Two posts ago we showed you the digit sound system for remembering numbers. This week we provide two Python programs (and one R program) to help you create mnemonics.

w2num.py – converts every word in the English language to a number
find.py – takes a longish number and gives you words that when put together translate into that number using the digit-sound system.

If you are not a programmer, or if you are lazy, you can get to the chase and just download the relevant outupt files here

wordNum.txt – every word in English translated into a number using the digit-sound system
numWords.txt – All the English words that translate into a given number

However, you have a long number that you want to make a mnemonic for, you’re better off using find.py or find.R below to find mnemonics. These handy programs break up long numbers into shorter ones which do have mnemonics.

NOTES:
These programs use phonetics (not spelling) to make the mnemonics. Read the small print on Jap Murre’s site to see why “letter” is 514 and not 5114 and why “ing” is 27. Thanks to Daniel Reeves for convincing discussions on the merits of using a purely phonetic system.

As usual, we’re open for suggestions on how to improve these programs … they’re both kind of slapdash jobs.

##w2num.py##
Prerequisites: You need to save a couple of wordlists locally. In particular
1) You need to save a copy of http://bit.ly/xz8YIC
as cmudict-0.4.scm. This is a mapping from words in English to phonemes.
2) You need to save a copy of http://scrapmaker.com/data/wordlists/twelve-dicts/2of12inf.txt
as wordlist.txt. This is a list of reasonably common English words (b/c some of the words in cmudict are too rare to be useful).
3) Put cmudict-0.4.scm and wordlist.txt in the same directory as
w2num.py and then do
python w2num.py > wordNum.txt
to generate the number corresponding to every English word. (If in the future these URLs break, just search around for copies of cmudict-0.4.scm and 2of12inf.txt …there are many copies of these dictionaries around.)

##find.py##
This is the program you use when you have a longish number and you want to find words that encode it. It uses recursive ™ technology to break up numbers a few different ways to increase your chances of finding a mnemonic you like.
Prerequisite: You must have created wordNum.txt (by running python w2num.py > wordNum.txt) in the previous step and saved it to the same directory as this program before running.
To turn a number like 8675309 into words do
python find.py 8675309

Lastly, for kicks I wrote a slightly dumber version of find.py in R. It lacks the cool “rec3″ method of finding mnemonics but it gets the job done.

##find.R##
Same as find.py
Prerequisite: same as find.py
To turn a number like 8675309 into words change the variable mstr to ’8675309′ and run the program.

At Decision Science News, we have all kinds of numbers memorized: IP addresses, passport numbers, phone numbers, bank account numbers, logarithms, etc. Once you have stuff like this memorized, you’ll  realize how much of a hassle it was to look all this stuff up all the time. Besides, it’s fun.

But how is it done?

It is done with the digit-sound method, which we learned from Professor Jaap Murre’s neuroMod site at the University of Amsterdam. According to Murre, this amazing method traces back to 1648!

You associate a digit with a sound or sounds like this

0 z, s, soft-c ‘z’ is the first letter of zero
1 t, th, d ‘t’ has one downstroke
2 n ‘n’ has two legs (downstrokes)
3 m ‘m’ has three legs (downstrokes)
4 r ‘r’ is the last sound in fouR
5 l ‘l’ or L is the Roman numeral 50
6 soft-g, j, sh, ch ‘g’ when turned around gives ’6′
7 k, q, hard-c, hard-g ‘k’ is made of two 7′s back-to-back
8 f, v ‘f’ when written in script looks like ’8′
9 p, b ‘p’ when turned around gives ’9′

Then, when you want to remember a number, you simply replace the digits with sounds and try to make a vivid phrase. Any vowels or sounds not included in the chart are ignored when you are reading back the phrase. For example, suppose you want to memorize Jenny’s phone number: 867-5309.

Here’s how we might do it:

• Change 867 to sounds, such as “f”, “sh” and “k”
• Change 5309 to sounds, such as “l”, “m”, “s” and “b”
• Now insert vowels and letters not on the chart in order to generate a vivid phrase from those sounds.

We came up with the following dialog involving a parent and child at the dinner table:

–”Fish, ok?”
–”Lame! *sob*”

In this mini play, we have “f”, “sh”, “k” “l”, “m”, “s” and “b” in order, and can easily reconstruct 867-5309. Remember, when reading it back, you ignore all vowels and letters that aren’t in the chart.

The only start-up cost is memorizing the chart, but that’s not hard. There’s even a training tool so you can practice until you have the associations down pat.

Of course, since there can be a few possible sounds for each letter, you need to play around a bit to find a phrase that works for you. To generate possibilities for long numbers, we’ve written the following python script.

NOTE 1: In this code, we’ve simplified the digit-sound method to include fewer letters since we always get confused with hard vs soft letters. We find the simplified method works just fine for us.
NOTE 2: This code is written in a hackish, randomized way that generates wasteful dupes. However, we don’t care since it doesn’t output the dupes, it runs super fast, and if you have a huge number of possibilities, you don’t want to see them all anyway. That said, if people have better implementations, we’re open for suggestions. Thanks to Dan Reeves for tightening up the code a bit!
NOTE 3: You can take the output of this program and hunt for words that fit the patterns with a linux call like this.
grep -E ‘^l[aieouy]*m[aieouy]*n[aeiouy]*$’ wordlist.txt
Where you are looking for words that match “l”, “m”, “n” and wordlist.txt is a list of words, one per line. You can find many wordlists here.

#Usage. Save this file as generate.py
#Then to use, do generate.py 867 (if 867 is the number for
#which you want to generate letters)
from cStringIO import StringIO
import random
from sys import argv
from time import strftime
mstr=str(argv[1])
def genString():
        numsounds = [['s','z'], ['t','th','d'],['n'],['m'],['r'],\
['l'],['j','ch','sh',],['k','q'],['f','v'],['p','b']]
        return [random.choice(numsounds[int(i)]) for i in mstr]
i=0
myDict={}
while i<10000:
        astr=' '.join(genString())
        if not myDict.has_key(astr):
                myDict[astr]=1
                print '%-20s ==> %2d : %20s' % \
(astr,i,strftime("%Y-%m-%d %H:%M:%S"))
        i=i+1

Photo credit: http://memory.uva.nl/memimprovement/eng/phon_peg_1.htm

Click to enlarge
The market cap of the countries comprising 90% of the world’s market cap (end 2010)

A famous finance professor once told us that good diversification meant holding everything in the world. Fine, but in what proportion?

Suppose you could invest in every country in the world. How much would you invest in each? In a market-capitalization weighted index, you’d invest in each country in proportion to the market value of its investments (its “market capitalization”). As seen above, the market-capitalization of the USA is about 30%, which would suggest investing 30% of one’s portfolio in the USA. Similarly, one would put 8% in China, and so on. All this data was pulled from the World Bank, and at the end of this post we’ll show you how we did it.

What makes life easy is that economists have grouped countries into regions and the market caps of each can be computed by summing the market caps of the constituent countries. See this MSCI page for the manner in which we’ve categorized countries in this post:

North America: USA, Canada
Emerging: Brazil, Chile, China, Colombia, Czech Republic, Egypt, Hungary, India, Indonesia, Korea, Malaysia, Mexico, Morocco, Peru, Philippines, Poland, Russia, South Africa, Taiwan, Thailand, Turkey
Asia: Japan, Australia, Hong Kong, New Zealand, Singapore
Europe: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom
Other: Everybody else in the world

What makes life easier still is that one can purchase ETFs that work like indices tracking the various regions of the world. For example at Vangard one can purchase

• Vanguard Total Stock Market ETF (Ticker VTI), which basically covers North America
• Vanguard MSCI Emerging Marketing ETF (Ticker VWO), which covers emerging markets
• Vanguard MSCI Pacific ETF (Ticker VPL), which covers Asia (minus the Asian emerging markets countries)
• Vanguard MSCI Europe ETF (Ticker VGK), which covers Europe

What makes life a bit more complicated is that not everybody agrees on market-cap weighting. Some favor Gross Domestic Product (GDP) weighting for the reasons given at this MSCI page.

• GDP figures tend to be more stable over time compared to equity markets’ performance-related peaks and troughs
• GDP weighted asset allocation tends to have higher exposure to countries with above average economic growth, such as emerging markets
• GDP weighted indices may underweight countries with relatively high valuation, compared to market-cap weight indices

Here are the countries that make up 90% of the world’s GDP:

Click to enlarge
The GDP of the countries comprising 90% of the world’s GDP (end 2010)

Because we love you, we’ve also computed how the classic market regions compare to one another in terms of GDP

If you track changes in market cap from month to month as we do, you’ll appreciate the stability that GDP weighting provides. For example, the market cap data in this post, from 2010, is already way out of date.

So, weight by market cap, or weight by GDP, or do what we do and weight by the average of market cap and GDP, but don’t wait to diversify your portfolio outside of a handful of investments you happen to own. Invest in the world.

For those who love R and who want to learn how to pull down World Bank data, the code is provided here. I got started with this by borrowing code from this very helpful blog post by Markus Gesmann and tip my hat to its author. Thanks also to Hadley for writing ggplot2 (and for visiting us at Yahoo the other week. And thanks to Yihui Xie for his code formatter.

R CODE

library(ggplot2)
library(car)
library(RJSONIO)
ftse=read.delim("201112_FTSE_Market_Caps.tsv",sep="\t")
##See www.msci.com/products/indices/tools/index.html
##Emerging::Brazil, Chile, China, Colombia, Czech Republic, Egypt, Hungary,
#India, Indonesia, Korea, Malaysia, Mexico, Morocco, Peru, Philippines,
#Poland, Russia, South Africa, Taiwan, Thailand, Turkey
##Europe::Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy,
#Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom
##Asia::Japan, Australia, Hong Kong, New Zealand, Singapore
recodestring=" c('USA','CAN')='North America';
c('BRA','CHL','CHN','COL','CZE','EGY','HUN','IND','IDN','KOR','MYS','MEX','MAR','PER','PHL','POL',
'RUS','ZAF','TWN','THA','TUR')= 'Emerging';
c('AUT','BEL','DNK','FIN','FRA','DEU','GRC','IRL','ITA',
'NLD','NOR','PRT','ESP','SWE','CHE','GBR')= 'Europe';
c('JPN','AUS','HKG','NZL','SGP')= 'Asia';
else='Other'"
#How much of the world do we want to consider (to keep graphs manageable)
CUTOFF=.9
#Some code from internet to read world bank data. Adapted from
#http://www.r-bloggers.com/accessing-and-plotting-world-bank-data-with-r/
#See also http://data.worldbank.org/indicator/NY.GDP.MKTP.CD
getWorldBankData = function(id='NY.GDP.MKTP.CD', date='2010:2011',
value="value", per.page=12000){
require(RJSONIO)
url = paste("http://api.worldbank.org/countries/all/indicators/", id,
"?date=", date, "&format=json&per_page=", per.page,
sep="")
wbData = fromJSON(url)[[2]]
wbData = data.frame(
year = as.numeric(sapply(wbData, "[[", "date")),
value = as.numeric(sapply(wbData, function(x)
ifelse(is.null(x[["value"]]),NA, x[["value"]]))),
country.name = sapply(wbData, function(x) x[["country"]]['value']),
country.id = sapply(wbData, function(x) x[["country"]]['id'])
)
names(wbData)[2] = value
return(wbData)
}
getWorldBankCountries = function(){
require(RJSONIO)
wbCountries =
fromJSON("http://api.worldbank.org/countries?per_page=12000&format=json")
wbCountries = data.frame(t(sapply(wbCountries[[2]], unlist)))
levels(wbCountries$region.value) = gsub("\\(all income levels\\)",
"", levels(wbCountries$region.value))
return(wbCountries)
}
### Get the most recent year with data
curryear = as.numeric(format(Sys.Date(), "%Y"))
years = paste("2010:", curryear, sep="")
##Get GDP for the countries (GDP in Current US Dollars)
GDP = getWorldBankData(id='NY.GDP.MKTP.CD',
date=years,
value="gdp")
#If a year has > 60 NAs it is not ready
while( sum(is.na(subset(GDP,year==curryear)$gdp)) > 60 )
{curryear=curryear-1}
#Keep just the most current year
GDP=subset(GDP,year==curryear);
## Get country mappings
wbCountries = getWorldBankCountries()
## Add regional information
GDP = merge(GDP, wbCountries[c("iso2Code", "region.value", "id")],
by.x="country.id", by.y="iso2Code")
## Filter out the aggregates, NAs, and irrelevant columns
GDP = subset(GDP, !region.value %in% "Aggregates" & !is.na(gdp))
GDP=GDP[,c('year','gdp','country.name','region.value','id')]
GDP=GDP[order(-1*GDP[,2]),]
GDP$prop=with(GDP,gdp/sum(gdp))
GDP$cumsum=cumsum(GDP$gdp)
GDP$cumprop=with(GDP,cumsum(gdp)/sum(gdp))
GDP$market.region=recode(GDP$id,recodestring)
resize.win <- function(Width=6, Height=6)
{
plot.new()
dev.off();
#dev.new(width=6, height=6)
windows(record=TRUE, width=Width, height=Height)
}
slice=max(which(GDP$cumprop<=CUTOFF))
smGDP=GDP[1:slice,]
smGDP$country.name=factor(smGDP$country.name,levels=unique(smGDP$country.name))
smGDP$region.value=factor(smGDP$region.value,levels=unique(smGDP$region.value))
smGDP$market.region=factor(smGDP$market.region,levels=unique(smGDP$market.region))
g=ggplot(smGDP,aes(x=as.factor(country.name),y=prop))
g=g+geom_bar(aes(fill=market.region))
g=g+opts(axis.text.x=theme_text(angle=90, hjust=1))+xlab(NULL)+
ylab( paste("Proportion of global GDP\nTop ",CUTOFF*100,"% countries",sep=""))
g=g+opts(legend.position=c(.8,.8))
resize.win(5,5) #might have to run this twice to get it to shift from the default window
g
ggsave("GDPbyCountry.png",g,dpi=600)
smGDPsum=ddply(GDP,.(market.region),function(x) {sum(x$gdp)})
smGDPsum$V1=smGDPsum$V1/sum(smGDPsum$V1)
smGDPsum=smGDPsum[order(-1*smGDPsum[,2]),]
smGDPsum$market.region=factor(smGDPsum$market.region,levels=c("North America","Emerging","Asia","Europe","Other"))
g=ggplot(smGDPsum,aes(x=as.factor(market.region),y=V1))
g=g+geom_bar(aes(fill=market.region))
g=g+opts(axis.text.x=theme_text(angle=90, hjust=1))+xlab(NULL)+ylab("GDP of Countries by Market Region\n")
g=g+opts(legend.position="none")
resize.win(5,5) #might have to run this twice to get it to shift from the default window
g
ggsave("GDPbyRegion.png",g,dpi=600)
##Get Market Cap data from World Bank CM.MKT.LCAP.CD
### Get the most recent year with data
curryear = as.numeric(format(Sys.Date(), "%Y"))
years = paste("2010:", curryear, sep="")
##Get MarketCap for the countries (MCP in Current US Dollars)
MCP = getWorldBankData(id='CM.MKT.LCAP.CD',
date=years,
value="mcp")
#If a year has > 60 NAs it is not ready
while( sum(is.na(subset(MCP,year==curryear)$mcp)) > 150 )
{curryear=curryear-1}
#Keep just the most current year
MCP=subset(MCP,year==curryear);
## Add regional information
MCP=merge(MCP, wbCountries[c("iso2Code", "region.value", "id")],
by.x="country.id", by.y="iso2Code")
## Filter out the aggregates, NAs, and irrelevant columns
MCP = subset(MCP, !region.value %in% "Aggregates" & !is.na(mcp))
MCP=MCP[,c('year','mcp','country.name','region.value','id')]
MCP=MCP[order(-1*MCP[,2]),]
row.names(MCP)=NULL
MCP$prop=with(MCP,mcp/sum(mcp))
MCP$cumsum=cumsum(MCP$mcp)
MCP$cumprop=with(MCP,cumsum(mcp)/sum(mcp))
MCP$market.region=recode(MCP$id, recodestring)
slice=max(which(MCP$cumprop<=CUTOFF))
smMCP=MCP[1:slice,]
smMCP$country.name=factor(smMCP$country.name,levels=unique(smMCP$country.name))
smMCP$region.value=factor(smMCP$region.value,levels=unique(smMCP$region.value))
smMCP$market.region=factor(smMCP$market.region,levels=unique(smMCP$market.region))
g=ggplot(smMCP,aes(x=as.factor(country.name),y=prop))
g=g+geom_bar(aes(fill=market.region))
g=g+opts(axis.text.x=theme_text(angle=90, hjust=1))+xlab(NULL)+
ylab( paste("Proportion of global Market Cap\nTop ",CUTOFF*100,"% countries",sep=""))
g=g+opts(legend.position=c(.8,.8))
resize.win(5,5) #might have to run this twice to get it to shift from the default window
g
ggsave("MCPbyCountry.png",g,dpi=600)
smMCPsum=ddply(MCP,.(market.region),function(x) {sum(x$mcp)})
smMCPsum$V1=smMCPsum$V1/sum(smMCPsum$V1)
smMCPsum=smMCPsum[order(-1*smMCPsum[,2]),]
smMCPsum$market.region=factor(smMCPsum$market.region,levels=c("North America","Emerging","Asia","Europe","Other"))
g=ggplot(smMCPsum,aes(x=as.factor(market.region),y=V1))
g=g+geom_bar(aes(fill=market.region))
g=g+opts(axis.text.x=theme_text(angle=90, hjust=1))+xlab(NULL)+ylab("Market Cap by Market Region\n")
g=g+opts(legend.position="none")
resize.win(5,5) #might have to run this twice to get it to shift from the default window
g
ggsave("MCPbyRegion.png",g,dpi=600)


Decision Science News has posted before on Zhao, Lynch, and Chen’s practical article on mediation analysis. John Lynch has written the following, re-emphasizing the article’s main points:

Meaningless Mediation
John G. Lynch, Jr., University of Colorado
January, 2011

In August of 2010, JCR [Journal of Consumer Research - Ed] published an invited paper by Zhao, Lynch and Chen on common abuses of mediation analysis.

Zhao, Xinshu, John G. Lynch, Jr., and Qimei Chen (2010), “Reconsidering Baron and Kenny: Myths and Truths about Mediation Analysis,” Journal of Consumer Research, 37 (August), 197-206.

In a note accompanying the paper, the editor suggested that authors either follow its recommendations or take them into account if they chose to use an alternative approach. The paper made four points. As I observe how the paper is being used and adopted by JCR authors and authors at other journals, the least original of our recommendations is the most widely adopted, so in this note I want to restate the recommendations in order of importance.

1. Consider the discriminant validity of the mediator. Our single most important point is stated on the last page of Zhao et al. Many, many reports of mediation tests in consumer research and psychology are utterly meaningless because the authors have not demonstrated that the mediator is distinct from the independent variable or the dependent variable. When it is not distinct, the data will appear to support “full mediation” in Baron and Kenny’s terms and “indirect only” mediation in the parlance of Zhao et al.

A great many meaningless mediations are published in leading journals in which the mediator M is essentially a manipulation check (and hence, no discriminant validity from X) or an alternative measure of the conceptual dependent variable (and hence, no discriminant validity from Y). Some reviewers looking for any evidence of process may give “partial credit” for even meaningless mediations; this would encourage defensive insertion of meaningless mediation analyses by authors. We could save a lot of page space by deleting reports of these mediation results from the pages of JCR, JMR, and JCP. Until very recently, I have not seen much evidence that the Zhao et al. paper has had any deterrent effect on this error.

2. Embrace partial mediation and use unexpected “direct” effects to stimulate theorizing about omitted mediators. Our second most important point was that X-Y relationships are likely to have multiple mediators, and we researchers are usually not smart enough to test for more than one. In that case, it is likely that the data will sometimes indicate “indirect only” mediation (or “full mediation in Baron and Kenny’s terms), but more often will support either the “competitive mediation” or “complementary mediation” outlined by Zhao et al. Here, an unexplained direct effect of X on Y accompanies a significant indirect effect X – M – Y as posited by the researchers. Followers of Baron and Kenny viewed those direct effects with mild embarrassment. We pointed out that the sign of the direct effect can often be a hint to the sign of some omitted mediator. I should note that model misspecification and omitted variable bias can lurk as easily in data that seem to be consistent with “indirect only” (“full”) mediation as in data where there is an unexplained direct effect. The great advantage of the latter case is that the sign of the direct effect gives the authors some tip that there is more to learn, and a hint of what direction to look – for omitted indirect paths matching the sign of the “direct” effect. Write to me for an easy-to-understand example of “indirect only” results hiding omitted variable bias due to an omitted second mediator.

3. Test only for the indirect effect X – M – Y and not for an “effect to be mediated.” The Baron and Kenny procedure required that authors show a significant zero order effect of X on Y to establish “an effect to be mediated.” We showed that this effect is algebraically equivalent to the “total effect” of X on Y: the sum of the indirect effect of X on Y through M and the direct effect of X on Y. We noted that this total effect test is meaningless or superfluous. If the signs of the direct and indirect effects are opposite, it is easy to fail to observe an “effect to be mediated” or to observe an “effect to be mediated” of the wrong sign despite strong evidence for the posited indirect pathway. If the signs of the direct and indirect effects are the same, the test of the zero order effect of X on Y will always be significant when the indirect effect is significant – hence the test is superfluous here. We pointed out how nonsensical it was to treat a result as publishable when a posited indirect effect matched the sign of an unexplained direct effect, but not in the equally likely case in which the unexplained direct effect was opposite in sign. Ironically, about the time our paper was coming out, I received a rejection from a top journal with an AE report citing, among other failings, the marginal significance of the “effect to be mediated” in one of two replications.

4. Use Preacher and Hayes boostrap instead of Sobel test. The least important and least original point in Zhao et al. is, ironically, the one that seems to have caught on: use bootstrap tests rather than Sobel tests for the indirect effect X-M-Y. This one is a “no brainer.” Bootstrap tests using the very simple-to-use Preacher and Hayes (2008) macro are almost always more powerful than Sobel tests for reasons explained in our paper. There are no published bootstrap tests of mediation of within-subjects effects, where Sobel tests can be used. But in the usual between-subjects case, authors should head to Andrew Hayes website http://www.afhayes.com/spss-sas-and-mplus-macros-and-code.html

Though many consumer researchers have started using bootstrap tests, I have had colleagues tell me that reviewers told them to remove bootstraps tests and replace with Sobel. AEs should be vigilant to contradict such clearly incorrect advice if it appears in JCR reviews.

Though not emphasized in Zhao et al., the other major advantage of the Preacher and Hayes (2008) macro is that it makes it easy to test multiple mediator models.[1] Most published mediation tests consider a single mediator, though we assert in Zhao et al. that most X-Y relations likely have multiple mediators. Authors who are insightful enough to posit dual mediators almost always test each one piecewise using the Baron and Kenny tests we criticized. That’s wrong. With the Preacher and Hayes macro, it takes the same single line of code in SPSS or SAS to specify a multiple mediator model as to specify a single mediator model.

[1] Use MPLUS to analyze latent variable versions of the same multiple mediator models.

This week Decision Science News and Peter McGraw’s blog team up for a joint post. — Enjoy

The New York Times just ran a piece called “A Case Against Helmets in Lacrosse“.

The hook of the article is that wearing helmets, which one would expect to make the game safer, could make the game more dangerous. Let’s review the quotes.

One:

It’s hard to absolutely prove, but what we’ve seen is that behavior can change when athletes feel more protected, especially when it comes to the head and helmets,” said Dr. Margot Putukian, Princeton’s director of athletic medicine services and chairwoman of > the U.S. Lacrosse safety committee. “They tend to put their bodies and heads in danger that they wouldn’t without the protection. And they aren’t as protected as they might think.”

Two:

Then again, other sports have spent the last several years realizing that safety equipment can bring dangers of its own. Checking in professional hockey became considerably more vicious with the adoption of helmets in the 1970s and ’80s, and football players felt so protected by their helmets and face masks that head-to-head collisions became commonplace at every age level.

Three (re: protective eyewear in Women’sLacrosse):

[Someone] said that after the move to make eyewear mandatory for the 2005 season, “It’s subconscious, but you see harder checking, and rougher play.”

Interesting topic! Let’s get Dan and Peter’s take on it:

DAN’S TAKE
This is an example of moral hazard, which the Wikipedia (at least during the last five minutes) defines as a situation in which “a party insulated from risk behaves differently than it would behave if it were fully exposed to the risk.” The top 10 Google Scholar papers with “moral hazard” in the title have over 10,000 collective citations. You hear a lot about moral hazards, for example, that people began driving more recklessly when seat belts were invented, at cocktail parties, coffee breaks, dinners with visiting speakers, and other moments in which people say what they really think.

Is it possible that requiring lacrosse players to wear helmets will increase risk to players? I doubt it. My grounds for skepticism? TCTBT: too cute to be true.

TCTBT (also “too clever to be true”) arguments survive not because they are correct, or supported by the best evidence, but because they are elegant, counter-intuitive, and make a person sound smart at a cocktail party. They fly well in Op-Ed pieces, keynote speeches, and other places where one is unlikely to be asked for evidence.

The smart thing to do when you hear a TCTBT explanation is to doubt it. Since repeating a clever explanation is clearly its own reward, how is one to say that the person offering it is well-informed or just trying to be conversationally brilliant?

While I feel that moral hazard is overhyped, I must admit that I’ve gone spelunking with a helmet and without a helmet, and yes, a person does let his head bump against the cave walls more often with the helmet on than off. However, the net impact to your head is less with the helmet on. This is what I would expect to happen in lacrosse. More helmets will bump against helmets, which seems more ‘violent’, but the net noggin impact will be less. And despite popular belief, they have rules in lacrosse, so the temptation to run helmet first into people may not even have an advantage.

Interestingly, the article says, “checking in professional hockey became considerably more vicious with the adoption of helmets in the 1970s and ’80s”. Again, correlation, is not causation. And is ‘viciousness’ measured in injuries?

I say let Moral Hazards join Prisoner’s Dilemmas, Tragedies of the Commons, and other cleverly constructed scenarios that don’t arise in proportion to the vast numbers of articles written about them. Pay more attention to boring things like default effects that exert large and demonstrable influences on hundreds of decisions in daily life.

That said, I know nothing about lacrosse except that I watched Pete play it once, so without further ado, I’ll turn the typing over to him. If you have empirical evidence of moral hazards we should be concerned about, please post in the comments.

PETE’S TAKE:
As someone with more than 30 years of (combined) lacrosse coaching and playing experience, my intuition leads me to believe that introducing helmets into the women’s game will increase the behaviors that put players at risk of injury. However, given that I regularly lecture on the fallibility of intuition, I also agree with Dan. I would like to see causal evidence before drawing a line in the sand. More important than just examining if helmets increase risky behavior, the analysis for deciding to institute helmets would need to balance the costs of the risky behavior against the benefits of the helmet. That is, the temporary bumps and bruises caused by more aggressive play may be worth incurring in order to reduce the risk of concussion. In that way, a thorough risk analysis would seem to be worth the price given the many thousands of women and girls who play the game.

The question of wearing a helmet creates a moral hazard in the game of lacrosse is complicated because the presence of the helmet would seem to influence more than just the offensive and defensive player’s behavior. The protection that a helmet provides could influence the way that the official monitors the game. Officials could more laxly (no pun intended) enforce the rules because they perceive the aggressive behavior as less risky, which in turn could further increase aggressiveness.

One last way to think about the helmet debate is to consider how much the debate is being colored by tradition (aka the status quo). And in this way, a useful question would be, if a new sport like lacrosse was created, would helmets be required given the emerging evidence about the risks of concussion?

Photo credit:http://www.flickr.com/photos/timailius/2472012607

This week, our former home, the Center for the Decision Sciences at Columbia University, has turned us on to this article about a fitness plan that charges you more if you work out less. Yes, it’s a business putting applied behavioral economics to work, not unlike Stickk.com or Beeminder.com (formerly known as Kibotzer.com and currently known as bmndr.com by those who like to abbrvt).

The plan works in conjunction with gyms. We have often heard people say, with great assurance, that gyms bet on people not making good use of their memberships, that they are businesses based on a human inability to commit. Sure, if you take “good use” to mean going every day, it’s clear that gyms could not handle the crowds that would result. However, we at DSN often doubt people who say things with great assurance without evidence. After all, people don’t make good use of their gym memberships may not renew their memberships, and that is bad for business.

SOFT CONTRACTS: AN IDEA THAT WILL NOT WORK IN THEORY BUT WILL WORK NONETHELESS
DSN once committed to patronize the gym 150 times a year. We did pretty well for the first half of the year and then were met by a long spell of illness and travel. After that, hitting the 150 number required more frequent workouts then we cared to commit to, so we gave up. It’s a problem when commitment contracts don’t make room for periods in which you cannot, should not, or do not want to honor your contract. Yes, we grouped “do not want” with the others. For instance, suppose we had been offered an all-expense-paid, two-month trip around the world that would leave no possibility for going to the gym. Taking the once-in-a-lifetime trip would be more important to us than going to the gym 150 times per year, so we would choose to lose the gym bet to take the trip. However, that’s not ideal. What’s ideal is to honor the gym bet when we are home and healthy and to suspend it otherwise. Since we can define these preferences a priori, it should be possible to define contracts that work within those rules. However, since it is impossible to think, a priori, of every possible thing that might make one want to suspend the bet, it seems reasonable to simply trust people to be honest about when they should be honoring the contract and when not. We realize this violates all the sacred principles of commitment contracts, experimental economics, law & order, and so on, but so what. History is filled with failed token economies and successful systems of soft guidelines.

OTHER DSN POSTS ON COMMITMENT DEVICES
Stop overeating with a turn of the wrist

Contracts to fight procrastination

Decision Science News does not read news often. (We took Herbert Simon’s advice that checking the news every week or so is enough and are much happier since). However, each time we do we see headlines of the following sort:

Want to live longer? Get a grip! (On the correlation between longevity and hand grip strength).

If you want to live longer, then walk faster (On the correlation between longevity and walking speed).

If you’re reading Decision Science News, we don’t have to tell you that there’s not a scrap of evidence in the research cited in these articles that walking faster or giving firmer handshakes makes you live longer. See a target article to see it’s all correlational, not causal.

This is serious. First, it’s saying something that isn’t true. The news shouldn’t do that. They seem to get away with it by virtue of the fact that most people can’t conduct research themselves. (If they lied about testable relationships, e.g., “Want to avoid a ticket? Park on the sidewalk,” people would stop believing them rather quickly). Second, the effects are pervasive. We’ve seen PhDs in every field get suckered by a bogus headline.

(Speaking of headlines, DSN finds it hard to believe that anyone doing science journalism for more than a week wouldn’t fully grasp the correlation/causation distinction, if they didn’t have it already. Thus, we suspect there might be a strange relationship between people who write the stories and the people who write the headlines. We will check with our sister.)

It is handy that there is a Wikipedia article entitled Correlation does not imply causation, but how can people stand a chance against media machines that propagate new stories of this type every day?

A simple step is to prevent these stories from getting credibility with search engines by putting rel = “nofollow” in the URL when we ever have to link to such articles (as we have done above). But admittedly, that’s pretty weak.

Can we think of something better? This is 2011. We’re not at the mercy of a few media giants anymore. Online, people can exert a ot of collective power. What can be done about this? Maybe a browser plugin (like xmarks) that can overlay ratings on top of hyperlinks? A collective that keeps a kind of blacklist, perhaps punishing a publisher with less traffic each time they post such a headline?

What have you got?

One thing Decision Science News particularly enjoys about being at Yahoo! Research is the brilliant colleagues. This week, two of them, Winter Mason and Sid Suri, presented us with this manuscript which is a guide to conducting research on Amazon’s Mechanical Turk.

Manuscript? Manuscript from heaven, we say, for here at DSN we are often being asked the ins and outs of this technology and now have a guide to link to. Read it now before it gets published:

A Guide to Conducting Behavioral Research on Amazon’s Mechanical Turk

See Decision Science News’ earlier posts on MT:

• Amazon’s Mechanical Turk meets classic JDM experiments
• How to run experiments on Mechanical Turk

Photo credit: The R language for statistical computing

A friend of Decision Science News, who is co-organizing a session on JDM (judgment and decision making research) for students, recently emailed a handful of JDM researchers:

Those of us in the JDM session are doing quite different research and couldn’t really see how we were more “JDM” than, say, someone doing “cognition”, which lead us to the question “What is JDM?”

If you have a few minutes in the next couple days to just shoot me a note about what makes JDM distinct, I’d really appreciate your thoughts. My goal is to give students a couple different (anonymous, of course) opinions about what JDM is from people more senior than those of us in the session.

Here is the opinion that Decision Science News gave:

This post from Decision Science News, based on a text analysis of conference programs, gives some insight into how what is currently being done in JDM is distinct from Social Psych

http://www.decisionsciencenews.com/2010/02/15/the-difference-between-spsp-and-sjdm/

Also, the first list does a pretty good job of showing the core topics of JDM: risk, uncertainty, choice, decision, probability, prediction, future, intertemporal choice. Missing from the list would be: heuristics, utility, forecasting, normative models, prescriptive models, and descriptive models.

The Society for Judgment and Decision Making (SJDM) was formally formed in 1986 (from a core who had been meeting less formally before that) and I’ve heard it was basically people interested in the exciting field of research opened up by Tversky & Kahneman. Their 1974 Science article still touches upon much of what is done today.

The oldest President’s letter to be found online, written by Barbara Mellers in 1996, speaks of “almost five decades” of JDM research, which would point to somewhere in the late 1940s. Well after Brunswik, a few years after Von Neumann and Morgenstern’s “Theory Games and Economic Behavior” and a few year’s before Ward Edward’s Psychological Bulletin article “The theory of decision making”, the abstract of which is (emphasis added):

This literature review of decision making (how people make choices among desirable alternatives), culled from the disciplines of psychology, economics, and mathematics, covers the theory of riskless choices, the application of the theory of riskless choices to welfare economics, the theory of risky choices, transitivity of choices, and the theory of games and statistical decision functions. The theories surveyed assume rational behavior: individuals have transitive preferences (“… if A is preferred to B, and B is preferred to C, then A is preferred to C.”), choosing from among alternatives in order to “… maximize utility or expected utility.”

And Meller’s President’s letter (emphasis added) describes what she saw as the big topics (in addition to her opinions about the focus, which we won’t touch upon here):

For almost five decades, researchers in judgment and decision making have explored human errors in judgment and choice. We have documented instances in which people violate fundamental principles and axioms. We have discovered cases in which people disobey the most basic rules of statistics, probability, and logic. We have identified factors that should be irrelevant, but aren’t, such as the response mode, the problem representation, and the decision frame.

What are the legacies of this research? We have probed the boundaries of human rationality. We have discovered important limitations of cognitive processing, and we understand how poor judgment makes people their own worst enemies. But somewhere along the way, we lost sight of everything else.

While walking across campus to a colloquium one afternoon, a colleague asked me whether the speaker was a member of the JDM Society. When I told him “yes,” he said, “Then give me a quick preview. What is the error of the day?” He was perfectly serious. We are well known for setting traps and taking delight at human failure.

Haven’t we reached the point of diminishing returns? Demonstrations of one more error for the sake of an error, or one more violation for the sake of a violation, are nothing new. Not only are they not new, they add to an already lopsided view of human competence. We need theories of decision making that predict not only errors, biases, and violations of axioms, but also broader themes of psychological and social functioning. We know very little about the effects of emotions on choice. We know very little about the relationships between decision making and signal detection, memory retrieval, or categorization. Not only that, we know very little about the impact of social context. Why are certain errors, and not others, attenuated in experimental markets, and possibly other institutional settings?

One of the reasons we may have become so preoccupied with errors is because we applied to our descriptive theories the organizing principles from our normative theories. In normative theories, we classify decisions depending on the assignment of probabilities to states of nature (decision making under certainty, risk, uncertainty, or conflict), and these categories may not be optimal for descriptive theorizing. In the animal literature, decisions are often classified on the basis of the animal’s activities, such as foraging and mating. Perhaps functional distinctions might be appropriate in the human literature as well. How often have you heard complaints that our theories apply to purchasing decisions, but not decisions about marriage or children? How often have you heard complaints that our theories of gambles don’t generalize to medical treatments, job opportunities, or even vacation sites? Perhaps the missing links in our descriptive theories would become more apparent with a different set of organizing principles that highlight our activities, goals, and desires.

We have gotten a great deal of mileage out of errors. Decision making is discussed in many psychology texts. It is also cited in marketing, organizational behavior, political science, and microeconomics texts. Philosophers, economists, and statisticians are also developing richer and more interesting definitions of rationality. Finally, psychologists have begun to study human strengths as well as human weakness, and this work should have important consequences for artificial intelligence systems designed to complement and aid human decision making.

To have a lasting impact, we should continue to go beyond errors, mistakes, and other human failures and adopt a broader perspective. As John Locke said, “It is one thing to show a man that he is in error, and another to put him in possession of the truth.”

The point is, for better or for worse, the majority of JDM research has always been about the difference between formalisms and human behavior. The formalisms are drawn from economics, mathematics, and psychology as Edwards said, and I’d guess that the following list of formal models (with examples of JDM research areas in parens) is close to complete:

• probability (base-rate neglect / conservatism, confidence),
• logic (Wason selection task),
• subjective expected utility (Prospect Theory, Support Theory),
• choice axioms (Independence of irrelevant alternatives, attraction / compromise effects)
• statistics
  • sampling (Representativeness, law of small numbers, probability weighting, decisions from experience)
  • inference (lens model, fast&frugal heuristics)
  • estimation (Availability, Anchoring, risk perception)

Outside of this, there is a bit of descriptive work (Naturalistic DM, individual differences) and a bit of prescriptive work, though the latter is usually taken up in the field known as Decision Analysis. Like Mellers quite a few JDM researchers have not been happy with the organization around axiomatic norms, but if we are to define JDM by what it is has primarily been in the past, this generalization is hard to deny.

Since Meller’s letter, attention has moved from documenting differences to building more formal models of what people do, with Prospect Theory being the field’s most successful export.

As to the differences with Social Psychology, I think the blog post above addresses the differences in current practice.

As to the differences with Cognitive Psychology, Barsalou’s textbook puts JDM as a field within Cognitive Psychology and I think this is right: judging, choosing, and deciding are thought processes. Cognitive Psych is defined as covering perception, memory, thinking, language, and problem solving. Barsalou’s chapters are: categorization, representation, executive control, working memory, long-term memory, knowledge, language structure, language process, and thought. JDM typically falls under “thinking” / “thought”.

If forced to choose two books that represent what the field is about, I’d go with:

• Rational Choice In An Uncertain World, by Reid Hastie and Robyn M. Dawes
• Judgment under Uncertainty: Heuristics and Biases, edited by Daniel Kahneman, Paul Slovic, Amos Tversky

Photo credit:http://www.flickr.com/photos/captkodak/272746539/

Defaults are settings or choices that apply to individuals who do not take active steps to change them (Brown & Krishna, 2004). Collections of default settings, or “default configurations” determine the way products, services, or policies are initially encountered by consumers, while “reuse defaults” come into play with subsequent uses of a product. At the finest level, a single question can have “choice option default”, which on electronic forms can take the shape of a pre-checked box (Johnson, Bellman, and Lohse, 2002).

Defaults have been shown to have strong effects on real-world choices in domains including investment (Cronqvist & Thaler, 2004; Madrian & Shea, 2001), insurance (Johnson et al, 2003), organ donation (Johnson & Goldstein, 2004), marketing (Goldstein et al, 2008) and beyond.

They have a wide appeal among marketers and policy makers in that they guide choice while at the same time preserving freedom to choose. They are often regarded as the prototypical instruments of libertarian paternalism (Sunstein & Thaler, 2003).

Through default-setting policies, choice architects exhibit influence over resulting choices. The palette of policies includes simple defaults (choosing one default for all audiences), random defaults (assigning a configuration at random, for instance, as an experiment), forced choice (withholding the product or service by default, and releasing it only after an active choice is made), and sensory defaults (those that change according to what can be inferred about the user, for example, web sites that change language based on the visitor’s IP address).

Products and services that are re-used can also avail themselves of persistent or reverting defaults (which, respectively, remember or forget the last changes made to the default configuration) and predictive defaults (which intelligently alter reuse defaults based on observation of the user).

Those setting defaults should be aware of the ethical risks involved (Smith, Goldstein & Johnson, 2010). The ethical acceptability of using a default to guide choice has much to do with the reason why the default has an effect in the first place. When consumers are aware that defaults may be recommendations in some cases and manipulation attempts in other cases (Brown & Krishna), they exhibit a level of “marketplace metacognition” that suggests they retain autonomy and freedom of choice. However, if defaults are effective because consumers are not aware that they have choices, or because the transaction costs of changing from the default are too high, defaults impinge upon consumer autonomy. An often prudent policy, though not a cure-all, is to set the default to the alternative most people prefer when making an active choice, without time pressure, in the absence of any default. Running an experiment on a sample of the greater population can determine these preferences, and can be done in little time and at a low cost in the age of Internet experimentation (Gosling & Johnson, 2010).

REFERENCES

Brown, Christina L. and Aradhna Krishna (2004), “The Skeptical Shopper: A Metacognitive Account for the Effects of Default Options on Choice,” Journal of Consumer Research, 31 (3), 529-539.
Cronqvist, Henrik and Richard H. Thaler (2004), “Design Choices in Privatized Social Security Systems: Learning from the Swedish Experience,” American Economic Review, 94 (2), 424-428.
Goldstein, Daniel G., Eric J. Johnson, Andreas Herrman, and Mark Heitmann (2008), “Nudge Your Customers Toward Better Choices,” Harvard Business Review, December, 99-105.
Gosling, Samuel D. and John A. Johnson (2010), Advanced methods for conducting online behavioral research. Washington, DC: American Psychological Association.
Johnson, Eric J., Steven Bellman, and Gerald L. Lohse (2002), “Defaults, Framing, and Privacy: Why Opting In Is Not Equal To Opting Out,” Marketing Letters, 13 (1), 5–15.
Johnson, Eric J. and Daniel G. Goldstein (2003), “Do Defaults Save Lives?” Science, 302, 1338-1339.
Johnson, Eric J., John Hershey, Jacqueline Meszaros, and Howard Kunreuther (1993), “Framing, Probability Distortions, and Insurance Decisions,” Journal of Risk and Uncertainty, 7, 35-53.
Madrian, Brigitte C. and Dennis F. Shea, D. F. (2001), “The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior,” Quarterly Journal of Economics, 116 (4), 1149-1187.
Thaler, Richard, Daniel Kahneman and Jack L. Knetsch (1992), “The Endowment Effect, Loss Aversion and Status Quo Bias,” in Richard Thaler, The Winner’s Curse, Princeton: Princeton University Press, 63-78.
Samuelson, William and Richard Zeckhauser (1988), “Status Quo Bias in Decision Making,” Journal of Risk and Uncertainty, 1 (1), 7-59.
Smith, N. Craig, Daniel G. Goldstein, and Eric J. Johnson (2010). Choice without Awareness: Ethical and Policy Implications of Defaults. Working paper.
Sunstein, Cass R. and Richard H. Thaler (2003), “Libertarian Paternalism Is Not an Oxymoron,” The University of Chicago Law Review, 70 (4), 1159-1202.