I establish four facts regarding the pattern of NLRB supervised representation
election activity over the past 45 years: 1) the quantity of election activity has fallen
sharply and discontinuously since the mid-70’s after increasing between the mid-1950’s
and the mid-1970’s; 2) union success in elections held has declined less sharply, though
continuously, over the entire period; 3) it has always been the case that unions have been
less likely to win NLRB-supervised representation elections in large units than in small
units; and 4) the size-gap in union success-rates has widened substantially over the last forty
years. I develop a simple optimizing model of the union decision to hold a representation
election that can account for the ﬁrst three facts. I provide a pair of competing explanations
for the fourth fact: one based on differential behavior by employers of different sizes and one
purely statistical. I then develop and estimate three empirical models of election outcomes
using data on NLRB elections over the 1952-98 time period in order to determine whether
the simple statistical model can account for the size pattern of union win rates over time.
I conclude that systematic union selection of targets for organization combined with the
purely statistical factors can largely account for the observed patterns.