actual votes for 3rd party candidates in the battleground states this year will be low (my belief, that I did not state, is that it will be ~0.5%). To the extent the polls currently indicate larger support for 3rd party candidates in the battleground states, I believe voters will not actually cast their votes that way (i.e., they are telling pollsters one thing, but when they get into the voting booth they will do something different).

I stated that I had not done enough research to really back that up numerically, but that I had looked at the actual Nader vote in FL in 2000 vs 2004, which indicated to me that after 2000, voters "learned their lesson" and decided it was better to vote for "the lesser of two evils".

To really make a numerical case for my hypothesis, you would have to go back and look at polls vs. actuals in the 2000 Battleground states, and look at polls vs. actuals in the 2004 Battleground states. You may, from that data, see a discrepancy in the polls vs. the actuals, esp. in 2004, which would bolster the case for the hypothesis.

I did not however suggest averaging the 2000 and 2004 results! If my hypothesis holds, then I using the 2000 3rd party vote to project the 2008 3rd party vote would be a mistake.

Josh Putnam said...

Allen,
You're right. I should probably add a sentence in there to clarify. I'm not suggesting that that was what you proposed. Instead, your comment made me think about adding those numbers in. Nothing more, nothing less.

And yes, to fully test the hypothesis you're suggesting you'd need to make the comparison between the predicted vs. actual results.

As I said in the post, adding in those 2000 numbers for Nader would inflate his standing some (In fact the range of the predictions increases from about .15 to around 1 point, which is less than I would have thought. And that upper end is skewed by Alaska. It serves not as a pivot point for the regression line, but as an anchor on that upper right corner.), but the regression isn't weighting those numbers in any way. That information is as valuable in that equation as the other variables. The result is an idea of what Nader support may look like given this increased amount of information.

But yes, to get at your hypothesis we'd need those 2000 polls, and I'm still looking for those for other purposes. If I get a full version of that data, I'll make that comparison and post the results.

Josh Putnam said...

Oh, and I agree with what has been said so far on the polling issue. I thought it was worthy of a discussion though. Thanks.

MSS said...

Is nobody polling Cynthia McKinney? I wonder how much of the Nader vote intention is people who think he is the Green nominee. Which he isn't in 2008 (and also was not in 2004).

Would McKinney and Nader simply split this 2% or so, or does McKinney have the potential to draw votes that would not come out for Nader?

I have no idea. I could speculate either way, and in the absence of any polling, I guess speculation is all I have.

Josh Putnam said...

MSS,
She has appeared in a few of the national polls lately. And she was in the CNN state polls that came out this week. I'll have those links up later tonight in the electoral college post. She was getting anywhere from 0-3 points in those four states.

MSS said...

Josh, thanks. I will look for that information. As I indicated above, I'll be very curious to know if those polls suggest she is drawing from the same pool as Nader, or a bigger one.

--Matthew Shugart (a.k.a. MSS)

Josh Putnam said...

Matthew,
I glanced through the polls that Pollster has listed and McKinney has only recently been added (end of July and on). On the national level, she's getting 1% consistently in the handful of polls in which she has appeared. The only state level polls are the CNN polls that came out in Pennsylvania, Colorado, New Mexico and Nevada last week.

On the state level, then, I'm hesitant to make any claims about how she may or may not be affecting Nader. Across those four states, he got 6-8% while, as I said yesterday, she received 0-3%. If anything Nader's numbers there are higher than they've been in some other state polls. Is her presence in the questionnaire actually augmenting his support? I doubt it. Both are running ahead of expectations in those polls.

At the national level there is only one example of a polling firm that asked about the four way race and subsequently asked about the five way race with McKinney included. And predictably that was the Zogby Interactive poll (...an internet-based poll). The four way race was the subject of a mid-August poll while McKinney was added in later to another poll. McKinney's 1% in August brought Nader's earlier 2% down to 1% later. That may certainly be overstating the relationship, as it could very well just be statistical noise causing that change, but that is how it looks.

CNN's new national poll for August (out today) shows Nader with 4% and McKinney with 0. Both are down from 6% and 1%, respectively, a month ago (and 4% and 1% just last week). But I don't like CNN's way of reporting these numbers. They focus on the two way race in the write up and don't include any link to the actual numbers. Oh well, that's what Pollster is for.

If I get a chance, I'll do a mock up of the McKinney numbers and the Green performance in '04 and put it up later on.

MSS said...

Thanks, Josh.

Obviously, drawing conclusions when the underlying percentages are so small is, well, inconclusive.

But interesting!

Josh Putnam said...

Sure, Matthew. No problem.