Black voter model final results: 82% accuracy in predicting winner of a state

[Godhumor]

Last edited Thu Jun 9, 2016, 09:06 AM - Edit history (1)

After only missing 1 state in previous contests, June 7 saw three misses. 2 of these were expected by me to be misses :

CA doesn't fit well into a single demographic model due to its size and diversity. In this case total non-white population would have been a better measure.

NM didn't fit the model, because Clinton's history with the region overwhelms demographic tendencies.

The only non-predictable miss was SD. I think largely in part because no focus was put on the state by either campaign. Different turnout than expected, for sure.

And before anyone asks, yes, I put DC in the correct column already. There is 0% chance DC goes against Clinton.

Anyway, the model outperformed an expected accuracy of 68% from using standard deviation as a measure, which means there was a very, very high correlation between black population and Clinton winning a state. This year, more than many, was all about demographics. And the black bloc was the most consistent factor in determining what states Clinton would win--taking the group by an average margin over 50% in every state, win or lose, will do that.

Original post and results below

-------


Hypothesis: Simply by analyzing the percentage of a state's population that is black, I believe you can fairly accurately predict the "winner" of a state in upcoming primaries and caucuses. There will be misses, I am sure, but I'm bored, so let's see how this works out. I'll bump the thread and fill in actual results (With insightful commentary like, "Boy, that one was wrong!&quot as primaries go on...assuming people are interested in my validation or humiliation, as the case may be.

So let's begin with states that have already voted

Your key for numbers below:
State Rank for Black Pop. State % of Pop. that is Black

All numbers from 2010 Census

Bernie Wins
44 NH 1.22%
33 CO 4.28%
31 MN 4.57%
26 OK 7.96%
49 VT 0.87%
29 KS 6.15%
32 NE 4.50%
47 ME 1.03%
16 MI 14.24%
48 ID 0.95%
43 UT 1.27%

Avg Black Pop 4.28%
At +1 Standard Deviation 8.38%

Hillary Wins
40 IA 2.68%
23 NV 9.00%
5 SC 28.48%
6 AL 26.38%
12 AR 15.76%
3 GA 31.4%
25 MA 8.1%
10 TN 16.78%
18 TX 11.91%
9 VA 19.91%
2 LA 32.4%
1 MS 37.30%
11 FL 15.91%
14 IL 14.88%
19 MO 11.49%
7 NC 21.60%
17 OH 12.04%
35 AZ 4.16%

Avg Black Pop 17.79%
At -1 Standard Deviation 7.80%


Prediction Methodology: If a state's black population is less than Bernie's 1 St Dev number, I predict he wins. If it is more than Hillary's 1 St Dev number, I predict she wins.

So my straight up, no commentary predictions (Note: this isn't a prediction of margin of victory, just who comes out on top as the state's winner. I also do not include territories):

34 AK 4.27% Bernie Correct
38 HI 3.08% Bernie Correct
36 WA 3.74% Bernie Correct
30 WI 6.07% Bernie Correct
42 WY 1.29% Bernie Correct
13 NY 15.18% Hillary Correct
21 CT 10.34% Hillary Correct
8 DE 20.95% Hillary Correct
4 MD 30.1% Hillary Correct
20 PA 10.79% Hillary Correct
27 RI 7.5% Bernie Correct
22 IN 9.07% Hillary Sanders--Clinton won black vote by 52%
37 WV 3.58% Bernie Correct
24 KY 8.2% Hillary Correct
41 OR 2.01% Bernie Correct
28 CA 6.67% Bernie Clinton
50 MT 0.67% Bernie Correct
15 NJ 14.46% Hillary Correct
39 NM 2.97% Bernie Clinton
46 ND 1.08% Bernie Correct
45 SD 1.14% Bernie Clinton
** DC 50.7% Hillary Correct

Now there will be some misses here, because the two data sets overlap in the 2nd Standard Deviation (Mean+2*StDev vs Mean -2*StDev), so the question will become which states and in which direction. That said, misses should favor Hillary as her Standard Deviation is over twice as wide as Bernie's (Wider standard deviation means more variation in the numbers. In this case, wider Deviation means Clinton has been more successful among a wider variation in black population than Bernie).

[texstad79]

1. Pretty good

for such a basic model. Tweaking to use non-white voters would probably have also nailed CA and NM.

[Godhumor]

6. Agreed. But I made this back when some people were insisting "the black gap" was divisive

I wanted to prove a point that it was real. I think this incredibly basic model showed it pretty well.

[woolldog]

2. Why didn't you link your OP?

What are you hiding?

[Godhumor]

3. Because I create new threads for results instead of bumping old ones

But I paste the original message in the OP.

Last thread before this one was in the AA group at
http://www.democraticunderground.com/118750891

[Raissa]

4. I've enjoyed following this since your first post.

It's been interesting to watch it unfold. I am curious as to how these kinds of margins could play out with future candidates, particularly as the non white electorate continues to grow.

[Godhumor]

5. This year it worked, because the black demographic broke consistently and heavily one way

Working in the future will depend heavily on if we see the same kind of demographic behavior.

[DemocratSinceBirth]

7. I told you Cali would come through for HRC

She walloped the Vermont independent in heavily Latino congressional districts. There was a strong positive correlation between the percentage of Hispanics in a congressional district and her performance.

[yardwork]

9. I see that too, and it's partly due to Trump.

And it's partly due to Bernie's campaign. Susan Sarandon did Bernie no favors.

[DemocratSinceBirth]

13. "Susan Sarandon did Bernie no favors."

And it's partly due to Bernie's campaign. Susan Sarandon did Bernie no favors.

"If you don't know your past, you don't know your future."

[Godhumor]

16. I always thought it would, too

And I think I stated as much. CA was always going to be a weakness for such a simple model. Even if it had been right for CA it would have been purely a coincidental result.

[DemocratSinceBirth]

17. I would never question your proficiency. You're our own Nate Silver or Alan Abramowitz.

It would be interesting to use the entire non white population as your independent variable.

[Godhumor]

18. Don't tempt me! Heh

I spend far too much time here, as it is.

[yardwork]

8. I've really enjoyed your posts on this.

Models are fascinating.

[auntpurl]

10. Awesome, thanks so much for putting this up

I love statistics.

Agree with those above who said adding Hispanics to the AA vote would have put CA and NM in the right column (and she'd already won AZ and TX). But I understand your rationale for wanting to show the AA vote by itself was a legitimate metric to discuss.

[auntpurl]

11. PS. SD is the one state I think the AP report really affected. nt

[Starry Messenger]

12. K&R

[geek tragedy]

14. NM and Cali can be explained by Latino voters,

who time and time again provided Clinton with big majorities in states like Texas, Florida and New York.

[NurseJackie]

15. That's very impressive!

You did a lot of work! I always enjoy reading your posts!