Measuring green jobs in online vacancies with the ILO Green Dictionary

A documented replication of the ILO methodology (Delaporte, Escudero and Adamczyk 2025; Adamczyk et al. 2025) on a public corpus of 55,825 English-language job postings: JobStreet job postings, Malaysia, 2023-2024 (Hugging Face, azrai99/job-dataset) (source, Apache 2.0 (dataset card); scraped from JobStreet). Code, dictionaries and this page: github.com/etorresram/ilo-green-dictionary-vacancies. Author: Eric Torres Ramírez.

Two corpora are available: United States (LinkedIn, Kaggle) and Malaysia (JobStreet, Hugging Face). Same code, same dictionaries, different exception-list status.

Vacancies analysed55,825
Green vacancies (at least one green task)9.1%
Darker green (share of green tasks above the mean)3.0%
Median posted wage, green vs non-green (MYR)51,000 vs 45,000

Read this first. Time coverage: January to June 2024 used for the monthly series (months with at least 500 postings). Vacancies come from one platform in one country and there are no applicants' profiles, so demand cannot be compared with supply. The skills variables use the selected keywords published in the ILO brief, not the full taxonomy. The exception list was validated on the US corpus and reused here; a corpus-specific context check is pending. All figures illustrate the workflow; none is an estimate of a national labour market.

1. What was done

  1. Data preparation. Duplicate and empty postings removed; posted salaries annualised; platform industry or category labels mapped to ISIC Rev. 4 sections with a reviewed file.
  2. Text pre-processing in the ILO sequence: tokenisation, normalisation, stop-word removal (keeping stop words inside dictionary expressions), lemmatisation (spaCy). Dictionary expressions go through the same pipeline, so matching is lemma-to-lemma on full words.
  3. Green task variables: 9 sustainability domains, binary and counts; a vacancy is green if it matches at least one domain.
  4. Skills variables: 15 subcategories (cognitive, socio-emotional, manual).
  5. Green share and shade: share of green tasks among all matched skills and tasks; non-green (0), lighter green (below the mean among green vacancies), darker green (at or above).
  6. Occupation coding: job title → O*NET-SOC → SOC 2010 → ISCO-08 through official crosswalks, with every admissible code kept and an explicit assignment rule.
  7. Validation: context check of frequent terms (75% rule), a blind hand-coding sample, and a boilerplate sensitivity test.
step n
raw rows 59306
after duplicate and empty-text removal 55825
with posted wage 23177
with ISIC section 55825

2. Green vacancies over time

Monthly share of vacancies with at least one green task and share of darker-green vacancies; hover for the number of postings per month. Tables t23 and t24 give the same by month and ISCO-08 major group.

3. Green vacancies by occupation and industry

Occupation coding coverage and ambiguity:

metric n share
postings 55825 1
matched to O*NET-SOC (exact) 16627 0.298
matched (exact after stripping modifiers) 3107 0.056
matched (fuzzy >= 90, same head noun) 3492 0.063
matched (keyword rule, lowest confidence) 16975 0.304
with an ISCO-08 code 39895 0.715
with more than one admissible ISCO-08 code 14984 0.268

Sensitivity of the occupational profile to the assignment rule for titles with more than one admissible ISCO-08 code (main rule vs alternative rule):

isco08_major_label n green_share green_share_alt_rule abs_diff_pp
Plant and Machine Operators, and Assemblers 263 0.190 0.188 0.220
Craft and Related Trades Workers 641 0.178 0.142 4
Elementary Occupations 473 0.156 0.148 0.890
Professionals 16609 0.107 0.113 0.590
Managers 10275 0.094 0.088 0.590
Technicians and Associate Professionals 6768 0.079 0.081 0.190
Clerical Support Workers 3917 0.050 0.049 0.160
Service and Sales Workers 899 0.049 0.063 1

4. Shade of green and sustainability domains

shade n share
non-green 50744 0.909
lighter green 3418 0.061
darker green 1663 0.030

5. Skills requirements in green versus non-green vacancies

6. Posted wages

Posted annual wages (MYR, 23,177 vacancies with a salary field, midpoint of the posted range, annualised). Nominal values; over a period of a few months deflation changes nothing material, and the code has a hook for a CPI deflator when the series is longer.

green_shade n median_wage mean_wage p25 p75
non-green 21,904 45,000 54,702 36,000 60,000
lighter green 843 54,000 65,711 40,800 81,000
darker green 430 50,164 59,577 40,350 69,000

7. Candidate occupation-specific technical green skills

Expressions over-represented in green vacancies relative to non-green vacancies of the same ISCO-08 sub-major group, excluding terms already in the ILO dictionaries. A starting point for a national green skills taxonomy, to be reviewed by experts.

ISCO-08 sub-major group Top candidate expressions (over-represented in green vacancies)
Administrative and Commercial Managers occupational safety, site safety, site safety supervisor, safety supervisor, yellow book, palm oil, safety environment, safety standard, sime darby, conduct safety
Business and Administration Associate Professionals external stakeholder, quantity surveyor, proficient microsoft, prove track, industry trend, key account, achieve sale, sale executive
Business and Administration Professionals actively contribute, closing activity, sale engineer, currently seek, relate activity, senior analyst, key stakeholder, end user, willingness learn, risk assessment
Information and Communications Technology Professionals joinhiltiasiaitservice look, touch joinhiltiasiaitservice look, touch joinhiltiasiaitservice, interview don, stay touch joinhiltiasiaitservice, second like, telephone interview don, joinhiltiasiaitservice look forward, exciting career thorough, career thorough people
Production and Specialized Services Managers product quality, raw material, operational excellence, corrective action, mechanical engineering, performance indicator, continuous improvement, industry standard, key performance, key performance indicator
Science and Engineering Associate Professionals product quality, maintenance activity, electrical mechanical, quality control, continuous improvement, preventive maintenance, mechanical engineering, mechanical electrical, electrical electronic, certificate diploma
Science and Engineering Professionals treatment plant, power plant, occupational safety, engineering sdn, engineering sdn bhd, safety standard, facility engineer, single line, line diagram, detailed engineering

8. Validation and quality assurance

Boilerplate sensitivity: company text repeated across many postings can inflate green matches. Results with repeated sentences removed:

indicator ilo_method without_boilerplate
green vacancies (share of all) 0.091 0.083
darker green vacancies (share of all) 0.030 0.028
green postings containing at least one boilerplate sentence NaN 0.328
green postings that stop being green without boilerplate NaN 0.086
boilerplate sentences identified (>= 20 postings) NaN 1,129

9. Reproduce

Clone the repository, download the vacancy dataset into data/, install requirements.txt and run the numbered scripts in src/ in order. Every intermediate file, threshold and rule is set in config.yaml; the README is written as an implementation guide for a national institution.