The Skill Content of Technological Change Autor et al' 2003

1
The Skill Content of Technological ChangeAutor
et al. 2003

• Joshua Ruben and Kasyn Stevenson
• April 22, 2008

2
Objective

• Explain the correlation between the adoption of
  computer based technologies and the increased
  demand for college educated labor.
• Prove that technological advance drives the
  demand for educated labor.
• Note This paper is innovative in that it looks
  at the tasks workers perform rather than the
  credentials workers have who perform certain
  tasks.

3
Background

• Computerization is a long-term trend that dates
  back to the advent of the Jacquard loom in 1801.
• Computerization increases the marginal
  productivity of non-routine tasks.
• Since 1801, there has been more than a
  trillion-fold decline in the real price of
  computing power.

4
Observations

• Computer based capital substitutes for workers
  performing routine tasks
• Computer capital complements workers performing
  non-routine tasks.
• Routine and non-routine tasks are imperfect
  substitutes.

5
Definitions

• A task is routine if it can be accomplished by
  machines following explicit programmed rules.
• A task is non-routine if the rules are not
  sufficiently well understood to be specified in
  computer code and executed by machines.
• Routine and non-routine tasks are divided into
  manual and cognitive categories.

6
(No Transcript)
7
Predictions

• Industries and occupations that are intensive in
  labor input of routine tasks will make relatively
  large investments in computer capital as its
  price declines.
• Decreased demand for routine labor.
• Increased demand for cognitive non-routine labor.
• More demand for educated workers who hold a
  comparative advantage in non-routine tasks.
• The model makes no prediction for manual
  non-routine labor demand.

8
Supporting Evidence

• Starting in the 1970s, labor input of routine
  cognitive and manual tasks in the US economy
  declined, and labor input of non-routine analytic
  and interactive tasks rose.
• Industry wide accelerating shifts in labor input
  favoring non-routine tasks in rapidly
  computerizing industries, insignificant before
  the 1960s.
• Occupations undergoing rapid computerization
  paralleled industry wide shifts in labor.
• The substitution away from routine labor input
  was pervasive across all education levels.

9
Experimental Design

• Ideal experiment compare two identical autarkic
  economies, one facing a dramatic decline in
  computing prices
• Next best option (in clear terms) gather data on
  who does what and how many people do it.
• In academic speak gather data on industry task
  inputs and cross reference it with labor
  demographics.
• Method Isolate how increased computerization
  effects job tasks.
• Controls for supply side factors such as rising
  educational attainment of the workforce,
  increased use of female labor, and occupational
  shifts.

10
The Data

• Dictionary of Occupational Titles (DOT)
• Details the task requirements of different jobs,
  Dept. of Labor
• Census and Current Population Survey
• Provides samples of employed workers from
  1960-1998, Census Bureau
• National Income and Product Accounts (NIPA)
• Measures industrial capital stocks, Dept. of
  Commerce
• Match these data sets together over time
• Resulting data set describes how many workers are
  performing what tasks each year, divided by
  industry

11
Summary Statistics
12
Trends in Routine and Non-routine Task Input,
1960-1998

• Figure is constructed using Dictionary of
  Occupational Titles 1977 task measures by gender
  and occupation paired to employment data for 1960
  and 1970 Census and 1980, 1990 and 1998 Current
  Population Survey (CPS) samples. Data are
  aggregated to 1,120 industry-gender-education
  cells by year and each cell is assigned a value
  corresponding to its rank in the 1960
  distribution of task input (calculated across the
  1,120, 1960 task cells). Plotted values depict
  the employment-weighted mean of each assigned
  percentile in the indicated year.

13
Task Model

• Main Assumptions
• A1 Computer capital is a substitute for human
  capital in carrying out routine tasks (perfect
  substitute).
• A2 Routine and non-routine tasks are imperfect
  substitutes.
• A3 Greater intensity of routine inputs increases
  the marginal productivity of non-routine inputs.
• Do these sound familiar????

14
Task Model II

• Labor Input Assumptions
• Workers have heterogeneous productivity
  endowments spread in varying levels between
  routine and non-routine, ri and ni.
• Workers decide how much input type to provide
  according to their comparative advantage.
• Workers respond to wage changes.
• Li?iri ,(1- ?i)ni where ?i?0,1 probability
  of endowment
• Li input endowment of worker i
• Eiri ,ni measure in efficiency units
• Ei input endowment of worker i
• Implications
• wrp where wr- wage per efficiency unit of
  routine task input.
• Worker self selection among occupations clears
  the labor market.

15
Task Model III

• Autor et al. assume a Cobb Douglas production
  function for tractability.
• Q (LRC)1-ß (LN)ß, ß?(0,1)
• LR Routine Labor
• LN Non-Routine Labor
• C Computer Capital
• Q Output
• Computer Capital is supplied perfectly
  elastically at market price p per efficiency
  unit.

16
Task Model IV

• Equilibrium conditions
• Recall the Romer (1990) model in which workers
  chose their sector such that the marginal wage
  across sectors was equal.
• We do pretty much the same thing here.
• Bottom Line The ratio of wn/wr increases as the
  price of computers declines.

17
Recap

• An exogenous decline in the price of computer
  capital raises the marginal productivity of
  non-routine tasks, causing workers to reallocate
  labor supply from routine to non-routine task
  input. Although routine labor input declines, an
  inflow of computer capital more than compensates
  yielding a net increase in the intensity of
  routine task input in production.

18
Empirical Implementation

• Sources of variation in data
• extensive margin changes in occupational
  distribution of employment over time.
• intensive margin changes in task content
  within occupations over time.
• Variables for routine/non-routine tasks
• Defined from the Handbook for Analyzing Jobs
• Cognitive Tasks
• Non Routine DCP (direction, control, and
  planning), GED-Math
• Routine STS (set limits, tolerances, standards)
• Manual Tasks
• Non-Routine EYEHAND
• Routine FINGDEX
• Note Above variables are measured as percentiles
  from the base year 1960.

19
Testable Hypotheses

• P1 A proportionate increase in the demand for
  routine task inputs is larger in routine task
  intensive industries.
• P2 A decline in the price of computer capital
  also raises demand for non-routine task input.
• P3 Occupations that make relatively larger
  investments in computer capital will show larger
  increases in labor input of non routine tasks and
  larger decreases in labor input of routine tasks.

20
P1 Test

• Did industries historically intensive in routine
  tasks adopt computer capital more rapidly than
  others as prices fell?
• First Autor et al. create an index of relative
  routine task intensity by industry
• Routine Task Sharej,1960100rj,1960/(rj,1960nj.1
  960)
• Now regress
• ComputerAdoptionj,1960-97aß(RoutineTaskSharej19
  60)
• This test is significant and thus proves P1.

21
P2 Test

• Begin by estimating a model for the within
  industry relationship between computer adoption
  and task change over 4 decades
• ?Tjkt a ? ?Cj ejkt
• ?Tjkt the change in industry js input of task k
  between years to and t
• ?Cj Annual change in the percentage of workers
  using a computer at their jobs from 1984-97 (CPS)
• Robustness Checks/controls
• Run the regression separately for each decade to
  ensure that computerization is the driving factor
  behind changes in industry task inputs.
• Principal Component Analysis is used to identify
  alternative component variables and confirms the
  main regression.

22
(No Transcript)
23
P2 Test Results Summary

• Changes in routine task input are uniformly
  negative in the 70s, 80s, and 90s. These are
  economically large and in most cases
  statistically significant.
• Converse is true for non-routine.
• Relationship between computerization and industry
  task change tends to become larger in absolute
  magnitude with each passing decade, suggesting a
  secularly rising relationship between
  computerization and task change.
• No significant relationship between
  computerization and task change in the 1960s.

24
P2 Proof of Causality

• Autor shows that task change is driven by
  computerization and not overall capital-skill
  complementarity (capital deepening).
• Constructs 2 variables to control for capital
  deepening
• Log of capital investments flow per worker
• Log capital to labor ratio
• Adds dummy variables for each decade.
• Runs same style regression as before and find the
  coefficients on the capital investment and
  intensity variables economically small and in
  most cases insignificant.
• This indicates that aggregate capital deepening,
  apart from computer investment, explains little
  of the observed change in task input.
• See page 1307 for full regression equation.

25
P2 Causality Check II

• Next Autor et al. show that task shifts have
  sparked the increase in educational attainment
  and not the other way around.
• This goes against the conventional wisdom that
  education spawns task changes.
• Recall that every laborer has skill endowment
  Eiri,ni.
• Also ?Tijkt a ? ?Cj eijkt
• Now i serves as an index for educational
  attainment.

26
P2 Education Results

• This regression shows that laborers tend to
  perform more non-routine tasks as computerization
  of the workplace increases.
• This effect is true across all educational
  levels, albeit to varying degrees.
• Effects are strongest in the 2 middle groups, and
  statistically significant.
• Effects for college graduates and high school
  dropouts have the correct coefficients but are
  insignificant.
• Results suggest task change is antecedent to
  educational upgrading rather than merely a
  reflection of it.

27
P3 Test

• Want to examine if increased computer usage has
  driven task shifts within occupations.
• Secretary example
• The authors match occupations from the 1977 and
  1991 revisions of the DOT to estimate the
  equation
• ?Tmkt a ? ?Cm emkt
• ?Tmkt is the change in occupational input of task
  k between 1977 and 1991 with 2 digit COC (Census
  Occupation Code) occupation m.
• ?Cm is the change in occupational computer
  penetration measured by the CPS.

28
P3 Results

• Occupations making relatively greater increases
  in computer use saw relatively greater increases
  in labor input of non-routine tasks and larger
  declines in labor input of routine cognitive
  skills.
• Each of these relationships is significant at the
  10 level or greater.
• Only in the case of routine manual tasks, where
  the point estimate is close to 0, do the authors
  fail to find the expected relationship.
• Prove that shifts away from routine labor are
  concentrated in industries and occupations that
  adopted computer technology most fully.

29
Modeling Economic Importance

• To quantify task shifts in concrete economic
  terms, we pool all of the preceding results
  together to calculate their potential
  contribution to the demand for college educated
  labor.
• Autor creates 3 new models that uncover some
  interesting statistical conclusions.
• Page 1316 has the full models, sort of.

30
Economic Importance Results

• Shifts favoring non-routine over routine task
  contributed 2.5 percentage points growth per
  decade to college equivalent employment from
  1980-1998.
• Task shifts attributable to computerization
  increased college employment by 3.7 percentage
  points per decade 1980-98.
• Almost 40 of the computer contribution to rising
  educational demand in the last 2 decades is due
  to shifts in task composition within nominally
  unchanging occupations.

31
Big Picture

• Changes in task demands accompanying workplace
  computerization are economically large and could
  have contributed substantially to relative demand
  shifts favoring educated labor in the United
  States since 1970.
• The causal force by which advancing computer
  technology affects skill demand is the declining
  price of computer capital.

32
HmmmSuspicious

• Many of their regression results generate
  coefficients that more than fully account for
  observed changes.
• The R2 terms for all regressions is extremely
  weak.
• The study omits advanced degrees (MBA, JD, etc)
  from the data.

33
Discussion Questions

• What would happen if computer capital could
  substitute for non-routine labor?
• What implications does Autors analysis have for
  the current debates over outsourcing,
  immigration, and income inequality?
• Do you see any links to Mokyrs/Professor
  Hueckels factor endowment arguments?
• This paper made no predictions on the effect of
  computerization on non-routine manual tasks. Do
  you have any predictions?
• Do computers always increase productivity? Is
  there an upper limit on gains to computerization?
  Is this limit affected by the education level of
  the labor force?
• Social Implications? Does computerization
  marginalize certain segments of the population,
  and what affect does this have on societal
  welfare as a whole?
• One model showed that increased computerization
  leads people to shift the usage of their skill
  endowment towards non-routine tasks for which
  they may not have a comparative advantage. Do
  you find this problematic?