Wednesday, October 28, 2020

What you believe depends on where you stand, apparently.

Or, talking your book on surveys. 

Political Polarization and Expected Economic Outcomes by Olivier Coibion, Yuriy Gorodnichenko, and Michael Weber is a fascinating working paper on the election. 

...despite wanting different things, voters should be able to broadly agree on the likelihood of different electoral outcomes..

Nope. 

87% of Democrats expect Biden to win while 84% of Republicans expect Trump to win. Importantly, this stark disagreement does not reflect two sets of partisan voters each foreseeing a close election that just barely breaks their way. Among Republicans, the average probability they assign to Trump winning is 76%, with more than one in five saying that Trump will win with 100% probability. Among Democrats, the average probability assigned to Biden winning is 74%, with almost 15% of them saying that Biden will win with 100% probability. 

Less surprising:

Republicans expect a fairly rosy economic scenario if Trump is elected but a very dire one if Biden wins. Democrats ... expect calamity if Trump is re- elected but an economic boom if Biden wins. 

Perhaps of course that economic forecast is why each group votes the way they do, and the conditional distribution should go the other ways -- given which president you think will be a calamity, you vote for the other one. 

Normally I am a bit skeptical about surveys -- they measure what people respond on surveys. Surely people don't mean 100% chance of my candidate wining in the same way they assess the probability of the car breaking down on the way to work. But here measuring what people respond on surveys is quite interesting! If people respond 100% chance of their candidate winning, the same people's response that 100% chance of the stocks they bought going up makes more sense. We learn what "likely" means in casual conversation, compared to "true-measure conditional probability." 

So, I can forecast with 100% probability, the libertarian revolution is coming in 2024! 

Update: 

It would be interesting to document the same pattern with media and pollsters. Informally they seem to confuse "is likely to win" with "who I want to win." 


11 comments:

  1. Some years ago I did a deep dive into Canadian street level politics - out there knocking on doors and talking to voters. I found that voters across the spectrum generally actually agreed in broad terms on the outcomes they wanted. However they disagreed about what the current facts were. And they disagreed about the mechanisms at work in the economy and in society so they disagreed on what the effects of various policies would be.

    The agreement on goals was encouraging. The disagreement about mechanisms was a challenge. The disagreement about the facts of the current state of the country was a shock.

    ReplyDelete
  2. "So, I can forecast with 100% probability, the libertarian revolution is coming in 2024!"

    From your mouth to God's ears.

    ReplyDelete
  3. I think you've got it right: this more likely proves the difficulty most people have thinking in terms of probability, as opposed to some deeper point about the nature of partisan thinking.

    In my own experience even very smart people struggle to think in terms of probabilities unless they work in a field where it matters, or have a hobby where it matters (e.g., poker).

    ReplyDelete
    Replies
    1. If you want enjoyable reads on this, Taleb's "Fooled By Randomness" is fun and Kahneman's "Thinking Fast, Thinking Slow" is thicker but super accessible for readers with all levels of technical training. In particular Kahneman's studies with Amos Tversky over the years looked at this topic and started with professors of statistics who themselves failed to apply probability correctly in making actual decisions.
      In my personal opinion, any economist would, can and probably does find their work completely legit without having to go down the entire behavioral econ rabbit hole.

      Delete
  4. Agree with the basic point that asking people to make probability assessments is a waste of time and never elicits what a statistician would remotely regard as a "probability". It is akin to going door to door and asking people for their view on whether space-time is really curved. How can they provide an answer to a question they clearly don't even grasp?

    The working paper is a good example of "research" that takes something that is completely obvious and then labels it as a "puzzle" because the result does not accord with a strict and logical interpretation even though the survey respondents do not use logic or any well-defined probabilistic construct in providing their responses.

    ReplyDelete
  5. The link to the cited paper at SSRN is broken.
    Try:

    https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3720679

    ReplyDelete
  6. Surely people do assess the probability of their automobile breaking down during the daily commute in the same fashion as they compute the likelihood of their favored political candidate winning at the polls on election day.

    If people have perfect information, they are more likely to offer a more informed guess as to the the probable outcome of the polls on election day, or the likelihood that their old clunker will make it to work and back that day. This is not to say that perfect foresight guarantees an unbiased estimate of the future outcome. It is the rare individual who is perfectly objective and unbiased, for our hopes and prejudices do intervene most of the time. "To err is human, to forgive is divine." (a. s.)

    Of course, that is what the paper did indeed find in the end: Better information improved the subject's estimate of the probability of success (or failure) of his/her favored candidate, or likelihood of the old clunker malvis it into the office on time that day!

    Quel surprise.

    ReplyDelete
  7. Yahoo Finance's "Daily Briefing" states this morning that "[t]he most important companies have never been more important."

    And one wonders why academics labor so long and so hard to educate us that we might better avoid the perils of the "information age". Is it, then, any wonder that researchers find such interesting gems as the one displayed in the bar chart featured so prominently in the article above?

    ReplyDelete
  8. This doesn't seem to match the "who do you think your neighbor is voting for?" question. More people think their neighbors are voting for Trump than Biden (even among Democrats). That should lower the perceived probability of Biden winning.

    ReplyDelete
  9. Worthy of a read

    https://pcl.stanford.edu/research/2015/iyengar-ajps-group-polarization.pdf

    ReplyDelete

Comments are welcome. Keep it short, polite, and on topic.

Thanks to a few abusers I am now moderating comments. I welcome thoughtful disagreement. I will block comments with insulting or abusive language. I'm also blocking totally inane comments. Try to make some sense. I am much more likely to allow critical comments if you have the honesty and courage to use your real name.