Published in 2004. Social Science Quarterly 85: 19-30.
Lee Epstein
Andrew D. Martin
In their article, Does Age Matter?, Carp, Manning, and Carroll
group a continuous variable, age, into three categories or
cohorts: 1= under the age of 46; 2=46-64; and 3=65 and older.
When they enter this (now ordinal-level, treated as if it were
interval-level) variable into a logistic regression model
designed to explain judicial decisions in age discrimination
cases (coded 0 if the judge ruled against the plaintiff alleging
age discrimination and 1 if in favor of the plaintiff), it
produces a positive and statistically significant coefficient;
when they enter this same cohort-age variable into a model
designed to explain outcomes in litigation alleging
discrimination based on race or gender, it yields an
insignificant coefficient. Taken together, these results lead the
authors to conclude that age "matters" in precisely those cases
in which they expect it to matter—age discrimination
suits—and that it does not matter in those cases in which
they expect it not to matter—race and gender discrimination
suits.
As it turns out, though, this conclusion—and, more
important, the inference that Carp and his colleagues reach based
on their sample of data—rests entirely on their grouping
decision. Specifically, if we respecify the models in precisely
the same way as the authors but vary the cut-off points on the
age-cohort variable, we find that for age discrimination cases,
only 2 of 23 (plausible) cut-off points produce results in line
with the authors’ expectation.
Click here
for the article (.pdf).
The data we used in this article are available from Bob Carp at the University of Houston.
Please email Lee
Epstein if you are unable to obtain them.