"No way this reflects real movement in public opinion -- too bumpy"

[bluewater]

says The Ecnomist's data journalist, G. Elliot Morris, about the RCP POLLS tracker.

Link to tweet

The RCP Poll tracker just graphs the simple, unweighted average of the last 5 polls or so, yielding that "bumpy" looking chart, as G. Elliot Morris colloquially described it, that over reacts to changes in a few recent polls.

The Economist's Poll tracker takes a more sophisticated approach:

Methodology
We estimate support for each candidate using a statistical method called Bayesian dynamic Dirichlet regression. The model aggregates polls over the course of the campaign, putting more weight on polls conducted recently, less on those with small sample sizes and accounting for “house effects”—the tendency for some polling firms to over- or underestimate support for certain candidates. We exclude polling firms that do not use rigorous methods. In the past, surveys conducted over the phone with a live interviewer or with online survey-takers that use well-thought-out methodologies have been more reliable than other methods.

Check out all the results at:

https://projects.economist.com/democratic-primaries-2020/

[TidalWave46]

1. This is funny. Truly funny. The "more sophisticated approach" has nothing to do with "bumpy".

The RCP Poll tracker just graphs the simple, unweighted average of the last 5 polls or so, yielding that bumpy chart that over reacts to changes in a few recent polls.

The Economist's Poll tracker takes a more sophisticated approach:

[bluewater]

2. Right, RCP's simplistic average results in a cruder "bumpy" chart that is less useful

in accurately assessing changes in public opinion.

Performing a regression analysis to fit the best line thru the data points is both more accurate and presents a "smoother", more realistic interpretation of how the polling is changing over time.

[TidalWave46]

3. You did it again. NT

[bluewater]

4. lol Taking issue with the colloquial use of the word "bumpy"

hey, take it up with G. Elliot Morris.

lol

And yeah, The Economist's approach using a statistical method called Bayesian dynamic Dirichlet regression to plot he best fit line thru the polling data is much more sophisticated than RCP's averaging the last 5 polls.

[TidalWave46]

5. He wrote this?

The RCP Poll tracker just graphs the simple, unweighted average of the last 5 polls or so, yielding that bumpy chart that over reacts to changes in a few recent polls.

The Economist's Poll tracker takes a more sophisticated approach:

[bluewater]

6. lol still going on about "bumpy", That's what Morris called it too. lol

Shall one call it CRUDE, or JAGGED, or SIMPLISTIC, or INACCURATE instead of "bumpy" when describing how RCP POLLS trendline graph looks?

Hey, take that up with G. Elliot Morris. He clearly was speaking colloquially -- i.e. informally -- when he described it as "bumpy"



Here, let me help... this is the definition of coloquially:

colloquially
[colloquially]

ADVERB
in the language of ordinary or familiar conversation; informally.
"the storms hit several states along a corridor colloquially referred to as “tornado alley”"

So, what was the point you were trying to make? That using an informal term in describing how the RCP POLLS tracker looks was "bad"?

Ok.....

[TidalWave46]

7. It's a blatant flaw in statistical analysis.

Letting it hang out there. It's amusing.

[bluewater]

8. Nonsense. Regression analysis is a well developed statistical tool.

You seem to want to take issue with the informal use of the term "bumpy", and ignore the fact of how crude RCP's approach actually is.

But hey, have fun.

[TidalWave46]

9. Love it.

That’s a smoothe wave.

[bluewater]

11. Thanks.

I never thought I would meet someone that thought regression analysis wasn't a valid statistical tool, and more sophisticated than a simple 5 point moving average filter.

Go figure.


Hey, did you know that the RCP's simple 5 point moving average filter results in a "bumpy" looking graph compared to one constructed using a statistical method called Bayesian dynamic Dirichlet regression?



thanks for the discussion

[TidalWave46]

13. You conflate things that Connor be conflated.

Where have I taken issue with regression analysis. Now you are being dishonest. Smoothe.

[bluewater]

14. You seem to take issue with the colloquial use of the word "bumpy"

and that's about it, as far as I can gather from this discussion.

But, hey, have fun.

Smoothe wave time...

[brooklynite]

10. The ECONOMIST model shows that Warren's rise has accelerated since September...

[bluewater]

12. But it doesn't include the HarrisX online poll of the Harris Panel

so it is perhaps a bit suspect...