What should I ask Nate Silver?

Nate will be doing a Conversations with Tyler, this February 16, 3:30 Arlington campus of GMU.

What should I ask him?  Please leave your answer in the comments, and the day of the event you also can use Twitter hashtag #cowensilver.

Here is the YouTube teaser for the event.  Here is the RSVP for the event.  By the way, the Kareem video and podcast will be ready  tomorrow (Tuesday) morning.


If it's all about the data, why did he get the UK 2015 election so badly wrong?

He got 2010 wrong too. The UK had lots of national polls, but they were all bad. They also had lots of constituency polls, but they were all really bad. In the USA, he had to predict 51 outcomes, and with states like Wyoming and Rhode Island in the mix, the difficult calls numbered about 12. The UK election had about 150 difficult calls: fifty Scottish seats, forty English/Welsh Lib Dem seats, five Ukip challenges and lots of marginals that could have gone Labour/Conservative.

Having said all that, a world with Silver in public life is better than a world in which he sits quietly at home making money.

I would ask: So, Donald Trump?

He doesn't say it's all about the data, he says it's better to consider the data along with other factors. Nate isn't perfect, but he gives a much needed perspective that others don't bring.

I would be interested in asking him about these results. Why the poor showing? Is the UK fundamentally "different" in some way?

Of course, the 2015 election caught everybody by surprise. I don't think he has covered a similar "upset" situation in the U.S.

The byline is not Nate, but one of the fivethirtyeight writers did two different "What We Got Wrong" write ups here:
and here:

I would guess that predicting outcomes in a parliamentary, multi-party system is just plain tougher than the US, with it's two party system. More variables in the UK system.

This goes to the question I would have Cowen ask- in 2008, 2012 whom did he want to win the general election in the US. Also, whom does he want to see win in 2016. Additionally, Cowen could ask him whom he wanted to win in the UK in 2010 and 2015.

And as a follow-up, How do you insulate your personal preferences and political/professional/peer pressures from your forecasts?

And how did he (and Clay Davenport) get the Matt Wieters projection so hilariously, obviously wrong? Seriously, that kicked off a whole meme.

Nate says in his book that when he was a professional poker player in Las Vegas, making money suddenly became much harder at the end of 2006. The "fish" disappeared, leaving only other poker professionals. He gave up professional poker in early 2007 because he had lost $135,000 in a few months due to no more fish to feast upon.

Does he now think the poker bubble of, say, 2004-2006 was a side effect of the housing bubble in 2004-2006 in the Sand States of Nevada, Arizona, California, and Florida? If he'd drawn the connection back then between the end of the poker bubble and the end of the housing bubble, could he have gotten in on the Big Short?


The fish went because online poker was effectively banned in the USA from 2006 onward. Technically you could still play but because it was illegal for US financial institutions to help it raised the bar to entry so much that Average Joe // Fish just stopped playing in big numbers.

source: had a friend who made millions pre-2006 and talks about this a lot.

That's what Nate Silver says too, but it doesn't make much sense that people would stop driving to Las Vegas for legal poker in December 2006 because online poker was made illegal. Doesn't the opposite sound more plausible: more people would drive to Las Vegas because they couldn't scratch their poker jones online anymore?

It makes more sense that the whole Las Vegas poker bubble of 2004-2006 was a side effect of the housing bubble in Nevada, Arizona, and California with its cash-out refinancings and the like. As I wrote in my Taki's Magazine review of Silver's "The Signal and the Noise:"

"The rise and fall of his poker prosperity provides the best chapter in the book, even though Silver misses the obvious-in-hindsight link between the poker bubble and the subprime bubble. In 2003, only 839 people participated in the World Series of Poker in Las Vegas. In 2004, as soaring home prices in what Michael Lewis called the Sand States of Nevada, Arizona, and California turned mortgages into ATMs, there were 2,576 entrants. By 2006, there were 8,773."

In other words, Nate Silver was privy from December 2006 onward that the housing bubble was popping, but he totally missed the insight that the poker bubble was an offshoot of the mortgage bubble.

That ought to provide some perspective on how hard it was to figure out what the characters in "The Big Short" figured out: Nate Silver, a very smart guy, was at the Mirage hotel in Las Vegas in the winter of 2007 while subprime companies like New Century were going under, but it never occurred to him that there was a connection between the mortgage crash and the disappearance of the poker fish he'd been feasting upon.

Eh, maybe it increased the size of the pots, but is there a real reason to think that increased access to credit or rising home values leads people to play poker more, as opposed to, say, black jack? I think it's more likely that the whole thing was just a fad. Lots of people watched Rounders, thought it was a cool movie, then figured out you could play it online, which was a relatively new concept, then ESPN started pushing it because poker tournaments were cheap content they could broadcast at 12:30 am on a Wednesday in February, and suddenly everyone wanted to be the guy with the big stack of chips they saw on TV. Of course, this didn't last, as people learned through bitter experience they were more likely to lose big than to win big, so poker got old and people moved onto the next fad, World of Warcraft, posting selfies on Facebook, etc.

Pro poker player (2004-2012) here, and I currently host a podcast which frequently discusses these issues. Basically you're both correct.

The UIGEA was a significant factor leading to the contraction of both live and online poker. Live and online poker have had a complementary relationship, not a competitive one. This has been studied academically but really any poker pro will tell you the same.

The 2006 financial crisis also harmed live and online poker significantly. So too has the overall improvement in the level of play among elite players, and a speeding up of the poker ecosystem as elite pros are learning to defeat mediocre pros and amateurs much more effectively than ever before (for myriad reasons too lengthy to get into here).

I would be interested in hearing Nate's thoughts about "explanation without prediction" and "prediction without explanation".

In his book he argues that for good predictions it helps to know the causal structure. He uses the example of weather forecasts for this. He even says that that's why he stayed away from economics forecasting, because the underlying model was much less clear.

Dani Rodrik argues in his recent book "Economics Rules" (p.184):
"However, no social science should claim to make predictions and be judged on that basis. The direction of social life cannot be predicted. There are too many drivers at work. To put it in the language of models, there are numerous models of the future, including those that have yet to be formulated! At best, we can expect economics and other social sciences to make conditional predictions: to tell us the likely outcomes of individual changes, taken on at a time, while other factors remain constant."

Thank you,


Looking internationally, does Mr Silver still trust opinion polls as prediction tools, as opposed to snapshots in time?

Do you think that the average academic could write a 538 calibre article in their area of expertise? Why or why not?

Adjusting electoral polling results for the effects of polling results on elections: is this is either desirable or possible?

Soon after fivethirtyeight came online, it became clear that its editorial voice wouldn't be as uniformly flattering to leftist sensibilities as vox. (The Pielke article was an example.) This probably came as a surprise to many leftist prospective readers, who saw you as a fellow traveler and even a hero of the Obama election cycles. Do you think, in this age of political polarization and Balkanization, that this cost you readers? Would "salon/slate plus analytics" have been a better business proposition?

A related question -

'Soon after fivethirtyeight came online, it became clear that its editorial voice wouldn’t be as uniformly flattering to rightist sensibilities as WSJ.

This probably came as a surprise to many rightist prospective readers, who saw you as a fellow traveler and even a hero of the election cycles where the Republicans gained control of the House and Senate.

Do you think, in this age of political polarization and Balkanization, that this cost you readers? Would “forbes/fox plus analytics” have been a better business proposition?'

For people who see things in binary, the world must be a place of great confusion.

Huh? Silver pretty clearly is on the left, but makes an effort to be balanced and focuses more on what is happening than value judgements about it. Wright's question makes sense, while the opposite would not.

"For people who see things in binary, the world must be a place of great confusion. "

The lack of self awareness in that sentence is almost incredible. Mercatus sucks! Koch Mouthpieces! Germany amazeballs!

I used to enjoy pointing out prior's stupidity but I have to admit, he's outlasted me. I give up, he's like hitting a piece of iron. You guys need to carry on for me.


Some people might say that pushing Obama and neglecting Trump are indicative of mental bias.

Yes, but for others indicative of mental health.

Only if we're talking over a cliff.

The obvious questions:

What is the future of polling since you can't just call people's homes anymore?

What effect will the NL DH have on baseball?

What are Nate's thoughts on Scott Adams' (Dilbert creator) writings and predictions of Trump (that a general election will be a landslide to Trump), based on Trump being a master persuader with the ability to land linguistic "kill shots" on other candidates?

Scott Adams has the best explanation for Trump's rise in the polls and has consistently picked him to win since April.

This has nothing to do with policies nor do I support Trump in any manner.

Adams' non-humor work is full of crackpot theories, and his over-analysis of Trump's mystical abilities is typical of that ("Trump is talking directly to people’s subconscious." Seriously? http://blog.dilbert.com/post/128474925371/how-to-spot-a-wizard)

Obviously you have never done sales or been good at it. There are ways of structuring an offering that encourages people to drop their defenses. It works sometimes, or better put, one way will work with some, another way others. Obama was very good at this, all successful politicians are. Trump has a warmed up audience for his shtick, and he is playing it masterfully. How he does it has many levels and as usual good luck is going with the current.

Adams is simply recognizing the technique.

Is 538 doomed at Disney without Simmons?

Was Simmons ever a part of 538?

No, but its a good question

I think what is meant is "What does the death of Grantland at ESPN/Disney portend for 538?

Since he's a baseball guy and (apparently) once made money playing poker, I'd be interested to know his take on Daily Fantasy Leagues. If he truly devoted himself to it, could he profit on daily fantasy baseball bets?

Are you familiar with his baseball work? It was largely about predicting player performance. https://en.wikipedia.org/wiki/PECOTA

I'm pretty sure he could make a killing on fantasy baseball bets.

Maybe he already is. If you were, would you be talking about it? I wouldn't. Not until after it's made illegal and the sites are shut down -- then I'd you how I made all that money in 2015-2017.

If I had to guess, I'd expect him to say, "No, it's impossible to make much money in daily contests; there's just too much volatility and the cut taken by the house is bigger than any advantage you can gain over the long-term".

1. Ask him about the differences between forecasting and futurism. Ask if he has any predictions for 2030 or 2050 or 2100.
2. Ask him about about faith and religion. Does he believe in a God? Does he practice Judaism in any way? How does that impact his thinking about the causality and modeling?
3. In the same vein, ask him whether being gay shaped his analytic worldview at all. It often seems that he has adopted the view that his sexuality is basically as boring as any heterosexuals' sexuality would be. That's a pretty cool picture of the world that I'd like to be true. But he grew up in a time when that was generally not the case!
4. Ask him about being a business-owner and boss. What drove the move to ESPN? Where might he go next?

The obvious question: what does he think of prediction markets, and under what circumstances (if any) does he think that they will generate more accurate predictions than the careful gathering of data?

A robust market would include those engaged in "the careful gathering of data."

It's in the "gathering":

Silver's election forecast model is one way of aggregating data from a bunch of sources into a single answer; a prediction market is another way.

I do think this is an interesting question.

Right; I'm meaning to contrast Silver's methodologies with that of a prediction market. If we assume that prediction markets take what he writes into account (a big assumption, of course, and one that may not really matter), should he have more confidence in them than his own results?


How does he think they interplay?
Are there any prediction markets he checks regularly?
Does he think he can trade profitably in prediction markets?
Does he think more media should be including prediction market figures in their reporting?

What does he see as their roll going forward as compared with "careful gathering of data," and data models? Does he expect both approaches to become more robust?

I will observe that there has been virtually no discussion of prediction markets on the first few episodes of the 538 Elections podcast.

This was pretty much what I'd planned on asking about.

How much of a decline in influence did you [Nate] suffer by leaving the NY Times?
Have you measured it?
What metrics did you use?

How much does his work (or polling work of others) influence election turnout and outcomes? What are the implications of a system where everyone votes for their favorite candidate vs. the current environment where many people consider "electability" in their calculus?

I believe he's written on the future of polling before, but I wouldn't mind trying a more extreme version of the question- if political polling were banned, or say subjected to the Do Not Call list, how would you replace that information in your model of election outcomes? Would you just stick with ideology, incumbency, and such, or do you think that fundraising and bumper stickers and yard signs could be weak but significant indicators of support?

What advice would you give a politician who is planning to run for President in 2020 and would like to maximize his/her electability?

What are some times that data science convinced you to change your mind? Could be about sports, I suppose, though that might be more controversial than politics.

What level does he cluster his standard errors at?

In his modeling does he account for trends in the poll time series continuing for several days after pools close or not? Or are polls just used as snapshots. Are there equivalents to momentum and value in polling or is the data too crappy?

Nate got eviscerated by the bloggers and editorialists on the left for his chapter on climate change in The Signal and The Noise. If he were writing that chapter today, what would he do differently?

How do his results compare to those of Phil Tetlock's "super forecasters"?
Has he or would he enter a forecasting tournament at Good Judgement
What does he think of Tetlock's statement:

"If you have an influential column and a Nobel Prize and big speaking
engagements and you're in Davos all the time -- if you have all the
status cards -- why in God's name would you ever agree to play in
a tournament where the best possible outcome is to break even?"

Smartphones are ubiquitous. Smartphones have revolutionized data collection and mass communication. In elections, they arguably make "ground game" much less important. So why haven't they improved polling? Why haven't Smartphones improved data collection?

Does polling make liars out of otherwise honest politicians or does polling merely facilitate lying by otherwise dishonest politicians? Speculative.

I'd be interested to hear what he thinks of prediction markets and betting markets.

The information from his old blog helped me to make quite a lot of money betting on American political results (I live in the UK where we're allowed to bet on things like that). If his predictions are wildly different to the consensus of the betting markets, does he question whether the "wisdom of the crowds" is seeing a fault in his model? If betting was legal where he lived, would he put money on the markets when he thought that the options were wildly mispriced?

People's stated preference for privacy imply they value it vastly more than their actions online suggest. How much better do you think our polling and polling analysis could get if polling firms had as much knowledge about their respondents as certain websites have about their visitors? Do you think there's a business model here that could produce a better product (e.g. "A survey USA/Facebook poll out today shows..." or "A CNN/ORC/Verizon poll..."), or does it get too invasive for the online data folks to chance a business on?

the website 538 publishes % probabilities of which team wins every game for sports leagues in the US. Despite the exercise being somewhat pointless to me (who cares if 538 think team A has a 61% chance of winning tonight), how does 538 judge whether its predictions are good? Does it compare results vs. Vegas odds? Something else?

Given that poll results can influence decision making (candidate donations, media coverage, debate access, as well as the desire of individuals to back a winner), how does he prevent his own biases from influencing his predictions? Is objectivity a higher value for him than getting good candidates elected? For example, some could read into his dismissal of Trump's leading position a desire for Trump to not succeed.

Can we please do one of these with Heckman?

How does one go about acquiring the skills that go into the modern data-savvy writer/analyst's toolbox? Asking as a recent polisci/IR student who laments the current state of the liberal arts' curriculum, at least the portions devoted to quantitative analysis.

The obvious 'tools' in the toolbox that goes into the typical 538 article include:

- data acquisition (data mining, APIs, Python, webscraping, finding published datasets, etc.)
- data crunching (Python, R, etc.)
- data visualization
- web/graphic design
- quantitative-focused analysis
- statistics

Do you see yourself as a Frequentialist or a Bayesian? And why?

How does he assess the overall accuracy of his predictions - i.e., how does Nate calibrate the proportion of times outcomes occur vs. his predicted probabilities that an outcome will occur. In yet other words, how do the probabilities of his models translate to observed proportions of events?

Who is going to win the election?

It would be *amazing* if you could avoid asking about Trump. That would be quite a feat.

In which situations do people most often miscalculate their odds of a favorable outcome? Is it ever advantageous to do so?

There are strong arguments that predicting the future--even data assisted--is mostly luck; that for example, supposed genius investors have just been extraordinarily lucky the last year or even the last several years. And that given a little bit of time, there luck always changes. Nate Silver was crowned with glory in 2012 because he was by far the "best" pollster. But how do we know he just wasn't the luckiest pollster? Especially since his prediction record since then has been less than stellar. And that his "return to the mean" is simply and accurately "predicted" by the "Law of Randomness?"

See if you can get any response from him on this Salon piece on how he whiffed so hard on Trump.

If there is any solace, he is not alone. The rules were ripe for change and Trump recognized an opportunity. And has the skill to play it.

Before polls journalists would count how many showed up at a candidate's event. Polling took advantage of the ubiquitous telephone without call displays or answering machines, as well as a social expectation to answer and be polite.

That no longer stands, and the challenge of getting a representative sample is high. But in 2008 and with Trump auditorium audience size seems a pretty good indicator.

If Seth Curry is leading the revolution of professional basketball, who is leading the revolution of other sports? Anyone?

How he thinks about Bernie Sanders, whats the percentage of Bernie will win the election(in his guess), and if Bernie wins the election, does he think Bernie's plan for reforming wall street is plausible?

Ask him if he still does everything in Excel. How does Excel compare to coding?

Yeah, this is what I was going to ask about. Tyler could also work in the angle that much economics modeling (and thus papers) rely on Excel, which is more difficult to inspect than, say, an R script.

Is this actually true? Speaking as a researcher doing quant work in a social science that economists look down upon, that is horrifying, if true.

I am not an economist myself, but I have economist friends, and when I see them ask people to check drafts, Excel is often involved. But "much" is vague on purpose--I don't pretend to know how Excel use compares to STATA use, for example.

How does he field-test the error bars he gives for election projections? Does he have empirical results showing that X% of actual election outcomes fell within his 2-sigma margins, where X% is between 95% and 96% (or at least between 94% and 97%, say)? Or does he use "resampling" ideas such as the Efron bootstrap or funkier things like (when good by-county polling data is available) randomly mashing up counties into different "states", running his model(s) to get projections for the mashed-up election, and seeing how often the same real outcome data would have produced the same winner in the mashup?

Ask him why is there is no Bayesian analysis on his blog

Why did he so vastly underestimate Sanders?

The breaking news today is regarding a deadly accident around the Delaware River last summer.

All the important information including order number, items packed,
final destination, etc. Latest technology 2013 This will
present you with the American IP address had to access Hulu's TV as
you're in Costa Rica, China, Russia, Europe, or anywhere else.

The event was said to be organized in August 2011 in Japan though the earthquake disaster
has cast a shadow of uncertainty over the big event. With
the many talk about being the i - Pad 3 and what it offers and doesn't have beneath the hood only time will tell on that

Biotin is an important component to new hair regrowth as well as healthy skin and nails.
This nutrient is recognized as one from the best vitamins for thinning
hair. Biotin suppliment These changes will enhance your overall health along
with the health of one's hair.

Genetics - It is common for those suffering from hair loss to have inherited an inherited condition from their biological loved
ones (birth parents or grandparents). Sweet potatoes may also be among
best foods that make your hair grow longer.

Comments for this post are closed