Okay, first off, here's the link to the data so that you know what I'm talking about.
I'll be analyzing the raw data of the PPP poll in the Missouri Senate race in three aspects:
- I'll be looking at another kind of intensity gap that benefits the GOP
- I'll be looking at where Robin Carnahan falls short of the Democratic potential in the state.
- I'll make an educated guess as to how undecided voters will break.
Follow me beyond the fold for that analysis.
You can also actively follow along if you have the spreadsheet opened and a little bit knowledge of Excel/GDocs (which is pretty much identical to Excel), I'll paste the a little bit more advanced formulas here in the paragraphs with italics. If you just want to read the conclusions, just ignore those paragraphs.
- The traditional definition of the intensity gap in this election is that Republicans are mad at the Democrats, want to take back their country and somehow push their party back to power- shorter, that the intensity gap is fueled by enthusiastic partisan Republicans, such as the intensity gap in 2006 and 2008 was fueled by enthusiastic partisan Democrats of all colors- not only Progressives, or minorities, or young voters- all Democrats were fired up that they could get their party back in power.
The first two aspects of that may be true, but the latter one- that the Republicans are fired up about their party and candidates is not really true.
More than 10% of PPP's sample report that they view BOTH Blunt and Carnahan unfavorably. Those aren't enthusiastic Republicans, they're mainly either Independents who want to keep a tight balance of power between both parties, or Tea Partiers who don't like the Republicans because they're too 'liberal' and politician-like.
Still, Blunt, the personification of a sleazy DC politician, is winning 40% of these voters, Carnahan takes home only 9%. The others split among Undecided and both minor party candidates, who take a combined 20%- both about as much as Carnahan.
Enthusiastic Republicans and Conservatives alone can't put Blunt past a Toss-Up with Carnahan, it's people who vote for Republicans as the lesser evil who put him over the top right now.
The way to calculate this in Excel is with a =COUNTIFS formula:
For the Blunt supporters who disapprove of both, use =COUNTIFS(F2:F1000,2,G2:G1000,2,H2:H1000,1), and substitute the last 1 for a 2 if you want the Carnahan supporters and 3-5 for Undecideds and the minor party candidates (3 is Undecided, I don't know who of 4 and 5 is the Libertarian and who is the Constitutionalist).
A lot of Democratic strategists say that the election won't be that bad because while people don't like Democrats, they like Republicans even less. That's true if you look at a favorability/unfavorability split, yes.
But what these binary questions don't convey is that there's an intensity gap in the dislike of the two major parties as well. A lot of people don't like the Republicans, but they HATE the Democrats. Which means that even with an approval rating of around 20-25% the Republican Party might still have a ceiling of above 50% of the vote, because a lot of the 75-80% who dislike them will still vote for them over us.
Which is a silver lining for the Democratic Party, because even if the Republicans should be able to sweep these elections, they still won't have a real mandate, and will be voted out as quickly as they might be voted in this November.
2) How high is the ceiling for Democrats in Missouri and where does Robin Carnahan fall short of it?
Even though a lot of people in Missouri straight out hate everything that's Democratic right now, most people are somewhere in the middle and therefore reachable for some Democratic candidate- just not necessarily for Robin Carnahan.
We can look at the Percentage of people to whom one of the following applies: They approve of Pres. Obama's, Senator McCaskill's or Governor Nixon's job performance, view Robin Carnahan favorably, plan to vote for her, voted for Pres. Obama in 2008 or consider themselves Democrats.
To make things easier for you if you're actively following along, the formula to determine if this applies to the respondant in Row 2 (as an example) is '=IF(OR(B2=1,D2=1,E2=1,G2=1,H2=2,K2=2,N2=1),1,0)'.
The result: Almost 63% of poll respondants are, even in this climate, potential Democratic voters. Yet, Carnahan gets only 38% of the vote.
Let's look at the people who are potential Democrats and don't support Carnahan.
If you're following along: If you put the formula above in Column S, the formula for that would be: '=IF(AND(NOT(H2=2),S2=1),1,0)'
How do they split on the various questions?
(The way to calculate this is by =COUNTIFS, as an example for Obama Approval, '=COUNTIFS(B2:B1000,1,U2:U1000,1)', if you put the last formula for 'potential Democrats but NOT Carnahan' in Column U)
Obama approval:
20% Approve
75% Disapprove
5% Undecided
Most of those are leaving the Democratic fold because of the national climate, I don't think there's much that Carnahan can do about that. But about 5% of the complete sample approve of Obama's performance and still won't vote for Carnahan, this is something that she has to work on.
Kit Bond Approval
53% Approve
26% Disapprove
20% not sure
Kit Bond isn't a very controversial Senator, a lot of people who might support some Democrats also approve of Bond's work. That doesn't necessarily mean that Carnahan can't get them, most of them are probably soft supporters of any politician who appeals to those in the middle.
Claire McCaskill Approval
23% Approve
65% Disapprove
11% not sure
Again, those former Democratic-sympathizers who don't support Carnahan are even overall a quite Republican bunch, but about 20-25% are still supporting McCaskill and Obama, but not Carnahan.
Jay Nixon Approval
71% Approval
19% Disapproval
11% not sure (doesn't add up to 100% due to rounding)
Ahhhh, that's it. Nixon seems to still have the support of lots of people who are generally not inclined to support Democrats, but still like the Governor (and they're not only so-so on Nixon, they LIKE him).
This might be key for Carnahan: Get Nixon on the campaign trail for her, and try to figure out how Nixon is rising above the general dislike for Democrats among these voters.
Okay, I guess that's the most important result of this analysis, but let me still give you the other crosstabs for transparency:
They like Blunt by a 58-25 margin.
And they dislike Carnahan by a 65-19 margin, which sounds bad, but it's even worse when you think about the fact that these are all voters who DON'T vote for Carnahan. Those 19% who like her still don't vote for her. Not a single one of those 19% (or 5% of the whole PPP sample) does vote for her. If she got all of them, she'd be very close to Blunt.
0% of them vote for Carnahan (Duh!), but 66% vote for Blunt, 10% are Undecided (so at least some of the pro-Obama, pro-McCaskill and even pro-Carnahan people have already decided AGAINST her, they're not even on the fence anymore), and almost 25% vote for the third-party candidates that I can't keep apart. Doesn't matter.
They REPORTEDLY went for McCain in 2008 by a 64-29% margin, but keep in mind that people lie to pollsters sometimes. In 1964, almost 70% of Americans who were polled maintained that they definitely voted for Kennedy in 1960 over Nixon- yeah, sure. The 64% of McCain voters and the 8% 'did not vote/for someone else' crowd might include some Obama voters who are ashamed of their vote now.
9% of them are liberal, 31% moderate and 60% conservative.
51% are women, 49% men.
48% are Republicans, 35% Independents, 17% Democrats.
On race, age and geographic breakdown they closely match the overall sample.
In conclusion, about 65% of those people that are open to supporting Democrats and yet don't support Carnahan are people who are generally Republicans or Republican-leaning Independents and yet approve of Governor Nixon's performance, 25% are Democrats who like all or most Democrats, but not Carnahan, and 10% are people who randomly like some Democrat other than Nixon- maybe they know them personally or got some good constituency service from McCaskill or whatever.
Carnahan must win back some of those mostly loyal Democrats who don't support her, and try to get some of the pro-Nixon Republicans by having him campaign a lot for her.
3) How might the Undecided voters break?
Okay, now we've reached the point where you can't really follow along because I'm not exclusively using Excel anymore, but also an external statistical program, Gretl. Yeah, I know, R and Stata are probably more mainstream, but I like Gretl, and it's good enough for what I do here.
Basically, I'm transforming the poll results into a lot of dummy variables: Binary variables that look like this: 'Approve of Obama? 1 (Yes)/0 (No)'. Disapprove of Obama? 1/0. No opinion on Obama? 1/0'... and so on, through all questions, including geographic breakdown, age and everything.
Then I'm tossing out two groups: The people who support one of the third party candidates (because I don't care in the least if the Undecided voters break for them- it doesn't affect the balance between Blunt and Carnahan), and the people who are Undecided, because I don't want their data to go into the calculation yet.
So I have only the people who support Blunt or Carnahan left. Now I run a so-called Probit regression model to find the probability that they support Carnahan (and obviously the probability that they support Blunt is 1-p(Carnahan)).
For those of you that can read regression output, here it is:
coefficient std. error t-ratio slope
---------------------------------------------------------
const -'0.221338 0.307203 -'0.7205
ObamaApp 1.11774 0.316734 3.529 0.399109
ObamaUnd 1.10207 0.499319 2.207 0.418328
BondApp -'0.464724 0.265323 -'1.752 -0.161167
BluntApp -'1.22358 0.269781 -'4.535 -0.409135
CarnApp 1.67362 0.268059 6.243 0.566181
McCain08 -'1.16402 0.287933 -'4.043 -0.401692
Mean dependent var 0.427574 S.D. dependent var 0.352339
McFadden R-squared 0.846807 Adjusted R-squared 0.828910
Log-likelihood -59.92011 Akaike criterion 133.8402
Schwarz criterion 164.2964 Hannan-Quinn 145.7205
Number of cases 'correctly predicted' = 555 (96.9%)
f(beta'x) at mean of independent vars = 0.352
Predicted
0 1
Actual 0 318 10
1 8 237
Now, extrapolating this to the voters that are still undecided, how many votes does Carnahan get? A bit less than 36%, to 64% for Blunt (keep in mind that we only looked at those two, some will not vote, some will vote third-party, but the relation Blunt-Carnahan should be about 2-1).
This isn't good at all. It might add another two points to Blunt's margin, just that this support hasn't materialized yet.
By the way, just for fun: Three profiles of three undecided voters- one is projected to break for Carnahan, one for Blunt, one should be on the fence until the bitter end (the model says that she has a 43% chance of going for Carnahan).
Carnahan's voter is from the Kansas City area, likes Obama and McCaskill and Carnahan, dislikes Bond and Nixon and Blunt, liked the bailout, isn't sure what to think about Prop B, is male, considers himself a moderate, voted for Obama, and is a white Independent aged 46-65.
Blunt's voter is somewhere from Eastern Missouri, dislikes Obama, McCaskill and Nixon, likes Blunt and isn't sure what to think about Bond and Carnahan. He's undecided on PropB, doesn't like the Bailout, is Conservative, voted for McCain, is a white Republican woman age 46-65.
The really, really undecided voter is also from Eastern Missouri, likes Bond, dislikes Obama, Nixon and McCaskill-- so far, everything's fine for Blunt. BUT, the voter dislikes Blunt and likes Carnahan. The voter is against the bailout, against Prop B, is a Conservative Independent who voted for McCain, a white Woman and also aged 46-65.
Okay, that's it for now. I had a lot of fun doing this, and it became a really, really long diary, but I hope you also had fun following my work.
Thanks for reading.