Tuesday, December 1, 2009

Reflections on an old Chinese statistical joke

I've mentioned in the past the value I'm seeing in a design book (Measurement, Design and Analysis, by Pedhazur & Schmelkin) I'm reading for a course. The authors reference a book by H. Zeisel (Say it with figures) who "pointed out that, according to an old Chinese statistical joke, the rate of mortality among people who are visited by a doctor is much higher than among those who are not visited by a doctor."

Reflect for a moment on this joke. Once you get it, think about how this applies to the world of GIPS(R) verifications, when a non-random approach is used.

If the verifier selects only a certain group of composites to review (e.g., "marketed"), might it be quite likely that they will conform with the standards, especially if the firm being verified knows that there's a greater likelihood of only these being checked?

These non-random verifications can be likened to what Pedhazur & Schmelkin refer to as "quasi-experimental designs," "that suffer, to a greater or lesser extent, from serious shortcomings and pitfalls...[and] that utmost circumspection be exercised in the interpretation of the results, and in conclusions...based on them."

Perhaps I'm beginning to sound like a broken record (whatever a "record" is), but by continuing to periodically bring this subject up I am hopeful that the GIPS Verification Subcommittee will take action to come out in opposition to such practices as they are fraught with problems.

1 comment:

  1. Dave... I couldn't agree more. With the CFA institute focusing on a standard forcing disclosure of being (or not being) verified, the value and credibility verification comes into question. The Nowel Group has firsthand experience in analysing the data of a firm (not to me mentioned) which has been consistently "verified" by a well known verifier (not to be mentioned), and found to be terribly OUT of comppliance. Giving the verifier the "benefit of the doubt", it most likely caused by only doing a sampling within the "marketed" compposites and using the "ostrich approach" on everything else. Before forcing a disclosure on the issue of verification, perhaps the CFA should initiate a process to actually establish standards for verification, a process to "test" compliance to these standards (verify the verifier), and a "license" to perform verifications which can be revolked for non-compliance. Then verification might actually have value and mean something. Otherwise forcing disclosure of an event that may have little or no value only serves to make the CFA look ridiculous.


Note: Only a member of this blog may post a comment.