The Indiana Jones of Economics, Part II
In the second installment of his adventure story about searching for the elusive Giffen good, Robert Jensen describes some of the setbacks they suffered along the way.
Raiders of the Lost Arc Elasticity, Part II
By Robert Jensen
Let me start at the beginning to explain how our search for a Giffen good evolved. About five years ago, I was using a large, publicly available data set of Chinese households to explore the link between income and health. My colleague Nolan Miller walked into my office, saw what I was doing and asked, half as a joke, if I’d looked for a Giffen good.
I looked at my data, and sure enough, there it was. Higher rice prices in southern provinces of China were associated with higher rice consumption. The same held in northern provinces with wheat (things like noodles). We giggled like idiots, and quickly wrote up the results in a short paper.
But then the ground began to rumble as a giant boulder rolled towards us: The Identification Problem. Readers of this blog will know this problem well, but I’ll describe how it applies here.
Remember, we’re looking for a positive correlation between price and consumption/demand — higher prices associated with higher quantity demanded, lower prices with lower quantity demanded. So, let’s say we see a bunch of towns, and people living in those towns with the highest rice prices consume the most rice. Case closed, right?
Not quite. Plain old economics tells us that if people want more of some good, its price goes up. So, we see high rice prices where there is high rice consumption, but did the high consumption cause the high price (economics as usual) or did the high price cause the high consumption (Giffen behavior)?
The usual solution to this problem is to find some outside factor that affects the price but does not affect demand (except through price), and then in effect look at how the change in price associated only with this factor affects demand.
So we tried matching the data to rainfall records (rainfall affects crop yields, and thus price). After spending months and months on this, there just wasn’t enough data to estimate the relationship well, so the procedure failed. And anyway, we began to think this wasn’t a valid strategy, since rainfall could also affect the demand for rice (by affecting wages, the prices of other goods besides rice, and a host of other factors).
So, we were thwarted — we had some evidence of Giffen behavior, and it was sort of believable, but not quite.
We were on the verge of tossing out the whole project when it hit us: why not go to China, give people subsidies to change the price they pay for foods and see what happens? This way, we would know that our price change caused the change in consumption, and the identification problem would be solved.
So we did just that. We teamed up with a respected Chinese economist, Sangui Wang, and I headed off to China, fedora on my head and trusty whip by my side (OK, Yankees cap and laptop).
Together, we set out across China, through deserts and forests, from half-empty villages to bustling cities, searching for clues by interviewing the poorest people about their diets. Theory told us where to look for Giffen behavior — the “Giffen conditions” included: households that were extremely poor, and consuming a simple diet of primarily a basic good (like rice or wheat), a little bit of a fancy good (like meat), and little else of budgetary or nutritional significance.
We chose one southern province, Hunan, where rice was the dominant staple, and one northern province, Gansu, where wheat was dominant. It became clear early on in our travels that the Giffen conditions seemed to hold almost everywhere we went. Thus, theory told us these guys should be Giffen consumers, so it was time to put the theory to the test.
We chose our sample sites and assembled a data collection team. We printed vouchers that households could use to purchase rice (in Hunan) and wheat (in Gansu) at subsidized rates for six months, and contracted with local shopkeepers to honor the vouchers in exchange for reimbursement. We gathered a pre-subsidy baseline survey for 1,300 households, chose at random which households would receive the vouchers, and returned for two more rounds of data collection, one during the subsidy period, and one after the subsidy had ended.
All that was left to do was wait. This made for a very tense half year.
As the months passed and the program ended, I’d wake up every morning and run to my computer to check if the data had come in. I knew it would take a while for our local team in China to enter and validate the data, but I couldn’t help checking over and over, like a rat pressing a lever hoping for a pellet of food.
And then one day, there it was, the e-mail from our team with a file attached. This was the culmination of a long journey, over five years of our lives, with our academic reputations on the line. Hearts pounding, faces flushed and hands trembling, we ran the first regression …
… and found the exact opposite of what we were looking for.
Next time: What went wrong?