On the distribution of job characteristics: an analysis of the DOT data

Abstract

We analyse the information in the Dictionary of Occupational Titles to characterize the structure of labour demand. Two dimensions, an intellectual factor and a dexterity factor, capture two-thirds of the variance in job requirements; the remaining (co-)variance cannot be easily structured. Simple linear relationships go a long way in describing the matching between job activities and required worker qualities (Intellect for complex relations to Data and to People, Dexterity for complex relations to Things). There is no dichotomy between mathematical and verbal required skills. Poor working conditions are not restricted to workers in low-level jobs; we find strong support for compensating wage differentials. At more intellectual jobs, men receive less wage compensation for working conditions, while in jobs requiring greater dexterity they receive more. Such a relationship is absent for women.

Notes

¹ Data collected by the O*NET consortium may soon be a good alternative data source. Starting in 1998 with a pretest and in 2001 with the actual project, the O*NET consortium operates a continuous data collection as a follow-up on the data collected in the DOT. The family of occupations has been reorganized to improve homogeneity within families, new variables have been defined, and there is a rolling schedule of updating occupational information. The dataset has information on some 1000 occupations. However, the recent nature of the data and the change in methodology prevents historical trend analysis, which is our long-term research objective. At present, this article aims for a picture of the labour demand structure at the beginning of the 1990s.

² A more detailed assessment of the DOT data is given in Vijverberg and Hartog (2005), which also surveys related earlier studies that use DOT data. An extensive data Appendix is available on the internet at www.utdallas.edu/~vijver/DOT.Appendix.pdf

³ Even so, when an occupational job content variable is extracted from DOT, the information content of this variable is subject to the within-occupation sampling probabilities of the jobs that make up the occupation. To give an example, let an occupation (‘baker’) consist of two types of jobs (‘bread baker’ and ‘pastry baker’). Let there be four times as many bread bakers as pastry bakers in the labour force. DOT may well include only one description of each baker job. The description of the average ‘baker’ job should assign four times more weight to the bread baker description. However, such weights are not available, and we are forced to assign equal weights to each job within an occupation.

⁴ When we started with this research, one of us worked on building a linkage. We became aware of the NOICC Master Crosswalk database when about 75% had been linked. In terms of the detailed 1990 Census occupational codes, 54.1% of these matched exactly. If one were to distinguish only 19 broad occupational groups, there is an 86.25% match. In reviewing some of the differences, it became apparent that the NOICC Master Crosswalk linkage could be questioned in some cases. On the basis of the DOT job description, 227 of the 12741 code links have been changed.

⁵ At www.utdallas.edu/~vijver/DOT.Appendix.pdf

⁶ Technically, the sum of all eigenvalues equals the number of variables entered in the analysis. The average value of the eigenvalues is therefore equal to 1.

⁷ Reduction of the DOT information by factor analytic methods has been done before. Gittleman and Howell (1995) apply cluster analysis to jobs covered in the 1980 Census and various editions of the CPS. They are interested in applying the segmentation view of the labour market and base their clusters not only on DOT variables, but also on hourly earnings, annual earnings and institutional features like union coverage. Shu et al. (1996) link the fourth edition DOT (1977) scores to the jobs in the 1960, 1970 and 1980 Censuses and stress that the underlying factor structure is highly stable over time. Gittleman and Howell discuss a few other applications of factor analysis; none of these uses the DOT 1992 edition. Hartog (1980) used DOT factors to explain the wage structure. Recently, Ingram and Neumann (2006) applied factor analysis to the DOT data, with results that are comparable (and complementary) to ours; they stress different implications.

⁸ We use the iterated principal factor method of Stata 8. The standard principal factor method yielded a series of negative eigenvalues that confuse the interpretation of the percentage of the variance explained by the selected factors.

⁹ Technically, a joint distribution between any pair of variables could be depicted as a scatterplot, but with over 12 000 data points, a scatterplot becomes an impenetrably dense mass of black dots. More importantly, a scatterplot is suitable only when observations are all weighted equally. With varying weights, some points should count more heavily than other points. A balloon plot solves both problems simultaneously. It parcels the two-dimensional area into cells and shows the frequency mass in each cell by the size of a balloon that sits on the cell's centroid.

¹⁰ The figure nicely illustrates the demand for worker traits as envisaged in Tinbergen (1956).

¹¹ The fourth eigenvalue is actually less than 1. However, the fourth factor explains 10% of the variance, which amounts to half of the variance unexplained by the first three factors. Furthermore, the fourth factor has a clear interpretation.

¹² The other variables, not yet mentioned, are: Feeling, Tasting/Smelling and Accommodation. Feeling and Accommodation load on the Precision factor with factor loadings that are half the size of the Near Acuity and Fingering variables. Tasting/Smelling appears to be a unique data component that we will not pursue further.

¹³ We do not present the factor analyses in extenso, to save both space and the reader.

¹⁴ For example, for the compensation group at a value of 1.21, there is one microscopically small balloon that has only 3/1000th of the size of the largest balloon in the Dexterity plot.