Occupational Classifications: A Machine Learning Approach
August 2018
Working Paper Number:
CES-18-37
Abstract
Characterizing the work that people do on their jobs is a longstanding and core issue in labor economics. Traditionally, classification has been done manually. If it were possible to combine new computational tools and administrative wage records to generate an automated crosswalk between job titles and occupations, millions of dollars could be saved in labor costs, data processing could be sped up, data could become more consistent, and it might be possible to generate, without a lag, current information about the changing occupational composition of the labor market. This paper examines the potential to assign occupations to job titles contained in administrative data using automated, machine-learning approaches. We use a new extraordinarily rich and detailed set of data on transactional HR records of large firms (universities) in a relatively narrowly defined industry (public institutions of higher education) to identify the potential for machine-learning approaches to classify occupations.
Document Tags and Keywords
Keywords
Keywords are automatically generated using KeyBERT, a powerful and innovative
keyword extraction tool that utilizes BERT embeddings to ensure high-quality and contextually relevant
keywords.
By analyzing the content of working papers, KeyBERT identifies terms and phrases that capture the essence of the
text, highlighting the most significant topics and trends. This approach not only enhances searchability but
provides connections that go beyond potentially domain-specific author-defined keywords.
:
payroll,
industrial,
employee,
employed,
job,
classified,
classification,
classifying,
department,
hiring,
workforce,
employing,
worker,
occupation,
clerical,
associate,
wage data,
employee data
Tags
Tags are automatically generated using a pretrained language model from spaCy, which excels at
several tasks, including entity tagging.
The model is able to label words and phrases by part-of-speech,
including "organizations." By filtering for frequent words and phrases labeled as "organizations", papers are
identified to contain references to specific institutions, datasets, and other organizations.
:
Bureau of Labor Statistics,
Social Security Administration,
National Science Foundation,
National Bureau of Economic Research,
Office of Management and Budget,
Current Population Survey,
Longitudinal Business Database,
Employer Identification Numbers,
Federal Register,
Department of Labor,
Standard Occupational Classification,
Longitudinal Employer Household Dynamics,
Information and Communication Technology Survey,
Business Register,
Protected Identification Key,
Ohio State University,
University of Michigan,
Integrated Longitudinal Business Database,
Person Validation System,
Center for Administrative Records Research,
Person Identification Validation System,
National Center for Science and Engineering Statistics
Similar Working Papers
Similarity between working papers are determined by an unsupervised neural
network model
know as Doc2Vec.
Doc2Vec is a model that represents entire documents as fixed-length vectors, allowing for the
capture of semantic meaning in a way that relates to the context of words within the document. The model learns to
associate a unique vector with each document while simultaneously learning word vectors, enabling tasks such as
document classification, clustering, and similarity detection by preserving the order and structure of words. The
document vectors are compared using cosine similarity/distance to determine the most similar working papers.
Papers identified with 🔥 are in the top 20% of similarity.
The 10 most similar working papers to the working paper 'Occupational Classifications: A Machine Learning Approach' are listed below in order of similarity.
-
Working PaperA Task-based Approach to Constructing Occupational Categories with Implications for Empirical Research in Labor Economics
September 2019
Working Paper Number:
CES-19-27
Most applied research in labor economics that examines returns to worker skills or differences in earnings across subgroups of workers typically accounts for the role of occupations by controlling for occupational categories. Researchers often aggregate detailed occupations into categories based on the Standard Occupation Classification (SOC) coding scheme, which is based largely on narratives or qualitative measures of workers' tasks. Alternatively, we propose two quantitative task-based approaches to constructing occupational categories by using factor analysis with O*NET job descriptors that provide a rich set of continuous measures of job tasks across all occupations. We find that our task-based approach outperforms the SOC-based approach in terms of lower occupation distance measures. We show that our task-based approach provides an intuitive, nuanced interpretation for grouping occupations and permits quantitative assessments of similarities in task compositions across occupations. We also replicate a recent analysis and find that our task-based occupational categories explain more of the gender wage gap than the SOC-based approaches explain. Our study enhances the Federal Statistical System's understanding of the SOC codes, investigates ways to use third-party data to construct useful research variables that can potentially be added to Census Bureau data products to improve their quality and versatility, and sheds light on how the use of alternative occupational categories in economics research may lead to different empirical results and deeper understanding in the analysis of labor market outcomes.View Full Paper PDF
-
Working PaperNOISE INFUSION AS A CONFIDENTIALITY PROTECTION MEASURE FOR GRAPH-BASED STATISTICS
September 2014
Working Paper Number:
CES-14-30
We use the bipartite graph representation of longitudinally linked em-ployer-employee data, and the associated projections onto the employer and em-ployee nodes, respectively, to characterize the set of potential statistical summar-ies that the trusted custodian might produce. We consider noise infusion as the primary confidentiality protection method. We show that a relatively straightfor-ward extension of the dynamic noise-infusion method used in the U.S. Census Bureau's Quarterly Workforce Indicators can be adapted to provide the same confidentiality guarantees for the graph-based statistics: all inputs have been modified by a minimum percentage deviation (i.e., no actual respondent data are used) and, as the number of entities contributing to a particular statistic increases, the accuracy of that statistic approaches the unprotected value. Our method also ensures that the protected statistics will be identical in all releases based on the same inputs.View Full Paper PDF
-
Working PaperAn 'Algorithmic Links with Probabilities' Crosswalk for USPC and CPC Patent Classifications with an Application Towards Industrial Technology Composition
March 2016
Working Paper Number:
CES-16-15
Patents are a useful proxy for innovation, technological change, and diffusion. However, fully exploiting patent data for economic analyses requires patents be tied to measures of economic activity, which has proven to be difficult. Recently, Lybbert and Zolas (2014) have constructed an International Patent Classification (IPC) to industry classification crosswalk using an 'Algorithmic Links with Probabilities' approach. In this paper, we utilize a similar approach and apply it to new patent classification schemes, the U.S. Patent Classification (USPC) system and Cooperative Patent Classification (CPC) system. The resulting USPC-Industry and CPC-Industry concordances link both U.S. and global patents to multiple vintages of the North American Industrial Classification System (NAICS), International Standard Industrial Classification (ISIC), Harmonized System (HS) and Standard International Trade Classification (SITC). We then use the crosswalk to highlight changes to industrial technology composition over time. We find suggestive evidence of strong persistence in the association between technologies and industries over time.View Full Paper PDF
-
Working PaperWork Organization and Cumulative Advantage
March 2025
Working Paper Number:
CES-25-18
Over decades of wage stagnation, researchers have argued that reorganizing work can boost pay for disadvantaged workers. But upgrading jobs could inadvertently shift hiring away from those workers, exacerbating their disadvantage. We theorize how work organization affects cumulative advantage in the labor market, or the extent to which high-paying positions are increasingly allocated to already-advantaged workers. Specifically, raising technical skill demands exacerbates cumulative advantage by shifting hiring towards higher-skilled applicants. In contrast, when employers increase autonomy or skills learned on-the-job, they raise wages to buy worker consent or commitment, rather than pre-existing skill. To test this idea, we match administrative earnings to task descriptions from job posts. We compare earnings for workers hired into the same occupation and firm, but under different task allocations. When employers raise complexity and autonomy, new hires' starting earnings increase and grow faster. However, while the earnings boost from complex, technical tasks shifts employment toward workers with higher prior earnings, worker selection changes less for tasks learned on-the-job and very little for high autonomy tasks. These results demonstrate how reorganizing work can interrupt cumulative advantage.View Full Paper PDF
-
Working PaperSqueezing More Out of Your Data: Business Record Linkage with Python
November 2018
Working Paper Number:
CES-18-46
Integrating data from different sources has become a fundamental component of modern data analytics. Record linkage methods represent an important class of tools for accomplishing such integration. In the absence of common disambiguated identifiers, researchers often must resort to ''fuzzy" matching, which allows imprecision in the characteristics used to identify common entities across dfferent datasets. While the record linkage literature has identified numerous individually useful fuzzy matching techniques, there is little consensus on a way to integrate those techniques within a single framework. To this end, we introduce the Multiple Algorithm Matching for Better Analytics (MAMBA), an easy-to-use, flexible, scalable, and transparent software platform for business record linkage applications using Census microdata. MAMBA leverages multiple string comparators to assess the similarity of records using a machine learning algorithm to disambiguate matches. This software represents a transparent tool for researchers seeking to link external business data to the Census Business Register files.View Full Paper PDF
-
Working PaperA Tale of Two Fields? STEM Career Outcomes
October 2023
Working Paper Number:
CES-23-53
Is the labor market for US researchers experiencing the best or worst of times? This paper analyzes the market for recently minted Ph.D. recipients using supply-and-demand logic and data linking graduate students to their dissertations and W2 tax records. We also construct a new dissertation-industry 'relevance' measure, comparing dissertation and patent text and linking patents to assignee firms and industries. We find large disparities across research fields in placement (faculty, postdoc, and industry positions), earnings, and the use of specialized human capital. Thus, it appears to simultaneously be a good time for some fields and a bad time for others.View Full Paper PDF
-
Working PaperThe Gender Pay Gap and Its Determinants Across the Human Capital Distribution
June 2023
Working Paper Number:
CES-23-31R
This paper links American Community Survey data and postsecondary transcript records to examine how the gender pay gap varies across the distribution of education credentials for a sample of 2003-2013 graduates. Although recent literature emphasizes gender inequality among the most-educated, we find a smaller gender pay gap at higher education levels. Field-of-degree and occupation effects explain most of the gap among top bachelor's graduates, while work hours and unobserved channels matter more for less-competitive bachelor's, associate, and certificate graduates. We develop a novel decomposition of the child penalty to examine the role of children in explaining these results.View Full Paper PDF
-
Working PaperMale Earnings Volatility in LEHD before, during, and after the Great Recession
September 2020
Working Paper Number:
CES-20-31
This paper is part of a coordinated collection of papers on prime-age male earnings volatility. Each paper produces a similar set of statistics for the same reference population using a different primary data source. Our primary data source is the Census Bureau's Longitudinal Employer-Household Dynamics (LEHD) infrastructure files. Using LEHD data from 1998 to 2016, we create a well-defined population frame to facilitate accurate estimation of temporal changes comparable to designed longitudinal samples of people. We show that earnings volatility, excluding increases during recessions, has declined over the analysis period, a finding robust to various sensitivity analyses. Although we find volatility is declining, the effect is not homogeneous, particularly for workers with tenuous labor force attachment for whom volatility is increasing. These 'not stable' workers have earnings volatility approximately 30 times larger than stable workers, but more important for earnings volatility trends we observe a large increase in the share of stable employment from 60% in 1998 to 67% in 2016, which we show to largely be responsible for the decline in overall earnings volatility. To further emphasize the importance of not stable and/or low earning workers we also conduct comparisons with the PSID and show how changes over time in the share of workers at the bottom tail of the cross-sectional earnings distributions can produce either declining or increasing earnings volatility trends.View Full Paper PDF
-
Working PaperBusiness Dynamics of Innovating Firms: Linking U.S. Patents with Administrative Data on Workers and Firms
July 2015
Working Paper Number:
CES-15-19
This paper discusses the construction of a new longitudinal database tracking inventors and patent-owning firms over time. We match granted patents between 2000 and 2011 to administrative databases of firms and workers housed at the U.S. Census Bureau. We use inventor information in addition to the patent assignee firm name to and improve on previous efforts linking patents to firms. The triangulated database allows us to maximize match rates and provide validation for a large fraction of matches. In this paper, we describe the construction of the database and explore basic features of the data. We find patenting firms, particularly young patenting firms, disproportionally contribute jobs to the U.S. economy. We find patenting is a relatively rare event among small firms but that most patenting firms are nevertheless small, and that patenting is not as rare an event for the youngest firms compared to the oldest firms. While manufacturing firms are more likely to patent than firms in other sectors, we find most patenting firms are in the services and wholesale sectors. These new data are a product of collaboration within the U.S. Department of Commerce, between the U.S. Census Bureau and the U.S. Patent and Trademark Office.View Full Paper PDF
-
Working PaperJob Tasks, Worker Skills, and Productivity
September 2025
Working Paper Number:
CES-25-63
We present new empirical evidence suggesting that we can better understand productivity dispersion across businesses by accounting for differences in how tasks, skills, and occupations are organized. This aligns with growing attention to the task content of production. We link establishment-level data from the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey with productivity data from the Census Bureau's manufacturing surveys. Our analysis reveals strong relationships between establishment productivity and task, skill, and occupation inputs. These relationships are highly nonlinear and vary by industry. When we account for these patterns, we can explain a substantial share of productivity dispersion across establishments.View Full Paper PDF
