Task Specific Human Capital

Size: px

Start display at page:

Download "Task Specific Human Capital"

Arlene Austin
5 years ago
Views:

1 Task Specific Human Capital Christopher Taber Department of Economics University of Wisconsin-Madison March 10, 2014

2 Outline Poletaev and Robinson Gathmann and Schoenberg

3 Poletaev and Robinson Human Capital Specificity: Evidence from the Dictionary of ational Titles and Displaced Worker Surveys, JOLE, 2008

4 The previous set of papers we have looked at have focused on and ation codes, but these are not ideal: codes are well defined, but not that closely related to what a worker actually does ation codes in principle seem better, but in practice don t work that well as many jobs are not that easy to classify Poletaev and Robinson push toward a better measure of job skills

5 Dictionary of ational Titles (DOT) Information on occupations or jobs Records complexity of job in dealing with data people things Also gives detailed information on many other things including reasoning development mathematical development time to learn techniques physical demands performance under stress Asks about the importance of these skills for the job

6 Factor Analysis They use factor analysis to try to get a sense of the jobs L characteristics K underlying skills (K < L) For each occupation j we observe the numerical score of the importance of each characteristic C lj For characteristic l on occupation j we write C lj =µ l + λ l θ j + ε lj where λ l and θ j are K dimensional vectors.

7 Can write C as an L dimensional vector Assuming ε and θ uncorrelated C j =µ + Λθ j + ε j Cov(C) =ΛΣ θ Λ + Σ ε After some normalizations (including that Σ θ = I) we can identify Λ and Σ ε sequentially by fitting the covariance as well as possible Given this the Cov(C) is across occupations. Choosing K can be done in different ways

8 Their factors Their main four factors (in order of importance) seem to pick up General skills Fine motor skills Physical strength Visual skills After knowing Λ we can then try to get an estimate of θ j for each 3 digit occupation code Think of these as like the π in the multidimensional human capital model They then use these and look at workers who lose their jobs via plant closings from the Displaced Worker Survey

9 Table 1 Displacement ing Patterns by, ation and Skill All Workers Males Females % 63.44% 64.96% ation % 66.48% 69.84% ation (45) % 59.66% 61.05% ation (13) % 50.49% 47.26% Main Skill (Order) % 38.96% 40.56% Skill PC % 23.98% 22.30% Skill PC % 28.05% 27.62%

10 Table 2 ation and Skill Portfolio Changes Skill PC1 Skill PC2 ation ation (45) ation (13) Males ation

11 Table 3 ation and Skill Portfolio ers by Status All Workers Males Females ation ation (45) ation (13) Skill PC Skill PC

12 Wage losses Next they want to look at the relationship between wage losses and switching This has all of the issues of selection in the returns to seniority literature, but since look at workers displaced through plant closing so it might not be quite as bad Still an issue for why one person switches industry after rehire and another does not

13 Table 4 Mean Log Wage Losses for, ation and Skill Portfolio ers and ers All Workers Males Females Unconditional Conditional Conditional Conditional (.0071) (.0079) (.0065) (.0087) (.0084) (.0112) (.0101) (.0138) ation (.0069) (.0080) (.0063) (.0092) (.0083) (.0117) (.0097) (.0149) ation (45) (.0074) (.0074) (.0067) (.0083) (.0087) (.0106) (.0104) (.0131) ation (13) (.0084) (.0068) (.0074) (.0073) (.0095) (.0096) (.0118) (.0112) Skill PC (.0120) (.0060) (.0107) (.0059) (.0137) (.0077) (.0172) (.0092) Skill PC (.0111) (.0061) (.0098) (.0061) (.0127) (.0079) (.0154) (.0095) Notes: Standard errors in parentheses. The conditioning variables are schooling, experience, pre-displacement tenure, years since displacement and weeks without work after displacement.

14 Table 5 Mean Log Wage Losses by ation and Skill Portfolio ers and er Status All Workers Males Females Skill PC1 Skill PC1 Skill PC1 Skill PC1 Skill PC1 Skill PC1 ation (.0092) (.0078) (.0107) (.0117) (.0103) (.0137) (.0149) (.0118) (.0172) ation (45) (.0084) (.0084) (.0539) (.0109) (.0107) (.0111) (.0744) (.0139) (.0132) (.0129) (.0775) (.0176) ation (13) (.0075) (.0097) (.0311) (.0114) (.0098) (.0126) (.0461) (.0144) (.0116) (.0152) (.0418) (.0188) Skill PC2 Skill PC2 Skill PC2 Skill PC2 Skill PC2 Skill PC2 ation (.0092) (.0082) (.0098) (.0117) (.0108) (.0127) (.0149) (.0125) (.0154) ation (45) (.0084) (.0089) (.0435) (.0101) (.0108) (.0117) (.0624) (.0130) (.0134) (.0137) (.0600) (.0160) ation (13) (.0076) (.0103) (.0252) (.0106) (.0099) (.0132) (.0358) (.0136) (.0118) (.0163) (.0351) (.0172) otes: Standard errors in parentheses.

15 Table 6 Mean Log Wage Losses by and Skill Portfolio ers and er Status All Workers Males Females Skill PC1 Skill PC1 Skill PC1 Skill PC1 Skill PC1 Skill PC (.0093) (.0077) (.0232) (.0121) (.0121) (.0100) (.0290) (.0155) (.0147) (.0118) (.0386) (.0192) Skill PC2 Skill PC2 Skill PC2 Skill PC2 Skill PC2 Skill PC (.0096) (.0079) (.0204) (.0112) (.0122) (.0104) (.0267) (.0144) (.0153) (.0122) (.0315) (.0177) Notes: Standard errors in parentheses.

16 We would like to look at this all based on pre-displacement tenure. Should be a bigger deal for workers with larger pre-displacement tenure We don t have that in the DWS, but we do have experience and can look at this difference.

17 Table 8 Incremental Wage Losses for More Experienced Workers er er (.0340) (.0288) Skill PC1 er Skill PC1 er (.0250) (.0480) er er er er (.0371) (.0337) (.0929) (.0567) Skill PC2 er Skill PC2 er (.0252) (.0458) er er er er (.0384) (.0395) (.0733) (.0549)

18 Outline Poletaev and Robinson Gathmann and Schoenberg

19 Gathmann and Schoenberg How General is Human Capital? A Task-Based Approach JOLE, 2011

20 Conceptual Framework Empirically there will be more than two dimension of skills but I follow them by presenting a model with two dimensions: Analytic (A) and manual (M) Let o index an occupation The skill of person i in occupation o at time t can be written as log (S iot ) =β o t A iot + (1 β o) t M iot

21 Task specific human capital grows (via a learning by doing technology) so t j iot =tj i + γ 0 H j it where the first part is unobserved skill and H j it will be observable (or at least a known function of observable stuff)

22 Human capital in accumulated in proportion to its importance in the occupation H A it = o β o O io t H M it = o (1 β o ) O io t Where O io t is the total lifetime experienct for individual i in occupation o at time t

23 This gives log (S iot ) =γ 0 [β o H A it + (1 β o ) H M it ] + β o ti A + (1 β o ) ti M γ 0 T iot + m i0 Let e po be the rental rate of skill in occupation o, then log (w iot ) =p o + γ o T iot + m i0

24 Putting this into any type of search model in which people care about income we get the following predictions 1 Individuals should make more moves earlier in lifecyle than later 2 They should move to occupations that are more similar to their previous ones 3 Tenure in the previous occupation should be more valuable when the current occupation is close to the previous one

25 Data They use two sources of data The German Qualification and Career Survey is a bit like the DOT Interview employees about their jobs (focus on men) Asks about 19 tasks that can be performed on their job Take average in answer across different jobs for task intensity Measure difference between jobs using angular seperation D oo 1 where j indexes a task [( J J j=1 q joq jo ) ( J )] j=1 q2 jo j=1 q2 jo

26 Table 1 Summary Statistics of Task Data Mean Std.Dev Example: Teacher (%) Example: Baker (%) Analytical Tasks: Research, evaluate or measure Design, plan or sketch Correct texts or data Calculate or bookkeeping Program Execute laws or interpret rules Analytical is Main Task Manual Tasks: Equip or operate machines Repair, renovate or reconstruct Cultivate Manufacture, install or construct Cleaning Serve or accommodate Pack, ship or transport Secure Nurse or treat others Manual is Main Task Interactive Tasks: Sell, buy or advertise Teach or train others Publish, present or entertain others Employ, manage personnel, organize, coordinat Interactive is Main Task Observations 52,718 1,

27 Table 2 Measuring Distances between ations Distance Measure (Angular separation) ation 1 ation 2 Distance Mean Standard Deviation Most Similar (all Education Groups): Paper and Pulp Processing Printer, Typesetter Wood Processing Metal Polisher Chemical Processing Plastics Processing Most Distant (all Education Groups): Banker Unskilled Construction Worker Banker Miner, Stone-Breakers Publicists, Journalist Unskilled Construction Worker Most Common ational Moves (Low-Skilled): Truck Driver, Conductor Store or Warehouse Keeper Unskilled Worker Store or Warehouse Keeper Assembler Store or Warehouse Keeper Most Common ational Moves (Medium-Skilled): Chemist, Physicist Electricians, Electrical Installation Sales Personnel Office Clerk Truck Driver, Conductor Store or Warehouse Keeper Most Common ational Moves (High-Skilled): Engineers Chemist, Physicist Entrepreneurs Office Clerk Accountant Office Clerk 0.080

28 The German Employee Panel is administrive income data >100,000 employees Administrative rather than survey From social security system (80% of Germans)

29 Table 3 Summary Statistics of West German Employee Panel Low-Skilled Medium-Skilled High-Skilled Percentage in Sample (%) Age (in Years) (6.03) (5.23) (5.27) Not German Citizen (%) Median Daily Wage (44.97) (43.33) (60.23) Log Daily Wage (0.45) (0.33) (0.42) Percentage censored Actual Experience (in Years) (5.34) (4.75) (4.58) ational Tenure (in Years) (4.09) (4.04) (3.88) Firm Tenure (in Years) (3.84) (3.66) (3.25) Task Tenure (in Years) (3.02) (2.92) (3.44) ational Mobility Distance of Move (0.23) (0.22) (0.18) Firm Mobility Most Common ations Warehouse Keeper Electrical Installation Engineer Assembler Locksmith Technician Conductor Mechanic, Machinist Accountant Unskilled Worker Office Clerk Office Clerk Office Clerk Conductor Researcher, Clergymen Number of Observations 244,759 1,003, ,930 Number of Individuals 20,846 79,396 16,735

30 Predictions They construct occupational human capital analogous to above (but normalized slightly differently) Lets look at predictions of model

31 Table 4 Observed Moves are More Similar than under Random Mobility Random Mobility Observed Mobility Mean th Percentile th Percentile th Percentile th Percentile th Percentile NOTE. The table repports selected moments of the distribution of observed occupational moves ("Observed Mobility") and compares it against what we would expect to observe under random mobility ("Random Mobility"). We calculate random mobility as follows: for each mover, we assume that the probability of going to any other occupation in the data is solely determined by the relative

32 Low-Skilled Medium-Skilled High-Skilled Y: Distance of Move (1) (2) (3) (1) (2) (3) (1) (2) (3) Experience (0.0007)*** (0.0007)*** (0.0016)*** (0.0005)*** (0.0005)*** (0.0012)*** (0.0009)*** (0.0010)*** (0.0028)*** Experience Squared (0.0000)*** (0.0000)*** (0.0001)*** (0.0000)*** (0.0000)*** (0.0001)*** (0.0001)*** (0.0001)*** (0.0001)*** ation Tenure (0.0004)*** (0.0006)*** (0.0002)*** (0.0003) (0.0005)*** (0.0008) Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes ation Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Individual Fixed Effects No No Yes No No Yes No No Yes Observations 45,124 45,124 45,124 99,390 99,390 99,390 13,680 13,680 13,680 Mean Distance of Move NOTE. The table reports results from a regression where the dependent variable is the distance between two occupations. The distance measure is the angular separation, based on 19 tasks. The sample consists of all occupational movers and results are reported separately by education group. Column (1) only includes experience and experience squared. Column (2) adds occupation tenure. Column (3) includes fixed worker effects to control for individual unobserved heterogeneity. All specifications include year and current occupation dummies. Robust standard errors clustered at the individual level are reported in parentheses. * p <.1. ** p <.05. *** p <.01. Table 5 Distance of Move Declines with Time in the Labor Market

33 Table 6 Similar Moves and the Correlation of Wages Across Jobs Low-Skilled Medium-Skilled High-Skilled Y: Log Daily Wage after Move (1) (2) (3) (1) (2) (3) (1) (2) (3) Wage Last Period (0.006)*** (0.006)*** (0.009)*** (0.004)*** (0.005)*** (0.006)*** (0.009)*** (0.013)*** (0.017)*** Wage Last Period*Distance (0.020)*** (0.016)*** (0.055)*** Distance of Move (0.090)*** (0.075)*** (0.262)*** Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes ation Dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 64,674 45,124 45, ,962 99,390 99,390 27,150 11,848 11,848 NOTE. The table reports results from wage regressions where the dependent variable is the log daily wages at the target occupation after swtiching both employers and occupations. Results are reported separately by education group. Column (1) uses the sample of firm movers as a benchmark for comparison. Column (2) repeats the analysis for the sample of joint occupational and firm movers. Column (3) adds the distance measure as well as the distance measure interacted with the wage in the last period. The distance measure is the angular separation based on all 19 tasks. All specifications include the log daily wage in the last period, actual experience, actual experience squared, year and current occupation dummies. For the low- and medium-skilled, we estimate OLS models. Standard errors in parentheses allow for clustering at the individual level. For the high-skilled, we estimate tobit models, and exclude censored observations at the previous occupation. Standard errors in parentheses are bootstrapped with 200 replications to allow for clustering at the individual level. * p <.1. ** p <.05. *** p <.01.

34 Low-Skilled Medium-Skilled High-Skilled Y: Log Daily Wage after Move (1) (2) (1) (2) (1) (2) Past ational Tenure (0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.002)*** (0.003)*** Past Tenure *Distance (0.004)* (0.002)*** (0.011)*** Distance of Move (0.010)*** (0.007)*** (0.030)*** Year Dummies Yes Yes Yes Yes Yes Yes ational Dummies Yes Yes Yes Yes Yes Yes Observations 45,124 45,124 99,390 99,390 13,680 13,680 * p <.1. ** p <.05. *** p <.01. Table 7 Past ational Tenure Matters for Wages NOTE. The table reports wage regressions where the dependent variable is the log wages in the target occupation after switching both employers and occupations. Column (1) in each specification controls for past tenure in the source occupation, experience, experience squared, as well as year and current occupation dummies. Column (2) additionally includes the distance measure and its interaction with past occupational tenure. The distance measure used is the angular separation based on all 19 tasks. For the low- and medium-skilled, we report results from OLS regressions. Standard errors in parentheses allow for clustering at the individual level. For the high-skilled, we estimate tobit models. Here, standard errors in parentheses are bootstrapped with 200 replications to account for clustering at the individual level.

35 Econometric Model Next they want to do something analogous to Altonji and Shakotko, Topel, Kambourov and Manovskii (but that Pavan might not like) log(w ioft ) =α o E it + γ o T iot + δ o O iot + λ o F ift + ε ioft where now T iot is task tenure, and everything else should be familiar Note at all coefficients are occupation specific Error term will involve individual, occupational match, task match, and firm match which will be correlated with pretty much everything

36 Approach: Use displaced workers during the first period they work after displacement (term tenure must be zero for these guys) Use fixed effects to deal with permanent characteristics (so people who were displaced twice) Also use control functions to deal with selection. Need exclusion restrictions age and age 2 for experience Altonji Shakotko instrument for occupational tenure local labor markets for task tenure I don t want to get into econometric details-not perfect, but makes sense

37 Table 8 Returns to Labor Market Skills: Least Squares Estimates Whole Sample Firm ers Displaced Workers (1) (2) (3) (4) (5) (6) Panel A: Low-Skilled Task Tenure (0.0010)*** (0.0015)*** (0.0031)*** ational Tenure (0.0006)*** (0.0006)*** (0.0010)*** (0.0009)*** (0.0016)*** (0.0017)*** Experience (0.0009)*** (0.0011)*** (0.0013)*** (0.0016)*** (0.0028)*** (0.0035)*** Experience Squared (0.0000)*** (0.0000)*** (0.0001)*** (0.0001)*** (0.0002)*** (0.0002)*** Firm Tenure (0.0006)*** (0.0006)*** Observations 244, ,759 64,674 64,674 9,275 9,275 Panel B: Medium-Skilled Task Tenure (0.0005)*** (0.0009)*** (0.0018)*** ational Tenure (0.0003)** (0.0003)*** (0.0004)*** (0.0004)*** (0.0007)*** (0.0008)*** Experience (0.0004)*** (0.0005)*** (0.0007)*** (0.0009)*** (0.0015)*** (0.0020)*** Experience Squared (0.0000)*** (0.0000)*** (0.0000)*** (0.0000)*** (0.0001)*** (0.0001)*** Firm Tenure (0.0002)*** (0.0002)*** Observations 1,003,823 1,003, , ,962 28,441 28,441 Panel C: High-Skilled Task Tenure (0.0029)*** (0.0042)*** (0.0090)*** ational Tenure (0.0010)*** (0.0012)** (0.0015)*** (0.0015)** (0.0033)*** (0.0031)** Experience (0.0015)*** (0.0026)*** (0.0024)*** (0.0037)*** (0.0059)*** (0.009) Experience Squared (0.0001)*** (0.0001)*** (0.0001)*** (0.0001)*** (0.0004)*** (0.0004)*** Firm Tenure (0.0011)*** (0.0011)*** Observations 172, ,930 28,982 28,982 2,919 2,919

38 Table 9 Returns to Labor Market Skills: Control Function Estimates Fixed Effects Control Function Control Function Displaced Sample Firm ers Displaced Sample (1) (2) (3) (4) (5) (6) Panel A: Low-Skilled Task Tenure (0.0049)*** (0.006)*** (0.010)*** ational Tenure (0.0030)*** (0.0030)** (0.0020) (0.003)*** (0.003)** (0.0040) Experience (0.0083)*** (0.0094)*** (0.002)*** (0.004)*** (0.003)*** (0.006)*** Experience Squared (0.0003)*** (0.0003)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Observations 9,275 9,275 64,674 64,674 9,275 9,275 Panel B: Medium-Skilled Task Tenure (0.0034)*** (0.004)*** (0.007)*** ational Tenure (0.0017)*** (0.0014)*** (0.001)*** (0.001)* (0.002)*** (0.002) Experience (0.0051)*** (0.0062)*** (0.001)*** (0.003)*** (0.002)*** (0.004)*** Experience Squared (0.0002)*** (0.0002)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Observations 28,441 28, , ,962 28,441 28,441 Panel C: High-Skilled Task Tenure (0.028) (0.010)** (0.027) ational Tenure (0.009) (0.008) (0.002)*** (0.005)*** (0.0013)*** (0.005)** Experience (0.031)** (0.056) (0.003)*** (0.022)*** (0.003)*** (0.022)*** Experience Squared (0.001) (0.002) (0.0003)*** (0.001)*** (0.0003)*** (0.001)*** Observations 2,919 2,919 28,982 28,982 2,919 2,919

SUPPLEMENT TO TRADE LIBERALIZATION AND LABOR MARKET DYNAMICS (Econometrica, Vol. 82, No. 3, May 2014, )

Econometrica Supplementary Material SUPPLEMENT TO TRADE LIBERALIZATION AND LABOR MARKET DYNAMICS (Econometrica, Vol. 82, No. 3, May 2014, 825 885) BY RAFAEL DIX-CARNEIRO APPENDIX B: SECTORAL DEFINITIONS