Developing measures to assess the extent of sustainable competitive advantage provided by business process reengineering

Journal of the Academy of Business and Economics, April, 2003 by R. Srinivasan

A multi-dimensional hypothesized SCORE model was examined using LISREL framework (Joreskog & Sorbom 1989) using the following model:

(1) X = [LAMBDA][xi] [delta]

Where X is a vector of Q observed variables [xi] is a vector of n (n

(2) [SIGMA] = [LAMBDA] [PHI] [LAMBDA] [phi]

where [PHI] is matrix of intercorrelations among the common factors and [phi] is a diagonal matrix of error variance ([[theta].sub.[delta]]) for the measures.

Maximum Likelihood (ML) parameter estimates for [LAMBDA], [PHI], [phi] and [chi square] GFI for the null model implied by equations (1) and (2) are obtained from the LISREL program. The probability level associated with the given [chi square] statistic indicates the probability (p) of attaining a large [chi square] value given that hypothesized model is supported. The higher the value of p, the better is the fit, and as a rule of thumb, value of p>0.10 are considered as an indication of satisfactory fit (Hawley and Maxwell 1971).

Researchers increasingly complement [chi square] statistic with Bentler & Bonnet's (1980) incremental fit index--which is an indication of the practical significance of the model in explaining the data, since exclusive reliance on [chi square] statistic is criticized for many reasons (Fornell & Larker 1981). The representations of the data are,

(3) [DELTA] = ([F.sub.o] - [F.sub.K]/[F.sub.o]

where [F.sub.o] = [chi square] value obtained from a null model specifying mutual independence among the indicators, and [F.sub.K] = [chi square] value for the specific model. The general rule of thumb is that should be greater that 0.90 (Bentler & Bonnet 1980). Table 3 summarizes the results of assessments for unidimensionality for the 14 dimensions. It provides the following model statistics for the assessment of GFI: The [chi square] statistic, its associated degrees of freedom, p-level of significance and the Bentler & Bonnet index [DELTA]. From the results in Table 3, one can conclude that each of the 14 dimensions achieve unidimensionality and convergent validity at the monomethod level of analysis.

4.2 Internal Consistency of Operationalizations (Reliability)

The above result for unidimensionality is not a direct assessment of reliability of the SCORE construct. A typical approach to measure reliability is in terms of the Cronbach Alpha coefficient (Cronbach 1951), which ranges between 0 and 1 has a desirable property of being a lower bound of reliability (Lord & Novick 1968). However, since Cronbach Alpha gives equal importance to all indicators, its application is limited. An alternative approach to measuring reliability is that it represents the proportion of measure variance attribute to the underlying trait. Thus with this approach, reliability ([P.sub.c]) can be calculated as,

(4) [[rho].sub.c] = [([SIGMA][[gamma].sub.1]).sup.2] Variance (A) / [([SIGMA][[gamma].sub.1]).sup.2] (A) [SIGMA][[theta].sub.[delta]])

where [[rho].sub.c] is the composite measure of reliability, n is the number of indicators, and [[gamma].sub.1] is the factor loading which relates item I to the underlying theoretical dimension (A). In other words, [[rho].sub.c] is the ratio of trait variance to the sum of trait and error variances. When it is greater than 50% it implies that the variance captured by the trait is greater than that of error components (Bagozzi 1981). Table 4 shows that all the [[rho].sub.c] indices are greater than 0.50, thus indicating existence of internal consistency or reliability.