448 H。 Malekitabar et al。 / Safety Science 82 (2016) 445–455

put in class 4 and the fourth decision block in Fig。 1 is never answered yes。

It is worth noticing here that not all accidents are necessarily caused by a violation of a safety rule。 Even in the cases that a vio- lation played a clear role, Behm suggested that a better design removing some direct cause from the accident scenario could have possibly prevented it (Behm, 2005)。 This is what Cooper et al。 (2005) recognize as ‘‘adopting enhanced  design  standards”, a known means of risk prevention among all risk treatment strate- gies。 The outputs of this research, however, will not be restricted to violations, nor to design deficiencies, since there might be also many hazardous situations with no affordable ‘‘prevention” strat- egy, identification of which would make a worthwhile contribution to safety risk  management。

2。3。Data assessment

The reliability and adequacy of the samples had to be ensured regarding two major questions: (1) Are the samples adequate in number, and (2) are they representing the whole population of incident types in the construction industry? In the first, Eisenhardt (1989) recommends to close the sampling when evi- dences of saturation have appeared, i。e。, observed phenomena are being repeated。 In this research, saturation was achieved at about the first one hundred cases for the cases containing incidents of electrocution, confined spaces and machinery, and about two hun- dreds for the cases of fall incidents。 The authors yet decided to con- tinue the study on all available cases to ensure a higher level of analysis comprehensiveness。 With respect to the second question, the universality of the results is inferred by a comparison of the two formal databases with the reports that the participants of the survey have provided, showing a reasonably similar distribu- tion of the accident types, which is summarized in Table 1。 In this table, the accident cases reported by each source are categorized into five types, and the number of cases belonging to each category are compared to show the distribution of accident types among different databases。 For example, accidents consisting of falls or falling objects, formed 39–47% of the accidents in all three data- bases, which is consistent with earlier findings, for instance a share of 47% announced by Li et al。 (2015) who conducted a content sta- tistical analysis on available safety information。

3。Theory

Fig。 1。  Classification of driver  rules。

that separate solid modeling or topological discussions will be needed。 Solihin and Eastman  categorize these rules  into class  3。 In fact, some of these class 3 rules may need additional data to be acquired from external sources and are noted in this research as a separate subclass 3b, as the location of municipal facilities can be determined using augmented satellite maps or governmen- tal web services, and equipment operational specifications can be queried from the manufacturer or open source catalogues。 Further- more, Some drivers that may require advanced technical simula- tions, for example the effect of a heavy equipment passing by a trench on soil stability, and possibly need the data to be piped from the Building Information Model to specialized software, are noted as subclass 3c。 Solihin and Eastman add a fourth class of rules that do not seek for a definite answer, but provide the best answer among possible solutions for the  case  (Solihin  and Eastman, 2015)。 Since this research is to find the drivers needed for risk identification, no rules about risk treatment are expressed to be

Although it may be common to use the concepts  of  risk ‘‘drivers” and ‘‘sources” interchangeably, examples of risk sources can be found dormant in the absence of risk drivers。 Referring to the International Risk Governance Council (IRGC), one can  think of a risk source as a fertile ground which contains seeds of count- less mysterious risks waiting to be cultivated by one or more dri- vers (Graham et al。, 2010b), like the first raindrops stimulate the growth of a beanstalk, and the next ones help it grow faster。 By def- inition, a risk driver is an event or condition, be it an error or not, that stimulates, triggers or even delays a potential risk (Bartlett, 2002; Berkeley et al。, 1991; Graham  et  al。,  2010a;  Vose, 2008)。 For example, the source of the risk ‘‘fall from elevation” is the grav-