Gallup Facepalm

[cthulu2016]

In the Gallup graphic above... the story that created all the analysis of what Obama polled on Sunday, etc... where is the day of the debate? (October 3rd.)

When I read in the Washington post article what the numbers were for the three days before the debate and the three days after the debate I assumed the day of the debate was one of the three days before the debate... it never occurred to me they were leaving out the day of the debate.

So we can calculate October 3rd at 52-46.

We still cannot, however, calculate 10/7 without knowing 9/30. It is likely that 10/7 was a very big one day sample for Obama but we cannot calculate it with the data available. We can, however, say that it was about 7 points higher than whatever 9/30 was.

This post supersedes all previous comment and calculation by me which was based on assuming that Gallup was speaking English.


http://www.gallup.com/poll/157907/romney-narrows-vote-gap-historic-debate-win.aspx

[reformist2]

1. The 7-day ending Oct. 6 has the race at 49-46. So...

Starting with 49*7 = 343, and assuming the pre- and post-debate 3-days are precisely 50 and 47, I get

343 - 3*50 - 3*47 = 52 for the missing October 3 daily.

But given the rounding errors possible, the Oct. 3 number could be anywhere from 47-57.

[BumRushDaShow]

2. Seems someone asked them how could it be manipulated to be a "tie"

and Whomp! There it was!

And if we "complained", then they could come back and say that "both sides do it" complaining about the polls.

[hrmjustin]

3. and RCP counts both in their averages and Romney numbers in the average go up.

[gcomeau]

4. Arg.... NO.

"So yes, Now that I recognize that there's a day missing in their "tie" story it is easy to calculate the missing day (10/3) as having been 52-46, and yesterday (10/7) Obama polled a crazy one-day sample of roughly 57-38."

Yes, that is a crazy one day sample. There is no way that happened and you cannot calculate any number like that from the available data... for one simple reason. You do not know the numbers from the day that dropped off on the same day as the 10/7 numbers were added. That has not been published anywhere and it is absolutely necessary data to perform the calculation you are describing here.

The *only* thing we can tell from the data provided is that 10/7 was better than 9/30. 9/30 could have been really great and 10/7 was fantastic... or 9/30 could have been kind of crappy and 10/7 was ok.

[cthulu2016]

5. Ironic. I've been arguing that all day and

then head-faked myself in light of realizing that we CAN calculate 10/3.

But you're right. We still cannot calculate 10/7 without knowing 9/30.

[JaneQPublic]

6. Glaring problems elsewhere with Gallup data

I downloaded the Gallup "complete trend" data file for their Obama Job Approval polls. This file has all results for their 3-day rolling survey from Obama's first few days to recent days. (It's available from the gray button on this page:
http://www.gallup.com/poll/113980/Gallup-Daily-Obama-Job-Approval.aspx

However, some glaring errors were immediately evident, where they obviously placed Approval and Disapproval data in the opposite columns. Check the bold rows in Jan 2009, where one day they have Obama's Approve number is 14 and the next is 65.

There is another obvious case of reversed data in the most recent data (again bold). I personally posted on DU about Obama's Approval numbers hitting 54 and then 52, after it was announced on the Gallup home page. But this data file of theirs has it bass-ackwards again!

These silly errors certainly undermine the Gallup organization's ability to process and manage data.

"Gallup Daily: Obama Job Approval",
"Each result is based on a three-day rolling average",
-------------------------------------------------------------------------,
Date,"% Approve","% Disapprove",
"01/21-23/2009","12.0000000000","68.0000000000",
"01/22-24/2009","13.0000000000","69.0000000000",
"01/23-25/2009","14.0000000000","67.0000000000",
"01/24-26/2009","65.0000000000","15.0000000000",
"01/25-27/2009","64.0000000000","16.0000000000",
"01/26-28/2009","64.0000000000","17.0000000000",
"01/27-29/2009","66.0000000000","17.0000000000",



"09/23-25/2012","43.0000000000","51.0000000000",
"09/24-26/2012","44.0000000000","50.0000000000",
"09/25-27/2012","45.0000000000","50.0000000000",
"09/26-28/2012","47.0000000000","48.0000000000",
"09/27-29/2012","46.0000000000","46.0000000000",
"09/28-30/2012","46.0000000000","47.0000000000",
"09/29-10/1/2012","45.0000000000","48.0000000000",
"09/30-10/2/2012","44.0000000000","50.0000000000",
"10/1-3/2012","42.0000000000","54.0000000000",
"10/2-4/2012","43.0000000000","52.0000000000",
"10/3-5/2012","45.0000000000","50.0000000000",
"10/4-6/2012","46.0000000000","48.0000000000",-------------------------------------------------------------------------,
SOURCE: http://www.gallup.com/,