August 11, 2005

Sampling for Country of Origin

On a Motley Fool forum, Shelbyboy is inviting us to undertake a distributed, nonrandomized survey (rr) of the country of origin of Wal-Mart goods:

I think it would be interesting to conduct a non-scientific random survey of products in a Wal-Mart Super Center to see where they are manufactured.

To that end, I'm proposing a group effort.

This is a great idea, and very interesting (I'd also write down the prices). But it may or may not be reliable. If they could get 20 people to count each department in different stores across the US, and then aggregate the results, I might believe the data are reliable.

But if one person is doing the analysis in one store, and if he breaks apart the Wal-Mart universe by department, he's still left with the problem of how to select goods within each department in an unbiased fashion. Since it has little formal layout, and an unavoidable bias to counting the goods most accessible, which might be correlated with country of origin, the "pick the first X" sampling method is not sufficiently randomized, yielding potentially useless and misleading results.

So how can we construct a reliable survey for one person/group to undertake in one store? In order to do this right, we would need a frame (a list that covers all goods sold in Wal-mart), from which the sampled items will be chosen in some systematic but random way.

Since a list of all individual products is not available, we could cluster-sample:, i.e. layout the floorplan of an entire store in a M-by-N sized grid, and randomly select K grid spots, counting ALL of the items in each grid. To ensure good coverage of all departments, we could require that all regions be included by at least one of the K spots.

(Note that this method will probably ignore services sold by Wal-Mart, like oil changes and vision testing, and will does not include gasoline, newspapers, and any items with no "Made in ..." label).

As you can guess, this process would be very difficult and time consuming, but it would be hard for me to accept seriously the results of a country-counting survey.

Posted by Kevin on August, 11 2005 at 11:12 AM