Autor, Levy and Murnane, “The Skill Content of Recent Technological Change: An Empirical Exploration.” QJE 2003
This is one of my favorite paper in the topic series of wage inequality distribution. It not only has creative ideas in gauging and measuring task components in each industry and occupation, but also has a very fine general equilibrium construction, which leads to a very logically cohesive development from theory to empirics. The intuition explanation from the model also shines bright. Overall, it also leads to a new branch in the discussion of skill biased technical change.
Previously discussions of SBTC focuses on drawing the correlation of computerization and labor demand change. This paper focuses on causal inference. It observes that: 1. computer capital substitutes for workers with limited and well-defined set of cognitive and manual activities, i.e. routine tasks; 2. It complements workers with problem-solving and complex communication tasks, i.e. nonroutine tasks. It shows that initially routine task intense industries will make relatively large investment in computer capital as its price declines, which reduces routine task workers and increases nonroutine task input, as represented by highly educated workers with comparative advantage in nonroutine versus routine tasks. Greater intensity of routine inputs increases marginal productivity of nonroutine inputs. Based on observations, the paper further splits routine and nonroutine tasks into manual tasks and analytic and interactive tasks.
The paper assumes CRS Cobb-Douglass production function with routine and nonroutine labor inputs and computer capital as production factors. Computer capital is perfectly elastic at market price, which is further exogenously determined. Assumption has perfect substitutability between computer capital and routine task labor. And these are all relative complements to nonroutine labor. Workers have heterogeneous endowment in routine and nonroutine tasks. Therefore, workers choose tasks composition based on comparative advantage. Market clearing conditions have that: 1. wage of routine tasks equal to price of computer capital; 2. Worker routine and nonroutine tasks clear labor market. From the model, we can see that decline in computer price reduces wage for routine tasks 1-1, and raises demand for routine task inputs, which further raises nonroutine tasks wage. Hence, marginal worker reallocate labor input from routine to nonroutine tasks, and demand increase for routine task is fulfilled by computer capital. From the model, the causal force is computer capital price.
Further detailing the model into industry level, it assumes all industry Cobb-Douglas technology, First order conditions return profit maxed wage rate and factor demand function. From these derivations, it proposes that: 1. each individual firm’s degree of adoption to computer capital depends on industry specific factor share of nonroutine tasks, despite the same computer price; 2. Decline in computer price raises demand for nonroutine task input, and the scale is larger in routine-task-intensive industries; 3. Larger investment in computer capital show larger increase in nonroutine labor input and larger decrease in routine labor input.
Empirically, the paper uses Dictionary of Occupational Titles to decode industry task composition. IPUMS and CPS Merged Outgoing Rotation Group are also used. Job task change over time is measured by: 1. Extensive margin: Occupational distribution of employment, holding constant task content within occupations at 1977 DOT level; 2. Intensive margin: change in task content measures within occupations over the period 1977 to 1991. The paper further transform DOT measure of tasks into percentile values corresponding to rank in 1960 distribution of task input.
To test the first proposal, it hypothesizes that historically routine task intensive industries should adopt computer capital more rapidly as its price fell. Simple predictive regression and robustness check confirm the hypothesis.
To test for the second proposal, it hypothesizes that declining computer price should reduce aggregate demand for routine task labor and increase for nonroutine task labor. By pairing task measures with employment data for each decade, it shows that share of labor force employment in nonroutine cognitive and analytical intensive task increased substantially, and the opposite for routine tasks. Secular decline is also found in nonroutine manual tasks. It shows that distribution of nonroutine analytic and nonroutine interactive task input grew substantially. And the routine cognitive and routine manual decreased substantially. When further split data by gender, both genders show similar trends; but female has numerically larger shift. When decompose the shift by extensive and intensive margin over the 20 year interval, both nonroutine analytic and nonroutine interactive task measures show strong and accelerating within-industry growth. Routine cognitive and routine manual acceleratingly declined since 1970s due to within industry shift. In summary, trends dominating the change is from within-industry shift.
Connecting these shift back to computerization, regression of change in computer on change in input task is conducted for each decade and for within industry and between industry. Results show that almost the entire observed within industry change in nonroutine task change is explained by computerization, so is for between industry changes. To further test the robustness of these conclusions, the paper also uses principal component analysis and computer capital investment data from National Income and Product Accounts. All confirms the original conclusion, and confirms the third proposal about larger computer capital investment.
Further, the paper claims that change in demand for workplace, from technological change, is the underlying cause of relative demand shifts favoring educated labor. As previously shown, computerization adaptation predicts increase in nonroutine workers and decrease in routine workers. The paper conjectures the reasons due to: 1. industry purchase computer needs better educated workers to master these tasks; 2. Industries change task assignment of workers with given educational attainment to more of a routine task assignment. Regression within group is estimated. However, holding computer adoption fixed, the model cannot predict within industry task change for high school and college group. The paper explains that the reason is due to “topping out” for college graduates, since they are already at the top of distribution, and high school dropouts have too less human capital to effectively adopt to new technology requirement.
To quantify the task shift, the paper draws task changes within industry, education groups, and occupations to calculate demand for college educated workers. With imputed elasticity of substitution between high skilled and low skilled labor inputs, it finds that extensive margin task change explains 20% to 25% of estimated demand shift for college versus noncollege labor. 40% of computer contribution to rising educational demand in the last two decades in sample period is due to shift in task composition within occupations.
Articonomics 