The Things That Hurt People Are Not the Same as the Things That Kill People: Key Differences in the Proximal Causes of Low- and High-Severity Construction Injuries

For nearly a century, the safety profession has been influenced by the “safety pyramid.” The theory introduced by Heinrich (1931) and later refined by Bird (1966) suggested that there is a constant ratio of minor to major injuries. When building the safety pyramid, Heinrich used insurance data to establish a ratio among accidents of varying severity levels. Specifically, he found that for every 300 near hits (incidents resulting in no injury) there are 29 minor injuries and one major injury (Heinrich 1941). Bird (1966) later expanded this theory to include unsafe acts and used the safety pyramid to support the thesis that minor and major accidents have similar causes, and that injury severity is fortuitous.

Although Heinrich did not suggest that the safety pyramid was causal, his work was cited to support the pervasive belief that preventing low-severity injuries leads to a corresponding reduction in high-severity injuries (Manuele 2002). This fundamental concept served as the foundation of many accident prevention programs (Khanzode et al. 2012; Marshall et al. 2018; Rebbitt 2014), which led to a generalized approach to safety management. Consequently, most traditional safety programs are structured to prevent all types of injuries without specific focus on more severe injuries.

There is an ongoing debate about the validity of the ratios in the safety pyramid. Some researchers found a correlation in the rates of minor and major construction injuries over long periods (e.g., Hubbard and Neil 1986) and a similar distribution of high-risk tasks preceding minor and major injuries (Lozada-Larsen and Laughery 1987; Laughery and Vaubel 1993). However, others disputed these studies, claiming that they failed to account for the confounding factors that influence injury severity such as the interaction with energy, age of the injured employee, nature of injury, and injured body part (e.g., Petersen 1989; Salminen et al. 1992; Manuele 2011; Cooper 2019; Busch et al. 2021). Furthermore, although ratios appeared to be stable, some studies (e.g., Zhou et al. 2012) have found differences in the frequency of causal factors that lead to near misses, less severe accidents, and serious accidents in construction. Thus, the correlations fail to pass as causational evidence (Rebbitt 2014; Dunlap et al. 2019). Although these studies have elucidated the limitations in the validity of the pyramid ratios, the extent to which these limitations justify the need for a differentiated approach to safety management in construction is still uncertain.

If the ratio of low- to high-severity injuries was indeed fixed, then long-term injury rates must trend in parallel. However, macrolevel trends have revealed that, while the rate of lower-severity injuries has declined over the last decade, the rate of fatalities has plateaued in the USA (BLS 2022). This divergent trend is similar when comparing the number of fatalities to the number of nonfatal injuries. To illustrate this trend, the authors created Fig. 1 using fatal and nonfatal injury data from 2010 to 2021 collected by the Occupational Safety and Health Administration (OSHA). The three-year rolling averages in Fig. 1 show that the rate of nonfatal construction injuries has declined dramatically while the fatality rate has plateaued. If the ratio of major to minor injuries were indeed fixed, this divergent trend would be impossible, especially at an industry level representing billions of worker hours. This counterintuitive trend supports the emerging viewpoint that focused injury prevention efforts are needed to prevent serious injuries and fatalities (e.g., Manuele 2008, 2011; Shanon and Manning 1980) and raised the following question: what if anything, is different about the causes of serious injuries and fatalities (SIFs)? Answering this question will reveal key aspects of a differentiated approach to prevent SIF in construction. This is aimed at addressing the disconcerting lack of improvement in the rate of fatal injuries shown in Fig. 1.

To establish the specific gap in knowledge, research related to the causes of high- and low-severity injuries and the proximal causes of SIF was reviewed. For consistency and brevity, the term serious injury or fatality (SIF) is used to refer to higher-severity injuries (i.e., life-threatening, life-altering, and life-ending injuries as defined by Bayona et al. (2023), while low-severity injury (LSI) is used to broadly refer to injuries that are less than serious (i.e., actual injuries that are not life-threatening, life-altering, and life-ending).

Divergent macrolevel trends in high- and low-severity incidents over the last decade suggest that the proximal causes of SIFs and lower-severity injuries are possibly different. Inferential evidence supports this hypothesis and suggests that the differentiating factors could be linked to specific types of equipment and tools used (Nussey 1980; Groves et al. 2007; Love et al. 2019), human–object/machine interaction (Shanon and Manning 1980; Laughery and Vaubel 1993; Salminen et al. 1992; Love et al. 2019), supervision factors (Salminen et al. 1992; Haslam et al. 2005; McSween and Moran 2017; Winge et al. 2019), or even demographic factors and body parts (Alizadeh et al. 2015; Salminen et al. 1992; Love et al. 2019). Thus, there is a growing consensus among researchers that SIFs must be managed differently (Busch et al. 2021; Krause and Murray 2012; Manuele 2008). However, despite this mounting evidence, there is some counterevidence provided by Bellamy (2015, p. 98) who found that, “fatal and (serious) nonfatal accidents share the same causes but not in the same proportions.”

This study aimed to contribute to this debate by undertaking a comparative analysis of SIF, PSIF, and LSI cases and determine whether there are statistically significant differences in the causal factors. This study builds upon but deviates from existing knowledge by examining the proximal causes using actual incidents, a blind assessment by experts, and statistical testing. Specifically, the study was designed to answer the following questions:

Answering these questions would provide a deeper understanding of how SIFs can be specifically addressed in modern safety programs. Including PSIFs in the study will elucidate the extent to which near misses can be used for SIF-related learning.

Since the main objective was to identify any differences in the proximal causes of SIFs, PSIFs, and LSIs (i.e., “SIF differentiators”), operational definitions of each group were created. The definitions of these key incident types are as follows:

In total, the focus group analyzed 38 incidents that were distributed as follows: 13 SIF, 13 LSI, and 12 PSIF. These incidents came from the utility sector, commercial and pipeline construction, and specialized construction. Aligned with OSHA’s Construction Focus Four (Brown et al. 2021), most cases involved struck by, caught-in or between, and falls. Tables 5 and 6 show the complete data sets for the 38 incidents analyzed.

As with any study based on human interpretation, there are several limitations of the results. First, a convenience sample of volunteering safety experts and companies limits the external validity of the results. Although the authors captured a broad range of industry sectors within the cases examined, a larger and more diverse set of incidents may produce more generalizable conclusions. Second, a larger repository of cases may allow for the testing of differences for categories of injury types, part of the body affected, and other confounding factors. Third, the retrospective data collection template may be subject to confirmation bias because answers were unavoidably reconstructed from previous incident investigation reports and worker interviews. Although strict rules and a standardized data collection tool were created, confirmation bias could be reduced, but not eliminated. Fourth, although the combination of focus groups and surveys has been effective in addressing assumptions and uncertainties (Abu-Taleb and Murad 1999), it was challenging to eliminate survey responses grounded in incorrect technical assumptions. In an ideal setting, a Delphi procedure would have been implemented to include controlled feedback and private reflections over iterative rounds to build consensus. However, Covid-19 pandemic restrictions made such controls practically infeasible. Fifth, this work does not provide causal explanations of the two SIF differentiators, nor does it examine potential confounding variables and extraneous factors. These are suggested as topics for future research. For example, researchers may wish to explore the reasons why some factors are more present in SIFs than in LSIs by examining a greater number of cases or by further analyzing PSIFs, which may exist in greater numbers than SIFs. Finally, future studies should challenge this work and test other potential differentiators and extend the study to geographies outside of North America.

Faced with compelling evidence that the ratio of fatal and nonfatal injuries is not constant, there has been a desire to identify what is different about SIFs. Within the safety profession, the idea that the things that hurt people are not the same as the things that kill people has begun to replace the antiquated view that SIFs will be prevented by addressing LSIs. To contribute to the body of knowledge, this study involved hypothesizing proximal factors that may differentiate SIFs from LSIs, collecting empirical data from real incident cases, evaluating the cases with an expert focus group, and conducting statistical tests to determine the extent to which any hypothesized factors differentiate SIFs from LSIs. Several key findings will help support SIF-focused prevention efforts that may address the plateau in fatality rates observed in North America.

Proper work planning and execution is well documented as a vital safety practice (Borys 2012; Mroszczyk 2015; Sawacha et al. 1999). In fact, issues with work planning, discipline, and execution have been repeatedly identified as contributing factors to SIFs (Cohen 1976; Alexander et al. 2017a, b; Winge et al. 2019; Mirzaei Aliabadi et al. 2020). However, this study was the first to identify this factor as being possibly unique to SIFs. In review of the cases, it was evident that in most SIF and PSIF cases, the issues in planning and execution revolved around the control of a specific and salient hazard. However, this was an issue for only one of the 13 LSI cases studied. The implication is that proper work planning and execution is needed to specifically address the control of high-energy hazards, which may be a cornerstone of a program that targets SIFs. This finding provides the impetus for further investigations that seek to elucidate on the antecedents (e.g., characteristics of a comprehensive work plan, understanding of a work plan across different stakeholders) to absent or poorly executed work plans, which could help in designing effective interventions.

Related to finding 1, a missing or insufficient direct control almost always precedes a SIF but is almost never involved in an LSI. That is, LSIs often occurred in situations where no specific identifiable control was missing. When dealing with hazards associated with environmental conditions (such as uneven surfaces or dust) or routine movements, a singular physical control was rarely identified. With most LSIs, it is difficult to propose a necessary physical safeguard and perform construction tasks in a dynamic environment.

Interestingly, the factors absent direct controls and absent or not followed work plan were almost always involved in the same SIF cases (i.e., they covaried). This suggests that they may be combined into one factor in future research and in practical application. To broaden the previous statement, the implication of these two key findings is that organizations should focus on ensuring the presence of direct controls against high-energy hazards through effective work planning and execution. This supports the recent research that suggests that safety should be measured by the extent to which life-threatening hazards are observed to be controlled (Oguz Erkal and Hallowell 2023). It also provides practitioners with an actionable factor to prioritize when focusing on SIF prevention efforts.

Perhaps most surprisingly, the other five hypothesized factors were not found to be significant. Although the expert panel had strong reasoning for hypothesizing unfamiliarity with the task, risk normalization, poor hazard recognition, and other human factors as differentiators, the study found that they were equally present in both SIFs and LSIs. This suggests that, although human factors are prevalent causal factors of injuries, they are not differentiators of SIF. The implication is that generalized safety practices that focus on human behavior might be important for injury prevention broadly, but that they may not help to target SIFs directly. This is interesting because of the historical emphasis of research and practice on human behavior and error precursors (e.g., Reason 1990; Senders and Moray 1991; Shappell and Wiegmann 1997).

Another important but unsurprising finding was that none of the factors examined differentiated SIFs and PSIFs. That is, the seven potential SIF differentiators were present in approximately equal proportion in both SIFs and PSIFs indicating that, at least for the factors studied here, these two incident types involved the same contributing factors. Although practitioners have long recommended studying PSIFs (e.g., Martin and Black 2015), the results presented here provide enhanced empirical evidence that PSIFs can be leveraged as part of a SIF prevention strategy. The implication is that organizations can greatly expand their SIF prevention programs by collecting and learning from PSIF incidents, which is extremely helpful since actual SIFs are exceedingly rare.

The findings presented here may be useful to those who attempt to combat the observed plateau in fatality rates by addressing SIFs directly. As companies focus their safety programs, the results suggest that the control of high-energy hazards through effective work planning, discipline, and execution should be a primary focus. Future researchers should challenge these findings and expand on this study by examining additional factors and testing the efficacy of different approaches to SIF prevention.

Some or all data, models, or code generated or used during the study are proprietary or confidential in nature and may only be provided with restrictions.

The authors would like to thank the Construction Safety Research Alliance (CSRA) and its members who participated in this study for providing continuous support, dedication, and insight.