Strategies for construction hazard recognition

89
Phase II – Hypothesis testing
Although the SMQM maturity model and cognitive retrieval mnemonics were developed from the input of an expert panel, we desired to empirically measure the impact of using the tool on the level of hazard recognition and communication in practice. As previously discussed, we aimed to test the null hypothesis that the use of the maturity model that is reinforced by the
mnemonic cognitive cues does not improve the proportion of hazards identified and discussed
prior to commencing work. In order to test the effects of such safety interventions, various research design methodologies have been discussed in literature. The purpose of such experimental or quasi-experimental designs is to allow researchers to infer causal relationships between the treatment variables (e.g. intervention) of interest and the dependent variable (e.g. observed effects) (Luftig and Jordan 1998; Shadish et al. 2010). This is often accomplished by deliberately selecting treatment variables to be studied and manipulating them in a controlled fashion to determine effects (Jordan and Luftig 2008). We considered three longitudinal methods for empirical hypothesis testing Pre/post testing, withdrawal design, and multiple baseline testing. Pre/post or before and after testing was dismissed because the method is negatively sensitive to typical changes that occur on construction sites and within project-based organizations that may cause severe threats to internal and external validity (Dimitrov and Rumrill 2003; Richards 1999). That is, when using these methods, changes in performance may occur due to factors that are unrelated to the intervention under investigation (internal validity, and may not be generalizable (external validity. On the other hand, the withdrawal design was dismissed because removing an intervention that positively affects human health and safety is unethical (Baer et al. 1968; Barlow et al. 2009;

90 Watson and Workman 1981). Thus, we selected multiple baseline design because, when properly conducted, it is immune to the effects of confounding factors and allows the researcher to leave the positive intervention in place once it has been introduced. Multiple baseline testing is effective for overcoming these challenges because of the requisite process of engaging multiple groups, staging the intervention within the study period, and inter- and intra-group statistical comparisons (Hawkins et al. 2007). Multiple baseline testing involves a series of longitudinal AB designs conducted concurrently that are replicated within a single study, where A refers to the baseline phase and B to the treatment phase (Barlow et al. 2009; Biglan et al. 2000; McGuigan 1997). The interventions are time-lagged or staggered for individual groups or subjects. Hence, while one group or subject receives the treatment, the other groups perform the role of a control group. We solicited large, stable projects from our expert group members. To minimize bias we ensured that the expert participants themselves were not involved in project management for the case site nor were they involved in the day-to-day work. Two project-sites were selected for two-week immersive case studies. Since we desired clear variation in the types of work being performed on the case sites, we selected a modular construction site with highly stable work processes and the construction of a natural gas power generating plant with dynamic work processes. Within each of the two sites, three independent crews were identified for field testing. Because the unit of analysis was the work crew, our study involved six baselines from two independent project sites. The sample size exceeds the minimum requirement of having two baselines suggested from

91 literature (Barlow et al. 2009; Blount et al. 1982; Fleece et al. 1981; Kazdin and Kopel 1975; Van Houten et al. 1985; Wolf and Risley 1971).

Download 2.75 Mb.

Share with your friends: