A quiz for econ graduate students, or anyone else with too much time on his or her hands

Usually our quizzes are more on the whimsical side, but here is a serious one that requires a real investment of time.

Some academics (at most one out of the three is an economist) in the United Kingdom recently published a paper in a journal called Applied Economics claiming that a 1 percent increase in the price of beer in the U.K. leads to a 2 percent decline in violent crime. I don’t know how to link directly to the paper, but if you go to this Cardiff University website and scroll down, you will find a link called “Violence-related injury and the price of beer in England and Wales.” That link will take you to the paper.

The paper has sensible theory behind it, is well written, and has seemingly careful econometrics. As I read the paper, however, I had the sort of uneasy feeling that Malcolm Gladwell, in the first chapter of Blink, describes some experts having when viewing a supposedly ancient sculpture that turned out to be a fake. I couldn’t figure out what was wrong with the paper and went on to something else. But I was so bothered by it that I went and read it again. I think on that second reading I found the fundamental flaw in the paper.

So, here is a different kind of quiz. The first one to identify the problem I have in mind wins a signed copy of Freakonomics, a Freakonomics t-shirt if the new batch really exists, and also a Freakonomics yo-yo if those are ready. If they are so inclined, I further encourage the winner to try to get the original data, test my/his/her hypothesis and send a short comment to the journal.

Hint: there may be many other problems with the paper, but my concern is very specific and easily testable.

(Thanks to Sam Richardson for bringing this article to my attention.)


Just have too much time on my hands. Can't seem to get the download though. Any chance of a mirror?


Scratch that. Link works, paper downloaded.


Get the paper directly here: http://www.cardiff.ac.uk/dentistry/research/phacr/violence/pdfs/Beer_violence.pdf


Perhaps there is some unaccounted for covariance between sporting events or the summer season and the price of beer. For example, extra refrigeration costs cause beers to be more expensive in the summer, which happens to be the time the hot weather makes people more irritable.


I would think that you would need to look at the real price of an alcoholic drink as a percentage of the real income (per capita) in the region.


The first flaw I saw was this one:

Selection criteria for inclusion of EDs into the study were: (1) availability of computerized violence data for a five year period starting from May 1995 to April 2000 and, (2) agreement from the hospital ED director to be included in the study.

What I recall from my research methods class is that self-selecting samples are unreliable, and since they only selected ones with computerized data (lazy buggers!), they totally hosed their study.


Is there a problem of stationarity? Trends in rising beer costs (i.e. inflation) vs. general reduction in violence? Not sure if this is somehow accounted for though.


I'm certainly not a graduate econ student, so I had to wade through this as best I could.

I think the previous poster is on to something, I just saw the post, and it goes along with what I've been thinking about problems with the price of beer.

The way they got the "Real Price of Beer" variable seems very fishy to me. It says that "the estimated parameters


My post got cut off...

I'm certainly not a graduate econ student, so I had to wade through this as best I could.

The way they got the "Real Price of Beer" variable seems very fishy to me. It says that "the estimated parameters


Sorry, I am having some posting issues for some reason...

The way they got the "Real Price of Beer" variable seems very fishy to me. It says that "the estimated parameters


The way they got the "Real Price of Beer" variable seems very fishy to me. It says that "the estimated parameters are used to generate a monthly series of regional beer prices which will mimic the seasonal pattern of the ONS UK-wide data. So if the the price of beer is moving based on season, then why are they using dummy variables for Autumn, Winter, and Spring? Wouldn't the price of beer be heavily dependent on season? Also, couldn't the price of beer also be vastly different depending on real house price, youth unemployment rate, and ethnic density? It just seems like the price of beer doesn't meet the assumption of independence.


I think the model is missing a supply curve.

The instruments in the paper might deal with the joint determination of violence and alcohol consumption, but the model does not deal with the joint determination of quantity of alcohol demanded and alcohol prices.

I think you need an instrument for price for this to work, unless you thought alcohol prices were exogenous.

The simple test I would run would be a test for orthogonality between the price of alcohol and the error term.


I am listening to the Freaconomics audio CD and today I went through the chapter which discusses the influence of gun control laws towards crime rate.
Does allowing guns decrease the crime rate? I find this question without merit as allowing carrying guns doesn't mean that people will. What if bearing arms at all times was made mandatory?
Well, if I remember correctly there was a country in the recent times that did that. After Chechnya declared its independence from the Russian Federation the Chechen government anticipating an armed conflict mandated all adult males to bear firearms at all times. Russia was busy with its internal power struggle so the war did not break out for some years but the crime rates to plummeted. A Polish newspaper ‘Gazeta Wyborcza' sent a journalist there to investigate why. The members of the underworld gave a simple answer that sounded something like: “ I don't know if he is reaching for a wallet of for the gun that I know he has…”
I read the above in a popular daily newspaper and this was before the first Chechen war in 1994, but it would be very interesting if some statistical data



The way price of beer is figured seems questionable. Taking regional figures for pints (served at pubs, see CAMRA) and then applying Retail Price Index (RPI).

How pints are priced is totally different than the way you price beer in the market.


My quick take also leads me to suspect that price of beer variable. Even at a pub, there are many different prices for different pints of beer -- which one is being used? If it is the cheapest or average pint, then it is likely that these are higher in richer regions. So what you will find is that richer regions have less violent crime -- not all that surprising. Attributing it to the price of alcohol (via a double proxy, some measure of the price of some beer) is a stretch.

The coefficient on that youth unemployment variable also is troubling -- who would have thought that higher youth unemployment would decrease violence? It looks like to me that there is severe multicollinearity, with a bunch of variables standing in for regional per-capita income.


There is something in there that bothers me. One problem is that the variable they are measuring is "beer price". This is not the same as saying when beer prices rise, violence falls. This could simply mean that violence is lower in regions that have higher beer prices to begin with. I was struck by the second largest factor they found -- violence declines as home prices increase. Surely, there is a correlation between housing prices and beer prices. I'm not sure you can separate the two. Have they reversed the analysis to show that beer prices are more correlated than violence than with housing prices?


I think noto is right. To put it another way, the authors use low beer prices as a proxy for high beer consumption. However, it could be that low beer prices are caused by low beer demand, in which case the quantity consumed will also be lower.


The fishiest thing (in my opinion) in that paper is the supposed two-way causation between V(iolence) and A(lcohol consumption).


Incidentally, as evidence that noto is right, note what Levitt writes: "Some academics (at most one out of the three is an economist) in the United Kingdom recently published a paper in a journal called Applied Economics..."

Levitt emphasizes the low number of economists because this supply-demand relationship is one that should be second nature to an economist. An economist shouldn't say, "Well, the price of beer is low, so consumption must be up."


Seasonality is included in their model but an annual variable is missing.

Since the data is from a 5 year period, they should've included some way of de-trending the model, either by including dummy variables per year or regular annual variable from 1 to 5.

This is a critical omission since prices and violence both are variables that have been shown to be correlated with time (yearly).