Michael Naaman and Kenneth Flamm Show that Sub-Regressions can be Statistically Unreliable

Categories: Studies

December 17, 2014 - Michael Naaman and Ken Flamm

Dr. Ken Flamm (photo credit: Steph Swope)

Ken Flamm

Michael Naaman

Michael Naaman

Christensen Associates’ Dr. Michael Naaman and Dr. Kenneth Flamm, a firm affiliate, have co-authored Sub-Regressions In Antitrust Class Certification Can Be Unreliable, which also appears on Law360.com.

Dr. Naaman and Dr. Flamm demonstrate that so-called sub-regressions can be statistically unreliable when based on subsamples in antitrust class certification.  They show that sub-regression models may be estimated with samples too small for the basic statistical properties of consistency and asymptotic normality to apply.  Slicing-and-dicing datasets into tiny subgroups violates these important large sample properties and, as a result, can produce specious evidence of lack of injury within subsets of class members.

Dr. Naaman and Dr. Flamm present a simple coin-flipping example to expose the fundamental methodological flaw. They then describe a simulation that they tailored to a recent antitrust class certification action.  Their simulation demonstrates conclusively that sub-regression techniques applied to too small samples can reach the false conclusion of differing impacts across class members when it is known with perfect certainty that impacts are the same across all class members.