Salas and Hallowell (2016) showed that four latent factors extracted from 20 contractor-reported proactive safety metrics statistically predict subsequent total recordable incident rate (TRIR) and severity rate (SR) across 261 large oil-and-gas construction contractors, while explicitly acknowledging that the relationships are correlational, rest on unverified self-reported metrics, and are untested outside that population. The open problem is to establish the predictive validity of safety leading indicators properly: to quantify and correct the reporting biases embedded in indicator measurement, to identify the causal — not merely correlational — effects of indicator-represented practices on subsequent TRIR and SR, and to characterize the domain (contractor sizes, sectors, and jurisdictions) over which the predictive relationship transports. A 2025 cross-industry scoping review of 48 studies (database searches through August 2023) found that no published study meets these standards — its authors report that the study designs were not appropriate for determining causality, internal validity was moderate to low, and no two studies assessed the same research question — and no study addressing this causal core is known in the literature to date, so the question remains open.

Background

Construction and oil-and-gas projects have traditionally monitored occupational safety with lagging indicators — outcome measures computed after injuries occur. The two used here are the total recordable incident rate (TRIR), the number of OSHA-recordable injuries per 200,000 worker-hours (OSHA is the U.S. Occupational Safety and Health Administration, whose recordkeeping rules define which injuries count), and the severity rate (SR), a companion measure of the lost-workday impact of those injuries per unit of exposure. Because lagging metrics reveal problems only after the fact, owners and clients have long sought leading indicators: measurable, proactive safety practices — near-miss reporting, job safety analyses (structured pre-task risk assessments), audits, site walkthroughs, stop-work authority, and corrective-action tracking — that vary before injuries do and could therefore provide early warning and guide preventive intervention.

A large survey-based program by Hinze, Thurman and Wehle (2013) identified the proactive metrics most often associated with good safety outcomes, but the resulting indicator lists were exploratory: small or convenience samples, subjective survey measures, and no quantitative, validated mapping from contractor practices to future injury rates. Salas and Hallowell (2016) supplied the first large-scale empirical attempt at such a mapping. Using enterprise reporting data from oil and gas owners, they assembled 20 contractor-reported proactive metrics for 261 contractors (each supplying at least 200,000 work-hours per year), reduced them by principal factor analysis (a statistical technique that collapses many correlated variables into a few latent factors) to four interpretable factors, and fit robust ordinary-least-squares regressions (regression with standard errors adjusted for unequal variance) on log-transformed factor scores to predict TRIR and SR. Trained on 2012–2013 data, the models predicted the held-out 2014 contractor data with predicted–observed Pearson correlations of about 0.72.

The authors were explicit about what this did and did not establish. Every predictor was contractor-reported to clients, so systematic reporting bias and heterogeneity in how metrics are recorded and verified across clients and contractors were unmeasured; the data were observational, so the estimated relationships are associations, not causal effects of the underlying risk-mitigation practices; and the sample was restricted to large oil-and-gas contractors in one enterprise and legal environment, leaving transportability to smaller contractors, other construction sectors, and other jurisdictions untested. The open question is precisely whether the predictive relationship survives these three threats: measurement bias, confounding, and domain restriction.

Problem Statement

Establish whether contractor-level proactive safety practices, measured as safety leading indicators, validly and causally predict subsequent lagging injury outcomes (TRIR and SR), and determine the domain over which that predictive validity holds.

The problem is to establish, jointly:

  1. Measurement validity — quantification and correction of the systematic reporting bias and heterogeneity in how leading indicators are recorded and verified across clients and contractors, so that indicator values reflect practices rather than reporting behavior.

  2. Causal identification — design-based estimates of the directed effects of indicator-represented risk-mitigation practices on subsequent TRIR and SR, including direct, indirect, and interaction effects among indicators within factors; or, under the same evidentiary standards, a demonstration that the Salas–Hallowell associations are artifacts of reporting behavior or confounding.

  3. External validity — a test of the predictive relationships outside the source population of contractors supplying at least 200,000 work-hours per year in the oil and gas sector — that is, on smaller contractors, other construction sectors, or other legal and regulatory regimes — with the domain of validity explicitly characterized (where the relationships transport and where they fail).

The source also proposed exploratory expansion of the candidate indicator set (safety-in-design, engineering controls, off-site manufacturing and automation, and other project-process variables). That expansion is an adjacent research direction rather than a component of this problem: the validity of the indicators already in use is logically prior to searching for new ones.

References

  1. Rico Salas and Matthew R. Hallowell (2016), 'Predictive Validity of Safety Leading Indicators: Empirical Assessment in the Oil and Gas Sector', Journal of Construction Engineering and Management 142(10):04016052, DOI 10.1061/(ASCE)CO.1943-7862.0001167, https://doi.org/10.1061/(ASCE)CO.1943-7862.0001167
  2. Jimmie Hinze, Samuel Thurman and Andrew Wehle (2013), 'Leading indicators of construction safety performance', Safety Science 51:23–28, DOI 10.1016/j.ssci.2012.05.016, https://doi.org/10.1016/j.ssci.2012.05.016
  3. Helen Lingard, Matthew Hallowell, Rico Salas and Payam Pirzadeh (2017), 'Leading or lagging? Temporal analysis of safety indicators on a large infrastructure construction project', Safety Science 91:206–220, DOI 10.1016/j.ssci.2016.08.020, https://doi.org/10.1016/j.ssci.2016.08.020
  4. Fawaz K. Bitar, Diane Chadwick-Jones, Matthew Lawrie, Marcin Nazaruk and Chan Boodhai (2018), 'Empirical validation of operating discipline as a leading indicator of safety outputs and plant performance', Safety Science 104:144–156, DOI 10.1016/j.ssci.2017.12.036, https://doi.org/10.1016/j.ssci.2017.12.036
  5. Daniel Kuok Ho Tang, Siti Zawiah Md Dawal and Ezutah Udoncy Olugu (2018), 'Actual safety performance of the Malaysian offshore oil platforms: Correlations between the leading and lagging indicators', Journal of Safety Research 66:9–19, DOI 10.1016/j.jsr.2018.05.003, https://doi.org/10.1016/j.jsr.2018.05.003
  6. Clive Q.X. Poh, Chalani Udhyami Ubeynarayana and Yang Miang Goh (2018), 'Safety leading indicators for construction sites: A machine learning approach', Automation in Construction 93:375–386, DOI 10.1016/j.autcon.2018.03.022, https://doi.org/10.1016/j.autcon.2018.03.022
  7. Katelyn Versteeg, Philip Bigelow, Ann Marie Dale and Ashok Chaurasia (2019), 'Utilizing construction safety leading and lagging indicators to measure project safety performance: A case study', Safety Science 120:411–421, DOI 10.1016/j.ssci.2019.06.035, https://doi.org/10.1016/j.ssci.2019.06.035
  8. Jing Xu, Clara Cheung, Patrick Manu and Obuks Ejohwomu (2021), 'Safety leading indicators in construction: A systematic review', Safety Science 139:105250, DOI 10.1016/j.ssci.2021.105250, https://doi.org/10.1016/j.ssci.2021.105250
  9. Elif Deniz Oguz Erkal, Matthew R. Hallowell and Siddharth Bhandari (2023), 'Formal evaluation of construction safety performance metrics and a case for a balanced approach', Journal of Safety Research 85:380–390, DOI 10.1016/j.jsr.2023.04.005, https://doi.org/10.1016/j.jsr.2023.04.005
  10. Hamidreza Golabchi, Abbey Dale Abellanosa, Lianne Lefsrud, Estacio Pereira and Yasser Mohamed (2024), 'A comprehensive systematic review of safety leading indicators in construction', Safety Science 172:106433, DOI 10.1016/j.ssci.2024.106433, https://doi.org/10.1016/j.ssci.2024.106433
  11. Deborah Watkins, Emma Bishop, Steven Naylor, Ruth Frankish, Malcolm Staves, John Dony, Paul Miller, Rachael McCool and Lavinia Ferrante di Ruffano (2025), 'A scoping review of the evidence base for the performance of leading indicators for improving safety outcomes: Available evidence, implications for practice and future directions', Journal of Safety Research 95:530–544, DOI 10.1016/j.jsr.2025.10.024, https://doi.org/10.1016/j.jsr.2025.10.024