Reconciling Occupational Mobility in the Current Population Survey - - PowerPoint PPT Presentation

▶
reconciling occupational mobility in the current
SMART_READER_LITE
LIVE PREVIEW

Reconciling Occupational Mobility in the Current Population Survey - - PowerPoint PPT Presentation

Oct 01, 2022 5 likes •555 views

Reconciling Occupational Mobility in the Current Population Survey Christian vom Lehn 1 Cache Ellsworth 2 Zach Kroff 3 1 Brigham Young University 2 Columbia University 3 U.S. Census Bureau QSPS Workshop September 19, 2019 Introduction Mobility

slide-1
SLIDE 1

Reconciling Occupational Mobility in the Current Population Survey

Christian vom Lehn1 Cache Ellsworth2 Zach Kroff3

1Brigham Young University 2Columbia University 3U.S. Census Bureau

QSPS Workshop September 19, 2019

slide-2
SLIDE 2

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

slide-3
SLIDE 3

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

slide-4
SLIDE 4

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

  • Current Population Survey (CPS) key data source for measuring rates (big sample, high

frequency observation, etc.)

slide-5
SLIDE 5

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

  • Current Population Survey (CPS) key data source for measuring rates (big sample, high

frequency observation, etc.)

  • Problem: Different survey measures within CPS generate different levels and trends of
  • ccupational mobility
slide-6
SLIDE 6

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

  • Current Population Survey (CPS) key data source for measuring rates (big sample, high

frequency observation, etc.)

  • Problem: Different survey measures within CPS generate different levels and trends of
  • ccupational mobility
  • March CPS (retrospective): 1 digit annual mobility 2-4%, trending down
slide-7
SLIDE 7

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

  • Current Population Survey (CPS) key data source for measuring rates (big sample, high

frequency observation, etc.)

  • Problem: Different survey measures within CPS generate different levels and trends of
  • ccupational mobility
  • March CPS (retrospective): 1 digit annual mobility 2-4%, trending down
  • Linked CPS files (longitudinal): 1 digit annual mobility 19-24%, trending up
slide-8
SLIDE 8

Introduction Mobility Measures Estimation Results Applications Appendix

Motivation

  • Occupations provide useful lens for understanding many economic phenomena –

inequality, trade, displaced workers, life cycle earnings, etc.

  • Accurate measures of the rate at workers switch occupations important for these

questions and others (labor market fluidity, etc.)

  • Current Population Survey (CPS) key data source for measuring rates (big sample, high

frequency observation, etc.)

  • Problem: Different survey measures within CPS generate different levels and trends of
  • ccupational mobility
  • March CPS (retrospective): 1 digit annual mobility 2-4%, trending down
  • Linked CPS files (longitudinal): 1 digit annual mobility 19-24%, trending up
  • QUESTIONS: What is the actual rate of occupational mobility? Is it rising or falling?

Implications?

slide-9
SLIDE 9

Introduction Mobility Measures Estimation Results Applications Appendix

What We Do

  • Use linked CPS data with multiple measures of occupational switching and estimate

actual rate of mobility using other labor market outcomes

  • Key assumption: Measurement error in each measure of switching is conditionally

independent

  • Estimation: overidentified GMM using multiple labor market outcomes
  • Also obtain estimates of magnitudes and trends in measurement error and correlated

worker characteristics

  • Applications:
  • Construct corrected time series of monthly occupational switching
  • Revisit findings in literature on worker level impacts of trade
slide-10
SLIDE 10

Introduction Mobility Measures Estimation Results Applications Appendix

Findings

  • Occupational mobility is falling over time, consistent with declining labor market fluidity

and migration

  • March CPS: right trend, but estimated actual rate is 70% higher (~2 pp)
  • Linked CPS: wrong trend, measurement error worsening over time
  • Measurement error in linked CPS correlated with workers who are male, nonwhite,

hispanic, young, and in certain occupations; but observables can’t explain upward measurement error trend

  • Trade applications:
  • Workers in tradable occupations less likely to switch occupations (contrary to

Ebenstein et al. (2014))

  • Slower worker adjustment implies lower welfare gains and slower transition to

steady state in a trade liberalization (vis a vis Artuc, Chaudhuri and Mclaren (2010))

slide-11
SLIDE 11

Introduction Mobility Measures Estimation Results Applications Appendix

Background Literature

  • Measuring Occupational Mobility: Moscarini and Vella (2003), Moscarini and

Thomsson (2007), Kambourov and Manovskii (2008), Lale (2012), Kamborouv and Manovskii (2013), Molloy et al. (2014), Lale (2017), Forsythe (2018)

  • Addressing Measurement and Misclassification Error: Aigner (1973), Mathiowetz and

Ouncan (1988), Mathiowetz (1992), Kane et al. (1999), Black et al. (2000), Bound, Brown and Mathiowetz (2001)

  • Economics of Occupational Mobility: Kambourov and Manovskii (2009), Groes et al.

(2014), Papageorgiou (2014), Molloy et al. (2014), Gorry et al. (2014), Artuc and Mclaren (2015), Wiczer (2015), Guvenen et al. (2015), Cortes (2016), Huckfeldt (2016), Gervais et

  • al. (2016), Traiberman (2017), Cortes and Gallipoli (2017), Cubas and Silos (2017),

Robinson (2018), Xu (2019), Carrillo-Tudela et al. (2019)

slide-12
SLIDE 12

Introduction Mobility Measures Estimation Results Applications Appendix

MEASURES OF OCCUPATIONAL MOBILITY

slide-13
SLIDE 13

Introduction Mobility Measures Estimation Results Applications Appendix

Background on Current Population Survey

  • Current Population Survey (CPS): monthly survey of 60,000 households, key source of

labor market data

  • Households surveyed for four consecutive months, out of sample for next eight months,

sampled again for four consecutive months

Month Interview

Wage Wage 1 2 3 4 5 6 7 8 14 15 16 9 10 12 13 11 5 6 7 8 1 2 3 4 Figure 1, Mueller (2017)

  • Additional supplements administered annually – annual socieconomic, job tenure,
  • ccupational mobility, displaced workers, fertility and marriage, voting, etc.
  • Large, representative, frequent sample makes it key data source for measuring
  • ccupational outcomes
slide-14
SLIDE 14

Introduction Mobility Measures Estimation Results Applications Appendix

Measuring Occupational Mobility in the CPS

  • Occupational mobility: fraction of workers employed presently and employed a year ago

who have different occupations

  • March CPS asks workers: “What was your longest job during [past year]?” (retrospective

measure)

  • Easy, convenient to compute – no linking required
  • Dependently coded – respondent must identify job description has changed
  • Relies on recall, and potentially imprecise timing (timing better in mobility

supplement)

  • Forces respondent to filter/decide what constitutes an occupational switch

(especially w/in firm)

  • Alternatively, longitudinally link individual responses:
  • Point-in-time comparison avoid recall/timing precision concerns
  • No dependent coding – independent coding errors could be large
  • Can’t observe movers; restricted to individuals remaining at same address
  • Can observe wage changes
slide-15
SLIDE 15

Introduction Mobility Measures Estimation Results Applications Appendix

Measurement Details and Sample Restrictions

  • Use responses in March CPS supplements 1980-2018, linked longitudinally (Rivera Drew

et al. (2014), Madrian and Lefgren (2000))

  • Drop all imputed observations (inc. whole sample) and linked responses responses with

inconsistent sex, race, age, educ.

  • Must be 18+ and employed this year and last year in non-gov’t industries
  • Occupational coding changes over time; apply consistent coding scheme following Dorn

(2009) and Autor and Dorn (2013)

  • (Talk) Report one digit outcomes (6 occupations); (Paper) Report one, two and three digit
  • utcomes (6, 17, and 325 occupations)
slide-16
SLIDE 16

Introduction Mobility Measures Estimation Results Applications Appendix

Comparing Occupational Mobility Measures (1 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 Annual Occupational Switching Rate March - Unrestricted March - Long. Restrictions JT Supplement

slide-17
SLIDE 17

Introduction Mobility Measures Estimation Results Applications Appendix

Comparing Occupational Mobility Measures (1 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 Annual Occupational Switching Rate March - Unrestricted March - Long. Restrictions JT Supplement 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.19 0.2 0.21 0.22 0.23 0.24 0.25 Annual Occupational Switching Rate

Two and Three Digit

slide-18
SLIDE 18

Introduction Mobility Measures Estimation Results Applications Appendix

ESTIMATION FRAMEWORK

slide-19
SLIDE 19

Introduction Mobility Measures Estimation Results Applications Appendix

Improving Our Estimates

  • GOAL: Use multiple (noisy) signals to improve aggregate estimates
  • Linked CPS allows us observation of “conflicting” responses – claim switch

longitudinally/retrospectively, but no switch retrospectively/longitudinally

  • Use extension of Kane, Rouse and Staiger (1999) – use relationship with other
  • bservables to evaluate relative contribution of noise and signal
slide-20
SLIDE 20

Introduction Mobility Measures Estimation Results Applications Appendix

Simple Regression Framework

  • Cond. exp. of labor market outcome Yit for individual i in year t, is given by:

E[Yit | SWit,Xit] = β0,t +β1,tSWit + Xitβ2,t

  • SWit: binary indicator for occupational switching
  • Xit: individual characteristics
  • SWit unobserved; only observe two noisy and conditionally independent signals ˜

SW

L it

(longitudinal) and ˜ SW

R it (retrospective):

P( ˜ SW

R it = 1 | SWit, ˜

SW

L it,Xit,Yit) = αR,0,t +αR,1,tSWit + XitαR,X,t

P( ˜ SW

L it = 1 | SWit, ˜

SW

R it ,Xit,Yit) = αL,0,t +αL,1,tSWit + XitαL,X,t

  • Probability of actual switch given by:

P(SWit = 1 | Xit) = δ0,t + Xitδ1,t

slide-21
SLIDE 21

Introduction Mobility Measures Estimation Results Applications Appendix

Estimation and Identification

  • Construct four indicator variables spanning all realizations of both signals

˜

Zit = [˜ Zi,1,t; ˜ Zi,2,t; ˜ Zi,3,t; ˜ Zi,4,t] (ex.˜ Zi,1,t = 1 if ˜ SW

R it = ˜

SW

L it = 1)

  • Estimate parameters year by year via two-stage GMM using moments E

˜

Zit

  • , E

˜

ZitYit

  • ,

E ˜

ZitXit

  • , E[X′

itYit]

  • Intuition:
  • When both signals indicate switch, estimated relationship with outcome closest to

“truth” (independence)

  • When signals disagree, gauge “accuracy” by relative magnitude of relationship with
  • utcome
  • Example:

All L=Y,R=Y L=N,R=N L=N,R=Y L=Y,R=N Switching FT/PT 9.6% 28.0% 8.7% 22.0% 10.8%

  • Note: more moments than parameters – overidentification allows for test of independence

assumption

slide-22
SLIDE 22

Introduction Mobility Measures Estimation Results Applications Appendix

Implementation

  • Potentially many outcome variables to choose from; how decide?
  • Require the following criteria:
  • Outcome has a priori plausible relationship with occupational switching
  • Outcome available for all samples, 1980-2018
  • Outcome ex ante plausibly uncorrelated with measurement error (exclude related

survey outcomes, i.e. industry)

  • Jointly estimate using following set of outcome variables
  • Indicator for whether part-time/full-time status differs across responses
  • Indicator for whether responses regarding prior year’s hourly wage differ by 10% or

more

  • Indicator for having more than one employer in the prior year
  • Indicator for whether number of weeks worked last year is more than 26 (half-year)
  • Controls: age, age squared, sex, white/nonwhite, hispanic, marital status, educational

attainment, two digit occ. fixed effects (in 2nd year)

slide-23
SLIDE 23

Introduction Mobility Measures Estimation Results Applications Appendix

Selecting Variables

  • Potentially many outcome variables to choose from; how decide?
  • Require the following criteria:
  • Outcome has a priori plausible relationship with occupational switching
  • Outcome available for all samples, 1980-2018
  • Outcome ex ante plausibly uncorrelated with measurement error (exclude related

survey outcomes, i.e. industry)

  • Jointly estimate using following set of outcome variables
  • Indicator for whether part-time/full-time status differs across responses
  • Indicator for whether responses regarding prior year’s hourly wage differ by

10% or more

  • Indicator for having more than one employer in the prior year
  • Indicator for whether number of weeks worked last year is more than 26

(half-year)

  • Controls: age, age squared, sex, white/nonwhite,hispanic, marital status, educational

attainment, two digit occ. fixed effects (in 2nd year)

slide-24
SLIDE 24

Introduction Mobility Measures Estimation Results Applications Appendix

RESULTS

slide-25
SLIDE 25

Introduction Mobility Measures Estimation Results Applications Appendix

Actual Occupational Mobility

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

  • Est. Occ. Mob.
  • Ret. Occ. Mob

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.05 0.1 0.15 0.2 0.25

  • Est. Occ. Mob.
  • Long. Occ. Mob

Two and Three Digit

slide-26
SLIDE 26

Introduction Mobility Measures Estimation Results Applications Appendix

Error Rates

False Positive Rate: P(SW = 0 | ˜ SW

j = 1); False Negative Rate: P(SW = 1 | ˜

SW

j = 0)

False Positive False Negative

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year

  • 0.4
  • 0.2

0.2 0.4 0.6 0.8 1 Longitudinal Retrospective 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 Longitudinal Retrospective

slide-27
SLIDE 27

Introduction Mobility Measures Estimation Results Applications Appendix

Actual Occupational Mobility by Outcome

Number of employers in past year > 1 Switched full-time/part-time work status

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

  • Est. Occ. Mob.

Joint Estimate 1980 1985 1990 1995 2000 2005 2010 2015 2020

Year 0.05 0.1 0.15

  • Est. Occ. Mob.

Joint Estimate

Weeks employed between year t − 1 and t > 26 % ∆ hourly wage > 10%

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year

  • 0.02

0.02 0.04 0.06 0.08 0.1 0.12 0.14

  • Est. Occ. Mob.

Joint Estimate 1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

  • 0.1
  • 0.05

0.05 0.1 0.15 0.2 0.25 0.3

  • Est. Occ. Mob.

Joint Estimate

slide-28
SLIDE 28

Introduction Mobility Measures Estimation Results Applications Appendix

Correlates with Error and Occupational Switching

  • Compare regression coefficients of individual characteristics across measures
  • 0.03
  • 0.02
  • 0.01

0.01 0.02 0.03 0.04 0.05 0.06 0.07 Male Hispanic Nonwhite Married Below HS HS degree Some Coll Actual Longitudinal Retrospective

slide-29
SLIDE 29

Introduction Mobility Measures Estimation Results Applications Appendix

Correlates with Error and Occupational Switching (cont’d)

  • Compare regression coefficients of individual characteristics across measures

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65

  • Occ. Switching Rate (All Else Constant)

Age

  • 0.04
  • 0.02

0.02 0.04 0.06 0.08 0.1 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 Occ Switching Rate (Demeaned, All Else Constant) Age Actual Longitudinal Retrospective

slide-30
SLIDE 30

Introduction Mobility Measures Estimation Results Applications Appendix

Correlates with Error and Occupational Switching (cont’d)

  • Compare regression coefficients of individual characteristics across measures
  • 0.15
  • 0.1
  • 0.05

0.05 0.1 0.15 Exec/Mananger Mgmt Related

  • Profess. Spec.

Technicians Sales Admin Support Housekeeping Protect Serv Other Serv Farm Mgmt Other Farm Mechanic/Repair Construction Extraction

  • Precis. Prod.

Operators/Assemblers Actual Longitudinal Retrospective

slide-31
SLIDE 31

Introduction Mobility Measures Estimation Results Applications Appendix

...But Observables Cannot Explain Upward Trend

  • Compute residualized measures of false positive rate for longitudinal occupational

switching – observables do not explain trend

1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

  • Avg. False Positive

False Pos. - Resid reported and act. switch False Pos. - Resid reported only

slide-32
SLIDE 32

Introduction Mobility Measures Estimation Results Applications Appendix

Total Occupational Mobility Estimates

  • Estimation sample necessarily removes movers; construct total mobility rate with and w/o

adjusting mover switching rates

1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11

Retrospective - Unrestricted Non-Mover Corrected All Corrected PSID Comparison

slide-33
SLIDE 33

Introduction Mobility Measures Estimation Results Applications Appendix

Robustness

  • Findings robust to:
  • Using Job Tenure and Occupational Mobility supplement

Results

  • Including additional outcomes/controls:

Results

  • Outcomes: alternate measures of income/hours changes, changes in

employer contributions to health insurance plans

  • Controls: past year FT status, usual hours worked, veteran status,

self-employed, state FE, disability status, proxy survey response, multiple job holding, no occ FE

slide-34
SLIDE 34

Introduction Mobility Measures Estimation Results Applications Appendix

Applications

  • Consider applications of our results to other measurement and literature using worker

flows in CPS

  • Adjust for error in monthly occupational switching rates
  • Revisit switching patterns and wages changes of workers in tradable jobs

(Ebenstein, Harrison, McMillan and Phillips (2014, Restat))

  • Revisit welfare gains from trade in structural discrete choice model (Artuc,

Chaudhuri, and Mclaren (2010, AER))

slide-35
SLIDE 35

Introduction Mobility Measures Estimation Results Applications Appendix

Application 1: Measurement Error in the Monthly CPS

  • Monthly occupational transitions subject to same types of measurement error
  • Pre-1994, independent coding (like longitudinal)
  • Post-1994, dependent coding (like retrospective)

1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

0.05 0.1 0.15 0.2 0.25 0.3 0.35

One Digit Two Digit Three Digit

slide-36
SLIDE 36

Introduction Mobility Measures Estimation Results Applications Appendix

Corrected Monthly Occupational Mobility

  • Compute predicted probability of switching, given observables P(SW = 1 | X,Y, ˜

SW):

  • bserve all X, only PT/FT switch and number of employers (since 1994)

Details

slide-37
SLIDE 37

Introduction Mobility Measures Estimation Results Applications Appendix

Corrected Monthly Occupational Mobility

  • Compute predicted probability of switching, given observables P(SW = 1 | X,Y, ˜

SW):

  • bserve all X, only PT/FT switch and number of employers (since 1994)

Details

  • Compare to Moscarini and Thomsson (2007) – rely on judgment calls about reasonable

switching patterns

slide-38
SLIDE 38

Introduction Mobility Measures Estimation Results Applications Appendix

Corrected Monthly Occupational Mobility

  • Compute predicted probability of switching, given observables P(SW = 1 | X,Y, ˜

SW):

  • bserve all X, only PT/FT switch and number of employers (since 1994)

Details

  • Compare to Moscarini and Thomsson (2007) – rely on judgment calls about reasonable

switching patterns

1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

0.01 0.015 0.02 0.025 0.03 0.035

Predicted MT Raw

slide-39
SLIDE 39

Introduction Mobility Measures Estimation Results Applications Appendix

Time Aggregation

  • Monthly switching rates only slightly lower than annual – suggests caution

in time aggregation

  • Example: Assume independent arrival rate of occupational switching

shock: SW A

t = 1−(1− SW M t )12

  • Implies annual mobility in the range 20-40%; too large
  • Procedure is sensitive to independence assumption and assumption of

minimal heterogeneity in switching rates.

slide-40
SLIDE 40

Introduction Mobility Measures Estimation Results Applications Appendix

Application 2: Estimates of Trade on Mobility and Wages

  • Ebenstein, Harrison, McMillan and Phillips (2014) study empirical measures of

trade/offshoring exposure

  • Identify losses to occupational displacement with IV regression:
  • Instrument: Tradability of an occupation (based on industry/occupation exposure)
  • Endogenous variable: Longitudinal occupational switching
  • Outcome variable: Log wage changes over time

E

  • ˜

SW

L i,o,t | Tradableo,Xi,o,t

  • = η0 +η1Tradableo + Xi,o,tη2

E

  • ∆ln(wi,o,t) | ˜

SW

L i,o,t,Xi,o,t

  • = ξ0 +ξ1 ˜

SW

L i,o,t + Xi,o,tξ2

  • IV won’t correct measurement error (non-classical) and instrument may be correlated with

error

slide-41
SLIDE 41

Introduction Mobility Measures Estimation Results Applications Appendix

Estimates of Mobility and Wages for Tradable Jobs

  • Jointly estimate their specification with measurement error model; get opposite results

Details

slide-42
SLIDE 42

Introduction Mobility Measures Estimation Results Applications Appendix

Estimates of Mobility and Wages for Tradable Jobs

  • Jointly estimate their specification with measurement error model; get opposite results

Details

Parameter One Digit Two Digit Three Digit IV estimate of occ. switching on wages, uncorrected (˜

ξIV

1 )

  • 0.119

(0.033)

  • 0.173

(0.022)

  • 0.106

(0.016) IV estimate of occ. switching on wages, corrected (ˆ

ξIV

1 )

0.403 (0.117) 0.559 (0.075) 0.348 (0.055) Difference

  • 0.522

(0.149)

  • 0.732

(0.094)

  • 0.454

(0.071) First stage, tradable occupation on

  • ccupational switching, uncorrected (˜

ηIV

1 )

0.045 (0.002) 0.070 (0.002) 0.091 (0.002) First stage, tradable occupation on

  • ccupational switching, corrected (ˆ

ηIV

1 )

  • 0.013

(0.002)

  • 0.022

(0.002)

  • 0.028

(0.003) Difference 0.058 (0.003) 0.091 (0.003) 0.119 (0.003)

slide-43
SLIDE 43

Introduction Mobility Measures Estimation Results Applications Appendix

Application 3: Structural Estimates of Welfare from Trade Liberalization

  • Artuc, Chaudhuri and Mclaren (2010) structurally estimate welfare consequences of trade

liberalization with worker mobility

  • Key parameter: cost of switching industries; disciplined by worker mobility rates
  • Acknowledge potential measurement error in retrospective measures and “inflate”
  • Assume that worker switching between industries is independent Poisson shock

and correct for time aggregation

  • Benchmark to annual flows in NLSY
  • Raises industry mobility rates by 130%
  • Occupational mobility rates suggest that needed correction is half as big
  • Re-estimate parameters of their model with correction consistent with findings for
  • ccupational mobility – lower flows imply higher costs of moving
slide-44
SLIDE 44

Introduction Mobility Measures Estimation Results Applications Appendix

Welfare Gains/Losses from Trade Liberalization

  • Results depend on discount factor, report for both β = 0.97 and β = 0.9; simulate 30%

reduction in tariffs on imported mfg goods

slide-45
SLIDE 45

Introduction Mobility Measures Estimation Results Applications Appendix

Welfare Gains/Losses from Trade Liberalization

  • Results depend on discount factor, report for both β = 0.97 and β = 0.9; simulate 30%

reduction in tariffs on imported mfg goods

  • Welfare gains for exposed workers on impact half as big and convergence much slower

β = 0.97 β = 0.90

  • 5

5 10 15 20 25 30 Time Period (Years) 1 1.005 1.01 1.015 1.02 1.025 1.03 1.035 Mfg Worker Welfare (=1 in Original SS) ACM Results Reparametrized Results

  • 5

5 10 15 20 25 30 Time Period (Years) 0.94 0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 Mfg Worker Welfare (=1 in Original SS) ACM Results Reparametrized Results

slide-46
SLIDE 46

Introduction Mobility Measures Estimation Results Applications Appendix

Conclusion

  • Estimate actual level of occupational mobility in CPS using GMM

approach

  • Occupational mobility trending down over time, but 70% higher than

retrospective measures

  • Measurement error in linked responses worsening
  • Use estimates of actual occupational mobility in several applications
  • Correcting monthly measures of occupational mobility
  • Worker mobility in response to trade shocks
slide-47
SLIDE 47

Introduction Mobility Measures Estimation Results Applications Appendix

Comparing Occupational Mobility Measures (2 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 Annual Occupational Switching Rate March - Unrestricted March - Long. Restrictions JT Supplement 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.24 0.25 0.26 0.27 0.28 0.29 0.3 0.31 0.32 0.33 Annual Occupational Switching Rate

Back

slide-48
SLIDE 48

Introduction Mobility Measures Estimation Results Applications Appendix

Comparing Occupational Mobility Measures (3 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 Annual Occupational Switching Rate March - Unrestricted March - Long. Restrictions JT Supplement 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.34 0.36 0.38 0.4 0.42 0.44 0.46 0.48 Annual Occupational Switching Rate

Back

slide-49
SLIDE 49

Introduction Mobility Measures Estimation Results Applications Appendix

Actual Occupational Mobility (2 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09

  • Est. Occ. Mob.
  • Ret. Occ. Mob

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.05 0.1 0.15 0.2 0.25 0.3 0.35

  • Est. Occ. Mob.
  • Long. Occ. Mob

Back

slide-50
SLIDE 50

Introduction Mobility Measures Estimation Results Applications Appendix

Actual Occupational Mobility (3 Digit)

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11

  • Est. Occ. Mob.
  • Ret. Occ. Mob

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

  • Est. Occ. Mob.
  • Long. Occ. Mob

Back

slide-51
SLIDE 51

Introduction Mobility Measures Estimation Results Applications Appendix

PSID Comparison

Compare our estimates of occupational mobility to PSID corrections in Kambourov and Manovskii (2009) (PSID has 50% more occupations, count switches among unemployed)

1980 1982 1984 1986 1988 1990 1992 1994 1996 1998

Year

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18

  • Est. Occ. Mob. - CPS
  • Est. Occ. Mob. - PSID

Back

slide-52
SLIDE 52

Introduction Mobility Measures Estimation Results Applications Appendix

Using Job Tenure and Occupational Mobility Supplement

Utilize biennial (since 2000) job tenure and occupational mobility supplement instead of March CPS; find similar results.

1980 1985 1990 1995 2000 2005 2010 2015 2020

Year

0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

Baseline Est. Occ. Mob. JT Est. Occ. Mob Back

slide-53
SLIDE 53

Introduction Mobility Measures Estimation Results Applications Appendix

Alternate Variable Combinations

Include additional controls or outcomes; similar results obtain

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.065 Baseline Alt Hrs/Inc

  • Inc. Health Ins.

1980 1985 1990 1995 2000 2005 2010 2015 2020 Year 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.065 Baseline No Occ. FE Add Variables, Full Sample

  • Add. Variables, Restrict Sample

Back

slide-54
SLIDE 54

Introduction Mobility Measures Estimation Results Applications Appendix

Details on Predicted Monthly Switching Rates

  • Compute monthly prediction of true switching rate as:

P(SW = 1 | X,Y, ˜ SW)

=

P( ˜ SW,Y | SW = 1,X)P(SW = 1,X) P( ˜ SW,Y | SW = 1,X)P(SW = 1,X)+ P( ˜ SW,Y | SW = 0,X)P(SW = 0,X)

=

P( ˜ SW | X,Y,SW = 1)P(Y | X,SW = 1)P(SW = 1 | X) P( ˜ SW | X,Y,SW = 1)P(Y | X,SW = 1)P(SW = 1 | X)+ P( ˜ SW | X,Y,SW = 0)P(Y | X,SW = 0)P(SW = 0 | X)

Use model estimates from annual switching to represent:

P(SW = 1 | X,Y = 1, ˜ SW = 1)

= (α0 +α1 + XαX )(β0 +β1 + Xβ2)(δ0 + Xδ1) (α0 +α1 + XαX )(β0 +β1 + Xβ2)(δ0 + Xδ1)+(α0 + XαX )(β0 + Xβ2)(1−δ0 − Xδ1)

  • Treat post-1994 switch as retrospective (unless missing data); treat pre-1994 switch as

longitudinal (and post-1994 with missing data)

  • Use pooled estimates of parameters; robust to using year by year estimates.
  • Construct all switching rates as 13 month moving averages.

Back

slide-55
SLIDE 55

Introduction Mobility Measures Estimation Results Applications Appendix

Details on Trade Estimation

  • Adjust our sample and covariates to match those of Ebenstein et al. (2014) – they use
  • utgoing rotation group data
  • Estimation robust to how two samples are treated
  • Treating two sample as disjoint (no correlation between parameters)
  • Using non-overlapping samples (literally disjoint)
  • Using strictly comparable sample and allow for correlation between parameter

estimates

  • Key difference from estimation is the first stage:

ˆ ξIV

1 = ˜

η1 ˆ η1 ˜ ξIV

1 = αL,1δ1(Trad.)+αL,X(Trad)

δ1(Trad.) ˜ ξIV

1

Back