Now from Nate Silver

[gabeana]

I do think people overthink things some times. In an 800-person poll, the margin of error for the difference between candidates is ~7 points. And 1 in 20 polls will fall outside that margin of error, i.e. be more than 7 points away from the true number.
Nate Silver
@NateSilver538
·
9m
Now, in practice, pollsters may sit on their outliers or rejigger them rather than publish as in. But *good* pollsters like Selzer or ABC/WaPo *do* go with their numbers. So you'll get the occasional Biden +17 in WI or Trump +7 in IA.

[elleng]

1. 'So you'll get the occasional Biden +17 in WI or Trump +7 in IA.'

[Bleacher Creature]

2. Great point.

We need to avoid the temptation to be giddy about an outlier that's good news just as much as we need to stop freaking out about a bad one.

[shrike3]

3. If I'm not mistaken, the margin of error is 7 points

(Unless I misunderstood.) So they may still be essentially tied. Not great for us, but not bad either.

[central scrutinizer]

4. Margin of error is a function of the sample size

With a sample size a little over 1000, the margin of error is about 3%. So 3% plus or minus gives you a width of 6%. The smaller the sample, the larger the margin of error. Most common is the 95% confidence level.

[shrike3]

7. Okay, this sample size is 800.

What would the margin of error be? I'm terrible with math.

[Dream Girl]

12. That not math, it's stats. There's a formula for calculating more, but most people use stat progrms

Both Sample size and score go into the MOE calculation

[shrike3]

13. Sorry, to me stats and math are the same thing. I'm bad at both.

Do you know what the MOE would be for 800 voters? I thought Nate Silver said seven; obviously I was wrong.

[central scrutinizer]

15. It would be 3.5%

Candidate A could be as much as 3.5% above the reported poll result, meaning Candidate B is 3.5% below giving Nate’s report 7% spread on the difference. ( I used to teach basic statistics). The biggest issue is the sampling technique. Bad sample means worthless results

[shrike3]

16. Ah. I see. Out of my wheelhouse. Thanks for sharing your expertise.

I imagine you were a good teacher. I would have given you a bad headache had I been your student.

[central scrutinizer]

19. Basic stats is barely math

A few formulas that are built into many calculators. But it’s a fun class to motivate discussion. Sociology, psychology, political science are all relevant.

[shrike3]

20. LOL. For me, it's math. But thanks for sharing your expertise.

[Dream Girl]

21. Sample composition, response rate, method of contact, sampling error, even survey design help

Determine a quality survey.

[Dream Girl]

5. For Iowa?

[shrike3]

6. Yes, Nate is commenting on Iowa.

Central scrutinizer seems to understand this much better than I do. Pay attention whatever he/she says. Don't pay attention to me.

[Tribetime]

8. 538 added a poll this week drumpf up 12 in Ohio

Whoever. Swayable polling is

[shrike3]

9. If you're talking about the D-rated poll, I think it was up eleven

I may be wrong.

Most of the other polls were fairly similar. Biden up in only one.

[Tribetime]

10. Yes I never heard of them but there other polls

Seemed fairly accurate. I don't think Ohio is anything close to that.

[shrike3]

11. I agree. Right now it's leaning to Trump, but not all that much.

[Tribetime]

14. Agree it will be close

[TexasBushwhacker]

17. Great. That will get the Dems who haven't voted

to show up on Tuesday.

[Tribetime]

18. 👍 hope so