View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

April 2013, EB13-04

Economic Brief
Job Search Behavior: Lessons from
Online Job Search
By Marianna Kudlyak and Jessie Romero

While there is a large body of theoretical work about the job search
process, there is relatively little empirical evidence about important
aspects of workers’ search behavior. A new database of online job
posting data sheds light on how workers search for jobs.
Although the labor market has improved
slightly in recent months, conditions are still
weak. The labor force participation rate and
the employment/population ratio have declined dramatically, and the long-term unemployment rate is at historic highs.1

little empirical evidence to answer important
questions such as: Do workers apply for jobs
randomly or systematically? How does search
intensity change with search duration? Do
workers’ reservation wages decline as search

In such a time of protracted labor market
weakness—especially when that weakness
includes very long unemployment spells—it
is especially important to understand how
workers search for jobs and how that behavior
changes over the course of their search. Studying this behavior can provide economists with
clues about the process of matching unemployed workers with vacant jobs.

To begin answering these questions, one of
the authors of this Economic Brief, Marianna
Kudlyak, and two colleagues have used a new
database from an online job search engine to
study how workers’ search behavior changes
during their search tenure.

Search-and-matching models of the labor
market provide a framework for studying this
process.2 These models are based on the idea
that there are “frictions” in the market—that is,
it takes time for workers to locate the right firm
and for firms to locate the right worker. An important concept in the matching process is the
idea of a “reservation wage,” which is the lowest
wage at which a worker is willing to work.
While there is a large body of theoretical work
about the search process, there is relatively

EB13-04 - Federal Reserve Bank of Richmond

Sorting by Education
Is job search random or directed? In a working
paper with Damba Lkhagvasuren of Concordia
University and CIREQ and Roman Sysuyev of the
National Exchange Carrier Association, Kudlyak
looks at whether or not workers with different
education levels apply for different jobs and
how the types of jobs they apply for change
with search duration.4
The data come from a private online job search
engine. They include daily records of all the applications a given job seeker sends to job postings on the site and all the applications that were
received for a given job between September

Page 1

2010 and September 2011. Kudlyak and her coauthors focus on the search behavior of workers
aged 25–64 and workers whose education level
ranges from high school completion to a master’s
degree. On average, younger workers apply to more
jobs; workers aged 25–34 send out 1.69 applications
per day, while workers aged 55–64 send out 1.34
applications per day. Younger workers also have
shorter search durations than older workers. To the
extent that the duration of search on the website
coincides with the duration of unemployment, this
is consistent with evidence that shows older workers tend to have longer unemployment spells.5 Of
course, the cessation of activity on the site does not
necessarily indicate that the worker has found a job;
he or she could be looking elsewhere or have given
up entirely.
To begin, the authors look at the distribution of job
applicants with different education levels across different job postings. They find that the distribution is
not the same across all jobs; some jobs have a higher
share of applicants with only high school diplomas,
for example, while others have a higher share of
applicants with college degrees. Formal statistical
tests reveal that the average education of applicants
in their first week of search on the website differs
from job to job. This suggests that at the beginning
of their searches, jobseekers sort themselves to jobs
according to education.
Based on this finding, the authors construct an
“education index” for each job, which is the average
education level of all workers who apply to that job
during their first week of searching. (The assumption is that workers apply to the jobs they find most
desirable at the beginning of their search.) The authors call jobs with higher index values—meaning
applicants have higher average education levels—
higher-type jobs. They call jobs with lower index
values lower-type jobs.
Kudlyak, Lkhagvasuren, and Sysuyev then look at
the jobseekers’ behavior to see if their degree of
sorting changes as their searches continue. At the
beginning of the process, jobseekers with more

education apply to higher-type jobs, and those with
less education apply to lower-type jobs. As their
searches continue, however, the degree of sorting
by education decreases, and workers apply to more
lower-type jobs than they did at the beginning.
These results imply that people become less choosy
the longer they have been searching for a job. More
precisely, the education index represents a tradeoff
between the expected wage and the probability
of getting a job; initially, workers apply to highertype jobs, which are likely to pay higher wages. As
their searches continue, however, workers apply to
lower-type jobs for which they have a better chance
of being hired. The fact that there is no single job
that workers of all education levels want to apply
for suggests that this tradeoff is important: jobseekers weigh not only the expected wages but also the
probability of being hired when deciding where
to apply.
The findings are related to the existing literature on
assortative matching, which examines the degree
to which the most productive workers match up
with the most productive firms.6 The findings suggest that observed firm-worker matches likely are
“mismatched” compared to those in a frictionless
labor market where workers find jobs right away.
The results also suggest that the private cost of job
search—that is, the cost of being unemployed—
increases with search duration. Both of these facts
may have important implications for the design of
models used to study labor markets.
Search Effort
Workers might apply to different types of jobs depending on how long they have been searching, but
do they apply to more jobs or fewer jobs? In a forthcoming paper, Kudlyak and Jason Faberman of the
Chicago Fed use the online job posting data to study
how search effort changes with search duration.7
Theoretically, search effort could follow several different patterns. It could remain constant throughout
the search. It could increase right before unemployment benefits expire, or it could decline as the search

Page 2

continues. With respect to the third case, there are
several reasons why observed search effort might
decline. One possibility is that the composition of
jobseekers changes: the workers who exert the most
effort find jobs sooner, leaving behind workers who
exert less effort. In this case, any given individual’s
effort might remain constant, but the effort of the
group as a whole appears to decline. Another explanation could be that there is a “stock-flow” effect,
in which the worker’s search effort depends on the
demand for labor. At the beginning of a search, a
jobseeker searches through the full stock of vacancies. After that, however, the jobseeker only looks
at newly posted jobs, so the level of search effort
depends on the size of the flow. Finally, search effort
could decline because the jobseeker’s reservation
wage has declined, which lowers the value of searching. While on the one hand a lower reservation wage
should increase the number of jobs a worker is willing
to accept, on the other hand those jobs also would be
worth less to the worker.
Some studies using cross-sectional data (which
contain one observation per worker and thus allow
comparing groups of workers at different durations
of search) have concluded that a worker’s reservation
wage remains constant throughout the search. This
result, however, could be due to the fact that workers with low reservation wages find jobs earlier, and
thus are not observed in the data at longer search
durations. A unique feature of the online job search
data set is that the researchers are able to follow the
same worker through time, rather than comparing
reservation wages of different workers at different
durations. Faberman and Kudlyak find that search
effort, as measured by the number of applications
sent out by jobseekers, declines as search continues,
even after controlling for composition and stock-flow
effects. Thus, it’s possible that the decrease in search
effort reflects the perceived lower value of the job
to the jobseeker or other factors that Faberman and
Kudlyak are continuing to study.
These studies use a large, novel database from an
online job search engine to study workers’ search

behavior. The results suggest that workers direct
their search based on education. The degree of
sorting decreases as search tenure increases, however, as does search effort. It appears that workers
become more willing to accept job offers at lower
wage rates after they have been searching for
a while.
The matching process involves both inputs and
outputs: the input is workers’ search effort, and the
output is getting hired at a certain wage. The online job search data shed light on important aspects
of the first half of the equation, but do not reveal
what happens to workers when their searches conclude. These outcomes are an important topic for
future research.
Marianna Kudlyak is an economist and Jessie
Romero is an economics writer in the Research
Department at the Federal Reserve Bank
of Richmond.

The unemployment rate has been below 8 percent for the
past six months after remaining stuck around 9 percent for
almost three years. But the labor force participation rate has
declined 2.5 percentage points since the beginning of the
recession, and the employment/population ratio has declined
4 percentage points. The long-term unemployment rate—the
share of unemployed workers who have been unemployed
for more than half a year—has been near or above 40 percent
since November 2009.


For an overview of search-and-matching models, see Diamond,
Peter, “Unemployment, Vacancies, Wages,” American Economic
Review, June 2011, vol. 101, no. 4, pp. 1045–1072; and
Mortensen, Dale T. , “Markets with Search Frictions and the DMP
Model,” American Economic Review, June 2011, vol. 101, no. 4,
pp. 1073–1091.


Emerging literature in this area includes Krueger, Alan B.,
and Andreas Mueller, “Job Search and Job Finding in a Period
of Mass Unemployment: Evidence from High-Frequency
Longitudinal Data,” CEPS Working Paper No. 215, January
2011; and Mukoyama, Toshihiko, Christina Patterson, and
Aysegul Sahin, “Job Search Behavior over the Business Cycle,”
Manuscript, February 2013.


Kudlyak, Marianna, Damba Lkhagvasuren, and Roman
Sysuyev, “Sorting by Skill over the Course of Job Search,”

Page 3

Federal Reserve Bank of Richmond Working Paper No. 12-03,
April 2012.

For example, according to the Bureau of Labor Statistics,
in February 2013 the average duration of unemployment
for workers aged 25–34 was 32.6 weeks, compared to 45.1
weeks for workers aged 45–54 and 45.6 weeks for workers
aged 55–64.


For example, see Eeckhout, Jan, and Philipp Kircher,
“Identifying Sorting—In Theory,” Review of Economic Studies,
July 2011, vol. 78, no. 3, pp. 872–906.


Faberman, Jason, and Marianna Kudlyak, “The Intensity of Job
Search and Search Duration,” Manuscript, November 2012.

This article may be photocopied or reprinted in its
entirety. Please credit the authors, source, and the
Federal Reserve Bank of Richmond and include
the italicized statement below.
Views expressed in this article are those of the authors
and not necessarily those of the Federal Reserve Bank
of Richmond or the Federal Reserve System.

Richmond Baltimore Charlotte

Page 4