【正文】 research, attributes that have a potential to cause project cost escalation were identified . In the past, several authors have examined the impact of isolated attributes on project cost However, no project management tool is available to account for the collective impact of all possible attributes. The attributes were divided into two groups, quantifiable and nonquantifiable attributes. Attributes that have a cost value associated with them in the project estimate were defined as quantifiable attributes, ., total material cost, total labor cost, total equipment cost, project management cost, and total cost of the project at end of work. Attributes that do not have a cost value associated with them in the project estimate were defined as nonquantifiable attributes. The need to differentiate between quantifiable and nonquantifiable attributes is elaborated later under modeling assumptions. FIG. 1. Example Influence Pattern Refers to the percentage cost escalation over the estimated project cost. To satisfy these requirements, a DSS such as COMPASS would be most suitable. MODELING ASSUMPTIONS The interrelationships between attributes, the resulting influence pattern, and the impact of attributes on the project cost have been structured by defining the five following modeling assumptions: Assumption 1 If an attribute, ., F (refer to Fig. 1) is influenced by a set of attributes, ., C and D, then the individual influence of the attributes in that set on F (., the influence of C on F and the influence of D on F) is considered to be independent, . p[(F n C)I(F n D)] =p(F n C) (Ia) :. p[(F n C) n (F n D)] 。 (2) module 2to determine the probable cost influence of attributes in a new project。) = [p(X = 11A = 1) 。 Berger 1985). Thus, mencing from the starting attributes A and B (refer to parts A and B of Fig. 5), the marginal probabilities for all of the attributes in the influence pattern are calculated (refer to part B of Fig. 5). Example calculations are shown in (12)(15) for obtaining p(D) in accordance with the modeling assumptions (also refer to part B of Fig. 5). The numbers have been individually rounded off by the spreadsheet. For example, the numbers in (12) are actually X = , and similarly for the other equations. The numbers shown in the equations are consistent with what appears in the figures. peA n D) =p(A)p(DIA) = X = (12) p(B n D) = p(B)p(DIB) = x = (13) ~ p(D) =p(A n D) U p(B n D) (14a) =p(A nD) +p(B n D) 163。 and (2) suggest changes if required, by providing their subjective (or judgmental) input with respect to the conditional relationship between attributes (refer to part B of Fig. 4, conditional probability, perception of the team members, columns 26). n the third step, the subjective input provided by the team members is analyzed by the PDM. The primary decision maker assigns weights to the input provided by the team members and also to the information extracted from the past project performance data (refer to part B of Fig. weight given to team member input). The weighting process is necessary to relate the varying knowledge and experience of the team members and thus the reliability of the information provided by them. In the fourth step, the system putes the group decision by taking into account the subjective input and the weights assigned by the PDM (refer to part B of Fig. 4, group perception). The important results obtained from the DPM are (I) the conditional probabilities of attributes。 they should have similar scope of work。 refer to Fig. 1)。 (2) the interrelationships between attributes established in the influence pattern。 and (2) they should have faced a cost escalation. For each historical project, the user subjectively identifies the state of attributes by using a binary mode, as explained earlier under the modeling concepts (refer to part B of Fig. 3). This information about the state of attributes in historical projects is processed by the DPM to determine The conditional probabilities, ., p(C =llA =1), p(E = 11 C = 1), and so forth [refer to part A of Fig. 3 and (2) and (3)] The individual cost influence of attributes (refer to part C of Fig. 3) The conditional probabilities are further calibrated in the GDM. This calibration is done with respect to the new project characteristics. The calibrated conditional probabilities and the individual cost influence of attributes are used as an input for the PWPCE model in module 2 to analyze a new project. p(C =llA =1) =p[(C =1) n (A =1)] i p(A =1) (2) p(C =llA =1) =2: [(C =1) and (A =1)]/ i 2: (A =1)/ where j = 1 ... n (n = number of past projects selected) (3) In the second stage of the DPM, a significant level of escalation is defined for each historical project (Le., a level of escalation that was accounted for in the contingency fund for the project). The critical line items and their associated escalation values are identified by considering the significant level of escalation thus defined (refer to part C of Fig. 3). Critical line items are those line items that had faced a cost escalation greater than the significant level of escalation defined for that project. Each critical line item is associated with a quantifiable attribute. For example, (refer to part C of Fig. 3), critical line item number 1 is associated with quantifiable attribute X (say, total labor cost) and critical line item number 2 is associatedwith quantifiable attribute Y (say, total material cost). After establishing the association between the critical line items and the quantifiable attributes, the user analyzes each critical line item with respect to the list of attribute analysis is conduc