7394 P.Xu et al.Energy Policy 39(2011)7389-7398 order of these factors is different.However,for other profes- sionals,apart from effective coordination and trust,those factors 700. including organizing skill of leader,project objectives control mechanism,and policy support were given much higher priorities than other factors.This is probably because people of the other professionals from governments,academics,consultancies,etc., paid more attention on the macro-factors,rather than micro- factors preferred by experts from industry. 昌刻语 4.2.Factor analysis 3昌 A long list of 21 CSFs is not very helpful to succinctly explain the success of a project.Factor analysis was used to explore and detect the underlying relationships among the identified CSFs 屋目昌程 This statistical technique can recognize a relatively small number of factors that can be used to represent relationships among sets 5N目 of many interrelated variables.The appropriateness of the factor analysis for the factor extraction needs to be tested in various ways.Factor analysis can be used either in hypothesis testing or 36星超3 8 in searching for constructs within a group of variables (Bartholomew and Knott,1999).Factor analysis is a series of methods for finding clusters of related variables and hence an ideal technique for reducing a large number of items into a more 时台 easily understood framework.It focuses on a data matrix pro- duced from the collection of a number of individual cases or respondents.In this paper,factor analysis is applied to explore the underlying constructs of the identified CSFs of EPC for sustainable BEER in hotel buildings. :01 In this research,21 CSFs were subjected to factor analysis using principal components analysis and varimax rotation.Prin- ciple components analysis is a common method in factor analysis, and involves the generation of linear combinations of variables in the way of factor analysis so that they account for as many of the variances present in the collected data as possible.Such an analysis summarizes the variability in the observed data by means of a series of linear combination of"factors".Each factor .604 can,therefore,be viewed as a "super-variable"comprising a specific combination of the actual variables examined in the 8 survey.The advantage of this method over other factor analytical approaches is that the mathematical representation of the derived linear combinations avoids the need for the use of questionable causal models (Johnson,1998). The first stage of the factor analysis is to determine the strength of the relationship among the variables,namely,the 21 identified CSFs,measured by the correlation coefficients of each pairs of the variables.Table 3 gives the matrix of the correlation 'sauiplinq coefficients among the CSFs.The correlation coefficients show 40 :2550 that the CSFs share common factors.The Bartlett test of sphericity is 807.409 and the associated significance level is 0.000,suggest- ing that the population correlation matrix is not an identity matrix.The value of the Kaiser-Meyer-Olkin measure of sam- pling accuracy is 0.752,which is higher than 0.5 and hence is considered acceptable.The results of these tests show that the sample data is appropriate for factor analysis. & In order to avoid confusion between the extracted factors and the CSFs,it is necessary to rename the extracted factor as a "cluster"in the interpretation of the results of the analysis. Six clusters with eigenvalues greater than 1 are extracted.Table 4 lists the cluster matrix after varimax rotation.Table 5 shows the final statistics of the principal component analysis,and the clusters extracted account for 66.618%of the variance. 4.2.1.Discussion and interpretation of clusters For further discussion,it is necessary to rename each of the 》 groupings.Based on an examination of the inherent relationshipsorder of these factors is different. However, for other profes￾sionals, apart from effective coordination and trust, those factors including organizing skill of leader, project objectives control mechanism, and policy support were given much higher priorities than other factors. This is probably because people of the other professionals from governments, academics, consultancies, etc., paid more attention on the macro-factors, rather than micro￾factors preferred by experts from industry. 4.2. Factor analysis A long list of 21 CSFs is not very helpful to succinctly explain the success of a project. Factor analysis was used to explore and detect the underlying relationships among the identified CSFs. This statistical technique can recognize a relatively small number of factors that can be used to represent relationships among sets of many interrelated variables. The appropriateness of the factor analysis for the factor extraction needs to be tested in various ways. Factor analysis can be used either in hypothesis testing or in searching for constructs within a group of variables (Bartholomew and Knott, 1999). Factor analysis is a series of methods for finding clusters of related variables and hence an ideal technique for reducing a large number of items into a more easily understood framework. It focuses on a data matrix pro￾duced from the collection of a number of individual cases or respondents. In this paper, factor analysis is applied to explore the underlying constructs of the identified CSFs of EPC for sustainable BEER in hotel buildings. In this research, 21 CSFs were subjected to factor analysis using principal components analysis and varimax rotation. Prin￾ciple components analysis is a common method in factor analysis, and involves the generation of linear combinations of variables in the way of factor analysis so that they account for as many of the variances present in the collected data as possible. Such an analysis summarizes the variability in the observed data by means of a series of linear combination of ‘‘factors’’. Each factor can, therefore, be viewed as a ‘‘super-variable’’ comprising a specific combination of the actual variables examined in the survey. The advantage of this method over other factor analytical approaches is that the mathematical representation of the derived linear combinations avoids the need for the use of questionable causal models (Johnson, 1998). The first stage of the factor analysis is to determine the strength of the relationship among the variables, namely, the 21 identified CSFs, measured by the correlation coefficients of each pairs of the variables. Table 3 gives the matrix of the correlation coefficients among the CSFs. The correlation coefficients show that the CSFs share common factors. The Bartlett test of sphericity is 807.409 and the associated significance level is 0.000, suggest￾ing that the population correlation matrix is not an identity matrix. The value of the KaiserMeyerOlkin measure of sam￾pling accuracy is 0.752, which is higher than 0.5 and hence is considered acceptable. The results of these tests show that the sample data is appropriate for factor analysis. In order to avoid confusion between the extracted factors and the CSFs, it is necessary to rename the extracted factor as a ‘‘cluster’’ in the interpretation of the results of the analysis. Six clusters with eigenvalues greater than 1 are extracted. Table 4 lists the cluster matrix after varimax rotation. Table 5 shows the final statistics of the principal component analysis, and the clusters extracted account for 66.618% of the variance. 4.2.1. Discussion and interpretation of clusters For further discussion, it is necessary to rename each of the groupings. Based on an examination of the inherent relationships Table 3 Correlation matrix of CSFs for EPC in sustainable BEER in hotel buildings.a CSF1 CSF2 CSF3 CSF4 CSF5 CSF6 CSF7 CSF8 CSF9 CSF10 CSF11 CSF12 CSF13 CSF14 CSF15 CSF16 CSF17 CSF18 CSF19 CSF20 CSF21 CSF1 1.000 CSF2 0.212 1.000 CSF3 0.124 0.413 1.000 CSF4 0.220 0.409 0.417 1.000 CSF5 0.019 0.554 0.458 0.311 1.000 CSF6 0.083 0.303 0.624 0.450 0.291 1.000 CSF7 0.119 0.184 0.026 0.159 0.040 0.045 1.000 CSF8 0.181 0.333 0.357 0.318 0.240 0.282 0.298 1.000 CSF9 0.126 0.355 0.321 0.253 0.271 0.232 0.197 0.667 1.000 CSF10 0.110 0.189 0.188 0.354 0.094 0.184 0.432 0.363 0.305 1.000 CSF11 0.269 0.356 0.357 0.172 0.340 0.200 0.142 0.430 0.405 0.264 1.000 CSF12 0.005 0.200 0.156 0.128 0.282 0.086 0.239 0.430 0.609 0.210 0.180 1.000 CSF13 0.172 0.229 0.246 0.062 0.344 0.056 0.139 0.289 0.109 0.185 0.198 0.264 1.000 CSF14 0.267 0.332 0.225 0.309 0.090 0.349 0.035 0.230 0.291 0.046 0.111 0.020 0.073 1.000 CSF15 0.109 0.222 0.083 0.189 0.038 0.180 0.020 0.286 0.283 0.240 0.054 0.207 0.122 0.148 1.000 CSF16 0.413 0.424 0.241 0.320 0.155 0.291 0.126 0.242 0.256 0.024 0.169 0.077 0.082 0.468 0.137 1.000 CSF17 0.153 0.400 0.292 0.221 0.461 0.454 0.076 0.213 0.270 0.025 0.263 0.087 0.059 0.272 0.106 0.331 1.000 CSF18 0.198 0.437 0.425 0.322 0.456 0.427 0.077 0.254 0.255 0.075 0.356 0.032 0.104 0.252 0.095 0.391 0.557 1.000 CSF19 0.138 0.620 0.480 0.345 0.678 0.266 0.041 0.286 0.387 0.186 0.593 0.147 0.181 0.267 0.006 0.245 0.522 0.461 1.000 CSF20 0.288 0.268 0.257 0.174 0.159 0.233 0.110 0.449 0.564 0.158 0.266 0.419 0.073 0.346 0.191 0.528 0.300 0.208 0.292 1.000 CSF21 0.102 0.130 0.151 0.161 0.182 0.239 0.202 0.333 0.359 0.294 0.145 0.370 0.376 0.135 0.363 0.185 0.167 0.060 0.054 0.479 1.000 a KaiserMeyerOlkin measure of sampling adequacy¼0.752; Bartlett’s test of sphericity¼807.409; degree of freedom¼210; significance¼0.00. 7394 P. Xu et al. / Energy Policy 39 (2011) 7389–7398