Previous Section
What I'd like to do now is essentially do some kind of regression model where we're thinking about the relationship between log of movement and price.
We measure the sensitivities (also known as price elasticity) if we standardize them. We come up with an average price sensitivity measure, (for a formal definition click here) the average price elasticity, and what I'm trying to do here is to illustrate the geographic context of the problem. Each of these individual thermometers is a store. Here you've got a store in Schaumberg, here's a store in Evanston, and here's one in Julliett. There are 80 different symbols on this -- the scale of the thermometer denotes the intensity of price sensitivity. You see some thermometers that are unfilled and other thermometers that are completely filled. The thermometers that aren't filled indicate those stores that are very price-sensitive. The thermometers that are filled up all the way denote those stores are really price-insensitive, not sensitive to price changes at all.
I'd like to go back and look at the geographic data. Let's go to Rob's point, that if you are looking at this and know anything about Chicago, you know that in the city people tend to have lower incomes, while in the suburbs people generally have higher incomes. On top of that, different suburbs look different. You get some suburbs that have been around for 50 or 100 years, they tend to be blue-collar workers, and you have other suburbs that tend to be white-collar suburbs. Just one indication of that would be to look at the percent of adults with college education. In some suburbs, like Evanston, close to 60% of the adults have a college education. In other suburbs, if you look in Chicago on the South Side, about 2% of the adults have a college education. The point is that you see a lot of diversity.
The point is, I could do a lot of things just to come up with some kind of estimate of whether the store is sensitive or not. The point is that now you see some of things matching up pretty well. Down here, for instance, you see that low college education results in stores that are very sensitive to price changes. In some of these northern suburbs, you see that in these stores of high education levels the indication is that people are not going to be price sensitive. The result we keep going after is this - first, since you know it's going to be important, let's measure price sensitivity. Then after we've done that, we ask, can we go back and improve this? Because we do have some good reasons to expect that price elasticity isn't just some random variable. And the fact that I'm college educated, I have a higher income and I've got a car, it implies something about my taste and my preferences.
We've got a lot of demographic variables. It's not just college education, it's also measurements of how many elderly people are in the area, ethnicity (which is a measurement of how many blacks and Hispanics), the log of the median income, and family sizes, for example. In this case, let me just read this off. It means that on the average in Chicago, 36% of households have working women. You know the standard deviation is around .05, so you know it's going something between 20 and 50 or 60 percent. Household value over 150 thousand, and some of these standard deviations are low - others are quite large. So if you think about the dispersion across these stores, there's actual dispersion in the housing market. It goes back to the question, can you really do this at the store level? Well, obviously, at the individual household level there's a lot of differences. But people tend to cluster with people that are similar to themselves. There are good reasons for that, and I think that's just trying to illustrate this fact.
Now I'll talk about the last four demographic independent variables that we have: our measurements of the distances between the stores and relative volume of the competitor's store compared to my overall sales. So the distance is measured in miles -- on average the distance to the nearest warehouse competitor like a Club store or a large warehouse store more similar to a Wal-mart, would be about 6 miles. Whereas on average the distance to my nearest competitor or actually a weighted average of my five competitors, is going to be about 2 miles. Now some stores are right next to each other, other stores are far away, so again you are just trying to capture the notion of what do I know about the market.
I want to pause here, because this sort of completes the first segment that I wanted to go through and explain what this data looked like to try to give you some flavor of what the problem is. Now we start thinking about how we are going to model demand. So our main question is about the data or about the problem itself.
For more description of competitive/demographic characteristics click here
Go to written version of paper