Skip to main content

Table 8 Unique variation explained by task measures

From: On the measurement of tasks: does expert data get it right?

Model

(1)

(2)

(3)

(4)

(5)

(6)

Abstract (Occ.)

12.4%

  

3.6%

  

Routine (Occ.)

5.4%

  

1.4%

  

Abstract (Ind.)

 

12.7%

 

3.2%

4.1%

13.5%

Routine (Ind.)

 

6.0%

 

1.1%

1.7%

6.4%

Abstract (Exp.)

  

13.8%

 

5.2%

 

Routine (Exp.)

  

5.0%

 

2.1%

 

Total (Occ.)

17.9%

 

18.9%

5.0%

7.3%

 

Total (Ind.)

 

18.6%

 

4.3%

5.8%

19.9%

N = 27,777

      
  1. The displayed values represent the unique variation in log wages associated with the task measure of interest, expressed relative to the R-squared of the full model. Results are based on computing the squared semipartial correlation between log wages and the task measure of interest. Models (1)-(3) correspond to specifications including occupation-level tasks from Survey data (“(Occ.)”), individual-level tasks (“(Ind.)”), and occupation-level tasks from Expert data (“(Exp.)”), respectively. Models (4) and (5) combine individual-level tasks with occupation-level tasks from Survey and Expert data, respectively. Lastly, model (6) includes individual-level tasks and occupational FE. The two bottom rows summarize the variance in low wages associated with task measures of interest, which has not been explained by all other covariates (including other task dimensions). All specifications include controls for gender, age, age squared, a dummy for living in an urban area, education dummies, occupational tenure, firm tenure, squared tenure for each dimension of experience, and indicator for firm size. I use BIBB/BAuA data that has been collected in 2011–12 and 2017–18, and data from BERUFENET, covering the years 2011–13. For the BIBB/BAuA data, see Hall et al. (2020b) and (Hall et al. (2020a), respectively. For the BERUFENET data, see Dengler et al. (2014)