Decision Science News

1.6 MILLION SQUARE KILOMETERS OF GARBAGE IN PERSPECTIVE

This Nature Scientific Reports article gives an update on the size of the Great Pacific Garbage Patch. It is estimated to be four to 16 times larger than previously estimated: 1.6 million square kilometers.

While the article does show the patch along with latitude and longitude lines (above) and the Hawaiian islands, it doesn’t provide much help communicating how large it is with familiar reference objects. (Incidentally, our research has found that people seriously underestimate the size of Hawaii).

How large is 1.6 million square kilometers? It’s about

• As big as Alaska
• One-fifth as big as the contiguous United States
• Half as big as India
• As big as Iran

Open to suggestions for how to put into perspective the estimated mass of the plastic: 80,000 metric tons.

REFERENCE

Lebreton et al. (2018). Evidence that the Great Pacific Garbage Patch is rapidly accumulating plastic. Scientific Reports.

h/t Stefano Puntoni

RUTH VS DIMAGGIO

We were looking up the wording of a joke we wanted to tell our kids:

A guy has a talking dog. He brings it to a talent scout. “This dog can speak English,” he claims to the unimpressed agent. “Okay, Sport,” the guys says to the dog, “what’s on the top of a house?” “Roof!” the dog replies. “Oh, come on…” the talent agent responds. “All dogs go ‘roof’.” “No, wait,” the guy says. He asks the dog “what does sandpaper feel like?” “Rough!” the dog answers. The talent agent gives a condescending blank stare. He is losing his patience. “No, hang on,” the guy says. “This one will amaze you. ” He turns and asks the dog: “Who, in your opinion, was the greatest baseball player of all time?” “Ruth!” goes the dog. And the talent scout, having seen enough, boots them out of his office onto the street. And the dog turns to the guy and says “Maybe I shoulda said DiMaggio?”

Hilarious, right? You’re welcome. For some reason, we remember this being told by Gabe Kaplan in an episode of Welcome Back Kotter but we can’t find evidence of this online.

The reason for this post, however, was the web page we found the wording on. Professor of Computer Science / Cognitive Science professor Justin Li thinks about the joke in terms of the Wason selection task. Have a look.

It does seem the talent agent is wary of the reasoning fallacy called affirming the consequent, the Wikipedia example of which is:

If Bill Gates owns Fort Knox then Bill Gates is rich.
Bill Gates is rich
Ergo, Bill Gates owns Fort Knox

In the context of this joke, you don’t want to reason

If the dog can speak English, the dog can answer the questions
The dog can answer the questions
Ergo, the dog can speak English

THAT’S ONE WAY TO GET “A” PAPERS FROM YOUR FACULTY

Table 5 Comparison of Average Amount of Cash Awards for a Paper Published in Selected Journals

When we first saw this image on Facebook, we thought it was a joke. But then we downloaded the paper and found that indeed, there’s a practice of paying authors based on the journal that publishes the paper.

“In summary, Chinese universities differentiate the amount of cash reward based on the JIF and JCR Quartile of journals in which the awarded papers are published. The average amount of cash award has increased over the past 10 years, except that the amount awarded to papers published in journals with low JIF has decreased. Publications in Nature and Science are awarded the largest amount of cash reward.” – Page 13

We’re now waiting for someone to write a web app that scrapes CVs and puts dollar values on them.

REFERENCE
Quan, Wei, Bikun Chen, and Fei Shu (working paper). Publish or impoverish: An investigation of the monetary reward system of science in China (1999-2016).

SEE ALSO
https://www.technologyreview.com/s/608266/the-truth-about-chinas-cash-for-publication-policy/
h/t Andreas Ortmann

REGISTRATION LIMITED, ACT SOON

On May 3–4, 2018, the CFPB will host its third research conference on consumer finance at the Crystal Gateway Marriott in Arlington, Va.

The goal of the conference is to highlight research on the topic of consumer finance that can inform researchers and policymakers. The conference will focus on high-quality consumer finance research, with academic and government researchers presenting their research papers.
Location

Location:
Crystal Gateway Marriott
1700 Jefferson Davis Hwy
Arlington, Va. 22202

Conference agenda

Register.

For logistical details and information about past conferences, please visit the conference webpage

REGIONAL DIFFERENCES

We came across this weather.com story on what people consider “sweater weather” in different states, and thought that the topic seemed just silly enough to be of interest to decision science news readers.

Not much value we can add here. How about some stats?

Mean sweater weather: 58.6 F (14.8 C)
Max sweater weather: Arizona, Nevada at 65 F (18.3 C)
Min sweater weather: South Dakota at 51 F (10.6 C)
Standard deviation: 2.8 F (1.6 C)

DATA IF YOU WANT IT

State,F,C
Arizona,65,18.3
Nevada,65,18.3
Florida,63,17.2
Alabama,62,16.7
California,61,16.1
Delaware,60,15.6
Georgia,60,15.6
Indiana,60,15.6
Iowa,60,15.6
Kansas,60,15.6
Louisiana,60,15.6
Maryland,60,15.6
Michigan,60,15.6
Mississippi,60,15.6
Missouri,60,15.6
New Jersey,60,15.6
North Carolina,60,15.6
Ohio,60,15.6
Oregon,60,15.6
Pennsylvania,60,15.6
South Carolina,60,15.6
Tennessee,60,15.6
Texas,60,15.6
Utah,60,15.6
Virginia,60,15.6
West Virginia,60,15.6
Wyoming,60,15.6
Arkansas,59,15
Illinois,59,15
New Mexico,59,15
Connecticut,58,14.4
New York,58,14.4
Rhode Island,58,14.4
Colorado,58,14.4
Massachusetts,57,13.9
Kentucky,56,13.3
Idaho,55,12.8
Maine,55,12.8
Minnesota,55,12.8
Montana,55,12.8
Nebraska,55,12.8
New Hampshire,55,12.8
North Dakota,55,12.8
Oklahoma,55,12.8
Vermont,55,12.8
Washington,55,12.8
Wisconsin,55,12.8
South Dakota,51,10.6


SOCIETY FOR JUDGMENT AND DECISION MAKING NEWSLETTER

The quarterly Society For Judgment and Decision Making newsletter is ready for download:

http://sjdm.org/newsletters/

NEW PAPER, NEW PRODUCT FEATURE USING PERSPECTIVE PHRASES

Click to enlarge

As mentioned a few times in past posts, we’ve been doing research on how “perspective sentences,” for example “Israel is about the size of New Jersey in area and population” helps Americans comprehend measurements beyond simply saying “Israel has an area of 20,770 square kilometers and a population of 8,793,000.” We’re happy to say that, due to the drive of Jake Hofman and others, this research has now been incorporated into Bing search engine results:

Chris Riederer, Jake Hofman and I have just published a new paper on this topic which will be presented at CHI 2018. In it, we propose and test methods for generating perspectives. For example, why do we feel that “roughly the population of California” is better than “roughly 10 times the population of Oklahoma,” even though they’re about equally accurate and even if the person you are talking to is from Oklahoma? It turns out that 10x not an ideal multiplier for people to work with (strange, we know) and Oklahoma is not ideal to use in examples.

Christopher Riederer, Jake M. Hofman and Daniel G. Goldstein. (2018). To put that in perspective: Generating analogies that make numbers easier to understand. In Proceedings of the 2018 ACM Conference on Human Factors in Computing Systems (CHI’18)

In the course of doing this research, we were able to generate some fun charts like the one at the top of the post. People underestimate the areas of pretty much all US states (above, top panel) but are especially bad at central and Western states. Furthermore, they underestimate the populations of Eastern states (above, bottom panel). The US is tricky for geographic inferences because many big states have small populations. This makes traditional election maps deceptive and have led to some (weird, we know) ways of rescaling them. [BTW, if you are interested in the psychology of demographic estimation, we can recommend Brown and Siegler (1993)].

Another thing we found is that people do a lot better estimating things when you give them a perspective sentence to help them (e.g., “The population of Poland is about as big as that of California. What is Poland’s population?”). The left chart below shows the improvement in area estimation (note that they still underestimate areas, even with hints) and the right chart shows improvement in population estimation. Click the chart to enlarge it.

Click to enlarge

If you’d like to read more, here are some popular articles on the research:

• Grasping Giant Numbers Is Far From Second Nature – Wall Street Journal
• How to Understand Extreme Numbers – Nautilus
• ‘Aha, now I get it!’ Microsoft is building technology to put numbers in perspective – The Microsoft AI Blog
• It’s Time for a Revolution in Investor Disclosures – Wall Street Journal

Here are some of the people who have worked on the research side of this project:

• Jake Hofman – Principal Investigator
• Dan Goldstein – me
• Chris Riederer – Intern
• Pablo Barrio – Intern
• David Bourgin – Intern
• Chinmay Singh – Engineer
• And many others at MSFT

APPLICATION DEADLINE MARCH 12, 2018

The 2018 Summer Institute on Bounded Rationality will take place on June 19 – 27, 2018, at the Max Planck Institute for Human Development in Berlin, Germany. The Summer Institute brings together talented young researchers and renowned scientists from around the globe and aims to spark a dialogue about decision-making under the real world constraints of limited time, information, or computational power.

It offers a forum for young scholars from various disciplines to share their approaches, discuss their research, and to inspire each other. The program will cover the fundamentals, methodology, and recent findings on bounded rationality. This year’s Summer Institute returns to its roots by focusing on how intelligent behavior arises from the interaction between of the structure of the environment combined with cognitive strategies used by the organism.The keynote address will be given by Ulrike Hahn, Professor at Birkbeck, University of London.

On behalf of the directors of the Summer Institute, Gerd Gigerenzer and Ralph Hertwig, we invite young decision-making scholars from all fields to apply.

Participation will be free, accommodation will be provided, and travel expenses will be partly reimbursed.

Applications are open until March 12, 2018.

Apply here: http://bit.ly/2npmmNT
Website: http://bit.ly/2DPGYcu

EXPERIENCE THE BINOMIAL AND NORMAL DISTRIBUTIONS

The Random Walker is an inexpensive mini quincunx

Every since taking Stephen M. Stigler’s class on the history of statistics at Chicago, we’ve been wanting to get our hands on a quincunx.

A quincunx, also referred to as a Galton Board or bean machine (*) is one of these, which you may have seen in a science museum:

Balls are released from the top and bounce off of the pins. Assume the ball goes left or right with equal probability at each bound. By the time they get to the bottom you’ll see that relatively few balls experience a series of exclusively left bounces or exclusively right bounces (the tails of the distribution) and most experience some mixture of left and right bounces and end up in towards the middle. The more equal the number of left and right bounces is, the closer the ball falls to the exact center. The balls are collected into bins and the count of balls in each bin should follow the binomial distribution. In one of the coolest moves math ever did, when the number of bins is large, the binomial distribution approximates the normal distribution.

Seeing this happen is a great way to teach probability without advanced notation. You can do it on paper by flipping a coin at each bounce, like I recently did with my 9 year old:

but that’s slow going. (If you do this, I recommend starting with a real coin then transitioning to the random.org smartphone app which flips coins quickly and easily. Boredom and frustration can cause the kid to lose interest before a pattern emerges).

In the past, when we’ve checked (and we’ve checked a lot) the cost of a quincunx was high. Hundreds or thousands of dollars.

But, as chance would have it, I stumbled across one on Twitter that turned out to be cheap. It’s called The Random Walker and was only $20 when we ordered it on Amazon.

(*) We don’t really like the name bean machine.

(**) We do really like LaPlace quote in the second photo. Before the modern day belief that minds don’t reason according to the theory of probability, it was thought that the theory of probability describes how minds reason.

HOW MANY GAMES WAS THAT?

AlphaZero is a reinforcement learning (RL) progam that can take a game like chess and given only the rules, can play games against itself and learn how to win.

According to several articles, it learned from scratch and surpassed human knowledge of chess in four hours. Specifically, it beat the leading chess computer in that time.

A friend of ours asked if it trained with more or less experience, in terms of games played, than a young human grandmaster has.

To look into this question, we read the paper.

About 30 people have become grandmasters before 15. Let’s overestimate and say they played 10 years or 3650 days and 100 games a day, that’s 365k games. From what I can tell, AlphaZero played about 20 million games at the point it beat a top rated chess engine called Stockfish (article, Table S3, noting it beat Stockfish at around 4 hours).

So it seems like AlphaZero needs more games to learn than a human grandmaster does. However, AlphaZero starts only with the rules and figures everything out from there. In contrast, people get coached and handed strategies which have been refined over millions of games. It makes sense that humans can learn from fewer games. Also RL systems explore patently ridiculous moves on the way to becoming good players and people can likely prune the space better. But on the other hand, the assumptions human bring to this pruning might be what causes us not to be as good at chess as AlphaZero.

Note that some say the real story here is that it taught itself not the four hours number, because of the serious difference in hardware between AlphaZero and Stockfish. Viswanathan Anand says on chessbase:

Obviously this four hour thing is not too relevant — though it’s a nice punchline — but it’s obviously very powerful hardware, so it’s equal to my laptop sitting for a couple of decades. I think the more relevant thing is that it figured everything out from scratch and that is scary and promising if you look at it…I would like to think that it should be a little bit harder. It feels annoying that you can work things out with just the rules of chess that quickly.

Photo credit:https://www.flickr.com/photos/mukumbura/4043364183/