Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

PEC: Where you’re the customer, not the product

October 21st, 2020, 10:04pm by Sam Wang

Today I was interviewed by a reporter for AP, a Mr. B. As PEC readers know, there are virtually no probabilities reported on this site for 2020. Still, he kept asking about them. Mr. B. seemed perplexed by my refusal to tell him a number, despite the fact that our scripts calculate it every day.

There’s an important reason to avoid talk of probabilities: At the Princeton Election Consortium, you’re not the product – you’re the customer. What does this mean?

To news outlets and political data consultants, you’re the product. Reporters want the traffic. I don’t blame them. It is tough to be in the news business. Anyway, a win probability is very good at getting eyeballs. See the success of tracker models like FiveThirtyEight and The Economist. But that’s in the same vein as spectator sport commentary, which is a type of entertainment. This is no year to be a spectator!

Then there are the political data consultants. Here, it’s a bit more individualized. Each one of you is a different product. Yay! But wait: What that means is that they model your personal likelihood of voting Democratic or Republican, how persuadable you are, and how likely you are to vote or make a donation. Then they sell that analysis of you to make money.

I feel differently. To me, elected officials and policies are the product. You are the customer. That is the central motivation behind this site – to give you the same kind of information that political parties get, except to empower you instead of them. This point seemed to confuse Mr. B.

As I explained to Mr. B., it is better to report races in terms of point margins. Point margins give you a sense of how close a race is, and lead you to think about how much your vote matters. I wrote about this in the New York Times after the 2016 election, and in the Columbia Journalism Review this summer.

This has the added advantage that the real problems this year have to do with the fairness and orderly conduct of elections. Those factors add uncertainties that reduce the probability of any win that is based on polling data alone. A straight probability calculation is so very…2016. That’s why we have done deep dives into which states count votes on Election Night and where election integrity is most at risk.

This year, our principal focus is to show where citizens are most powerful. For President and Senate, this is easier because there are polls. For state legislature, one has to rely on indirect measures such as expert ratings. (Though note that I recently learned from a data consultant that our ratings, which we get from CNalysis, are fairly good!) Anyway, for legislatures we calculate voter power in any state where voters have a shot at getting bipartisan governance – and thereby affect redistricting in 2021. That’s all featured in our Redistricting Moneyball feature. I estimate that as few as 150,000 votes make up the margins to determine whether over 100 Congressional seats are drawn on a bipartisan basis – or are at risk of being gerrymandered. Those are valuable voters!

I am being a little disingenuous. Like Mr. B. and the data pundits, I also want you to do something. I want you to vote, to canvass, and to give. You have it in your power to help repair our semi-broken democracy. Go to the sidebars and give to Senate or state legislative races. And did I mention that you should vote?

Tags: 2020 Election

14 Comments so far ↓

  • Sam Wang

    P.S. Don’t you guys dare tell Mr. B. where to find the Biden win probability.

  • PJC

    A bit snarky but please hear this. Most people who claim to be different aren’t. What’s really different is not to be the mostly the same and claim to be different but to be somewhat the same and up front about it.

    One of the things that *IS* different about this web site is that it helps rationally target money spent in elections.

    We would all be lying if we tried to pretend that ActBlue was a non-partisan entity, but ….. it’s interesting that ActBlue has shattered records for money collected, and that is really, really, really important for several reasons.

    First, it gives littler people more equal clout relative to the usual few, huge donors. And, second, it allows us to precisely target donations to desired outcomes.

    It un-blows up Democracy by ironically putting Citizens United to work in the favor of the smaller guy.

    I doubt the Framers believed that House races in Texas were of national concern, yet we in all states can now target them with precision, because we now know that they are. Though I think spending a dollar is different than “speech” per se, if the Supremes are not clever enough to notice the distinction, then we little people can spend our money on downballot state races in Texas.


    Texas is a California wannabe, but they just don’t quite have the money. If the Supremes think money is speech then its time California had a word with Texas. And y’all are invited to join the conversation.

  • George Bussey

    So, I guess he wasn’t real interested in random walks and Bayesian priors either?

  • ArcticStones

    Well said, Dr Wang.
    Thank you for your continuing service!

  • alexander i rudnicky

    The legend for your “meta-margin for house control” graph is unreadable. Can you more it up or down? That seems to work elsewhere on the page.

  • Froggy

    A little-known side benefit of voting early: in these days of campaigns sucking up data about voters, once you get everyone in your household to vote early, the campaigns stop bugging you with calls as part of their GOTV effort. Your meals are interrupted less, and the campaigns can turn their GOTV effort to other people who might really need those reminders.

  • Randy L Haugen

    I have worked hard to fully understand the meta-margins and other data,I have also worked even harder to get people to Vote and get active.Our Democracy is on the brink.

  • Matthew J. McIrvin

    Glad you came around on this.

    You undoubtedly know this, but the “win probability” is always the single weakest assertion that can come out of these models anyway, because it involves assuming an essentially unknowable distribution for correlated systematic error, which is probably radically different in every election cycle. The Meta-Margin is at least conditional: “if you assume X error, the result is Y”.

    • 538_Refugee

      But it is dramatic and leads to nice theatrics. Like insect ingestion. Best to keep your mouth shut on a windy day. ;)

    • ArcticStones

      Indeed! Moreover, this time around there is the unknowable distribution of the correlated systemic cheating.

      We are seeing plenty of concrete examples: voter roll purges, reduction of polling places in minority-dominant communities, too few drop-off boxes (Texas et al), and other disenfranchisement, the intentional slowdown of the US Postal Service, the 10-day delay in printing many Ohio & Pennsylvania ballots at a MAGA-owned printshop, the ban on free transport to the polls…

      Nonetheless, I should think that’s still exceedingly difficult to quantify and predict the cumulative effect of these and other efforts to hinder pandemic-hit America from voting, voting safely – and having everyone’s ballot counted.

    • Sam Wang

      I don’t think it’s correct that systematic error is unknowable. The range of possibilities can be estimated from past elections.

      However, it is true that “win probability” combines (1) overall margin with (2) the range of systematic error. To my thinking that hides assumptions in a way that is not helpful. My own sense is that adding the error assumption – thus creating a win probability – is useful for passive consumers of politics.

      There is plenty of demand for coverage of win probability as if the election is a sports event. For that, there are other sites for you to visit!

  • Amitabh Lath

    I am looking at the red/yellow bars on the Biden/Trump EV plot, as well as the House Meta-Margin plot. I presume the red (yellow) bars are the 1 (2) sigma limits.

    It would seem that Dems losing the House is *more* likely (edge of 2 sigma, or 5% ish) than Biden losing the election.

    That seems counter-intuitive. Do I have it right? I would have thought R’s winning back the House was significantly less likely.

  • Vicente Piedrahita

    I know it’s been a week since this article, but wanted to note how great the focus of your site post-2016 is.

    Yesterday I learned I had some extra cash available to deploy, and the place I turned to was your Senate moneyball table. It’s super useful to have a place where, as you say, we’re the customers, thank you! And hopefully the results are as good as (or better) than they were with the same approach in 2018.

Leave a Comment