1. Is there a signal from education?
Evidence from Polish LFS data.
Wojciech Hardy*
*Big thanks to: Joanna Tyrowicz, Stanisław Cichocki
and my colleagues at GRAPE
2. Presentation plan
1. The theory of signaling
2. Why even bother?
3. The empirical literature
4. My study
5. Conclusion and ideas for the future.
3. The theory of signaling
• Spence, 1973, Quarterly Journal of Economics,
Volume 87, Issue 3:
• „To hire someone is frequently to purchase a
lottery. (…) the employer cannot directly observe
the marginal product prior to hiring. What he
does observe is a plethora of personal data in the
form of observable characteristics and attributes
of the individual, and it is these that must
ultimately determine his assessment of the lottery
he is buying.”
4. The theory of signaling, Spence (1973)
You’re a non-discriminating owner of a banking company.
Consider three situations – who would you hire?
A)
•
•
•
•
Female
Age 25
Graduated in Psychology
No previous experience
A)
•
•
•
•
•
Male
Age 26
Graduated in Economics in 2012
2 years of previous experience
Finished a music school in 2012
A)
•
•
•
•
•
Male
White
Age 26
Graduated in Economics in 2012
2 years of previous experience
B)
•
•
•
•
Female
Age 25
Secondary school
No previous experience
B)
•
•
•
•
Male
Age 26
Graduated in Economics in 2012
2 years of previous experience
B)
•
•
•
•
•
Female
White
Age 26
Graduated in Economics in 2012
2 years of previous experience
5. The theory of signaling, Spence (1973)
Situation 1
Indices
Signal
A)
•
•
•
•
Female
Age 25
No previous experience
Graduated in Psychology
B)
•
•
•
•
Female
Age 25
No previous experience
Secondary school
1. Signaling induces costs. These are related to the individual’s skills.
2. Individual A reports being able to „afford” finishing studies.
3. He signals being more productive.
6. The theory of signaling, Spence (1973)
Situation 2
Signal
A)
•
•
•
•
•
B)
Male
•
Age 26
•
Graduated in Economics in 2012 •
2 years of previous experience
•
Finished a music school in 2012
Male
Age 26
Graduated in Economics in 2012
2 years of previous experience
1. It is impossible to assess the productivity with a 100% certainty.
2. Individual A reports having done more in 2012.
3. He signals being more productive.
7. The theory of signaling, Spence (1973)
Situation 3
Signal ?
A)
•
•
•
•
•
Where’s the signal?
B)
Male
•
White
•
Age 26
•
Graduated in Economics in 2012 •
2 years of previous experience
•
Female
White
Age 26
Graduated in Economics in 2012
2 years of previous experience
1. If I don’t discriminate, then A=B, right?
2. Over time, there’s a self-regenerating equilibrium, which changes
incentives based on the distribution of education.
3. The COSTS might vary based on indices if others discriminate.
4. Therefore the signal is there, based on the state of the market.
8. Why even bother?
If a signal of education exists
1. There’s a bias on all
estimates of education’s
role
in
productivity,
employment and entry
wages.
2.
Inequalities
in
education levels affect the
efficiency of employerworker matching.
9. Empirical literature
• Required skills versus necessary skills – self-report studies
(e.g. Dolton & Vignoles, 2000; Chatterji et al., 2003).
– Findings regarding differences in signaling for men and women.
• Natural experiments – education reforms
(e.g. Lang & Kropp, 1986; Chevalier et al., 2004).
• The self-regenerating cycle and the distribution of education
(Kroch & Sjoblom, 1994; Rosenbaum, 2000).
– The Ranks.
10. The idea
The education distribution can be used as an identification instrument for
signal’s strength.
Besides education level, the following are important:
• The Rank (i.e. the place in the distribution of education)
• Equals (i.e. how much our education level sets us apart from others)
• Group size (i.e. how many similar people are there on the job market)
11. The data
• Polish Labour Force Survey.
• Self-reported flow information.
• Years: 2004-2012.
– (prior no information on fields of study).
• Representative of the whole population.
– Both an advantage and a necessity.
12. Education level patterns across years
(24-26 year olds)
100%
90%
80%
70%
60%
Tertiary
High school
50%
High school vocational
Vocational
40%
Elementary
30%
20%
10%
0%
2004
2005
2006
2007
2008
2009
2010
2011
2012
13. Overrepresentation of occupations
after particular studies
Occupation
Field of study
Management Specialists Technicians
Engineering &
construction
+
Farmers
+
Pedagogy
Office Sales and
workers services
+
+
+
+
Machine
Low-skilled
operators
+
+
+
+
Humanities
Artisans
+
Science
+
+
+
+
Social sciences
+
+
+
+
General
+
Agriculture, etc.
+
+
Services
+
Health & care
+
+
+
+
+
+
+
+
+
+
14. Signal variables construction
• The groups:
–
–
–
–
–
–
voivodship,
age groups (but most concentrated below 30),
field of study,
gender,
year of the survey,
industry (for the wage equation).
• The signaling variables:
– The Rank = Lower education in group / Whole group
– The Equals = Equal education in group / Whole group
– The Group Size = Whole group
15. The results
Probit mfx by fields of study
Prob(working)
All
Age
Gender (1=Female)
Highschool
vocational
Highschool
Tertiary
Rank (% of lower
educ.)
Equals (% of equal
educ.)
Group size
N
0.002
-0.054***
Engineering
&
construction
0.004
-0.160***
0.018
0.087***
0.018
0.091***
Pedagogy
Humanities
Science
Social
science
Agriculture
Services
Health &
care
0.008
-0.013
0.009
0.003
0.006
-0.113***
0.001
0.015
0.004
-0.107**
0.003
-0.073***
0.010
-0.030
-0.178
0.048
0.068
0.100**
-0.024
0.081**
0.006
0.219***
0.213***
-0.071
0.181
0.291
0.155
0.221
0.079
0.180***
0.056
0.096
0.120**
0.151**
0.172***
0.006
-0.126
0.225
-0.163
-0.070
-0.031
-0.049
-0.160
-0.128
0.077***
0.124**
0.085
0.037
0.131**
0.089**
0.211**
0.028
0.099
0.000***
32,292
-0.000
6,485
0.000
1,305
0.002
1,312
-0.001**
2,461
-0.000
6,458
0.003
970
0.000
3,616
-0.000
1,206
** p<0.05; *** p<0.01
No signal controls:
Highschool
vocational
Highschool
Tertiary
0.020
0.025
0.013
0.084***
0.118**
0.089***
-0.283
0.040
0.051
0.080
0.014
0.037
0.016
-0.008
0.161
0.208
0.108
0.161
0.042
0.140***
0.006
0.021
0.035
0.047
0.129***
** p<0.05; *** p<0.01
dummies for years and fields of study (where appropriate) included but not reported
16. Additional: Wage regression
Wage regression
Age
Gender (1=Female)
Vocational
High school vocational
High school
Tertiary
Rank (% of lower
educ.)
Equals (% of equal
educ.)
Group size
Constant
R2
N
Change
observed +
Signal
(Male) Change
observed +
Signal
(Female)
Change
observed +
Signal
Last-year
students
Change
observed
Last-year
students +
Signal
24.908***
-230.618***
226.033***
269.447***
280.745***
447.067***
17.542***
-198.788***
306.456***
329.715***
302.565***
465.818***
25.357***
-227.299***
239.467***
284.125***
293.666***
465.321***
18.531***
-192.672***
321.845***
343.861***
313.863***
479.426***
9.689*
28.918***
345.702***
394.376***
342.596***
646.140***
278.870***
254.220***
248.866***
318.747***
-10.141
19.797
-22.415
51.972*
-76.465***
-133.679***
-190.577***
-49.548
-0.001***
40.273
0.24
7,575
-0.001***
34.378
0.26
3,467
0.001
193.791
0.27
1,520
-0.001***
-346.045***
0.25
1,947
1.213
0.24
7,575
-17.133
0.26
3,467
* p<0.1; ** p<0.05; *** p<0.01
dummies for years and fields of study included but not reported
17. Conclusions
• No observable signal in education, when
determining employment.
– Possible bias due to unobservable real job-search.
– Possible bias due to overeducation.
– Large heterogeneity between fields of studies.
• Some evidence of signal in education, when
determining first wages.
• Differences between genders (reinforcing the
findings of Chatterji et al., 2003)
18. Ideas for the future
• Combining the LFS dataset with Household
Survey and calculating the Heckman selection
equation.
• Comparison with a longer period of data (at
the expense of the ’field of study’ variable).
• Cross-country analysis.
• Deeper analysis of the signal differences
between genders.