538's Awards for Best and Worst Data Stories of 2016

  • Thread starter Thread starter BillTre
  • Start date Start date
  • Tags Tags
    2016 Data
Click For Summary

Discussion Overview

The discussion revolves around the awards given by FiveThirtyEight for data stories in 2016, particularly focusing on predictions related to the presidential election outcome. Participants explore the accuracy and implications of various predictions made by FiveThirtyEight and other prediction sites, as well as the broader context of data interpretation in political forecasting.

Discussion Character

  • Debate/contested
  • Exploratory
  • Technical explanation

Main Points Raised

  • Some participants note the various awards given by FiveThirtyEight, including categories like "Best Use of Data to Speak Truth to Power" and "Boldest Sacking of Experienced Humans in Favor of Untested Algorithm."
  • Several participants criticize FiveThirtyEight for not acknowledging their prediction of the presidential election outcome, suggesting it was among the worst predictions.
  • Others argue that FiveThirtyEight's prediction was not the worst, highlighting that they assigned a 30% probability to Trump's victory, which was higher than many other prediction sites.
  • A participant points out that FiveThirtyEight provided scenarios for Trump's potential path to victory and cautioned against overconfidence in predictions.
  • There is a discussion about the interpretation of a 30% probability, with one participant comparing it to a .300 hitter in baseball, indicating that such a probability does not imply certainty of failure.
  • Some participants reference other prediction models, such as those from the Princeton Election Consortium, which predicted a much higher probability of a Clinton victory.
  • Disagreement arises regarding the accuracy of FiveThirtyEight's predictions over time, with some participants asserting that their model had Trump at a 50% chance at one point, while others contest this claim.
  • There is mention of the need for additional modeling to calculate probabilities of electoral college victory from raw polling data.

Areas of Agreement / Disagreement

Participants express differing views on the accuracy of FiveThirtyEight's predictions, with some asserting they were among the best while others argue they were among the worst. The discussion remains unresolved, with multiple competing perspectives on the effectiveness of various prediction models.

Contextual Notes

Participants reference specific probabilities and predictions from FiveThirtyEight and other sources, but there are unresolved questions about the accuracy and interpretation of these figures, as well as the modeling techniques used to derive them.

BillTre
Science Advisor
Gold Member
Messages
2,769
Reaction score
12,185
Link
awards:
Statistical Fortitude
Best Use of Data to Speak Truth to Power
"Word of the Year" of the Year
Trudeau Prize for Governance
The Barest Minimum of Progress Achieved
Boldest Sacking of Experienced Humans in Favor of Untested Algorithm
The "Are We Still Doing This for Willful Misinterpretation of Government Statistics
Prescient Data of the Year
The Volkswagon Prize for Insidious Data Manipulation
Ashley Madison Memorial User Data Leak of the Year
Significant Digit of the Year
FiveThirtyEight Person of the Year
 
Physics news on Phys.org
Seems like they forgot to give themselves an award for "worst prediction of the presidential election outcome".
 
It's not like they didn't have a lot of company on that.
 
Vanadium 50 said:
It's not like they didn't have a lot of company on that.
True
 
phinds said:
Seems like they forgot to give themselves an award for "worst prediction of the presidential election outcome".

538? There's was far from the worst.
 
phinds said:
Seems like they forgot to give themselves an award for "worst prediction of the presidential election outcome".

Of all the major prediction sites, Fivethirtyeight gave Trump the highest probability of winning (~30%) (https://www.buzzfeed.com/jsvine/2016-election-forecast-grades?utm_term=.ptOkdDd4Xx#.km7PV7VxlD), so much so that other media sites were making fun of them for emphasizing the uncertainty of the outcome (http://theconcourse.deadspin.com/i-think-nate-silver-is-broken-maybe-1787096019).

If you want to make fun of someone, you should choose Sam Wang of the Princeton Election Consortium who predicted a 99% probability of a Clinton victory and promised to eat a bug on national television if Trump won (he delivered: http://www.cnn.com/videos/politics/2016/11/12/pollster-eats-bug-after-trump-win-smerconish.cnn)
 
  • Like
Likes   Reactions: phinds and dkotschessaa
Ygggdrasil said:
Of all the major prediction sites, Fivethirtyeight gave Trump the highest probability of winning (~30%)

Correct. Nate also laid out plenty of scenarios for Trump's (then potential) path to victory and plenty of stern warnings about being too sure of anything.

A lot of people failed to understand that 30% didn't mean 0%.

-Dave K
 
30% chance is the same as a .300 hitter getting a hit.

I've eaten various kinds of bugs before.
 
Ygggdrasil said:
Of all the major prediction sites, Fivethirtyeight gave Trump the highest probability of winning (~30%) (https://www.buzzfeed.com/jsvine/2016-election-forecast-grades?utm_term=.ptOkdDd4Xx#.km7PV7VxlD), so much so that other media sites were making fun of them for emphasizing the uncertainty of the outcome (http://theconcourse.deadspin.com/i-think-nate-silver-is-broken-maybe-1787096019)...
That 30% call was the highest 538 share given to Trump. Most of the time they predicted far lower. See Rasmussen for polling closest to the outcome.
http://www.rasmussenreports.com/pub...n_2016/election_2016_white_house_watch_trends
 
  • #10
mheslep said:
That 30% call was the highest 538 share given to Trump. Most of the time they predicted far lower.
Wrong, the Fivethirtyeight polls-only model had Trump up at 50% at the end of July. https://projects.fivethirtyeight.com/2016-election-forecast/

Over the long term, the average probability of a Trump win was probably only slightly lower than the 30% number on election day, in the 25-30% range.

See Rasmussen for polling closest to the outcome.
http://www.rasmussenreports.com/pub...n_2016/election_2016_white_house_watch_trends
The site you provide gives national poll numbers from a polling firm (which is the raw data sites like Fivethirtyeight et al. use in their models). Calculating a probability of electoral college victory from the poll numbers requires additional modeling.
 
  • Like
Likes   Reactions: BillTre
  • #11
Ygggdrasil said:
Wrong, the Fivethirtyeight polls-only model had Trump up at 50% at the end of July. https://projects.fivethirtyeight.com/2016-election-forecast/
Well that was the nomination bump.

Over the long term, the average probability of a Trump win was probably only slightly lower than the 30% number on election day, in the 25-30% range.
Yes, my mistake, as you say. I was familiar with 538 low predictions (~12%) in most of Oct, but missed the late September Trump surge.