Reuters/Ipsos 5-day through 9/25 is Clinton 43, Sanders 24 and Biden 18

[Godhumor]

http://polling.reuters.com/#!response/TR131/type/smallest/filters/PARTY_ID_:1/dates/20150808-20150925/collapsed/false

And this is why I hate the Reuters model--the volatility is insane.

Kenn3d posted the 5 day period that Reuters writes up each week, which is through 9/23. That had Clinton 40 and Sanders at 30. That thread is http://www.democraticunderground.com/1251622864

So why the major change in two days of data? Because Reuters has a continuously open poll every single day and they average the most recent 5 days of respondents. Between that and the fact that the sample is non-representative of the general population, candidates are subject to "bad" polling days frequently.

Look at the chart. On Sept 17th, Clinton was at 48% and Sanders at 24%. On September 19th, a good day for Sanders and a bad day for Clinton rolled into the average. Clinton dropped to 43 and Sanders went up to 29%. September 23rd, the end of the period pulled by Pollster, was the last day that the good Sanders day from the 19th was included in the rolling average. And boing! The 24th and 25th, Sanders drops back and Clinton bumps back up. Now we wait to see who has the next good or bad polling day and do it all over again.

Following Reuters is an exercise in futility, regardless of candidate. There is such volatility in their methods that literally day to day random walk theory says things can go bonkers in one direction or the other.

And this will be the rant I link back to every time someone posts something from Reuters, regardless of which candidate is suddenly doing well.

On edit: Kenn3d rightly pointed out that I misidentified Pollster as downloading the numbers themselves as opposed to working from the Reuters release.

[karynnj]

1. Three or five day averaging is a way to remove volatility, not increase it

that is why it is often called "smoothing". While it is true that a very good (or bad) day affects the series for the number days it is in the average, no smoothed estimate will be as high (or low) as this day would be if you just plotted each day.

[Godhumor]

2. But this version of rolling polls, the effect of 1 day is massive

Traditional polling weights respondents and samples to population norms while Reuters does not. Because of this, it is theoretically possible to have, say, a day where 80% of respondents are Clinton supporters even though that would never actually happen in real life. That kills the next five days of data in terms of meaningful conclusions.

Always open, non-normalized polling is so problematic that a rolling 5 day average does very little to control volatility.

[DCBob]

3. That is a strange poll. The volatility is crazy.

[upaloopa]

4. That poll has three people who are not running. When they drop them off the list Hillary will be

above 50%.

That is because they take points away from Hillary. Hillary is actually leading by over 30% if you include the three people running.

[oasis]

10. Makes sense to me.

[OilemFirchen]

5. It's an online poll.

Before Citing a Poll, Read the Fine Print

By Nate Silver January 15, 2012 5:35 pm January 15, 2012 5:35 pm

On Saturday, a survey came out showing Mitt Romney with a large, 21-point lead in South Carolina. The poll is something of an outlier relative to other recent polls of the state, all of which show Mr. Romney ahead, but by margins ranging from 2 to 9 points.

The poll, conducted by Ipsos for Reuters, has already attracted more than 200 citations in the mainstream media. Most of these articles, however, neglected to mention a key detail: in a break with Ipsos’ typical methodology, the survey was conducted online.

It hasn't changed:

This Reuters / Ipsos poll began in January 2012 and since then continuously polled between 2,000 and 3,000 people a week. Over that period, we have asked hundreds of questions ranging from presidential politics to the Oscars, from the Syrian civil war to the perception of social networks, such as Facebook and Twitter.

Unlike almost all mainstream polls, the data is entirely collected via online surveys. Online surveys allow us to collect far more data and to be more flexible and fast-moving than phone research, and online is also where the future of polling lies.

[Godhumor]

6. Absolutely. But just saying ignore it because it is an online poll doesn't carry as much weight now

Compared with back then. I tried that in a thread earlier this week and had to disengage quickly as to not be the angry guy shouting at the clouds.

This time around I can show the effects with the big blip on September 19.

[OilemFirchen]

7. Any online poll is, by its nature, flawed.

Sorry to hear about your experience, but your interlocutors likely were typing from within their sphincters.

While it is true that online polling has improved over the years, it's still an exercise in dart-throwing. Far too many "controls" to obtain a proximate random sample. And, of course, the intrinsic flaw of online polling is in its basic nature - it relies on access to the Internet, which is inherently tilted to certain subgroups.

Any online poll can be "right" at times, but none can deliver consistent results. Thus, the volatility you've observed.

[twii]

13. False

It's only flawed if it's unscientific.

[OilemFirchen]

14. I see.

What does that mean exactly?

[kenn3d]

8. ...weekly tabs that Pollster decided to grab?

Ok. I had written off the Reuters Rolling 5day poll after your previous comment about it. But following your link I could not see any data points on the Rolling 5day chart which matched the numbers in the weekly Ipsos/Reuters poll as compostited by HuffPolster. So I followed the links from Huff and found the following breakdown (which does not match the numbers in your OP).

(open in separate window to read large or link to Ipsos site here)
http://www.ipsos-na.com/download/pr.aspx?id=14901

These numbers break down by Democrat (608), Independent (218) and Total (968) and also limit candidates to the top 3 (Clinton, Sanders, and Biden). These breakdowns indicate even smaller spreads than the raw data composited in the HuffPollster chart.
Are you sure this is the same poll/data that's being shown in the 5day Rolling charts at your link?

[Godhumor]

9. Misstated by me apparently. Reuters releases an in-depth look at one 5 day roll up each week

This week they tagged it at September 23rd. Go to the chart in my OP and move the line that scrolls with your cursor to 9/23/15, and you will see the same numbers for Clinton, Biden and Sanders as in Pollster's chart. If you highlight the bubbles it will give you the min and max confidence range for the candidate in the 5 day period ending on that day. Pollster does't grab the numbers themselves, which I need to correct in my OP.

First edit: Also, my chart is slightly off in sample size and Sanders is at 29 instead of 30 (The other major candidates are dead on). More than likely that is due to either Reuters prepping data for write-up and release, or I missed a filter that they applied.

[DemocratSinceBirth]

11. Why is there such a discrepancy when I change the parameters to likely voters?

http://polling.reuters.com/#!response/TR131/type/smallest/filters/LIKELY_PRIMARY15:1,LIKELYD:1,PARTY_ID_:1/dates/20150808-20150925/collapsed/false

[Godhumor]

12. Look at how much the sample size drops

And more than likely it is because they tag likely as 1) needing to be a registered Democrat and 2) within the population parameter for must likely to actually go out to vote. Both of these facts would primarily penalize younger voters who tend to be more favorable to Bernie.