【正文】 ience in site safety were identified. They represent various stake holders in the construction value chain such as contractors, publicsector client, government safety department and safety auditing consultancy firm. All respondents have more than 5 years of working experience in the construction industry. They are considered subject matter experts because they have the necessary knowledge and working experience in handling construction projects. Data were collected through facetoface interviews using the structured questionnaire. Each interview lasted for approximately 2 h. Respondents were instructed to refer to Fig. 3 showing the four factors (level 1) and the 17 subfactors as the survey progressed in order to understand what they were paring. The respondents were further reminded that during the parison of the variables, they had to relate them to the enhancement of SMS on construction worksites. The points were given as genuinely and honestly as possible based on respondents’experience and no influence over any variables were induced.The relative importance ratings from the 30 respondents were input into Expert Choice 2000 software. The programme makes use of the respondents’ data to crosspare all variables to determine the weights and inconsistency ratios. Inconsistency ratio is a measure of the percentage of time decision makers are inconsistent in making judgement. The considered ‘‘a(chǎn)cceptable’’ inconsistency ratio is approximately 10% or less but‘‘particular circumstance’’ may warrant the acceptance of a higher value [19]. An inconsistency ratio of 100% is however unacceptable because the ratings are as good as random judgements. 14 of the 30 experts had inconsistency ratios above 15%. This was too high and their responses were discarded. Of the remaining 16, 14 Experts had low inconsistency ratios eo5%T and two had ratios between 10% and 15%. These two respondents (Experts 4 and 5) were given another chance to relook at their ratings and determine if they would like to change their decisions.Caution was taken to ensure that respondents do not change their previous decisions just to fulfil the inconsistency ratio target. Eventually, one respondent did not change his rating (Expert 5) while another (Expert 4) made some adjustments on his own free will.The inconsistency ratio for Expert 5 on the section of policy aspect was 38%. This is considered very high and Expert 5 had chosen to keep this score. Nevertheless, Expert 5’s data were included in the analysis of weightage because the higher than usual inconsistency ratio was due to his extreme judgement rather than a clerical error. Thus, Expert 5’s ratings were accepted even though the inconsistency ratio was greater than 10%. According to Saaty [19], an accurate judgement is more important than consistently inaccurate judgement.The first and second level weights were puted by averaging the weights for the 16 remaining respondents. As the 3P+I Model may be licensed, the actual weights are not shown in this paper. Nevertheless, the relative importances of the factors, in ascending order are:? Personnel Factor, ? Incentive Factor,? Process Factor,? Policy Factor.. Importance weights for lower level attributes using Likert scale (step 10)Due to the large number of third and lower level attributes, it was not practical to use AHP to determine the weights. As such, the 5point Likert Scale was used to elicit the importance weights. A questionnaire showing all the lower level attributes was designed. Respondents were asked to rate the extent to which each attribute contributed to the effectiveness of SMS on construction sites on a 5point scale where 1 ? not important, 3 ? neutral, 4 ? important and 5 ? very important (critical).To determine the importance of lower level attributes, 17 experts were randomly selected from the following types of organisations: clients (public and private)。building contractors (local and foreign)。 safety auditing and consulting firms。 and the MOM which is the safety regulatory body.Among the 17 respondents contacted, 12 expressed interest to take part in the questionnaire survey. They prised four clients, two safety auditing and consulting firms, five large building contractors and MOM. Data were collected using the structured questionnaire, through facetoface interviews. All interviewees were senior management and had many years of experience in the construction industry.From the ratings of the 12 interviewees, mean importance weight for each lower level attribute was calculated. These importance weights were also normalised. For the same reasons as given above, the individual importance weights are not provided in this paper.. Rating the construction site for each attribute (step 11)The next element of the MAVT model is the rating method that auditors are required to use to rate the different attributes. The rating method was first designed, and then verified with five of the industry experts who participated in the AHP described earlier. The principles adopted in designing the rating method were ease of rating and objectivity during assessment.The ideal rating method is one that allows safety auditors to allocate points to the attributes in an objective and straight forward manner. This is to minimise the probability of having two auditors getting vastly different results when evaluating the same construction worksite at the same time.Before the actual design of the rating methods, the five experts were interviewed on what appropriate rating methods could be adopted. Based on their feedback, the research team designed a set of rating methods, which were then shown to them. The five experts agreed with the rating methods, which are now described. Four possible rating options were designed:? 0/1 which means 0 or 1,? 0–1 which means fraction between ‘‘0’’ and ‘‘1’’,? 0/1/NA which means 0 or 1 or not applicable,? 0–1/NA which means fraction between 0 and 1 or not