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
SUMMARY

The forum discussion centers on FiveThirtyEight's predictions for the 2016 U.S. presidential election, specifically their assessment of Donald Trump's chances of winning. FiveThirtyEight assigned Trump a 30% probability of victory, which was the highest among major prediction sites. Participants in the discussion highlight the importance of understanding that a 30% chance does not equate to a certainty of loss, contrasting it with other predictions that were significantly more optimistic for Hillary Clinton. The conversation also critiques other prediction models, notably those from the Princeton Election Consortium.

PREREQUISITES
  • Understanding of statistical probability and its implications in forecasting
  • Familiarity with election forecasting models, particularly FiveThirtyEight's methodology
  • Knowledge of the 2016 U.S. presidential election context and key players
  • Awareness of media influence on public perception of polling data
NEXT STEPS
  • Research FiveThirtyEight's election forecasting model and its historical accuracy
  • Explore the Princeton Election Consortium's prediction methodology and its outcomes
  • Study the impact of statistical literacy on public understanding of election forecasts
  • Analyze the role of media in shaping narratives around polling data during elections
USEFUL FOR

Data analysts, political scientists, journalists, and anyone interested in the intersection of data science and electoral politics will benefit from this discussion.

BillTre
Science Advisor
Gold Member
Messages
2,739
Reaction score
11,974
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.