One of the fundamental questions in labour economics is why some workers are paid more than others within the same industry. This column uses data from adverts on a large US job website to investigate what's behind these wage differences. The job titles used in adverts capture more variation between jobs than standard occupational classifications. By failing to recognise this, the previous literature has attributed too much of wage inequality to luck and too little to differences in worker and firm characteristics.
Most modern labour markets exhibit a substantial degree of wage inequality (e.g. Mortensen 2003, Horstein et al. 2007). One of the fundamental questions in labour economics is understanding the sources of this inequality. Why are some workers paid more than others? Do wage differences reflect meaningful differences in worker and firm characteristics (such as occupation, the worker’s education level or the firm’s sector of industry)? Or are wage differences the result of sheer luck in getting the right job? The literature in this area finds that worker and firm characteristics explain at most half of the variation in wages across workers (e.g. Abowd et al. 1999). This suggests that luck plays an important role.
In a recent study, we shed new light on this question by analysing job adverts on the large US employment website CareerBuilder.com (Marinescu and Wolthoff 2016). A big advantage of these data compared to traditional data sources is that they contain the job titles that firms choose to describe their positions, such as “Java programmer”. We document that these job titles contain a lot of relevant information. In particular, the job titles are much more detailed than standard occupational classifications in describing differences in the required experience or area of specialisation, distinguishing between senior and junior accountants, or between Java and C++ programmers. We demonstrate that this information is crucially important. Unlike standard occupational classifications, job titles explain nearly all of the wage variation across job openings; with the job title in hand, we can guess very accurately how much a job pays. Therefore, a job seeker cannot count on luck to find a high-wage job – high-wage jobs are just different types of jobs.
Our data set includes all job ads on CareerBuilder for the Chicago and Washington, DC areas in January 2011. Only approximately 20% of these job ads include information on the wage that the firm plans to pay.
This fact may raise the question of whether the posted wages are representative of the wages that are earned in the US labour market more broadly. We establish that this is the case by showing that the distribution of posted wages does not systematically differ from the distribution of earned wages in representative data sets such as the Current Population Survey (CPS). One of the dimensions in which the wages in the two data sets are strikingly similar is the explanatory power of occupations as captured by the Standard Occupational Classification (SOC). In both data sets, the finest version of this classification can account for just under half of the wage variation.
While no finer occupational information is available in the CPS, the CareerBuilder data allow us to use job titles instead. It turns out that the explanatory power of job titles greatly exceeds that of SOC codes – job titles explain more than 94% of the cross-sectional wage variation. An alternative way to assess the importance of job titles is to first analyse to what extent the identity of the firm can explain wages, i.e. whether some firms systematically pay more than others. We find that there are large differences in pay across firms, and that firms that pay higher wages than other firms primarily do so because they employ workers with different job titles.
The power of words
To better understand these results, we analyse which words in the job title are particularly important. We identify two different groups.
- First, we find that words that indicate a level of seniority within an occupation are important: not surprisingly, job titles that include words like “manager”, “senior”, “executive” or “director” pay significantly higher wages than job titles with words like “coordinator”, “assistant”, “entry” or “junior”.
- Second, words that indicate particular areas of specialization have significant explanatory power as well: for example, “sales”, “engineer”, “consultant” or “java” are associated with higher wages, while “accountant”, “marketing”, “recruiting” or “network” indicate lower wages.
One concern in interpreting these results is that firms might choose certain job titles to justify paying a higher or a lower wage. That is, perhaps “senior accountant” and a “junior accountant” positions are fundamentally the same, except for their wages. We show that this concern is unfounded by analysing workers’ application behaviour. The argument is as follows. If these positions only differ in their wages, then we would expect two things. First, applicants to either position should be similar in terms of their characteristics. Second, the position offering the higher wage – i.e. the senior accountant position – should attract more applicants because it pays more.
Neither implication holds in our data. The characteristics of applicants differ across job titles within an occupation, with “manager”, “senior”, “executive”, or “director” positions attracting more experienced applicants. Moreover, the association between wages and the number of applicants is negative within an occupation. This latter fact may seem somewhat surprising at first sight, but it is consistent with the findings of a small literature that tries to establish the effect of a firm’s wage offer on its number of applications. Our data reveal that there is an intuitive reason for these results: we find that job titles with words like “manager”, “senior”, “executive” or “director” pay higher wages but attract fewer applicants, while words like “coordinator”, “assistant”, “entry” or “junior” pay lower wages but attract more applicants (see Figures 1 and 2). Hence, different job titles within the same occupation are in fact fundamentally different positions, and the relation between wages and applicants should be considered within a job title.
Figure 1 Word cloud of the words in job titles that are associated with a lower wage for a given occupation
Note: The size of a word represents its frequency; the shade represents the magnitude of the effect, with a darker colour indicating a more negative effect on the wage.
Figure 2 Word cloud of the words in job titles that are associated with more applicants for a given occupation
Note: The size of a word represents its frequency; the shade represents the magnitude of the effect, with a darker colour indicating a more positive effect on the number of applicants.
Summarising, our results indicate that there exists more variation between jobs in the US labour market than captured by even the finest level of SOC codes. By failing to recognise this, the previous literature has attributed too much of the wage inequality to luck and too little to meaningful differences in worker and firm characteristics. We find that the role of luck in determining a worker’s wage is in fact quite small. This has important implications for understanding the job search behaviour of unemployed workers. For example, US labour market data indicate that the average duration of unemployment is not very long (e.g. Shimer 2012). This is somewhat puzzling if one believes that luck plays an important role in the determination of wages. Why do workers not search more for a better-paying job? However, it is perfectly consistent with the idea that the role of luck is limited – continued job search is not that useful if most better-paying jobs are out of reach because they require more experience or a different area of specialisation.
Abowd, J, F Kramarz and D Margolis (1999), “High Wage Workers and High Wage Firms,” Econometrica 67(2), 251-333.
Hornstein, A, P Krusell and G Violante (2007), “Frictional Wage Dispersion in Search Models: A Quantitative Assessment”, NBER Working Paper 13674.
Marinescu, I and R Wolthoff (2016), “Opening the Black Box of the Matching Function: The Power of Words”, NBER Working paper 22508.
Mortensen, D (2003), Wage Dispersion: Why Are Similar Workers Paid Differently? MIT Press, Cambridge, Massachusetts.
Shimer, R (2012), “Reassessing the Ins and Outs of Unemployment,” Review of Economic Dynamics 15(2), 127-148