Positive Assortative Matching: Evidence from Sports Data
Author | Antonio Filippin,Jan C. Ours |
Date | 01 July 2015 |
Published date | 01 July 2015 |
DOI | http://doi.org/10.1111/irel.12096 |
Positive Assortative Matching: Evidence from
Sports Data
*
ANTONIO FILIPPIN and JAN C. VAN OURS
It is difficult to establish empirically whether or not there is positive assortative
matching in the labor market. We use longitudinal data from a 24-hour relay mara-
thon in Belluno, Italy, in which participants are affiliated with teams, to study group
dynamics in a manner that closely resembles workers’accessions to and separations
from firms. In our dataset the productivity of the individual agents is measured and
we exploit this by investigating the determinants of accessions, separations, and
assortative matching. We find support for the existence of positive assortative
matching; i.e., better runners moving to better teams in subsequent years.
Introduction
Firms can use hiring and separations as tools to increase productivity and
profits. Firms may hire young, inexperienced workers who have just entered
the labor market or more experienced workers who have quit or been let go
from another job. Separations may occur through layoffs, quits, or retirement.
That there is positive assortative matching, i.e., better workers tend to move to
better firms, looks like a natural consequence. However, this fact is not well
established in the literature. The way firms and workers form a match is the
topic of both theoretical and empirical research. If there are complementarities
in the production function, it is optimal when the best firms match with the
best workers. Then, a random allocation of workers across firms would imply
a loss of output. Similarly, if characteristics of firms and workers are substi-
tutes in production a random allocation is not optimal either.
It is not easy to establish whether or not there is assortative matching and if so
whether it is positive or negative. Originally, studies used wage data to investigate
*The authors’affiliations are, respectively, Department of Economics, University of Milan, Italy, and IZA
(Bonn). Email: antonio.filippin@unimi.it; Department of Economics, CentER, Tilburg University, the Neth-
erlands; Department of Economics, University of Melbourne, Parkville, Australia; CEPR (London); CESifo
(Munich); IZA (Bonn); Email: vanours@uvt.nl.
JEL: J14, J24, J31.
The authors thank the organizers of the San Martino Marathon for making the data available, and Lorenzo
Cappellari and Giovanni Pica as well as participants to the XXXVI SAEe conference (Malaga) for helpful
suggestions. The usual disclaimers apply.
INDUSTRIAL RELATIONS, Vol. 54, No. 3 (July 2015). ©2015 Regents of the University of California
Published by Wiley Periodicals, Inc., 350 Main Street, Malden, MA 02148, USA, and 9600 Garsington
Road, Oxford, OX4 2DQ, UK.
401
the matching between workers and firms. Worker fixed effects and firm fixed
effects are derived from wage equations estimated on matched worker–firm data.
The correlation between the two types of fixed effects is interpreted as being infor-
mative about the existence and the nature of assortative matching. Abowd,
Kramarz, and Margolis (1999) initially used this idea to analyze French worker–
firm data and found that the correlation between the two types of fixed effects is
small or negative. Abowd et al. (2009), applying a similar method to U.S.earnings
data, found that the correlation between worker and firm effects is close to zero.
Andrews et al. (2008) argued that estimation errors can cause a downward
bias in the estimated correlation between worker and firm fixed effects. An over-
estimate of a worker effect on average leads to an underestimate of a firm fixed
effect. This bias is bigger the fewer workers move between firms. Using German
matched worker–firm data they found that this bias can be considerable but not
sufficiently large to remove the negative correlation between worker and firm
fixed effects. Andrews et al. (2012), also using German data, addressed the
“limited mobility bias”by focusing on firms for which the number of movers is
large and found that there is significant positive assortative matching.
Eeckhout and Kircher (2011) and Lopes de Melo (2013) argued that the
method of Abowd, Kramarz, and Margolis (1999) did not properly measure sort-
ing because wages are not monotone in the firm type. Wages of a given worker
have an inverted U-shape around the optimal allocation. Wages can be lower
not only when the worker matches with a bad firm, but also, more surprisingly,
with a very good firm. The reason is that in the case of complementarities higher
productivity firms pay a very high cost if they match with a bad worker, because
it destroys their opportunity to match with a better one. Hence, they need to be
compensated for a suboptimal match via a lower wage. This implies that the
compensation is the highest when a worker meets the right firm and that the
wage schedule is not monotonically increasing in the firm type. Consequently,
standard firm fixed effects are not correlated with the true type of the firm.
Lopes de Melo (2013) suggested that the correlation between the fixed
effects of workers and the average fixed effects of his/her coworkers should be
used as an indicator for assortative matching. Applying this idea to Brazilian
matched worker–firm data Lopes de Melo (2008) found evidence of sorting.
However, this procedure and the use of wage data in general can only be
informative about the intensity of assortative matching, but not about its sign.
More recently, Bartolucci and Devicienti (2013) exploited the movers in a
matched worker–firm dataset, showing that there is positive assortative match-
ing. They used firms’profit and workers’wage to build a ranking of their
respective types. Relying upon such proxies they found that the direction of
assortative matching is positive, while the opposite would emerge estimating
the fixed effects of a wage equation.
402 / ANTONIO FILIPPIN AND JAN C. VAN OURS
To continue reading
Request your trial