Is there a "corrected constraint" measure?

Post date: Sep 11, 2014 8:7:48 PM

We recently heard the idea that the constraint measure in UCINET is not identical to the Burt 1992 measure but is instead a "corrected" measure. This is incorrect. The measure used is Burt's. Three theories about why this perception might exist.

First, this paper,

• Borgatti, S. P. 1997. Structural holes: Unpacking Burt’s redundancy measures. Connections, 20(1): 35‑38. [pdf]

is sometimes interpreted as finding errors in Burt's measure. It doesn't. The paper shows that, for binary data, there is a simpler, nicer formula that gives the same results -- for the effective size measure, not constraint. The paper does quibble with a particular example in Burt's book, and it does have a regrettably snarky tone, but does not provide a corrected formula for constraint.

Second, UCINET provides an option for including ego's alters alters in the calculation of constraint. If you were to choose this option -- which is not the default -- you would not be doing what Burt does. Even so, the formula remains unchanged.

Third, occasionally people have noticed that constraint is not constrained to the range 0...1. It can in fact have values larger than 1. Since this seems undesirable, perhaps it was thought that UCINET would have corrected that. It doesn't. The overage does no harm, and it's the way Burt wrote it, so why change it?