∗

Online Job Search and Matching Quality

Constantin Mang
Ifo Institute †

May 14, 2012

PRELIMINARY DRAFT

Abstract
The Internet has fundamentally changed the way workers and firms
are matched on the job market. Compared to newspapers and other
traditional employment resources, online job boards presumably lead to
better matches by providing a wider choice of job advertisements and more
sophisticated methods to find suitable vacancies. This study investigates
the association of online job search and matching quality using individual-
level data from the German Socio-Economic Panel (SOEP). Our results
show that job changers who found their new job online are better matched
than their counterparts who found their new job through newspapers,
friends, job agencies and other channels.

JEL classification: L86, J64

Keywords: Internet, job search, matching quality

∗ Comments by Oliver Falck, Ludger Woessmann and participants at the ITS Conference

in Budapest and the ICTNET workshop in Mannheim are gratefully acknowledged. The
research underlying this paper has been partially supported by Deutsche Telekom AG. The
views expressed are those of the author and do not necessarily reflect the views of Deutsche
Telekom AG.
† Ifo Institute, Poschingerstr. 5, 81679 Munich, Germany. Email: [email protected]
1 Introduction
An increasing number of people use the Internet to look for new jobs. One rea-
son why online job search became so popular is that it considerably changed the
search process. Employment websites like Monster.com allow job seekers to ac-
cess thousands of job offers and use intelligent filter mechanisms to find suitable
vacancies. Additionally, online job descriptions provide more detailed informa-
tion than traditional help-wanted ads in newspapers and magazines. Employers
benefit from the better targeting options of Internet job advertisements and are
able to screen online applications more efficiently. As a result, the matching
process on the labor market does not only become more efficient but the quality
of job matches should increase.
This paper provides first evidence that online job search is associated with higher
matching quality. Using micro level data from the German Socio-Economic
Panel (SOEP), we compare employees who found their job online with those who
found their job through newspaper advertisements, friends and other channels.
We show that Internet job finders can make better use of their skills, are more
content with the type of their work, believe to have a higher chance of promotion
and higher job security.
Our results indicate that the Internet is an especially valuable job search tool
for workers who are distant to the labor market. Job seekers with employment
interruptions have significantly better matching outcomes if they find their new
job through the Internet. While women with children below the age of 16 have
generally inferior results after starting a new job, this negative association is
alleviated for those of them who use online job search. We find a similar rela-
tionship for job seekers in rural areas. The disadvantage due to remoteness is
remedied if they find their job through the Internet. For workers who were un-
employed before they found a new job, we do not observe a positive association
with online job search. It seems that unemployed workers are not able to use
the Internet to their advantage, maybe due to a lack of necessary skills.
By restricting our sample to workers who found their previous job offline and
their current job either offline or online, we can compare the improvement of
the match quality after a job change, conditional on the job search channel. We
show that our results hold even if we compare online job seekers to only those
who find their job through newspapers and friends. When we compare differ-
ent search channels to the employment office, we find that only the Internet
is associated with significantly higher matching quality. We are able to miti-
gate numerous selection concerns by conducting robustness tests and providing
additional evidence from the German Internet job search market.
The reason why the Internet has a profound impact on the job matching process
goes beyond the wider selection of job opportunities, better search possibilities
and cheaper access to information. The Internet introduces new ways of passive
job search and allows firms to easily search for applicants. Career oriented

2
social networks like LinkedIn and online job boards like Monster.com allow
users to maintain online CVs that can be found by interested employers. Before
the advent of the Internet, the direct targeting of talent by firms was only
feasible through headhunters and mainly used for executive positions. Allowing
firms to tap into the large pool of passive job seekers with detailed information
about their skills and experiences, leads to more informed hire decisions and
contributes to better match quality.
Krueger (2000b) was one of the first to note that by reducing the cost of in-
formation, the Internet will induce workers and employers to learn more about
each other and will thereby improve the quality of job matches. Autor (2001)
points out that due to the Internet, workers and firms are able to consider more
potential match partners which will raise the minimum match quality they are
willing to accept. The higher match quality will in turn lead to higher out-
put and earnings. While Autor acknowledges that better match quality should
reduce job separations, he also states that the wider use of on-the-job search
will potentially offset this effect. The increasing popularity of CV databases
and career networks like LinkedIn, which was launched one year after Autor
published this article, gives his on-the-job search argument additional weight.
Freeman (2002) argues that better job matches might be the strongest macro-
economic consequence of online job search. Regarding unemployment durations,
he suspects that the lower cost of search might ultimately lead to longer search
times as workers and firms will consider more possible matching partners. Kuhn
(2003) draws on classic partial-equilibrium search models and hypothesizes that
by increasing the arrival rate of offers and decreasing search costs, online job
search should lead to shorter unemployment durations and higher quality of job
matches.
The empirical literature on Internet job search is mainly concerned with the
characteristics of online job seekers as well as the effect of online job search on
unemployment. Kuhn and Skuterud (2004) use US Current Population Sur-
vey (CPS) data from the years 1998 and 2000 to show that once observable
characteristics are held constant, Internet job search does not lead to shorter
unemployment durations and might even prolong them. Besides offering the
explanation that the Internet is indeed an inferior job search tool, the authors
raise selection concerns and hypothesize that the longer search time is compen-
sated by improved job quality. Stevenson (2006) agues that limiting the focus
on the unemployed can be misleading as the main effect of online job search
could be improved matching outcomes through on-the-job search. Using sim-
ilar data as Kuhn and Skuterud (2004), Stevenson finds that the Internet has
lead to higher employer-to-employer worker flows which could indicate better
job match quality for the employed. Kroft and Pope (2008) use data from the
classified advertisements community Craigslist.com and find that the website’s
local expansion has to some degree crowded out newspaper advertisement but
did not have an effect on unemployment rates. Using an instrumental variable
approach, Czernich (2011) finds no evidence that broadband Internet affects
unemployment rates. Replicating Kuhn and Skuterud (2004) with newer data,

3
Kuhn and Mansour (2011) find that in the period between 2008 and 2009, online
job search reduced unemployment durations by about 25 percent. Even though
the empirical evidence on the effect of online job search on unemployment re-
mains inconclusive, many studies point out that there should be a substantial
effect on matching quality. To our knowledge, however, there has been no study
which tested this claim directly.
This paper is structured as follows: In the next section, we are going to describe
the data as well as the estimation model. In section 3 we will present our results
and in section 4 we thoroughly discuss possible selection concerns. Section 5
concludes.

2 Individual-level Data on Job Search Methods

and Matching Quality
Using individual level data, we can investigate the relationship between finding
a job through the Internet and the matching quality. There are several different
approaches to measuring the quality of a job match. One indirect approach
is to use the employment duration as an indicator for the quality of a match
(for example Centeno, 2004). Measures that rely on job tenure assume that
”good matches endure” (Bowlus, 1995), a statement that certainly holds in
many cases but does not necessarily imply that all enduring matches are of high
quality, especially when we examine relatively rigid labor markets like Germany.
Another approach is to use the wage of a job as an indicator for matching quality
(e.g. Simon and Warner, 1992, van Ours and Vodopivec, 2008). However, the
wage of a job changer is typically determined before the employment contract
is closed and imperfect information will make it impossible to know the match
quality ex ante. A way to circumvent this problem would be to consider wage
increases in the years after a job change. A problem is that variations in wage
are to a large extent driven by supply and demand as well as other factors
that are not necessarily related to the matching quality. Ferreira and Taylor
(2011) find that the match quality explains less than 1 percent of wages and
Kuhn and Mansour (2011) cannot find an effect of Internet job search on wage
growth between jobs. We therefore use a different approach and take subjective
matching quality measures as our outcome variables.
Our estimation model has the following form:

Mi = β0 + β1 interneti + β2 Xi + β3 countyi + β4 industryi + β5 yeari + i

Where Mi are the subjective matching outcome variables of a person. Specif-

ically, we use the ability to apply own skills, the satisfaction with the type of
work, the perspectives, job security, social benefits, work load, commute and
working hours in the new job. Note that the dependent variables always indi-
cate the evaluation of a person’s new job compared to his prior job. interneti

4
is a dummy indicating whether a person found the job through the Internet.
Xi are individual-level covariates including gender, age, migration status, edu-
cation, number of job changes between 2000 and 2007, and a dummy indicating
whether the person was unemployed during the last year. In order to limit
the risk that we are merely observing a correlation based on the usage of the
Internet in general, we include a dummy that indicates the availability of Inter-
net in the household. Additionally, we include county, industry and year fixed
effects. Since our dependent variables are binary, we estimate a probit model
with robust standard errors.
The data used to estimate the model comes from the German Socio-Economic
Panel (SOEP). The SOEP is a representative annual panel survey of almost
11,000 households and more than 20,000 individuals in Germany. The Panel was
started in 1984 and covers a wide range of topics including many employment
related aspects. Most importantly for our analysis, people who changed their
job are not only asked how they learned about their new job but also how
their new job compares to their former job. There are several advantages of
focusing on job changers instead of also including first-time employees. First
of all, workers who had a job before have a reference point with which they
can compare the new job. Their expectations about how well they can use
their skills at work, for example, are likely to be more realistic than those of
respondents who have not had substantial work experience before. Secondly,
by comparing a new job with an old job of the same person, we can limit some
selection problems, as discussed in section four of this paper. And thirdly, we
can ignore the peculiarities of a person’s first job which is not only related
with occupational training but also associated with a high degree of insecurity
concerning the occupational choice.
In the SOEP, the question ”How did you find out about your new job?” leaves
several answer possibilities, from employment office to the Internet, of which
the respondent has to choose one. The variables which measure the matching
quality are construed based on the question ”How do you view your current
position compared to your previous one?”, followed by several sub-questions
regarding, for example, ”the type of work”, ”chances of promotion”, ”work hour
regulations”, ”work load” and ”commute” with the answer choices ”improved,
”about the same” and ”better”. There is a separate question which reads ”Are
you able to use your professional skills and abilities today more, about the same,
or less than in your previous position?”. The comparative variables are coded
1 when the answer is ”improved” and 0 otherwise. Alternatively, we estimate
an ordered logit model with all three answer choices and find similar results as
in the binary choice model1 .
By limiting our focus on job changers, we substantially reduce our sample to
about 2,000 observations per year between 2000 and 2007. As shown in table 1,
the share of people who found their new job through the Internet rises from less
than 1 percent in the year 2000 to above 6 percent in 2007. These relatively low
1 Results are available upon request

5
numbers are owed to the fact that we do not take into account the young and
often Internet savvy workers who found their first job through the Internet. The
numbers also do not reflect how many job seekers actually used the Internet at
any point during their job search, but more conservatively identify those who
learned about the job through the Internet and chose to sign a contract with the
employer later on. It is likely that for these two reasons, our estimates represent
a lower bound of a potentially larger association. The relatively small number
of observations makes it necessary to pool the observations between 2000 and
2007. Unfortunately, later years can not be used in our analysis because the
SOEP did not include all relevant questions in the years following 2007.
Table 2 shows the sample means by the channel which was used to find a new
job. Online job finders in our sample are on average 32.5 years old which is
slightly younger than workers who found their new job through other channels,
even though the difference is not significant. There are also more men among
the online job finders compared to those who used the newspaper, for example.
A large share of employees who returned to their former employer are women,
a finding which is probably driven by mothers who were on parental leave.
Surprisingly, Internet job finders are on average slightly better educated than
those who found a job in the newspaper. The share of formerly unemployed
job changers who used the Internet is slightly higher than the one of employees
who found their job through friends or newspapers. Although some of these
differences between groups are interesting, few of them are very large or even
statistically significant.

3 The Association of Internet Job Search and

Matching Quality
Table 3 presents the results of regressing different matching outcome variables on
the Internet search dummy as well as other covariates. All reported coefficients
are probit marginal effects. In the first column, the positive and significant
Internet coefficient indicates that online job seekers are more than 6 percent
more likely to use their skills better in their new job. They are also significantly
more likely to be satisfied with the type of work they do, as the high Internet
coefficient in the second column shows. The dependent variable with the highest
Internet coefficient is the perspective variable in the third column. It shows that
online job seekers are more than 8 percent more likely to have better chances of
promotion in their new job. Finding a job online is also associated with better
job security as shown in column 4. Surprisingly, in column 5 we see that also
social benefits are significantly better for online job seekers.
In the last three columns of table 3 we see results for dependent variables which
are not significantly correlated with online job search. Column 6, indicating the
satisfaction with the work load in the new job, has an Internet coefficient which

6
is positive but below 1 percent and insignificant. In column 7, Internet even has
a very small negative coefficient. This could mean that online job seekers are
more likely to find a job that is further away from home than their previous job.
A possible interpretation would be that the Internet opens up job opportunities
outside the regional boundaries and thereby increases work mobility. Online
job search is also hardly associated to work time, as shown in the last column.
Unlike for the other dependent variables in the table, men are significantly less
likely to improve either their commute or working time compared to women.
This might indicate that these dimensions are less important to men when they
change jobs.
In most cases, the coefficients of the control variables have the same sign across
dependent variables even though their size differs. Being male is positively
associated with our first five outcome variables. This means that male job
seekers evaluate their new job on average better than female job seekers. One
explanation could be that men either get better jobs or are more optimistic
about a recent job change. The opposite can be observed for older job changers
compared to younger ones. It can be argued whether older people get less
attractive jobs, are matched worse or simply see a job change in a more negative
light. A similar negative correlation can be observed for people with a migration
background as well as job changers with a period of unemployment before they
find a new job. The higher the education of a person, the better their matching
outcome.
It should be emphasized that the Internet coefficient is positive for all dependent
variables with the exception of commute. The reason why we see quite small
and insignificant coefficients on the work load and working time variables could
be that they are relatively poor measures of the matching quality. When a
person changes to a challenging job that is different to his prior one, the work
load might initially increase as the person has to become familiar with new tasks
and processes. It is therefore unclear whether higher work load is a good or a
bad sign for the matching quality. The working time in a particular job is not
necessarily determined by the individual work contract but rather by firm-wide
or union-wide agreements. Again, it is not surprising that the association of
online job search with this variable is rather small. The four variables which are
much more clearly measures of the matching quality, namely skill use, work type,
perspective and job security all have high and significant Internet coefficients. In
the next section, we will analyze heterogeneity effects on the skill use variable
which is arguably the most interesting measure of job match quality. In the
remaining parts of this paper we will usually refer to the first four dependent
variables because we believe that they are most relevant for assessing matching
quality.

7
3.1 Effect heterogeneity

Until now we were concerned with the average association of online job search
and matching outcomes among all job changers. Table 4 shows that the strength
of this association differs depending on the subgroup we are looking at. Each
line in table 4 represents one least squares regression with ”skill use” as the
dependent variable and the same control variables as in table 3. Additionally,
each regression contains the variable in the lead column and an interaction effect
of this variable with the ”found via Internet” dummy. The first column tells
us the association of finding a job online and being able to use personal skills
better in the new job, for workers that do not belong to the group described in
the lead column. Column two shows the main effect of the variable in the lead
column on the ability to use own skills. The third column reports the estimates
for the interaction term of Internet job search and the respective variable in the
lead column.
Workers who just re-entered the employment market are 10 percent less likely
to feel that they can use their skills better in their new occupation, as shown in
the second column of the first row. These workers were not unemployed before
they found a new job. Although we do not know the exact reason for their em-
ployment interruption, the high proportion of women in this group points into
the direction of parental leave. Other possible reasons for such career breaks
include educational leaves, national service, voluntary work, travel or rest. The
literature on employment interruptions argues that the worker’s human capital
stagnates or even decreases during career breaks, with the exception of educa-
tional leaves2 . The skills acquired in school and during previous occupations
become increasingly outdated and depreciate during employment pauses. Ac-
cording to Williams (2000), even career breaks due to self-employment can have
adverse effects as sector specific human capital decreases over time. Mincer and
Polachek (1974), who underline the importance of work history in human cap-
ital models, also point out that during periods of childbearing, the erosion of
market skills might lead women to revise their expectations and commitment
towards employment. The strong negative coefficient for women with children
in the second row of table 4 could support this finding. Besides the human
capital effect, there is a signalling effect induced by career breaks. Employers
could interpret an employment interruption as a sign of low commitment or
reliability. Consequently, they might be reluctant to offer jobs with high re-
sponsibilities to workers with career breaks. Together with the depreciation of
skills, this could explain the negative association of re-entry and being a mother
with our outcome variable.
The third column of table 4 shows a significantly positive interaction effect for
workers who just re-entered the employment market and found their job online.
The same reversal takes place for women with children. This could mean that
the Internet is an especially valuable job search tool for workers with employ-
2 for a recent overview of the literature on career breaks see Theunissen et al. (2011)

8
ment interruptions. One explanation for this finding could be that the negative
signalling induced by career breaks is less severe when the job is intermediated
through the Internet. Another, probably more convincing explanation is that
the Internet is especially important for workers who are more distant to the
labor market. Women who are caring for their children instead of working are,
for example, less likely to hear of current employment opportunities in the or-
ganization or industry they worked in before. The third row seems to confirm
the hypothesis that distance to the market matters. For those job changers
who live in a county that has a population density below average, denoted as
a rural county in table 4, the Internet interaction term is sizable and signifi-
cant. This finding relates to the ”death of distance” hypothesis which became
popular with the book of the same title by Cairncross (1997). He argues that
modern telecommunication networks will improve the access of rural areas to
larger markets. The disadvantage of job seekers in remote areas is alleviated by
the Internet which opens up new supraregional employment opportunities. It
should be stressed that in all specifications of table 4, we control for Internet
availability in the household. The positive coefficient for online job seekers in
rural areas is therefore more than a sign of being better connected due to Inter-
net access. It indicates that online job search makes a difference for those who
are distant from urban centers.
Workers who were unemployed before they changed jobs do not seem to benefit
from online job search to the same extent as other job changers. This casts
some doubts on the Internet’s ability to match unemployed workers more effi-
ciently. There are different possible explanations why in general job seekers who
are distant from the labor market seem to benefit from online job search but
unemployed job seekers do not. Stevenson (2006) argues that the Internet leads
to an increase in on-the-job search which reduces transitions from employment
to unemployment as workers can more easily find a new job online before their
current job terminates. Especially in light of the passive job search opportu-
nities enabled by the Internet, this hypothesis appears to be very plausible. It
would, however, mean that those who become unemployed are negatively se-
lected with respect to their ability to use online job search to their advantage.
In other words, someone who becomes unemployed nowadays, might not have
the capabilities to benefit from online job search in the first place.
This incapacity could be explained by a lack of exposure to the Internet at the
former workplace. Krueger (2000a) argues that the digital divide with regard
to race might be partially caused by a under-representation of minorities in
positions that use computers. Similarly, unemployed job seekers might be less
successful with online job search because they were less likely to use the Internet
at their former workplace. As a consequence of their lack of knowledge, they
might use inferior Internet search tools. If, for example, they only visit the
website of the employment agency but ignore many of the valuable resources
that other job seekers use, this difference in usage could explain why they do not
benefit from online job search. Hence, the problem for the unemployed could be
insufficient knowledge and lack of the complementary skills that are necessary

9
to use the Internet to their advantage.
Another possible explanation for the unpromising results of the unemployed
might be that the increased transparency and information available through
the Internet can in some cases work against an applicant. Even if the matching
quality of workers and employers improves, this does not necessarily mean that
every worker will get a better job. If a low qualified worker is hired for a
position that entrusts him with too much responsibility, the match is certainly
not ideal. However, the worker might like this job and feel comfortable with
his competencies. When he is matched to a new employer who is able to better
observe his abilities, the worker could end up being less content even though the
matching quality has objectively improved. The Internet offers numerous tools
that allow employers to learn more about their online applicants. Application
forms on websites, for example, are used by companies to request information
from applicants that they would not necessarily reveal in an offline application.
Candidates can then be easily compared on the basis of this information. In
some cases, candidates are asked to perform an online aptitude test during the
application process. Additionally, social networks like LinkedIn allow employers
to easily contact former employers of an applicant. In sum, the Internet could
reveal unfavorable information about workers and thereby increase the matching
quality but decrease the worker’s satisfaction with his new job. Unemployed
workers could be more prone to such revelation of negative information.
It is interesting to see that young job seekers below the age of 30 are not bene-
fitting disproportionately from online job search. This indicates that the digital
divide between younger and older workers does not have adverse consequences
when it comes to using online job search tools. Although young people are on
average able to use their skills better after a job change, this is not reinforced
by the technology they used to find the job. Another concern that is often
raised in context of the digital divide debate is that minorities are disadvan-
taged when it comes to the use of the Internet (e.g. Hoffman and Novak (1998),
Fairlie (2004)). While we do observe a negative association of having a migra-
tion background and our outcome variable, online job search is not less effective
for migrants. In fact, the respective coefficient in table 4 is relatively large and
positive, but not significant at the 10% level. Similarly, workers with tertiary
education do not benefit more than the average from online job search. This
result can be interpreted as a sign that the higher educated are not necessarily
the main beneficiaries of online job search.

3.2 How Internet Job Search Compares to other Search

Methods

In the previous sections we compared the Internet with all other means of find-
ing a new job. But what if the positive correlations shown in table 3 are mainly
driven by the comparison with job search tools that lead to especially poor
matching results? One channel that could lead to mediocre matching results

10
are public employment services. Holzer (1988), for example, finds that searching
a job through family, friends and newspapers is associated with a higher proba-
bility of receiving an offer than searching through the state employment agency.
Clark (1988) shows that the retention rates of job matches that were facilitated
by the public U.S. Employment Service (USES) are lower than those which
were facilitated by other intermediaries. Using data from Portugal, Addison
and Portugal (2001) also find that the public employment service is associated
with shorter job retention. Additionally, they show that rewards for observable
characteristics of job seekers are smaller in jobs found through the public em-
ployment service. One possible explanation for these differences could be that
public employment services have smaller incentives for successful matching than
private intermediaries, as Zweifel and Zaborowski (1996) argue.
We therefore want to analyze how different job search channels compare to
finding a job through federal and local employment offices as well as so-called
”Personalserviceagenturen” (PSAs) which are sort of temporal work agencies at-
tached to employment offices. Table 5 shows that the Internet is the only channel
with significant positive coefficients across all outcome variables in comparison
to the employment office. The friends variable which indicates that the job was
found through friends or family, is even negative for two of the four dependent
variables. One possible reason could be that if someone finds a new job through
personal connections, the formal job screening process which should normally
assure a good matching result is less effective. The fact that the coefficient for
job security is significantly positive does not contradict this hypothesis. In fact,
someone who was offered a job by a friend who works for the same organization,
might feel that his job is especially secure since it is protected by his supporter.
Private job agencies also perform worse than the employment office for two of the
four outcome variables, as shown in the fourth row. The ”other” category has
positive but insignificant coefficients for most of the matching outcomes. One
explanation could be that headhunters and other personal match makers fall in
this category. The newspaper coefficients are also positive, although insignifi-
cant and much smaller than the Internet coefficients. We can therefore conclude
that compared to the employment office, the Internet is the only channel that
is associated with significantly better matching results.
One could argue that people who find their job through the employment office,
private job agencies or undefined means are not of primary interest for our anal-
ysis. After all, these methods are very different from job search on the Internet
and job seekers who use private job agencies, for example, might have different
characteristics from those who use the Internet. In order to test this argument,
we exclude all job changers who have used other channels than the Internet,
friends and newspapers and repeat the estimations from table 3. In table 6 we
see that the Internet coefficients remain relatively stable for the skill use variable
while they are considerably lower for the other three variables. Apart from the
job security measure, however, they all remain significant at the 5 percent level,
at least. This robustness of the Internet coefficients demonstrates that online
job seekers are better matched not only compared to all other search meth-

11
ods taken together but also compared specifically to the most similar channels,
namely newspapers and friends.

4 Discussion of Potential Selection Issues

In the last section we found that our results are robust to a wide range of controls
and sample restrictions. However, there are some obvious selection issues which
could bias our results. Possible problems might arise if Internet finders were
fundamentally different from the workers we compare them with. A second
cause of concern is that online job seekers search differently from job seekers
who use other channels and are therefore able to find better matches. A third
issue is that the kind of companies that use online job tools could be different
from companies that advertise in newspapers or through other more traditional
channels. In the following section we want to tackle these concerns.

4.1 Selection on Unobservables

In the specifications of the previous section, we have seen that the observed
associations are not driven by factors like age, job position or industry. But
what if there was a selection on unobserved characteristics? There are plenty of
reasons why one could assume that workers who use the Internet for job search
are different from workers who prefer to read job advertisements in newspa-
pers. Online job seekers could, for example, be generally more open to new
technologies and adapt better to technological changes. This could in turn be a
characteristic which is valued by employers and correlated with better matching
outcomes.
One reason why the problem of selection on unobservables is less severe than
it might appear at first, is that we basically compare two jobs of the same
person. In our analysis, we focus on workers who have changed jobs and can
therefore judge their current job in comparison to their former job. They are
asked, for example, whether they can use their skills better, equally well or
worse than in their previous job. Since we have the same person making assess-
ments of two jobs, many typical selection issues are mitigated. It might well
be the case that workers who find their job over the Internet are better able
to use changes to their advantage and will therefore always have better career
perspectives than offline job seekers. But as long as this personal characteristic
is constant over time, there is no reason to believe that the same person would
have better perspectives in one job compared to another job for reasons that
are unrelated to the job itself. In other words, the problem of selection on un-
observed personal characteristics does not play a dominant role in many parts
of our analysis.
There are, however, certain selection criteria that could still potentially bias

12
our results. If online job seekers would generally see a job change in a better
light than offline job seekers, they could also believe that their new job fits
better to them, simply because they changed it recently. Although testing for
the personal attitudes towards a new job is very hard, there are some questions
in the SOEP that can be used as indicative evidence for a possible systematic
difference in attitudes between online and offline job seekers. In one question of
the SOEP, respondents are asked whether the statement ”When I think about
the future, I’m actually quite optimistic” applies to them. A second question
is concerned with the overall life satisfaction and reads: ”How satisfied are you
with your life” (scale 0-10). The first two columns of table 7 show that peo-
ple who found a new job over the Internet are generally not more optimistic
about the future and even slightly less satisfied with their life in general. Ad-
ditionally, the SOEP also incorporates rough measures of the so called ”Big
Five” personality traits: openness, conscientiousness, extraversion, agreeable-
ness, and neuroticism. These personality traits are fairly stable over time and
are expected to change very little after the age of 30 (Terracciano et al., 2010).
For our purposes, openness seems to be the most interesting of these factors.
In the SOEP, openness is measured in terms of being original and coming up
with new ideas as well as having an active imagination3 . The third and fourth
column of table 7 show that there is no significant difference between online
and offline job seekers along these dimensions. This gives some indication that
online job seekers are neither more open nor do they view the future or their
life situation in general more favorably.
Although there are good reasons to believe that selection on unobservables is not
the primary concern in our analysis, we can apply an additional test to eliminate
possible selection biases. If people who find their job online are systematically
different from others, this difference should not only affect the variables where
we see a significant positive association, namely the usage of skills, the satis-
faction with the type of work, the career perspectives and job security. The
difference between online and offline job seekers should in the same way affect
other variables, like the satisfaction with the working time. Let us assume, for
example, that online job seekers view their new job more positively than others
because of differences in personal attitudes. Then the generally better assess-
ment of the job should make these workers more content with their tasks but
also with their working time. Since there are little objective reasons why online
job seekers should have more comforting working times, we can use the assess-
ment of the working time as a reference point for all of the person’s judgements.
By including the ”working time” variable in the specifications of table 8, we
control for a possible selection effect on the evaluation of a given person. While
the association of working time with our outcome variables is highly significant,
the coefficients of the Internet variable remain stable. It therefore seems that
the relationship between Internet job search and matching quality is not largely
biased by a selection on unobserved characteristics.
3 The original questions from the SOEP read: ”I see myself as someone who is original,

comes up with new ideas” and ”I see myself as someone who has an active imagination”.

13
4.2 Selection into Search Intensities

Another selection concern is that online job seekers spend more time than offline
job seekers looking for a job. If workers who use the Internet for job search spend
many hours looking for fitting vacancies while others just open up the newspaper
and apply to the next best job advertisement, it not be too surprising if we saw
the latter having inferior matching results. But is it realistic that online job
seekers are the ones who search more intensively or are more serious about their
search? In order to answer this question we need to differentiate the various
kinds of offline job seekers in our analysis. It seems likely that people who
find their job through friends and family, for example, spend considerably less
time comparing different job offers than people who use the Internet for this
purpose. Table 5 shows that finding a job through friends and family is not
generally associated with superior matching quality. One possible reason why
friends and family do not perform very well could be that the affected job seekers
put only little effort into the search process. However, the positive association
of Internet job search and matching quality is not only driven by the comparison
to job search through friends. As shown in table 5, there is no other channel
besides the Internet that is associated with significantly higher matching quality
compared to our base category, the employment office.
Arguing that Internet job seekers are searching more intensively than workers
who apply through newspaper advertisements and other offline media does not
seem to be very convincing. One reason why it appears very unlikely that online
searchers are in any way more serious about their search are the low search
costs of the Internet. Looking for a job in a newspaper is much more costly
than using online job boards. First of all, newspapers themselves cost money.
Secondly, it is harder and takes more time to find advertisements that match
own qualifications. And thirdly, compiling a classical job application including
a printed photograph4 , a nice folder and a postal stamp is much more expensive
than filling out an online application form or sending an email. These costs
are one reason why the use of postal applications declined steadily over the last
couple of years while the use of electronic applications increased over time, as
shown in figure 1. Taking the low costs of online job search into account, it seems
unlikely that online job seekers are generally more serious about their search
compared to those who use newspapers as their search channel of choice.
Besides the lower costs, the Internet also offers passive job search opportunities,
as discussed above. Over 135m people maintain online CVs on the worldwide
largest professional network LinkedIn5 . Xing, a German competitor of LinkedIn,
has over 11m members6 . Online CVs allow recruiters and headhunters to ac-
tively search for job candidates. Both LinkedIn and Xing offer premium mem-
berships that are targeted to heavy users like recruiters and allow for more
4 in Germany it is standard practice to have a photograph on the CV
5 http://press.linkedin.com/node/1016
6 http://corporate.xing.com/deutsch/investor-relations/basisinformationen/qas/#c322

14
sophisticated search and filter options. LinkedIn tries to attract recruiters with
the slogan ”Find the World’s Best Passive Talent” 7 and charges up to d359.95
per month for a ”Talent Pro” premium subscription. Also online job boards like
Monster offer the possibility to upload CVs that can be viewed by recruiters.
Since uploading a CV to Monster clearly signals interest in new job opportuni-
ties, users can choose to hide sensitive information like their name and address
so that current employers do not realize that their employees look for a new
job. The possibility of passive job search clearly contradicts the argument that
online job seekers spend more time or look more intensively for new job oppor-
tunities. As figure 2 depicts, a majority of people who are interested in career
opportunities already use online career networks and CV databases. This shows
that online job search can be almost completely effortless, which is not true for
most other job search channels.
In order to further test whether people who use the Internet for job search are
more active than others, we use additional information from the SOEP. Re-
spondents who changed their job are asked: ”Were you actively looking for a
job when you received your current position, or did it just come up?”. Ta-
ble 9 shows that the active search coefficient is positive and highly significant,
indicating that people who search actively have considerably better matching
outcomes than people who find a job by chance. The coefficient of online job
search, however, remains positive and significant for the first three of our four
dependent variables in table 9. By controlling for active job search, we provide
additional evidence that workers who find their job through friends or news-
papers are not only matched worse because they find a job without actively
looking for it. The lower cost and lower effort necessary on the Internet implies
that, if anything, online job searchers are less serious about their search.

4.3 Selection of Advertised Jobs

Although we intensively discussed the possible selection of job seekers, we have

not addressed the employer side yet. Even if job seekers are neither selected on
unobservables nor into search intensities, our results could be biased if only a
certain kind of companies use the Internet for recruitment purposes. Like before,
we have to differentiate between the various offline search channels when com-
paring their exposure to selection. Companies who advertise vacancies through
the employment office, for example, might be less attractive than companies
who use online job boards. This could lead to a negative bias unrelated to
the matching process itself. However, we have seen in table 6 that our results
remain stable if we exclude the employment offices and job agencies from our
analysis. When comparing the Internet with other offline channels it becomes
more difficult to think of similar selection arguments. There is evidence that
many high-end jobs are not advertised over the Internet but only through other
7 http://talent.linkedin.com/Recruiter

15
channels. Specialized headhunters and HR consultancies definitely play an im-
portant role for executive jobs. Many people also believe that high salary jobs
are more often advertised in newspapers. Some companies prefer newspaper
advertisements because they look more expensive and signal the value that the
company puts into the position. This kind of selection would however lead to
an underestimation of the association we observe. If anything, the selection of
companies make our results look more conservative.
Figure 3 provides further evidence that companies who advertise online are not
necessarily the more attractive employers. It is striking that 40 percent of the
companies who advertise on Monster are temporal work companies. Usually,
these kind of companies are not among the favorites of job seekers and often
pay less than other employers. At Stellenanzeigen.de the share of temporal work
companies is significantly lower but still twice as high as the share of companies
listed on the German DAX stock index. The dominance of the generally less
attractive temporal work companies in online job boards would again lead to
an underestimation of our results.
Companies choose the advertising channel where they hope to reach the best
audience for a specific job advertisement. Different kinds of job positions are
therefore advertised through different channels, a fact that gives rise to another
selection concern. It leads to the question how job positions advertised through
the Internet compare to job positions advertised offline. Figure 4 shows how
well the search results at Monster.de fit to different job titles as search terms.
On average, almost 80 percent of the search results are exact matches to the re-
quests. There seems to be a tendency that jobs which require high qualifications,
like General Manager, HR Director, lawyer and engineer, produce inferior re-
sults compared to more mid-range jobs like controller or project manager. This
could be an indication that online job boards perform less well when it comes
to top jobs. If the Internet serves as a better channel for mid-level jobs and top
jobs are more often advertised through other channels, we would underestimate
our results.
Anecdotal evidence also suggests that while online advertisements lead to a high
quantity of applications, many of them are of lower quality. This could be re-
lated to the low application costs that we discussed above. Even if applicants
do not match the qualifications requested in an job advertisement, they still try
their luck and send out an costless online application, no matter how low their
chances of success are. From an employer perspective, this problem obviously
becomes larger, the higher the requested qualifications are. Figure 5 shows the
results of a study in which identical job advertisements for the position of a pro-
curement director were placed in eight newspapers and eight online job boards.
The applicants were subsequently rated according to their qualifications. Over-
all, the number of applications in response to the online advertisements was
more than 2.5 times higher than the number of applications in response to the
print advertisements. However, more than 50 percent of the Internet applicants
were not suitable for the job at all while this share was considerably lower for

16
newspaper applicants. Although these numbers are purely descriptive and by
no means representative, they reflect the cliche that online recruitment is more
about the quantity and less about the quality of applicants. As a result, we
might see less high qualification jobs online which again would lead us to an
underestimation of our results.

5 Conclusion
In this paper, we investigate whether online job search is associated with higher
job matching quality. While the question of how the Internet affects unemploy-
ment durations and other labor market outcomes has been repeatedly studied
in the literature, this paper provides first empirical evidence on the quality of
resulting job matches. We find that matching outcomes of online job seekers are
superior along the dimensions of making better use of own skills, being more
content with the type of work, having higher chances of promotion and higher
job security. These results are not driven by the comparison of the Internet
with inferior search channels like the employment office. Online job search is
associated with better matching quality even if it is directly compared to news-
papers, friends and family. Our results avoid bias from many possible sources
of selection. As we focus our analysis on workers who found their previous job
offline and their current job either offline or online, we are virtually able to com-
pare two matching outcomes of the same worker by using retrospective data.
Additionally, we tackle several selection issues by conducting robustness tests
and we provide some descriptive evidence in order to alleviate remaining selec-
tion concerns. Even though we are able to rule out the most obvious threats to
a causal interpretation of the associations presented in this paper, more work
is needed to identify a clear causal relationship between online job search and
matching quality.
Our results indicate that the Internet is especially important for job seekers
who are distant to the labor market. Workers with employment interruptions
are particularly well matched if they used the Internet to find their new job.
Online job search also seems to play an important role for mothers with chil-
dren. As gender inequality remains an issue within many labor markets, it is
important to know that the Internet helps to alleviate possible negative conse-
quences of a maternity leave. Our results also show that job seekers in areas
with lower population densities are better matched when they find their job
online. This finding has important policy implications with regards to the ex-
pansion of broadband Internet in rural areas. It is remarkable that online job
search seems to compensate many disadvantages of job seekers who are distant
to the labor market. Formerly unemployed job seekers, however, do not appear
to benefit from online job search to the same extent. One possible reason for
this finding is that the unemployed lack the complementary skills and knowledge
to use the Internet for their advantage. If this was the case, it could be worth

17
training unemployed job seekers in applying new technologies during their job
search process. Further research is needed in order to better understand the
relationship between online job search and Internet related skills.

18
References
Addison, J. T. and P. Portugal (2001): “Job Search Methods and Out-
comes,” .
Ai, C. and E. C. Norton (2003): “Interaction terms in logit and probit
models,” Economics Letters, 80, 123–129.
Autor, D. H. (2001): “Wiring the Labor Market,” Journal of Economic Per-
spectives, 15, 25–40.
Bowlus, A. J. (1995): “Matching Workers and Jobs: Cyclical Fluctuations in
Match Quality,” Journal of Labor Economics, 13, 335.
Cairncross, F. (1997): The death of distance: How the communications rev-
olution will change our lives, Boston, USA: Harvard Business School Press.
Centeno, M. (2004): “The Match Quality Gains from Unemployment Insur-
ance,” The Journal of Human Resources, 39, 839.
Clark, W. (1988): “Production costs and output qualities in public and private
employment agencies,” Journal of Law and Economics, 31, 379–393.
Czernich, N. (2011): “Broadband Infrastructure and Unemployment-
Evidence for Germany,” Munich Discussion Papers.
Fairlie, R. W. (2004): “Race and the digital divide,” The BE Journal of
Economic Analysis & Policy, 3, 15.
Ferreira, P. and M. Taylor (2011): “Measuring match quality using sub-
jective data,” Economics Letters, 113, 304–306.
Freeman, R. B. (2002): “The labour market in the new information economy,”
Oxford Review of Economic Policy, 18, 288–305.
Hoffman, D. L. and T. P. Novak (1998): “Bridging the Digital Divide:
The Impact of Race on Computer Access and Internet Use,” Science, 280,
390–391.
Holzer, H. J. (1988): “Search Method Use by Unemployed Youth,” Journal
of Labor Economics1, 6, 1–20.
Kroft, K. and D. Pope (2008): “Does online search crowd out traditional
search and improve matching efficiency? Evidence from Craigslist,” Unpub-
lished paper, University of Pennsylvania.
Krueger, A. (2000a): “The digital divide in educating African-American stu-
dents and workers,” Princeton University Industrial Relations Section Work-
ing Paper.
Krueger, A. B. (2000b): “The Internet is lowering the cost of advertising and
searching for jobs,” .

19
Kuhn, P. J. (2003): “The Internet and Matching in Labor Markets,” in New
Economy Handbook, ed. by Derek C. Jones, Amsterdam: Academic Press,
508–523.
Kuhn, P. J. and H. Mansour (2011): “Is Internet Job Search Still Ineffec-
tive?” IZA Discussion Paper.

Kuhn, P. J. and M. Skuterud (2004): “Internet job search and unemploy-

ment durations,” The American Economic Review, 94, 218–232.
Mincer, J. and S. Polachek (1974): “Family investments in human capital:
Earnings of women,” The Journal of Political Economy Political Economy,
82, 76–108.
Simon, C. and J. Warner (1992): “Matchmaker, matchmaker: The effect
of old boy networks on job match quality, earnings, and tenure,” Journal of
Labor Economics, 10, 306–330.
Stevenson, B. (2006): “The impact of the internet on worker flows,” .

Terracciano, A., R. R. McCrae, and P. T. Costa (2010): “Intra-

individual Change in Personality Stability and Age.” Journal of research in
personality, 44, 31–37.
Theunissen, G., M. Verbruggen, A. Forrier, and L. Sels (2011): “Ca-
reer sidestep, wage stepback? The impact of different types of career breaks
on wages,” Gender, Work & Organization, 18, 110–131.
van Ours, J. C. and M. Vodopivec (2008): “Does reducing unemployment
insurance generosity reduce job match quality?” Journal of Public Economics,
92, 684–695.

Williams, D. (2000): “Consequences of self-employment for women and men

in the United States,” Labour Economics, 7, 665–687.
Zweifel, P. and C. Zaborowski (1996): “Employment Service: Public or
Private?” Public Choice, 89, 131–162.

20
 Table 1: Number of job changers and Internet finders in sample

Year Observations Job Changers Changers / Internet Internet /

Observations Changers
2000 24,576 2,560 10.42 22 0.86
2001 22,351 2,400 10.74 45 1.88
2002 23,892 2,263 9.47 71 3.14
2003 22,611 1,922 8.50 68 3.54
2004 22,019 1,744 7.92 81 4.64
2005 21,105 1,603 7.60 84 5.24
2006 22,358 1,791 8.01 99 5.53
2007 20,886 1,939 9.28 119 6.14
Total 179,798 16,222 9.02 589 3.63

Table 2: Sample means by job search method

Internet Newspaper Friends Private Agency Job Center Other Back to former
age 32.51 35.02 33.43 35.71 33.11 34.60 36.31
(9.35) (10.55) (11.23) (11.22) (11.95) (10.38) (10.75)

male 0.57 0.40 0.48 0.59 0.51 0.51 0.36

(0.49) (0.49) (0.50) (0.49) (0.50) (0.50) (0.48)

migrated 0.15 0.18 0.22 0.22 0.19 0.14 0.16

(0.36) (0.38) (0.41) (0.42) (0.40) (0.35) (0.37)

education 4.16 3.61 3.34 3.90 3.24 4.02 3.66

(1.49) (1.42) (1.45) (1.52) (1.27) (1.53) (1.33)

unemployed 0.36 0.26 0.24 0.29 0.54 0.17 0.20

(0.48) (0.44) (0.43) (0.45) (0.50) (0.38) (0.40)
N 512 1,774 3,902 167 894 2,093 1,274
Means of the respective subsample. Standard deviations in parentheses.

21
 Table 3: The association between online job search and several matching outcome variables
(1) (2) (3) (4) (5) (6) (7) (8)
skill use work type perspective job security benefits work load commute work time
found via Internet 0.0645∗∗∗ 0.0798∗∗∗ 0.0812∗∗∗ 0.0455∗∗ 0.0687∗∗∗ 0.0095 -0.0070 0.0150
(0.0214) (0.0236) (0.0202) (0.0212) (0.0205) (0.0220) (0.0235) (0.0237)

male 0.0301∗∗∗ 0.0682∗∗∗ 0.0726∗∗∗ 0.0615∗∗∗ 0.0366∗∗∗ 0.0044 -0.0337∗∗∗ -0.0726∗∗∗

(0.0106) (0.0115) (0.0103) (0.0106) (0.0104) (0.0108) (0.0112) (0.0115)

age -0.0074∗∗∗ -0.0072∗∗∗ -0.0096∗∗∗ -0.0049∗∗∗ -0.0030∗∗∗ -0.0016∗∗∗ -0.0010∗∗ -0.0021∗∗∗

(0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005) (0.0005)

migrated -0.0304∗∗ -0.0292∗∗ -0.0453∗∗∗ -0.0271∗∗ -0.0016 -0.0053 0.0070 -0.0296∗∗

(0.0130) (0.0142) (0.0124) (0.0127) (0.0127) (0.0131) (0.0140) (0.0140)

education 0.0327∗∗∗ 0.0070∗ 0.0395∗∗∗ 0.0104∗∗∗ 0.0138∗∗∗ -0.0069∗ -0.0035 -0.0103∗∗

(0.0037) (0.0041) (0.0036) (0.0037) (0.0036) (0.0038) (0.0040) (0.0041)

unemployed -0.0305∗∗∗ -0.0388∗∗∗ -0.0534∗∗∗ -0.0340∗∗∗ -0.0232∗∗ -0.0426∗∗∗ -0.0239∗∗ -0.0469∗∗∗

22
(0.0107) (0.0116) (0.0103) (0.0106) (0.0104) (0.0106) (0.0113) (0.0115)

Internet available 0.0573∗∗∗ 0.0376∗∗∗ 0.0152 -0.0028 0.0179∗ 0.0057 0.0157 0.0326∗∗∗
(0.0109) (0.0119) (0.0108) (0.0109) (0.0106) (0.0110) (0.0116) (0.0117)

job changes 0.0051 -0.0100∗∗ -0.0037 -0.0128∗∗∗ -0.0094∗∗∗ -0.0008 -0.0018 -0.0057
(0.0036) (0.0040) (0.0036) (0.0037) (0.0036) (0.0037) (0.0039) (0.0040)

county-fixed Yes Yes Yes Yes Yes Yes Yes Yes

industry-fixed Yes Yes Yes Yes Yes Yes Yes Yes

year-fixed Yes Yes Yes Yes Yes Yes Yes Yes

N 9045 9162 8888 8901 8922 9058 9076 9107
Pseudo R2 0.098 0.076 0.121 0.079 0.073 0.058 0.050 0.067
Robust standard errors in parentheses.
Dependent variables take on the value 1 if the new job is evaluated better than the former one with respect to: the ability to use own skills, the type
of work, the chances of promotion (perspective), the security against job loss, social benefits, work load, commute, and work time regulations.
Probit estimation with average marginal effects
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
 Table 4: Effect heterogeneity with the dependent variable ”skill use”
Internet (1) main effect (2) interaction (3) N
Re-entry into employment 0.0523* -0.1031*** 0.1397* 9037
(0.0267) (0.0130) (0.0732)
Female and children below age 16 0.0489* -0.0991*** 0.1463** 9037
(0.0269) (0.0141) (0.0713)
Rural area 0.0228 -0.0285*** 0.1152** 9037
-0.0334 -0.0109 -0.0487
Unemployed during last 12 months 0.0850*** -0.0313*** -0.0284 9037
(0.0325) (0.0113) (0.0474)
Younger than 30 0.0833*** 0.0472*** -0.0255 9037
(0.0294) (0.0160) (0.0507)
Tertiary education 0.0825** 0.0075 -0.0183 9037
(0.0336) (0.0227) (0.0477)
Migration backround 0.0580** -0.0344** 0.1109 9037
(0.0261) (0.0139) (0.0683)
Dependent variable ”skill use” takes on the value 1 if the new job is evaluated better than the former one with
respect to the ability to use own skills. Every line represents one probit estimation with average marginal
effects according to Ai and Norton (2003), with column (1) showing the effect of the ”found by Internet”
variable, column (2) showing the main effect of the variable in the respective row and column (3)
showing the interaction effect of that variable with the ”found by Internet” variable. Robust standard errors
in parentheses. Regressions control for: male, age, migrated, education, unemployed, Internet at home.
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

23
 Table 5: Different job search channels compared to the employment office

(1) (2) (3) (4)

skill use work type perspective job security
found via Internet 0.0486∗ 0.0782∗∗∗ 0.0659∗∗∗ 0.0605∗∗
(0.0256) (0.0280) (0.0245) (0.0255)

found via friends -0.0213 0.0135 -0.0169 0.0378∗∗

(0.0170) (0.0183) (0.0166) (0.0171)

found via newspaper 0.0193 0.0266 0.0165 0.0233

(0.0188) (0.0203) (0.0182) (0.0189)

found via agenccy -0.0531 0.0410 -0.0077 0.0126

(0.0410) (0.0438) (0.0389) (0.0402)

found via other -0.0006 0.0194 0.0061 0.0258

(0.0183) (0.0198) (0.0179) (0.0186)

individual Yes Yes Yes Yes

county-fixed Yes Yes Yes Yes

industry-fixed Yes Yes Yes Yes

year-fixed Yes Yes Yes Yes

N 9045 9162 8888 8901
Pseudo R2 0.103 0.086 0.130 0.087
Robust standard errors in parentheses.
Probit estimation with average marginal effects. Dependent variables take
on the value 1 if the new job is evaluated better than the former one with
respect to: the ability to use own skills, the type of work, the chances of
promotion (perspective), security against job loss and benefits. Individual-level
covariates: male, age, migrated, education, unemployed, Internet at home and
number of job changes.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

24
Table 6: Reduced sample for comparison of Internet with newspaper and friends
(1) (2) (3) (4)
skill use work type perspective job security
found via Internet 0.0562∗∗ 0.0574∗∗ 0.0637∗∗∗ 0.0193
(0.0229) (0.0247) (0.0217) (0.0232)

male 0.0224 0.0652∗∗∗ 0.0824∗∗∗ 0.0677∗∗∗

(0.0140) (0.0150) (0.0135) (0.0141)

age -0.0072∗∗∗ -0.0078∗∗∗ -0.0098∗∗∗ -0.0052∗∗∗

(0.0006) (0.0007) (0.0006) (0.0006)

migrated -0.0221 -0.0337∗ -0.0409∗∗ -0.0364∗∗

(0.0168) (0.0181) (0.0159) (0.0166)

education 0.0332∗∗∗ 0.0111∗∗ 0.0429∗∗∗ 0.0136∗∗∗

(0.0049) (0.0053) (0.0047) (0.0050)

unemployed -0.0267∗ -0.0387∗∗ -0.0456∗∗∗ -0.0318∗∗

(0.0142) (0.0153) (0.0136) (0.0143)

Internet availability 0.0508∗∗∗ 0.0193 0.0142 -0.0151

(0.0142) (0.0153) (0.0139) (0.0144)

job changes 0.0024 -0.0174∗∗∗ -0.0118∗∗∗ -0.0163∗∗∗

(0.0048) (0.0052) (0.0046) (0.0049)

county-fixed Yes Yes Yes Yes

industry-fixed Yes Yes Yes Yes

year-fixed Yes Yes Yes Yes

midrule N 5341 5443 5265 5255
Pseudo R2 0.108 0.093 0.152 0.098
Robust standard errors parentheses.
Probit estimation with average marginal effects. Dependent variables take
on the value 1 if the new job is evaluated better than the former one with
respect to: the ability to use own skills, the type of work, the chances of
promotion (perspective), and the security against job loss.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

25
 Table 7: Internet job finders and personality traits
optimism hapiness originality phantasy
found via internet 0.1105 -0.1395 -0.0109 0.1806
(0.1002) (0.0771) (0.1808) (0.1905)

male -0.1049∗ -0.0739∗ 0.2503∗∗ -0.0662

(0.0481) (0.0314) (0.0862) (0.0985)

age 0.0140∗∗∗ -0.0197∗∗∗ 0.0013 -0.0101∗

(0.0023) (0.0016) (0.0042) (0.0047)

migrated -0.0790 0.0432 -0.0259 -0.1261

(0.0657) (0.0424) (0.1179) (0.1298)

education -0.0733∗∗∗ 0.1149∗∗∗ 0.0490 0.0066

(0.0181) (0.0117) (0.0312) (0.0348)

unemployed 0.1075∗ -0.3825∗∗∗ -0.1721 -0.2463∗

(0.0532) (0.0368) (0.0974) (0.1086)

Internet available -0.0828 0.0981∗∗ 0.2099∗ -0.0237

(0.0558) (0.0335) (0.1051) (0.1141)

job changes -0.0097 -0.0715∗∗∗ 0.0398 0.0485

(0.0179) (0.0118) (0.0316) (0.0374)

state-fixed Yes Yes Yes Yes

N 1038 11581 1021 1019
R-sq 0.060 0.053 0.027 0.022
Robust standard errors in parentheses. Ordinary least squares estimation.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

26
 Table 8: Controlled for working time
(1) (2) (3) (4)
skill use work type perspective job security
internet 0.0668∗∗∗ 0.0748∗∗∗ 0.0793∗∗∗ 0.0443∗∗
(0.0241) (0.0230) (0.0200) (0.0208)

working time 0.0657∗∗∗ 0.1773∗∗∗ 0.0881∗∗∗ 0.1565∗∗∗

(0.0103) (0.0099) (0.0092) (0.0090)

individual Yes Yes Yes Yes

county-fixed Yes Yes Yes Yes

industry-fixed Yes Yes Yes Yes

year-fixed Yes Yes Yes Yes

N 9108 9118 8857 8869
Pseudo R2 0.100 0.130 0.105
Robust standard errors in parentheses.
Probit estimation with average marginal effects. Dependent variables take
on the value 1 if the new job is evaluated better than the former one with
respect to: the ability to use own skills, the type of work, the chances of
promotion (perspective), and the security against job loss. Individual-level
covariates: male, age,migrated, education, unemployed, Internet at home and
number of job changes.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

27
 Table 9: Controlled for active search
(1) (2) (3) (4)
skill use work type perspective job security
found via Internet 0.0469∗∗ 0.0557∗∗ 0.0608∗∗∗ 0.0322
(0.0215) (0.0238) (0.0204) (0.0214)

active search 0.0669∗∗∗ 0.0688∗∗∗ 0.0635∗∗∗ 0.0370∗∗∗

(0.0100) (0.0108) (0.0098) (0.0100)

covariates Yes Yes Yes Yes

county-fixed Yes Yes Yes Yes

industry-fixed Yes Yes Yes Yes

year-fixed Yes Yes Yes Yes

N 8993 9114 8841 8854
Pseudo R2 0.101 0.080 0.126 0.079
Robust standard errors clustered at the household level in parentheses
Probit estimation with average marginal effects. Dependent variables take
on the value 1 if the new job is evaluated better than the former one with
respect to: the ability to use own skills, the type of work, the chances of
promotion (perspective), and the security against job loss. Individual-level
covariates: male, age, migrated, education, unemployed, and Internet at home.
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

28
Figure 1: Preference of job searchers for electronic and postal applications over
time
100

90
Electronic Postal
80 78
71
70 65
64
60
52 53 53
50 50
50 48 47
41
40 37

30 28
23
20 16

10

0
2003 2004 2005 2006 2007 2008 2009 2010

Source: Bewerbungspraxis 2011, Centre of Human Resources Information Sys-

tems (CHRIS). Based on 10,227 individuals interested in career opportunities.

Figure 2: Usage of CV databases and online career networks for passive job
search

Public CV Database 74.2 16.4

Yes
Career Network Profile 63.7 27.7 No
Planned

Company CV Database 40.9 49.8

Source: Bewerbungspraxis 2011, Centre of Human Resources Information Sys-

tems (CHRIS). Based on 10,227 individuals interested in career opportunities.

29
Figure 3: Company types that advertise on two major German online job boards

Monster 5 52 40 3

Stellenanzeigen.de 7 71 14 8

DAX Companies Non-Dax Companies Temporal Work Companies Consultancies

Source: Jobbörsen im Vergleich 2011, Fachhochschule Koblenz.

Based on 1,500 radomly selected German job advertisements per website.

Figure 4: Percentage of search results that match the search request on Mon-
ster.de
0% 20% 40% 60% 80% 100%
General Manager
HR Director
Lawyer
Engineer
Average
Sales Director
Administrator
Product Manager
Assistant
Marketing Manager
Team Leader
Application Engineer
Procurement Manager
Sales Manager
Accountant
Consultant
Software Developer
Project Manager
Controller

Exact match Not exact match

Source: Jobbörsen im Vergleich 2011, Fachhochschule Koblenz.

Based on 86,023 results of the ”quick search” function on Monster.de.

30
Figure 5: Quality of applications in response to online and print job advertise-
ments for ”Head of Procurement”
121

75
Online
39 41 Print
23 26

very good good not suitable

applicants applicants applicants

Source: Medialeistungstest 2010, WESTPRESS GmbH & Co. KG.

Responses to job advertisements in 8 German newspapers (among others
”Süddeutsche Zeitung” and ”Der Tagesspiegel”) and 8 German online job boards
(among others monster.de and stellenanzeigen.de).

31