Look at that, Overtime Politics caught just making shit up in their "polls"

[Godhumor]

Last edited Thu Jan 7, 2016, 02:50 AM - Edit history (1)

I posted about this outfit before at:
http://www.democraticunderground.com/1251951829

Basically, they're not real pollsters but they're posting interesting results from random dialing people in states. Well, I'm not even sure they're doing that, anymore.

See, the creator decided to start collecting demographic information for his "polls" beginning with the new year.

The first to come out was Michigan. GDP regular Robbins was so good as to immortalize the results in the Bernie group here:
http://www.democraticunderground.com/128093066

Or you can see the PDF of the original results here:
http://docdro.id/D4bnL2I

I saw the results and went to Overtime and saved a copy of the PDF for a more in depth analysis I've been working on for the past two days. Simply put, the math for their results didn't work. They're actually impossible to obtain based on the demographics given.

But a funny thing happened just now before I started writing up the math analysis showing results are being fabricated. Overtime Politics quietly changed the demographic breakdowns. And not just a little but a lot. In other words, the person realized the first results were impossible, so he went and tweaked them. You don't even have to take my word for it, in the Sanders Group thread above, Robbins provided the demographic breakdown. If you then follow the link he posted to Overtime, you'll see the breakdowns have all changed:

Let's start with the most minor changes:

Category HRC/BS/MOM/Unknown (Total)

Original
Male 43/42/3/10 (98)
Female 52/37/1/9 (99)

Revised
Male 43/43/4/11 (101)
Female 52/38/1/10 (101)

So, a little change. Unusual, but whatever. So let's get to the good stuff:

Original
White 41/43/4/12 (100)
Black 63/31/2/4 (100)
Hispanic 48/40/1/11 (100)
Asian 45/44/1/10 (100)
Nat. Am. 44/42/1/13 (100)

Now there is a huge number of problems with these results, which was going to be the focus of my post just a few minutes ago. However, I don't need to do that thanks to the idiot(s) at Overtime just going in and changing the data. One thing I will highlight is that rounding error says you should not always see totals at exactly 100%. Every subcategory other than male and female in the original results totaled exactly 100%. Every income bracket, every ethnicity bracket and every age bracket. Quite literally it is impossible with the sample size given.

By the same token, if O'Malley only got 10 responses and only accounts for 2% of the total, there is literally no way he can be above zero in every sub-demographic.

But both of these points were apparently realized, so they went and edited:

Revised
White 41/43/3/12 (99)
Black 62/32/1/5 (100)
Hispanic 49/37/0/12 (98)
Asian 49/43/0/6 (98)
Nat. Am. 49/49/0/12 (110!)

So where to start. How about Clinton mysteriously gaining 4 points in Asian support? How about the Hispanic, Asian and Native American O'Malley supporters disappearing? How about Clinton getting 5 extra points in Native American and Bernie getting 7?

They're simply making shit up.

Oh, and the typo in Native American support equaling out to 110%? That's really damning as well. That means someone was manually inputting values, as opposed to them being formulaic from a respondents tabulation. And someone apparently forgot to carry the one.

I could go on, but you'll have to take my word for it that the same thing happened for age and income.

You can see the revised Overtime Politics results here:
http://overtimepolitics.com/hillary-clinton-leads-bernie-sanders-by-7-points-in-michigan-47-40/

In conclusion, Overtime Politics makes numbers up and now I even doubt if they call anyone at all. Their "polls" are bullshit of the highest order.

[JonLeibowitz]

1. Great work! If you could, a posted PDF would be most appreciated.

Then we don't have to rely on the numbers posted in the Bernie Sanders group. And the proof would be so damning as to make their inevitable demise amusing in the extreme.

Consider scribd.

[Godhumor]

6. I'll have to look around

I don't want to upload to anything I have to log into for reasons. Might try out FileTea.

[JonLeibowitz]

7. Right. FileTea is very nice when you need it.

[Godhumor]

9. Done, link to the original PDF is now in the OP. N/t

[JonLeibowitz]

10. Great, thanks.

And now for posterity, I added the current version to The Internet Archive: https://web.archive.org/web/20160107065110/http://overtimepolitics.com/hillary-clinton-leads-bernie-sanders-by-7-points-in-michigan-47-40/

[HerbChestnut]

2. Looking at the results pdf

it looks like they just typed up a quick spreadsheet in Excel

[Godhumor]

3. Yup, exactly what they did

Last edited Thu Jan 7, 2016, 09:52 AM - Edit history (1)

Basically in their attempt to look better, they ended up casting a huge amount of doubt on whether they even call people in the first place.

I hope to never see another Overtime Politics "poll" on DU.

[HerbChestnut]

4. Also

In their descriptions of some polls showing Hillary and Bernie 5-9 points apart, they call Hillary's lead "significant" despite the fact their polls typically have MoE's of around 5%. In fact, the Michigan one has a standard error of 4.9% with only a 7 point gap. I know I'm preaching to the choir here, but ain't a significant difference if we're talking statistics.

[Godhumor]

5. I wouldn't worry about statistics anymore. This is definitive proof

That numbers are pulled out of where the sun don't shine. Every poll they've done or will do is now rendered invalid by proxy.

[MADem]

8. Gee, did they have a guy named Uretsky on their payroll.....?

[OilemFirchen]

11. ...

http://www.democraticunderground.com/1251964880#post7

[Godhumor]

13. Good call

Nice of them to conveniently F' up the charade with their very first "poll" including demographics.

[Scuba]

12. Many (most?) polls are designed to influence opinion, not measure it.

[Scuba]

15. Nope, none whatsoever ...

http://www.newsmax.com/surveys/ShouldObamaGetGunControl/Should-Obama-Get-Gun-Control-/id/120/kw/default/?dkt_nbr=urq2kwcz

Newsmax.com, one of America's leading online news services, is conducting an urgent national online poll about President Obama’s plans for stricter gun controls.

We are asking Americans about new calls for gun and ammunition controls, and whether they believe their rights are infringed.

[DemocratSinceBirth]

16. WoW/nt

[Godhumor]

17. I know, right?

This was a fun write up for me.

[Godhumor]

18. An afternoon bump n/t

[sufrommich]

19. I think the fact that there is no information on who

runs the site says it all.

[Godhumor]

21. It is so easy to overlook the red flags when it shows something you want to see

I fully expect to see more polls by them posted here. I truly hope I am wrong on that. This incident should be the last time anyone talks seriously about Overtime Politics.

[riversedge]

20. K and R--and thanks

[Jim Lane]

22. Many thanks! I've linked to this thread in an ongoing discussion on Wikipedia.

The Wikipedia articles about the Democratic and Republican primaries were edited to add Overtime Politics reports. This prompted a discussion about this source, consolidated in this thread on the Talk page for the article about the Republicans. I've started a subthread with a link to this DU thread and a summary of your criticisms.

Several Wikipedia editors had expressed doubt about including Overtime Politics but no one had gone into as much detail as you did.

Feel free to weigh in over there yourself. You can create a pseudonymous account and post immediately, or you can just post, in which case your IP number will be displayed. PM me if you encounter any problems with the Wiki markup, but the basic procedure is just to click on the Edit button. I'll be glad to clean up any newbie formatting errors you make.

[DemocratSinceBirth]

23. Good catch