Computerization, Obsolescence and the Length of Working Life Pter - - PowerPoint PPT Presentation

Computerization, Obsolescence and the Length of Working Life

Péter Hudomiet (RAND) Robert J. Willis (University of Michigan) 2020 Working Longer and Retirement Conference SIEPR October 1, 2020

Hudomiet & Willis Computerization and retirement October 1, 2020 1 / 27

Introduction

Technological change and retirement

Technological change affects labor market

disappearing and new jobs growing inequalities make us productive or redundant?

Technological change may affect retirement decisions

productivity may increase (retirement ↓)

• bsolescence: hard to keep up with the young (retirement ↑)

Hudomiet & Willis Computerization and retirement October 1, 2020 2 / 27

Introduction

Research question

Track obsolescence induced by computerization, 1984-2018

track computer use at work by detailed occupations, years, and age estimate knowledge gap

◮ diff. in computer use between young and old ◮ proxy for obsolescence

Effect of knowledge gap on retirement/wages/job satisfaction

controlling for skill & year fixed effects

Hudomiet & Willis Computerization and retirement October 1, 2020 3 / 27

Introduction

Contribution to the literature

Large literatures on

effect of IT investments, computer use on employment and wages

◮ Aubert et al. (2006), Gaggle and Wright (2017), Freeman (2018), Burstein et al. (2019), Friedberg (2006); Schleife (2006)

wages and employment in STEM vs. regular jobs

◮ Ahituv and Zeira (2010), Bartel and Sicherman (1993), Deming and Noray (2018), Weinberg (2002)

Not much on the effect of knowledge gap on retirement

time-series (cohort) variation

Big data

CPS computerization supplements (~600 thousand obs.) monthly CPS (~7 million obs.) on labor force status March CPS (~2 million obs.) on wages HRS (~12 thousand obs.) on expectations, depression, etc.

Hudomiet & Willis Computerization and retirement October 1, 2020 4 / 27

Introduction

History of computerization

Early phase (low penetration into jobs, homes)

1940-50s: digital computers invented 1950-60s: mainframe introduced commercially 1960-70s: minicomputer introduced

Late phase (growing penetration into jobs, homes)

1970-90s: personal computers introduced 1990-00s: growth of the internet 2000-10s: smart phone, tablets

Hudomiet & Willis Computerization and retirement October 1, 2020 5 / 27

Introduction

PCs and internet use by cohort and experience

Computer use at work Internet use at work

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Fraction 10 20 30 40 Labor market experience 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 2000-2009 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Fraction 10 20 30 40 Labor market experience 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 2000-2009

Computer & net use asymptotes to ~60%. Varies by birth cohorts

youngest cohorts always used

• lder cohorts learned on the job
• ldest cohorts did not learn

Hudomiet & Willis Computerization and retirement October 1, 2020 6 / 27

Introduction

Computer use at work by education

High school dropouts College graduates

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Fraction 10 20 30 40 Labor market experience 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 2000-2009 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Fraction 10 20 30 40 Labor market experience 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 2000-2009

Computer use at work increases with education Cohort gaps larger among highly educated

Hudomiet & Willis Computerization and retirement October 1, 2020 7 / 27

Introduction

Our theory

• 1. Price of computers gradually falls

higher skilled occupations adapt first (engineer)

• 2. Computers enhance productivity

but need to pay a fixed cost of learning

• 3. Younger workers adapt first

many older worker do not adapt ⇒ knowledge gap (temporary)

• 4. This “knowledge gap” harms older workers

skill obsolescence lower wages

• ccupation changes

early retirement

Hudomiet & Willis Computerization and retirement October 1, 2020 8 / 27

Introduction

Computer use at work

Computer use and knowledge gap in a hypothetical occupation

.5 1 .5 1 .5 1 1980 1990 2000 2010

Age 40-49 Age 60-64 Knowledge gap year

The gap varies by occupations & time

Hudomiet & Willis Computerization and retirement October 1, 2020 9 / 27

Patterns in computer use

CPS Computerization supplements

Computerization supplement available 1984-2017

14 waves N = 50, 000 per wave (600, 000 in total) after restrictions

Main measures:

computer use at work: 1984-2003 internet use at work: 1997-2017 computer use at home: 1984-2017 internet use at home: 1997-2017

Hudomiet & Willis Computerization and retirement October 1, 2020 10 / 27

Patterns in computer use

Computer use, high school dropouts

• Low use of computers

slowly growing not much difference by age

Hudomiet & Willis Computerization and retirement October 1, 2020 11 / 27

Patterns in computer use

Computer use, high school graduates

• Moderate use of computers

trending upward less use among older workers, mostly 60-69 young-old gap largest in the mid-late 90s

Hudomiet & Willis Computerization and retirement October 1, 2020 12 / 27

Patterns in computer use

Computer use, some college

• Higher use of computers

less use among older workers, mostly 60-69 young-old gap largest in the late 80s

Hudomiet & Willis Computerization and retirement October 1, 2020 13 / 27

Patterns in computer use

Computer use, college or more

• Very high use of computers

less use among older workers,

◮ both 50-59 ◮ and 60-69

young-old gap largest in the 80s

Hudomiet & Willis Computerization and retirement October 1, 2020 14 / 27

Patterns in computer use

Computer use summary

Large variation in computer/internet use by skill and age

more skilled workers more likely to use more skilled started using earlier

• lder workers started using later

higher skill older workers faced the largest gap

Next step

formalize the knowledge gap (between young and old) test if knowledge gap affected labor market outcomes

Hudomiet & Willis Computerization and retirement October 1, 2020 15 / 27

Patterns in computer use

Formalize the knowledge gap

Fit a model on computer/internet use at home/work

series of probit models internet use at work ⇒ computer use at work model has good fit

Recover knowledge gap

• diff. in computer use among older (50-69) and younger (40-49)

workers by year and skill groups (occupation x educ x race x gender x age) can be merged to other datasets by skill & year

Hudomiet & Willis Computerization and retirement October 1, 2020 16 / 27

Patterns in computer use

Predicted knowledge gap by education

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year HS dropout High school Some college College 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year HS dropout High school Some college College

Age 50-54 Age 55-59

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year HS dropout High school Some college College 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year HS dropout High school Some college College

Age 60-64 Age 65-69

Hudomiet & Willis Computerization and retirement October 1, 2020 17 / 27

Patterns in computer use

Predicted knowledge gap by occupation

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year Professionals Office Precision prod. Food and clean 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year Professionals Office Precision prod. Food and clean

Age 50-54 Age 55-59

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year Professionals Office Precision prod. Food and clean 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 Gap 1980 1990 2000 2010 2020 year Professionals Office Precision prod. Food and clean

Age 60-64 Age 65-69

Hudomiet & Willis Computerization and retirement October 1, 2020 18 / 27

Results

Estimation methods

The DD model yist = β0 + β1 ˆ Gapst + β2zt + γs + δt + εist

γs skill fixed effects (occupation, education, etc.) δt year dummies, zt other controls estimated on age 50-69

Hudomiet & Willis Computerization and retirement October 1, 2020 19 / 27

Results

Estimation methods

The DD model yist = β0 + β1 ˆ Gapst + β2zt + γs + δt + εist

γs skill fixed effects (occupation, education, etc.) δt year dummies, zt other controls estimated on age 50-69

The DDD model yist = β0 + β1 ˆ Gapst + β2zt + θst + εist

θst is γas fully interacted with years estimated on age 40-69 (or 25-49) 40-49 (or 25-49) are the control group (Gap = 0)

Hudomiet & Willis Computerization and retirement October 1, 2020 19 / 27

Results

Yearly transitions out of labor force

Monthly CPS, 1984-2018 Link short panels; estimate yearly transition probabilities

Employment → Nonparticipation Age 50-69 Age 40-69 DD DDD [1] [2] Knowledge gap 0.103 0.122 [0.010]** [0.009]** N 2,635,562 4,782,190 Mean outcome 0.080 0.059

Strong effects on employment → non-employment transitions

0.1 knowledge gap increases transitions by ~1 percentage point

Hudomiet & Willis Computerization and retirement October 1, 2020 20 / 27

Results

Subgroup models

Employment → Nonparticipation, DDD, Age 40-69 Subgroups coef s.e. Total sample 0.122 [0.009]** Males 0.073 [0.010]** Females 0.178 [0.016]** High school 0.132 [0.019]** Some college 0.188 [0.020]** College 0.106 [0.011]** Management 0.170 [0.015]** Professionals 0.129 [0.017]** Sales 0.130 [0.021]** Office/Administration 0.237 [0.023]** Precision production 0.097 [0.043]* Operators/laborers 0.025 [0.064]

Larger effects among females & middle skilled workers

Hudomiet & Willis Computerization and retirement October 1, 2020 21 / 27

Results

Sensitivity: alternative skill groups

Preferred definition of skill group

• ccupation x educ x race x gender x age

we tried simpler alternatives

Employment → Nonparticipation, DDD, Age 40-69, Skill groups measured by . . . Education Aggregate occ Full model [1] [2] [3] Knowledge gap 0.146 0.119 0.122 [0.009]*** [0.007]*** [0.009]*** N 4,782,190 4,782,190 4,782,190 Mean outcome 0.059 0.059 0.059

Results robust to definition of skill groups

Hudomiet & Willis Computerization and retirement October 1, 2020 22 / 27

Results

Sensitivity: controlling trends in occupations

Controling for restructuring of labor markets

share of occupations, normalized to 1 in 2002

Employment → Nonparticipation, DD, Age 50-69 Original V1 V2 [1] [2] [3] Knowledge gap 0.103 0.105 0.103 [0.010]*** [0.010]*** [0.010]*** Occupation share among 25-49 0.005 [0.001]*** Occupation share among 40-49 0.002 [0.001]

No change in estimated effects

Hudomiet & Willis Computerization and retirement October 1, 2020 23 / 27

Results

Other yearly transitions

No effects on yearly employment → unemployment transitions Some effects on transitions into part-time jobs No effects on occupation switches (but noisy estimates)

Hudomiet & Willis Computerization and retirement October 1, 2020 24 / 27

Results

Log annual income

Use log annual wage and salary income

CPS Annual Social and Economic Supplement (March CPS), 1984 - 2017

• ccupation in the longest job last year

Age 50-69 Age 40-69 Age 25-69 DD DDD DDD [1] [3] [4] Knowledge gap

• 0.646
• 0.744
• 0.730

[0.043]** [0.044]** [0.038]** N 674,794 1,361,999 2,429,273

Large effect: 0.1 gap decreases wages by ~7%

larger effects larger among females, middle-skilled workers

Hudomiet & Willis Computerization and retirement October 1, 2020 25 / 27

Results

Health and Retirement Study outcomes

Use longitudinal HRS data

use occupations at age 55

Works FT at 65 Years worked P65 ln(wage) Enjoy work CESD [1] [2] [3] [4] [5] [6] Knowledge gap

• 1.157
• 7.844
• 0.661
• 0.955

0.050

• 0.141

[0.492]** [3.367]** [0.214]** [0.412]* [0.325] [1.648] R-squared 0.288 0.305 0.251 0.515 0.272 0.300 N 6,162 6,162 11,761 11,041 9,366 9,597 Mean outcome 0.242 7.299 0.311 2.673 0.877 1.192

Retire earlier Lower expectations to work after age 65 (P65) Lower wages No effects on enjoying work/depression

Hudomiet & Willis Computerization and retirement October 1, 2020 26 / 27

Conclusion

Conclusion

We tested if computerization affected older workers

very large dataset(s)

Effect of computerization

increased probability of leaving the labor force increased moving into part-time jobs lowered wages

We found reasonable heterogeneity by subgroups

larger effects on females, some-college, office workers

Implications for the future?

computer use at work converged across age groups (no longer a pressure) but other technologies may have similar effects in the future

Hudomiet & Willis Computerization and retirement October 1, 2020 27 / 27