Polling Paradoxes

The Wibbly-Wobbly, Timey-Wimey World of Election Uncertainty

Dr. Greg Chism

U of A InfoSci + DataLab

😮‍💨

Playing

You have a 90% change of winning…

Sorry, you lost. 🙂

How does that make you feel?

We are bad at judging uncertainty

  • You had a 10% chance of losing
  • One in ten playing this game will lost
  • 90% chance of winning is nowhere near a certain win

Uncertainty

Lost ❌

✅ Won

Uncertainty

How can we understand this result?

Lost ❌

✅ Won

Probability vs. Certainty: A 71.4% chance for Clinton meant she was favored, but not guaranteed to win.

Why have polls become so bad?..

Well, its not actually that bad…

Well, its not actually that bad…

Well, its not actually that bad…

Polling has change a lot

Polling has change alot

  • Public opinion polls. Survey the opinions of respondents on any number of topics.

  • Baseline/benchmark polls. Baselines levels of voters’ perceptions, knowledge, opinions.

  • Brushfire polls. Voter sentiment during a race: “favorable” and “unfavorable”.

  • Tracking polls. Shorter, smaller daily polls that track how perceptions, attitudes, and opinions change.

  • Exit polls. Exiting polling locations on Election Day, to learn how they voted.

  • Push polls. Worded to lead the respondent toward a certain response.

  • Straw polls. Unofficial ad hoc vote.

😵‍💫

The dangers of favorability: Outcome

So favorability comes from polls?

So favorability comes from polls?

Not always… 😮‍💨

Election uncertainty

Election uncertainty: 538 Dashboard

Accuracy and precision matter most

Simulations

Purpose: present a range of possibilities

  • Pros: give a sense of build-up and possibilities

  • Cons: data overload can lead to interpretation difficulty

The dangers of favorability: Maps

Obscurity/Blurriness

Purpose: Fogginess is a powerful visual metaphor

  • Pros: Intuitive metaphor

  • Cons: Subject to different interpretation; hard to show levels of uncertainty

Error Bars/Lines

Purpose: indicate the uncertainty in a quantifiable measure

  • Pros: Useful to compare multiple estimates; easily understood

  • Cons: Details in the data can get lost if not represented properly.

Distributions

Purpose: Show the spread of possible values

  • Pros: Judgments made on sample vs. means/median

  • Cons: Distributions may need additional explanations

Multiple outcomes

Purpose: helpful to see various outcomes for projections/forecasts

  • Pros: Uncertainty is displayed more explicitly

  • Cons: Too many possibilities can lead to confusion

Now what?.. 🫨

Conclusions

  • Know that 28.6% still gives a significant likelihood for that candidate’s success.

  • Understand that polls have changed a lot, but are overall still good.

  • Polling is more accurate the closer the election is.

  • Pick the visual that is clearest to you.

Take a breath, its almost over 🙏

🔗 electionssuck.com