When do you suppose that the following sentences were written:

Should we worry about a computerized creation that plays to our unconscious? How vulnerable are we to these increasingly refined sales pitches?

They come from Michael J. Weiss, on p.25 of his book The Clustering of America. It is a mostly-favorable treatment of the use of big data to sort American zip codes into socioeconomic clusters, to help businesses make better use of direct-mail marketing and local advertising. The data also were used by political organizations to target efforts to get out the vote, solicit donations, and tailor messages.

The book appeared almost thirty years ago, in 1988. I read it when it first came out, and I recently ordered it so that I could read it again. I also ordered a follow-up book that Weiss wrote in 2000, called Our Clustered World. I will have more to say about the two books when I have finished. I am interested in what they contribute to the project of disaggregating the economy, meaning treating the U.S. as a collection of diverse economies that trade with one another.

One side note: In the late 1990s, when I was running my commercial web site providing information to people who were relocating, we contacted a company that had a similar cluster analysis, in order to enable users to search for particular types of towns. For example, you could select a place where you lived (or wish you lived) near Baltimore and then look for the three most similar towns near, say, Los Angeles. The application would take the socioeconomic cluster that you started with and match you with a part of Los Angeles that had a similar socioeconomic cluster.

The company provided us with their data on a couple of CD’s, and for us, loading it and putting up a front-end that could do the searches the way we wanted was a technical project. Probably the biggest challenge was creating a way to search by town name as well as by zip code.

Shortly after the application went live on the web, I received a very angry note from a Civil Rights organization. The data for each socioeconomic cluster included the two or three consumer items that were purchased much more in that cluster than in other clusters. Our application spat out that information, along with the other data about location. It turned out that one cluster’s unusually strong consumer propensities included fast food fried chicken. Someone evidently had done a search that caused this cluster description to appear and contacted the The Civil Rights group about it. The note that they sent us accused us of stereotyping the location as African-American, so that we were promoting segregation and redlining.

Of course, the company was not using racial stereotyping to speculate on consumer propensities. All of the consumer propensities that the company identified were data driven. If this was a stereotype, it evidently had a basis in reality.

We decided that it was appropriate to edit out that particular example, and just leave in the consumer propensities that did not have any racial connotations. As I recall, we looked in the cluster descriptions for other examples of consumer propensities that might have ethnic connotations, but we did not see any.