«Thu Thi Nguyen*, Lokman Mia, Allen Huang Department of Accounting, Finance and Economics, Griffith Business School Griffith University, Australia ...»
An invitation letter with a URL link to the web-based survey was sent to department managers through a variety of methods, such as managers’ email addresses, LinkedIn groups and management forums. Managers, who received the invitation letter, agreed to participate by clicking on the link to the web-based questionnaire. This method has weaknesses of response errors (respondents provide inaccurate data) and non-response errors (target respondents do not reply). In order to deal with these errors, appropriate procedures were carried out (Smith, 2003).
9 In particular, the Mann-Whitney U Test for possible non-response bias was run. No major non-response bias was indicated since differences between early and late respondents were insignificant. Every fortnight, for 2 months after the initial invitation letter, reminder letters were sent to encourage responses. There were 707 managers who clicked on the link to participate: 291 filled in the questionnaires and submitted them; 109 responses from managers in small enterprises were excluded; three outlier cases were removed from the data; finally, 182 cases were used for data analysis.
Table 1 presents the demographics of sample respondents. Of the 182 cases collected in this survey, 95 (52%) were from Ho Chi Minh, the largest and most dynamic city in Vietnam and 87 (48%) were from other cities and provinces. Nearly half (46%) of responses were from POEs, 60 (33%) from FOEs and 38 (21%) from SOEs. Managers in SOEs and FOEs tended to be better qualified than those in POEs: 35% of respondents in SOEs and 30% in FOEs had a Master’s degree or higher qualifications compared to 19% in POEs. In POEs and FOEs, managers had more years of managerial experience in their companies: 23% and 25% respectively, compared to only 18% in SOEs, who had more than 9 years’ experience.
Insert Table 1 here
3.2 Variable measurement Well-developed instruments from previous research were used to measure the variables of the study.
Managerial performance The instrument developed by Mahoney et al. (1963) was used to measure managerial performance. Participants rated their performance on eight dimensions of a 7-point Likert-type self-evaluation scale. Even though self-rating has been criticised for being subjective, this instrument has been widely used in management accounting research (Burkert et al., 2011, Agbejule, 2005, Chong and Eggleton, 2003, Gul and Chia, 1994, Etemadi et al., 2009) because of the significant correlation between self-rating and super-rating. Moreover, the adoption of objective measures is very difficult due to the complexity of the information. There are a number of studies which have rejected the criticism of subjective measures (Alam and Mia, 2006, Mia et al., 2005, Dunk, 2003).
Managers rated their own performance by placing a point from 1 “very poor” to 7 “excellent” on a eight dimensions, namely planning, investigating, coordinating, evaluating, supervising, staffing, negotiating, and representing. However, two items (negotiating and representing) were dropped because they were loaded on multi-factors (cross-loadings) with the cross-loadings differing by less than 0.2 (Hair et al., 2010) and the values of ‘Cronbach’s alpha if item deleted’ were higher than the final alpha value (Pallant, 2011). The remaining six dimensions were used to examine managerial performance, which satisfied the reliability level with a Cronbach’s Alpha of 0.92 (see Table 2).
The use of MAS information The instrument developed by Chenhall and Morris (1986) has been widely used in most studies on MAS. Following previous studies, the present study adopted a 7-point Likert-type scale on six items developed by Chenhall and Morris (1986), and Mia and Chenhall (1994) to measure the use of MAS information. The respondents rated the extent to which they use each item for making decisions by placing a point from 1 “not used at all” to 7 “used to a great extent”. The average scores for the six items (including future events, probability estimated information, non-economic information, external information, non-financial information, and non-financial 10 market information) represent the overall score for the use of MAS information. This instrument satisfied the reliability level with a Cronbach’s Alpha of 0.86 (see Table 2).
Reward systems This study characterises a reward system in terms of department managers’ perceptions of the link between reward systems and performance targets (performance measures). Questions used to measure the perceptions were adapted from Schulz (2010), Chow et al. (1999), and Shield and Young (1993). Five items are measured with a 7-point Likert-type scale ranging from 1 “not at all” to 7 “to a great extent”: (1) rewards are directly tied to individual performance; (2) rewards are directly tied to performance measures; (3) people’s rewards increase as their performance increase; and (4) individuals whose performance ranks in the top 25% receive higher rewards than those in the bottom 25%. The instrument satisfied the reliability level with a Cronbach’s Alpha of 0.93 (see Table 2).
Insert Table 2 here
3.3 Model analysis This survey tests the hypotheses of multiple relationships between a set of continuous variables (dependent, mediating, and independent); therefore, structural equation modelling (SEM) was used (Hair et al., 2010). SEM is a multivariate technique which can test complex relationships amongst variables. It can also find and test relationships amongst observed and unobserved variables. Other analytical techniques are also included in SEM, such as regression, path analysis and factor analysis. Many software packages, such as Amos, LISREL, PLS and Mplus, are available for estimating SEM models. The SmartPLS 2.0 (M3) Beta software package (Ringle et al., 2005) was used to test the hypotheses of the partial least square (PLS) model of the research. This technique is appropriate for this study because it estimates path models with many latent variables (construct variables) indirectly measured by multiple indicators (manifest variables), and it can be used for analysing data without specifying any distribution assumption (Hair et al., 2011).
As with other component-based SEM techniques, PLS allows the simultaneous examination of both the measurement model (outer model - the relationship between the latent variable and its indicators) and the structural model (inner model - the relationship between the constructs).
According to Ringle et al (2005), in the SmartPLS, the assessment of measurement model is similar to the principal components analysis (using the PLS algorithm with 300 maximum iteration, standardised values and centroid weighting scheme), while the structural model with path coefficients is comparable with ordinary least squares regression (using bootstrapping of 5000 resamples). These two models are discussed in the following sections (see Figure 2).
4. RESULTS Insert Figure 2 here
3.4 Measurement model We examined the reliability and the convergent and discriminant validity of the constructs by analysing the measurement model (see Table 2 and Table 3). The reliability was confirmed since the composite reliability values and Cronbach’s Alpha for all constructs exceeded the critical value of 0.7 (Hair et al., 2010). The convergent validity was demonstrated with all manifest variables loading on the constructs exceeding 0.6, with cross-loadings differing by less than 0.2 and with the average variance extracted (AVE), and communality values exceeding 0.5. Finally, the discriminant validity of the measures was demonstrated, since the square root of AVE for each construct (diagonal elements in Table 3) exceeded the correlations 11 between that construct and the others (Hulland, 1999, Hair et al., 2011, Fornell and Larcker, 1981).
Insert Table 3 here
3.5 Structural model Figure 2 presents the results from PLS analysis, including the measurement model with manifest variables loading on the latent variables and the structural model with standardised path coefficients. In this analysis, significance was based on one-tailed t-test and the amount of variance explained (R2). Standard errors were calculated based on the bootstrapping of 5000 resamples to obtain t-statistics in order to assess the path coefficients’ significance (Hair et al., 2011). The bootstrapping standard error is similar to the standard deviation (Ringle et al., 2005). First, we analysed the effect of reward systems and the use of MAS information on managerial performance (H1, H2, and H3, see Table 4 and Figure 2). Second, we tested the mediating role of MAS information based on procedures recommended by Cohen and Cohen (1983). See Table 5. Finally, to compare the differences among three groups of enterprises, we did PLS-based multi-group analysis as suggested by Henseler (2012). This is a non-parametric approach and there is no requirement for any distributional assumption (Figure 4). Following Tabachnick and Fidell (2007), the statistical significance of the difference between path coefficients (β values) was tested by converting them into z scores and calculating the observed value of z (zobs value). See Table 7 and Figure 5.
Table 4 and Figure 2 show that all hypothesised paths (H1, H2 and H3) were strongly supported, since all path coefficients (β = 0.26; 0.51, and 0.21 respectively) are positive and significant at the.01 level (one-tailed). They indicate that the reward system has significant positive effects on the use of MAS information and managerial performance. Moreover, there is a significantly positive relationship between the use of MAS information and managerial performance. The results are consistent with discussions in previous studies (van Veen-Dirks, 2010, Schulz et al., 2010, Sprinkle, 2003, Sprinkle, 2000). In the model, reward systems explained 7% (not significant) of the variance in the use of MAS information, while 36% of the variance in managerial performance was explained by reward systems and the use of MAS information.
Overall, the research model is an acceptable fit since the squared multiple correlation (R2) of our dependent variable (managerial performance) is reasonable (0.36). In the PLS path model, R2 values of 0.75, 0.50, and 0.25 are substantial, moderate, and weak, respectively. A “moderate” R2 may be acceptable if an endogenous latent variable is explained by a few (e.g., one or two) exogenous latent variables (Hair et al., 2011).
Insert Table 4 here In addition, Table 5 presents the results of testing the mediating effects of the use of MAS information on the relationship between reward systems and managerial performance, based on Cohen and Cohen (1983). First, we tested the significance of correlations among latent variables (column 2, Table 5). Then, we assessed the level of mediation in the full model (columns 3 to 6, Table 5). All correlations and paths are significant at the.01 level (one-tailed).
The results show that the use of MAS information partially mediated this relationship (H3 is supported), since (1) all latent variable correlations are significant; (2) all paths in the full model are significant; and (3) the β3 in the full model (0.21) was lower than the correlation between reward systems and managerial performance (0.36).2 According to Hair et al. (2010), 2 Mediating level is assessed as follows: (1) there is significant correlations among all three constructs; (2) paths from the independent variable to the mediator (β1) and from the mediator to the dependent variable (β2) are significant; (3a) if the path from the independent to the dependent variables (β3) remains significant and equal to the correlation between independent and dependent variables, then mediation is not supported; (3b) if β3 remains significant but less than the correlation between independent and dependent variables, then partial effect is
It can be seen from Table 6 that the use of MAS information had moderate effects on managerial performance with the f2 of 0.37, while the effect of reward systems on managerial performance was small with the f2 of 0.06.
Insert Table 6 here The influence of ownership type on the research model was examined. The Kruskal-Wallis test in SPSS was used to compare the mean scores of the three groups of enterprises, SOEs, POEs, and FOEs (these calculations are not presented in this paper, but are available from the authors).
Figure 3 presents the comparison of mean values of the constructs among the groups. There were statistically significant differences among groups of enterprises in two constructs: reward systems and managerial performance. Department managers’ perceptions of the link between reward systems and performance in POEs were higher than those in SOEs and FOEs (significant differences at.05 level, one-tailed); and department managers’ perception of their performance in SOEs was lower than that of their counterparts in POEs and FOEs. H4a was supported.
To test the effects of ownership type on hypothesised relationships, a PLS-based multi-group analysis approach suggested by Henseler (2012) was employed. First, the PLS algorithm was run for the full model with all data to generate latent variable scores (Lvs) for subsequent analysis (Hair et al., 2011). Second, the data with Lvs were divided into three subgroups of SOEs (n=38), POEs (n=84), and FOEs (n=60). Finally, PLS models were run separately for each group. Table 7, Figure 4 and Figure 5 present the results of the analysis with significant and positive relationships among latent variables except for the relationship between reward system and managerial performance in SOEs. It was expected that the hypothesized relationships in POEs and FOEs would be significant and positive, while those relationships would not be significant in SOEs. The results in Table 7 and Figure 4 suggest that in SOEs, the relationship between reward system and managerial performance was not significant; moreover, the relationship between reward systems and the use of MAS information was not significant at the 0.1 and.05 level. Therefore, H4b was supported.
Insert Figure 3 here Insert Figure 4 here Insert Table 7 here evidenced; and (3c) if the path from the independent variable to the dependent variable is not significant, then full mediation is supported Cohen, J & Cohen, P 1983. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences, Lawrence Erlbaum Associates, Mahwah, NJ, Hair, JF, Anderson, RE, Tatham, RL & Black, WC 2010. Multivariate Data Analysis, Pearson Education, Inc., Upper Saddle River. New Jersey 07485..