Zogby Poll Math: 1/15, Kerry beating Dean by 19% in Iowa!

[DjTj]

So I was doing a little bit of math in this thread:

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=132&topic_id=117468

and this result really surprised me, so I wanted to start a new thread. I'll explain how I got this number:

The numbers come from Zogby:

http://www.zogby.com/news/ReadNews.dbm?ID=781

Zogby's numbers are the average of three single day numbers, but he tells us oone day that the Monday numbers were 25% for Kerry and 18% for Dean:

http://www.zogby.com/news/ReadNews.dbm?ID=779

You can also see that the Monday-Wednesday numbers were 22% Kerry and 21% Dean.

Zogby gets its numbers like this: On Friday he releases a number which is: (Tuesday + Wednesday + Thursday)/3 = Average.

Using this equation, we can put in the information we know:

Kerry's 1/12-1/14 average:
(25% + X + Y)/3 = 22%
Dean's 1/12-1/14 average:
(18% + A + B)/3 = 21%

Our variables are:
X: Kerry's Tuesday poll number
Y: Kerry's Wednesday poll number
A: Dean's Tuesday poll number
B: Dean's Wednesday poll number

We solve these equations and we get:
X + Y = 41%
A + B = 45%

So this means that the sum of Kerry's Tuesday and Wednesday numbers is 41% and for Dean it's 45%

Zogby told us today that the Tuesday-Thursday averages are Kerry 24% and Dean 19%.

We can then write two new equations to average the Tuesday, Wednesday, and Thursday numbers:

Kerry's 1/13-1/15 average:
(X + Y + Z)/3 = 24%
Dean's 1/13-1/15 average:
(A + B + C)/3 = 19%

The new variables are:
Z: Kerry's Thursday poll number
C: Dean's Thursday poll number

Substituting and multiplying:
41% + Z = 72%
45% + C = 57%

Subtracting:
Z = 31%
C = 12%

Thus, on Thursday, Zogby must have polled Kerry at 31% and Dean at 12%. The error is 7.8%

(Since the error for 3 days is 4.5%, the error for one day should be 4.5% * Sqrt(3) = 7.8%.)

Surprising? I knew that Algebra would come in handy one of these days...

Edit: Oops! Looks like Zogby reported two different Monday numbers and the other one makes more sense. Look at this response below:

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=132&topic_id=123328#123517

Dean Mon 27, Tues 18, Wed 18, Thurs 21
Kerry Mon 20, Tues 25, Wed 21, Thurs 26

[hellhathnofury]

1. I'm not following, I admit I'm not a math person but this is just...well

UNconvincing.

:wtf:

[rfjockey]

2. Looks like the math is accurate to me, but I don't think the results are c

Looks like the math is accurate to me, but I don't think the results are correct. The problem is with Zogby. He reported two different daily numbers for Monday. You used the second set that he reported on 1/14. I think these were actually Tuesdays daily numbers. The first set he reported on Tuesday 1/13 I think are the correct ones.... Dean 27 Kerry 20. If you use those in your equations, then yesterday's daily numbers come out to Kerry 26 and Dean 21. Can't say for sure, but these numbers make more sense to me.

This would be the daily results for the week.

Dean Mon 27, Tues 18, Wed 18, Thurs 21
Kerry Mon 20, Tues 25, Wed 21, Thurs 26

[DjTj]

5. that makes more sense.

Zogby must've misspoke. I'm surprised nobody pointed that out in the thread yesterday. If this was noticed people wouldn't have been predicted Dean reclaiming the lead today.

The cool thing is that we have enough info to find out his numbers on any successive day. Unfortunately, there aren't many days left.

[rfjockey]

7. Math is useful

It does make waiting for todays polling numbers more interesting though since Dean will be dropping one of his daily lows and Kerry will be dropping one of his highs.... as Zogby has been saying, it's pretty much a statistical dead heat right now.

[arcos]

8. yes, but there will be tracking polls for NH, Feb 3 states, WI, others n/t

[newyawker99]

15. Hi rfjockey!!

Welcome to DU!! :toast:

[DjTj]

3. Nobody believes my fuzzy math...

...I invented the internet, I swear!

Seriously though, I checked it again and it's right. Kerry must've had a huge number on Thursday to gain on Dean like that.

[creativelcro]

9. The math is right, but the premise is wrong

Each full poll on 504 people is conducted over the course of 3 days. It is not 3 complete polls conducted on 3 different days. Also, the error measures are questionable. The only way to do it right is to have more than 1 poll in parallel (504 people each, say), and measure the variability. That would be the serious way to do it, and it would not be prohibitively expensive either. Just test 4 independent groups each time. That would give you a good measure of reliability. Which in this case, I suspect, is very low...

[Feanorcurufinwe]

14. Yes, the poll should really be read as it's intended.

Which is as a three-day rolling average poll. The sample for any one day is too small.

[Closer]

4. ANYway

back in the REAL world... :eyes:

[sandnsea]

6. okeedoke

I SERIOUSLY doubt this. Dean has way more support than that, it hasn't fallen off that much. There are alot of people who really do believe in him and his message, as crazy as that may seem to me! Crazy in a nice way. Really. :-)

[kerry-is-my-prez]

10. He might "bounce back" a little.... but not enough - IMO

[FrenchieCat]

11. Zogby is suspect with his polling....

Beware of Zogby polls. They are more often used to influence the vote than the record it.

Dean and Zogby probably manipulating the polls, so that when Dean wins, it will give him losta of MO......which he wouldn't have gotten otherwise....it's called lowering expectations. Don't be surprised if Dean wins by a few points and makes headlines because of it.

Trippi is most likely managing the message.......John Zogby has been Dean's ally throughout....don't be fooled.

[drfemoe]

12. SAMPLING ERROR

Since the error for 3 days is 4.5%, the error for one day should be 4.5% * Sqrt(3) = 7.8%.

"Sampling Error, often referred to as the Margin of Error, is the percentage that survey results are likely to differ from the actual due to the size of the sample drawn. If a survey were conducted of all the members of a population, the sampling error would be zero. There are other sources of possible error in survey research such as sample design error and measurement error."
http://www.zogby.com/methodology/index.cfm

Sampling Error has nothing to do with number of days included in poll results. It is based on (calculated from) the SIZE of the sample.

[creativelcro]

13. That's why the sampling error means nothing...

Because it means something only for truly RANDOM samples. To be truly random, they should just get a phone book, open at random, and call people. Even that would exclude people without a phone. Is that what they do ? It is better to get multiple, independent samples, which would give the effective sampling error.