In order to examine the proposed research model, the study employed the Partial Least Square which is a Structural Equation Modelling (SEM) technique. This technique is well suited for predictive models using small samples (Chin, 1998). Typically, SEM approach is used to develop a causal model with an objective of model validation. SEM is a second generation multivariate technique gaining popularity in management research. It combines multiple regression and factor analysis to estimate the interdependence relationships simultaneously. There are two approaches to SEM namely Covariance approach and Partial Least Square based approach. The Covariance-based approach for SEM needs a large sample. The definition of large size varies from one author to another. Some define it as a sample having more than 100 subjects and some others define it as a sample having three characteristics namely more than 200 subjects, at least three indicators, and typically requiring reflective mode. On the other hand, Herman Wold initiated the component-based approach to SEM in 1982, under the name “PLS” as an alternative to covariance based approach. Partial Least Square Path Modelling (PLS-PM) is generally meant as a component-based approach to SEM. Further, PLS does not make assumptions about the population or scale of measurement and there are no distributional requirements (Fornell et al 1995).
Since Partial Least Square makes no distributional requirements, the structural model was evaluated using the R-square for the dependent constructs and the size, t-statistics and significance of the path coefficients. Path coefficients in Partial Least Square are standardised regression coefficients (Staples et al 1998 in Cohen 2001). In order to ensure whether path coefficients are statistically significant or not, bootstrap and Jackknife re-sampling procedures are used to estimate standard errors for calculating t-values (Fornell and Barclay, 1983). The results are examined at 5 per cent significance level and the t-statistic value at the 0.05 level is 1.96. If the t-statistic value is greater than 1.96, the path is significant (Efron 1979, Efron and Gong 1983). Table 4 and Figure 2 present the results of the test of the model.
The path linking perceived environmental uncertainty to the extent of use of risk analysis techniques was found to be positively significant at 0.05 level (Beta = 0.241, t = 3.051). This reveals that greater the perceived environmental uncertainty, higher the level of risk analysis techniques in SIDs. The path linking perceived company performance to the extent of use of RAT was found to be positively significant at 0.05 level (Beta = 0.630, t = 12.828). This reveals that higher the perceived company performance, higher the level of risk analysis techniques in SIDs. The coefficient of determination i.e. R2 indicates the predictive power of the structural model i.e. these two variables have a good predictive power (0.526 i.e. 52.60 per cent) of the extent of use of risk analysis techniques in SIDs. In addition to the above, the analysis indicates that all the firms formally analyse project risk for almost all the projects.
It also reveals that the firms use multiple techniques simultaneously to evaluate the investment projects. The most often used method is sensitivity analysis, followed by probability analysis, CAPM, and Monte-Carlo-type probabilistic simulation techniques (see table 2).To examine the difference between automobile and ancillary companies with regard to the extent of use of risk analysis techniques, perceived environmental uncertainty and company performance, this study employed t-test for equality of mean score. Table 4 shows that there is no significant difference among the automobile and ancillary companies with regard to the extent of use of risk analysis techniques, perceived environmental uncertainty and company performance, This indicates that companies’ risk assessment practices in SIDs do not vary with their size and types of investments.