occupational tasks and changes in the wage structure sergio firpo, eesp-fgv, nicole fortin, ubc...
TRANSCRIPT
OCCUPATIONAL TASKS AND CHANGES IN THE WAGE STRUCTURE
Sergio Firpo, EESP-FGV,Nicole Fortin, UBCThomas Lemieux, UBC
SOLE/EALE 3rd World ConferenceJune 17-19, 2010London, UK
2
Goal of this paper Assess whether occupation-based explanations can
account for some of the change in U.S. wage inequality and explain the U-shaped pattern of wage changes / polarization.
Follows the recent introduction of occupation-based explanations in the wage inequality literature, 1. Nuanced view of technological change (ALM, 2003, AKK,
2006, Goos and Manning, 2008), Based on routine vs. non-routine tasks
2. Trade vs. offshoring Trade in labor services lead to classification of occupations
as offshorable/non-offshorable by Blinder (2007) and Jensen and Kletzer (2007).
3. Top-end occupations Piketty and Saez (2003) and Gabaix and Landier (2008)
3
Changes in Wage Inequality
0 10 20 30 40 50 60 70 80 90 100-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
Figure 4: Change in Real Wages by Percentile, Men
1974 to 1989
1989 to 2004
Source: Lemieux (2008)
4
Occupation-based Explanations Limited Empirical Evidence? But despite recent surge of interest for occupational-based
explanations, there is only limited evidence of how they have contributed to changes in wage inequality.
Autor, Levy, and Murname (2003) show that “routine jobs” tend to be in the middle of the wage distribution. Suggests this is where wages should decline most rapidly.
Autor, Katz, and Kearney (2006) make this argument more explicitly and present evidence of a U-shaped change in wages (well-established) since the late 1980s that coincides with U-shaped changes in the distribution of occupations.
But, showing that shapes coincide is not like formally showing that computerization does account for a specific fraction of changes in inequality, and how they compete with traditional explanations.
5
Contribution We assess the role of occupational tasks in three steps: 1. Provide some motivating evidence by
i) presenting a simple skill pricing model at the occupational level,
ii) showing the changes in the level (between) and dispersion (within) of wages by occupation (2-digit occupations).
2. Look at whether changes in occupational wages (level and dis-persion) are linked to occupational tasks measured in the O*NET(technological content, offshorability, etc.)
3. Put all this together with other factors in a decomposition using reweighting and re-centered influence function regressions (3-digit occupations)
6
Wage Setting Model Linear skill pricing equation for worker i in occupation j at time
t:
Sik is a set of K cognitive, non-cognitive, manual skills, etc.
The key difference relative to “skill-based” studies of inequality
is that the rjkt vary across occupations because of the
“unbundling” problem (Rosen, Heckman and Scheinkman).
The returns to these skills rjkt change over time because of
technology, offshoring, and other factors, with implications for both the level of wages across occupations and dispersion of wages within occupations.
7
Regression Model One simple way of capturing both the level (between) and
dispersion (within) effect is to run regressions of wage changes on the base period wage for each percentile of the within-occupation wage distribution
We estimate the following regressions:
where q indicates percentiles of the wage distribution.
Comparing αj by occupations will show “between” occupation changes
Sign and magnitude of βj will tell us if dispersion is increasing “within” occupation
9
Data Processing We use the 1983-85 and 2000-02 Outgoing Rotation Group
(ORG) Supplements of the Current Population Survey. Years chosen to keep consistent occupation coding and
have union status. Focus on males age 16-64, non-allocated, weighted by
hours (see Lemieux (2006))
We extract the nine deciles of the within-occupation wage distribution, i.e. wjt
q for q=10,20,...,90
At this point, we work with 2-digit occupations (45 of them) to have precise enough estimates of each decile
11
Descriptive Evidence: First step: U-Shaped Changes in Wages by Decile
AdminSup2
Serv_FoodFarm2
Serv_PersHand1Sales1Hand2Serv_Clean
Serv_Healht
Farm1Forest
Constr2
Transp1
Sales1
Produc2Secretairies
AdminS2
Record
Serv_Protect
Produc3
Health3Supervi
Transp2
CompOpConstr1
Mechanics
Produc1
Profes
EnginTech
Teachers
SalesCommSalesFinan
Health2
AdminSup1
ManagersOtherTechProfes
Health1
AdminExecuNatSc
MathScEngineers
Lawyers
-.2
0.2
.4W
age
chan
ge
.5 1 1.5 2 2.5 3 3.5Base period wage
Wage change Quadratic fit
1983-85 to 2000-02 change for each decile
Figure 1: Change in wage by 2-digit occupation
Descriptive Evidence:First step: U-Shaped Changes in Wages by Decile
AdminExecuManagers
EngineersMathScNatSc
Health1
Health2
ProfesTeachers
Lawyers
Profes
Health3EnginTech
OtherTech
Supervi
SalesFinanSalesComm
Sales1
Sales1
AdminSup1
CompOpSecretairiesRecord
AdminS2
AdminSup2
Serv_Protect
Serv_Food
Serv_Healht
Serv_CleanServ_PersMechanics
Constr1Produc1
Produc2
Produc3
Transp1
Transp2
Constr2
Hand1
Hand2
Farm1
Farm2
Forest
-.2
0.2
.4W
age
chan
ge
.5 1 1.5 2 2.5 3 3.5Base period wage
1983-85 to 2000-02 change for each decileFigure 1: Fitted change in wage by 2-digit occupation
12
13
Occupational TasksNon-routine vs. Automated vs. Offshored
ON-
SITE
14
Occupational Tasks Using the O*NET We combine various “Work Activities” and “Work Context” elements from
the O*NET 13 to construct five measures of occupational tasks
Technology/Offshorability 1) Information Content: occupations with high information content that
are likely to be affected by ICT technologies; they could also be offshored as in Jensen and Kletzer (2007) (JK)
2) Automation: occupations with high degree of potential/actual automation of jobs and is similar in spirit to the manual routine index of Autor et al. (2003).
Non-Offshorability Designed to capture aspects of job making it unlikely to be offshored 3) Face-to-Face Contact: if a job requires face-to-face personal interactions
with clients and/or co-workers, it is unlikely to be offshored 4) On-Site Job: reflects the first criteria used by Blinder (B), does the job
need to be done at a U.S. work location? 5) Decision-Making: again, jobs where frequent decision-making is
required will be less likely to be offshored, and is similar in spirit to the non-routine analytical index of Autor et al. (2003).
15
Occupational Tasks Using the O*NET
Technological Change/Offshorability Information Content
4.A.1.a.1 Getting Information (JK) 4.A.2.a.2 Processing Information (JK) 4.A.2.a.4 Analyzing Data or Information (JK) 4.A.3.b.1 Interacting With Computers (JK) 4.A.3.b.6 Documenting/Recording Information (JK)
Automation/Routine 4.C.3.b.2 Degree of Automation 4.C.3.b.7 Importance of Repeating Same Tasks 4.C.3.b.8 Structured versus Unstructured Work (reverse) 4.C.3.d.3 Pace Determined by Speed of Equipment 4.C.2.d.1.i Spend Time Making Repetitive Motions
16
Occupational Tasks Using the O*NET
Non-Offshorability Face-to-Face Contact
4.C.1.a.2.l Face-to-Face Discussions 4.A.4.a.4 Establishing and Maintaining Interpersonal
Relationships (JK, B) 4.A.4.a.5 Assisting and Caring for Others (JK, B) 4.A.4.a.8 Performing for or Working Directly with the Public (JK, B) 4.A.4.b.5 Coaching and Developing Others (B)
On-Site Job 4.A.1.b.2 Inspecting Equipment, Structures, or Material (JK) 4.A.3.a.2 Handling and Moving Objects 4.A.3.a.3 Controlling Machines and Processes 4.A.3.a.4 Operating Vehicles, Mechanized Devices, or Equipment 4.A.3.b.4 Repairing and Maintaining Mechanical Equipment (*0.5) 4.A.3.b.5 Repairing and Maintaining Electronic Equipment (*0.5)
17
Occupational Tasks Using the O*NET
Non-Offshorability Decision-Making
4.A.2.b.1 Making Decisions and Solving Problems (JK) 4.A.2.b.2 Thinking Creatively (JK) 4.A.2.b.4 Developing Objectives and Strategies 4.C.1.c.2 Responsibility for Outcomes and Results 4.C.3.a.2. Frequency of Decision Making
For each occupation, O*NET provides information on the “importance” and “level” of required work activity and on the “frequency” of five categorical levels of work context.
We follow Blinder (2007) in arbitrarily assigning a Cobb-Douglas weight of two thirds to the “importance” and one third to the “level” in a weighed sum for work activities. For work contexts, we simply multiply the frequency by the value of the level.
22
Occupational TasksSecond-Step: Regressions of Tasks on αj and βj
Call these five measures of task content (in each occupation j) TCjh, for h = 1, .. 5.
The second step regressions are
and
23
Occupational TasksSecond-Step: Regressions of Tasks on αj and βj
Table 2- Estimated Effect of Task Requirements on Intercept and Slope of Wage Change Regressions by 2-digit Occupation
(1) (2) (3) (4) (5) (6) (7) (8)TechnologyInformation content 0.010 0.027 0.015 0.010 0.017 0.040 0.046 0.030
(0.012) (0.011) (0.017) (0.018) (0.008) (0.012) (0.010) (0.015)Automation -0.035 -0.043 -0.025 -0.023 -0.046 -0.056 -0.055 -0.036
(0.013) (0.011) (0.017) (0.018) (0.009) (0.010) (0.015) (0.015)OffshorabilityNo Face-to-Face -0.040 -0.013 0.044 0.037 -0.052 -0.057 -0.043 -0.024
(0.017) (0.016) (0.023) (0.026) (0.011) (0.012) (0.019) (0.017)No On-Site Job -0.001 0.001 0.024 0.024 0.017 0.019 0.031 0.024
(0.007) (0.006) (0.009) (0.009) (0.005) (0.005) (0.007) (0.006)No Decision making -0.025 -0.008 -0.032 -0.025 -0.025 -0.055 -0.053 -0.020
(0.018) (0.018) (0.025) (0.029) (0.011) (0.015) (0.015) (0.021)Base wage no yes no yes no yes no yesR-square 0.400 0.550 0.450 0.440
Intercept Slope Intercept SlopeTasks included together Tasks included separately
25
Occupational TasksWithin and Between Changes and Automation
CompOpProcess
Serv_Healht
Produc1Produc2
Produc3
Opera1
Opera2
Hand1
-.2-.1
0.1
Wag
e ch
ange
.5 1 1.5 2 2.5 3 3.5Base period wage
1983-85 to 2000-02 change for each decileFigure 2b: F itted change in wage: Top-10 Auto
27
Occupational Tasks Within and Between Changes and On-Site
ManagersMathSc
Professor
Lawyers
ProfesSalesFinanSalesComm
Sales2
CompOpSecretairiesRecord
-.20
.2.4
Wag
e ch
ange
.5 1 1.5 2 2.5 3 3.5Base period wage
1983-85 to 2000-02 change for each decileFigure 2d. F itted change in wage: Bottom-10 On-Site
29
Main FindingsSecond-Step Regressions Occupational task requirements correlate in a reasonable
way with slopes and intercepts Occupation-specific slopes very important for fitting the
model/capturing curvature. Reflects heterogeneity in changes in within-occupation
inequality
But leaves two questions open: 1) What is the precise quantitative contribution of
occupations to changes in overall inequality? 2) What happens when we control for other factors
(education, union status, etc.)?
30
Reweighted RIF-Regression Decomposition:Third-Step Methodology:
We use a decomposition similar in spirit to Oaxaca-Blinder mean decomposition between time period 1 and time period 0,
ΔμO = E [Y|D = 1] - E [Y|D = 0]
= E [X|D = 1](β1-β0) + (E [X|D = 1] -E [X|D = 0]) β0
= ΔνS + Δν
X
wage structure effect composition effect but that works with other distributional statistic ν besides the mean i) the outcome variable, Y, has been replaced by by the recentered
influence function RIF(y; ν) of the statistics of interest ν (Firpo, Fortin, and Lemieux, 2009) For example, for the τ-th quantile of the distribution, qτ , we have :
RIF(y;)= qτ + IF(Y; qτ) = qτ + [τ –1(Y≤ qτ) ]/f(qτ)
Using the coefficients γν from a regression of RIF(y; ν) on X,
ΔνS = E [X|D = 1](γν
1- γν0) and Δν
X = (E [X|D = 1] -E [X|D = 0]) γν
0
31
Reweighted RIF-Regression Decomposition:Third-Step Methodology:
ii) Even in the case of the mean, if the true conditional expectation is not linear, the OB decomposition is biased (Barsky et al., 2002). Almost surely the case for other distributional statistics.
We address this issue using the modified decomposition:
ΔνS = E [X|D = 1](γν
1- γν01)
and
ΔνX = (E [X|D = 1] -E [X|D = 0]) γν
0+ Rν
where γν01 are the period 1 coefficients estimated in sample 0
reweighted to look like period 1, and the approximation error R is the difference between composition effects estimated by reweighting and RIF regressions.
iii) We address the omitted group issue by trying to pick a reasonable base group: HS graduates, non-union, 15-19 years of experience, and one standard deviation below the mean of the task measures.
32
RIF-Regressions Results The γν
0 and γν1
Table 1. Unconditional Quantile Regression Coefficients on Log Wages
Years: 1983/85 2000/02Quantiles: 10 50 90 10 50 90Explanatory VariablesUnion covered 0.208 0.406 -0.055 0.112 0.278 -0.073
(0.003) (0.004) (0.004) (0.003) (0.005) (0.006)Non-white -0.090 -0.140 -0.055 -0.040 -0.131 -0.071
(0.006) (0.004) (0.004) (0.006) (0.004) (0.006)Non-Married -0.162 -0.122 -0.015 -0.072 -0.111 -0.066
(0.004) (0.004) (0.004) (0.004) (0.004) (0.005)Education ( High School omitted) Primary -0.278 -0.392 -0.156 -0.390 -0.378 -0.070
(0.008) (0.007) (0.004) 0.012 (0.009) (0.005) Some HS -0.301 -0.146 -0.023 -0.396 -0.188 0.034
(0.007) (0.004) (0.004) (0.01) (0.005) (0.004) Some College 0.045 0.129 0.086 0.031 0.118 0.053
(0.005) (0.005) (0.004) (0.004) (0.004) (0.004) College 0.142 0.316 0.375 0.102 0.386 0.561
(0.004) (0.006) (0.007) (0.004) (0.005) (0.011) Post-grad 0.088 0.337 0.559 0.066 0.403 1.025
(0.005) (0.006) (0.011) (0.004) (0.006) (0.018)O*NET MeasuresInformation Content 0.048 0.072 0.015 0.038 0.061 0.030
(0.002) (0.002) (0.002) (0.002) (0.002) (0.003)Automation 0.021 -0.025 -0.047 0.009 -0.053 -0.035
(0.002) (0.002) (0.002) (0.002) (0.002) (0.003)No Face-to-Face 0.109 0.132 0.121 0.083 0.110 0.107
(0.003) (0.002) (0.003) (0.002) (0.002) (0.004)Non On-Site Job -0.007 0.021 0.037 -0.016 0.014 0.051
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)No Decision-Making -0.123 -0.127 -0.125 -0.109 -0.135 -0.118
(0.003) (0.003) (0.003) (0.003) (0.003) (0.004)Number of obs. 274,625 252,397
Note: 6 Experience classes also included in the regressions. Boostrapped standard errors (100 reps) are in parentheses.
34
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.2
0.2
.4
0 .2 .4 .6 .8 1Qu a n tile
2 0 0 0 -0 2
1 9 8 3 -8 5
Union
-.2-.1
5-.1
-.05
0
0 .2 .4 .6 .8 1Qu a n tile
Non- W hit e
-.5-.2
50
.25
.5.7
51
0 .2 .4 .6 .8 1Qu a n tile
Elem ent ar y
-.5-.2
50
.25
.5.7
51
0 .2 .4 .6 .8 1Qu a n tile
Dr op- O ut-.5
-.25
0.2
5.5
.75
1
0 .2 .4 .6 .8 1Qu a n tile
Som e College-.5
-.25
0.2
5.5
.75
1
0 .2 .4 .6 .8 1Qu a n tile
College
-.5-.2
50
.25
.5.7
51
0 .2 .4 .6 .8 1Qu a n tile
Post - G r aduat e
0.0
2.0
4.0
6.0
8.1
0 .2 .4 .6 .8 1Qu a n tile
I nf or m at ion
-.08
-.06
-.04
-.02
0.0
2
0 .2 .4 .6 .8 1Qu a n tile
Aut om at ion
.04
.06
.08
.1.1
2.1
4
0 .2 .4 .6 .8 1Qu a n tile
No Face2Face-.0
4-.0
20
.02
.04
.06
0 .2 .4 .6 .8 1Qu a n tile
No O n- Sit e
-.16
-.14
-.12
-.1-.0
8-.0
6
0 .2 .4 .6 .8 1Qu a n tile
No Decision- M aking
35
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.2
0.2
.4
0 .2 .4 .6 .8 1Quantile
2000-02
1983-85
Union
36
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.2
-.15
-.1-.0
50
.05
0 .2 .4 .6 .8 1Quantile
Non-Married
37
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.2
-.15
-.1-.0
50
0 .2 .4 .6 .8 1Quantile
Non-White
38
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.7
-.6-.5
-.4-.3
-.2-.1
0 .2 .4 .6 .8 1Quantile
Experience < 5
39
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
0
.02
.04
.06
.08
.1
0 .2 .4 .6 .8 1Quantile
Information
40
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.0
8-.0
6-.0
4-.0
20
.02
0 .2 .4 .6 .8 1Quantile
Automation
41
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
.0
4.0
6.0
8.1
.12
.14
0 .2 .4 .6 .8 1Quantile
No Face2Face
42
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.0
4-.0
20
.02
.04
.06
0 .2 .4 .6 .8 1Quantile
No On-Site
43
Figure 1. Unconditional Quantile Regression Coefficients: 1983/85-2000/02
-.1
6-.1
4-.1
2-.1
-.08
-.06
0 .2 .4 .6 .8 1Quantile
No Decision-Making
44
Decomposition Results Δν
O = ΔνS+ Δ ν
X and ΔνX = (E [X|D = 1] -E [X|D = 0]) γν
0
Wage structure effects essential to account for decreasing inequality at the bottom and increasing inequality at the top
For composition effects, only de-unionization has the right signs
Table 3. Decomposition Results 1983/85-2000/02 90-10 50-10 90-50 Variance GiniTotal Change 0.0622 -0.0830 0.1452 0.0443 0.0085
(0.0149) (0.0147) (0.0034) (0.0013) (0.0004)Wage Structure -0.0208 -0.1287 0.1079 0.0170 0.0041
(0.0114) (0.0113) (0.0043) (0.0015) (0.0004)Composition 0.0829 0.0457 0.0373 0.0273 0.0043
(0.0055) (0.0051) (0.0031) (0.0008) (0.0003)Composition Effects:Union 0.0215 -0.0159 0.0375 0.0080 0.0046
(0.0039) (0.0036) (0.0007) (0.0002) (0.0001)Education -0.0009 0.0111 -0.0120 -0.0019 -0.0041
(0.0022) (0.0018) (0.0018) (0.0003) (0.0001)Experience 0.0141 0.0186 -0.0045 0.0032 -0.0022
(0.0011) (0.0009) (0.0009) (0.0003) (0.0001)Technology 0.0030 0.0073 -0.0043 0.0430 0.0182
(0.0022) (0.0018) (0.0018) (0.0002) (0.0001)Offshorability 0.0073 0.0033 0.0040 -0.0204 -0.0054
(0.0008) (0.0007) (0.0007) (0.0002) (0.0001)Other 0.0103 0.0009 0.0093 0.0048 0.0025
(0.0016) (0.0014) (0.0016) (0.0002) (0.0001)
45
Decomposition ResultsΔν
S = E [X|D = 1](γν1- γν
01)
For wage structure effects, both technology and offshorability have the right signs and sizeable comparable “magnitude”
Table 3. Decomposition Results 1983/85-2000/02 90-10 50-10 90-50 Variance GiniWage Structure Effects:Total -0.0208 -0.1287 0.1079 0.017 0.0041
(0.0114) (0.0113) (0.0043) (0.0015) (0.0004)Union 0.0111 -0.0048 0.0160 0.0040 0.0031
(0.0031) (0.0025) (0.0021) (0.0006) (0.0001)Education 0.0890 0.0169 0.0721 0.0324 0.0072
(0.0029) (0.0024) (0.0025) (0.0019) (0.0005)Experience -0.0337 -0.0081 -0.0257 -0.0097 -0.0052
(0.009) (0.0098) (0.0092) (0.0037) (0.001)Technology 0.0484 -0.0322 0.0806 0.4520 0.2067
(0.0178) (0.0154) (0.0176) (0.0027) (0.0008)Offshorability 0.0500 -0.0218 0.0718 0.3334 0.0888
(0.0115) (0.0131) (0.0113) (0.0024) (0.0007)Other -0.0516 -0.0274 -0.0242 -0.0213 -0.0074
(0.0112) (0.0094) (0.0079) (0.0017) (0.0004)Residual -0.1339 -0.0511 -0.0828 -0.7736 -0.2890
(0.0059) (0.0054) (0.0048) (0.0054) (0.0015)
46
Figure 5. Decomposition of Total Change into Composition and Wage Structure Effects
-.05
0.0
5.1
.15
Log
Wag
e C
hang
e
0 .2 .4 .6 .8 1Quantile
Total ChangeWage StructureComposition
Change in Log Wages 2000/02-1983/85
48
Figure 6. Decomposition of Composition Effects
-.05
0.0
5.1
.15
Log
Wag
e C
hang
e
0 .2 .4 .6 .8 1Quantile
UnionEducationExperienceTechnologyOffshorability
Composition Effects
50
Figure 7. Decomposition of Wage Structure Effects
-.05
0.0
5.1
.15
Log
Wag
e C
hang
e
0 .2 .4 .6 .8 1Quantile
UnionEducationExperienceTechnologyOffshorability
Wage Structure Effects
51
Figure 5. Decomposition of Total Change into Composition and Wage Structure Effects
-.0
50
.05
.1.1
5Lo
g W
age
Cha
nge
0 .2 .4 .6 .8 1Quantile
UnionEducationExperienceTechnologyOffshorability
Total Effects
52
Conclusion Yes, occupational tasks help explain the U-shape feature of
changes in the wage distribution
But this is not just technology related. Offshoring works just as well, and so do
i) union wage effects, an ii) de-unionization (composition effect) iii) education wage effects, iv) demographics wage effects (non-married, non-white,
experience< 5)
Rare case where we seem to be able to explain all that needs to be explained!
THANK YOU!