Chapter 4 Research Instrument Development and Content Validity
4.2 Stage 2 Empirical Scale Refinement and Validation
4.2.4 The background of data collection
The present study is based on the Taiwanese IT industry, and mainly consists of elec-tronics manufacturers and semiconductor-related manufacturers (see Table 4.4), for two rea-sons:
1. Taiwan is a major player in and contributor to the world IT industry
Taiwan has achieved outstanding results in IT over the past two decades (Chang and Yu, 2001). Taiwan-made IT products dominate the world market in many categories. The world market share exceeds 50 %. The semiconductor manufacturing and electronics manufacturing industry (IT industry) especially have evolved to prominence in Taiwan’s recent economic development. The country currently ranks third in computer manufacturing and fourth in the semiconductor industry (Foundry ranked No.1; IC design ranked No.2) in the world (ITIS Project, MOEA, 2008). Keeping pace with thaw in the political relations between Taiwan and China, Taiwan’s IT industries have been playing a key role in that country — which, of course, has earned a reputation as “the world's workshop.” For instance, Taiwan’s leading notebook computer manufacturers — who have, at times, enjoyed a world market share in excess of 70% — began moving their production sites to China in 2001. All of Taiwan’s notebook pro-duction lines have now been relocated to China. Most of China’s share of the global notebook market can thus actually be attributed to the contributions by Taiwanese firms. In addition, the 2008 global IT industry competitiveness report issued by Britain's Economist Intelligence Unit (EIU) ranked Taiwan sixth out of the 64 countries rated in terms of IT industry competi-tiveness, behind only the U.S., Japan, South Korea, Britain, and Australia. As for IT industry labor productivity, Taiwan leads the world, with output value of US$386,413 per IT industry employee. The industry’s structure is the predominant reason for this high productivity (EIU, 2008).
57
2. Taiwan has delivered the best practices of SCM
In the IT industry, product life cycle is extremely short. Companies need to deliver new products before they have any market value. In the Taiwanese IT industry, the main type of business is original equipment manufacturing (OEM) and original design manufacturing (ODM). An OEM/ODM business is different from an own brand manufacturing (OBM) busi-ness in many respects. With OBM, companies can entirely control their marketing activities.
In the case of OEM/ODM, on the other hand, firms are not involved in their OEM/ODM cus-tomers’ sales/marketing activities. Companies isolated from the end-customer base still need to satisfy customer needs and react to new ones immediately. They are compelled to closely cooperate with all of the members in the supply chain so as to be able to react to unexpected changes. To cope with the rapid changes in customer needs and the extremely short product life cycles, the cross-functional cooperation of information systems in the IT industry may be more important than in those industries with a longer product life cycle. In today's fast-changing business environment, the Taiwanese IT industry depends heavily on its high-ly-effective SCM to achieve superb performance. It is important to understand the supply chain network in Taiwan, since that supply chain network may influence organizational effec-tiveness. For instance, Foxconn, which is a contractor for such world-famous products as the iPod and iPhone, relies on the support of its ERP system (SAP) to perform varied, high-quality, low-cost production tasks. Foxconn is also the major supplier to leading brand name companies such as Cisco, Dell, HP, Nokia, Sony, etc. (Foxconn, 2008).
Many scholars have conducted research into the SCM of firms in some developed countries (Benton and Maloni, 2005; Mabert, et al., 2001; Lim and Palvia, 2001). These stu-dies cover many types of industries, such as the chemical, pharmaceutical, bioengineering, automobile, etc. They also include a wide range of high-technology firms. The IT industry in developing countries, such as Taiwan, China, and Korea, has not, however, been comprehen-sively studied. The present study therefore presents the results of an empirical study of the
58
impact of implementation of ERP on SCM competencies by IT manufacturers in Taiwan.
Survey data is collected from a sample of Taiwanese IT companies listed in the Taiwan Stock Exchanges (TSE), mainly on electronics manufacturers (including: PC systems, peripherals, communications, consumer electronics, and computer components) and semiconduc-tors-related manufacturers (including: foundry, IC design, packaging and testing, mask, and equipment/material provider), and screened according to whether they have operational ERP systems (see Table 2). Refined scales employing items drawn from constructs and measure-ment items referred to in the relevant literature are used to conduct empirical, confirmatory analyses. Each item’s scale has measurement properties that fit into the commonly accepted guidelines for reliability and validity.
59
CHAPTER 5
ANALYSIS AND RESULTS
Once the data was collected, it was analyzed with the following objectives in mind: pu-rification, factor structure (initial validity), unidimensionality, and reliability. The methods that were used for each analysis are corrected-item total correlation (for purification), explo-ratory factor analysis (for factor structure and initial validity), structural equation modeling (for unidimensionality), Cronbach’s alpha (for reliability).
5.1 The Measurement Model
The need to purify the items before administering factor analysis is emphasized by Churchill (1979). Purification is carried out by examining the corrected-item total correlation (CITC) score of each item with respect to a specific dimension of a construct. The CITC score of each item with respect to a specific dimension of a construct. The CITC score is a good in-dicator of how well each item contributes to the internal consistency of a particular construct as measured by the Cronbach’s alpha coefficient (Cronbach, 1951). Items were deleted if their CITC scores were below .30 (Dunn et al., 1994), unless there are clear reasons for keeping the items in spite of low item total correlation (see Table 5.2). On the other hand, certain items with CITC scores above .30 may also be removed if their deletion can dramatically improve the overall reliability of the specific dimension. This can be determined by examining the
“alpha if deleted” score.
The reliability (internal consistency) of the items comprising each dimension was ex-amined using Cronbach’s alpha. Following the guideline established by Nunnally (1978), an alpha score of higher than .70 is generally considered to be acceptable.
60
After purifying the items, an exploratory factor analysis (EFA) was first conducted to check whether the proposed factor structures are indeed consistent with the actual data. The factor structures suggested by the EFA match the one proposed in the research model. The various loadings are shown in Table 5.1. Second, multiple regression also was conducted to verify the impacts of ERP benefits on SCM competencies (Table 5.5). The measurement model was estimated using SPSS 14.0. The properties of the measurement model are summa-rized in Table 5.2. Third, confirmatory factor analysis (CFA) was conducted to assess the measurement model; then, the structural relationships were examined. In this measurement model, no unidirectional path was specified between any latent variables. Instead, a cova-riance was estimated to connect each latent variable with every other latent variable. This measurement model was estimated using AMOS 7.0.
EFA is useful in discovering potential latent sources of variance and covariance in ob-served measurements. Items with good measurement properties should exhibit high factor loadings on the latent factor of which they are indicators, and small factor loadings on the factors that are measured by differing sets of indicators. Therefore, such results provide some evidence of initial validity of measurement items (Segars and Grover, 1998). To ensure the high quality of instrument development process, .50 was used as the cutoff score for factor loading. Items with loadings lower that .50 and items with serious cross-loadings (i.e. an item loaded very close to .50 on more than one factor) were removed. To streamline the final re-sults, factor loadings lower than .40 were not reported (Table 5.1). Moreover, the stability of the factors was analyzed by measuring the ratio of respondents to items, and the Tinsley and Tinsley (1987) guideline of having a minimal ratio between 5 and 10 was followed.
Even though EFA is useful at identifying underlying factor structure and thus providing initial unidimensionality (convergent validity) and discriminant validity, EFA initially as-sumes that the measurement errors of the items are uncorrelated. In practice, however, there is always some degree of error correlations among items and this cannot be detected by EFA
61
(Raghunathan et al., 1999). On the other hand, according to Gerbing and Anderson (1988) and Segars and Grover (1998), EFA does not provide an explicit test of unidimensionality. Unidi-mensionality can be defined as the existence of one latent trait or construct underlying a set of measures (McDonald, 2000; Hattie, 1985). In fact, Gerbing and Anderson stat that “factors in an exploratory analysis do not correspond directly to the constructs represented by each set of indicators because each factor from exploratory analysis is defined as a weight sum of all ob-served variables in the analysis” (p.189). More recently, the structural equation modeling (SEM) has been gaining increasing popularity due to its robustness and flexibility in estab-lishing unidimensionality. The research will thus use SEM to test unidimensionality of each construct.
One of the most widely used SEM software is AMOS. Using AMOS, it is possible to specify, test, and modify the measurement model. Model-data fit was evaluated based on mul-tiple fit indexes. The Chi-square is perhaps the most popular index to evaluate the goodness of fit of the model. It measures the difference between the sample covariance and the fitted co-variance. However, the Chi-square index is sensitive to sample size and departures from mul-tivariate normality. Therefore, it has been suggested that it must be interpreted with caution in most applications (Joreskog and Sorbom, 1989). Some of the other measures of overall model fit are goodness of fit index (GFI), adjusted goodness of fit index (AGFI), comparative fit in-dex (CFI), normed-fit inin-dex (NFI), and Root Mean Squared Error of Approximation (RMSEA) indicates the average discrepancy between the elements in the sample covariance matrix and the model-generated covariance matrix. RMSEA values range from 0 to 1, with smaller values indicating better model; values below .08 signify good fit (Browne and Cudeck, 1992).
62 Table 5.1: Exploratory factor analysis loading
63
Table 5.2: Summary of the measurement model
Construct
Indicator Mean Std. Dev. Principal com- ponents scores Item-to total correlation Standard Load- ing Cronbach alpha Composite Re- liability Average Variance Extracted esti- mates
Operational
64
65
5.2 Instrument Reliability and Validity
To validate our measurement model, content validity, construct validity (including Cronbach alpha), convergent validity, and discriminant validity were assessed. Content valid-ity was established by ensuring consistency between the measurement items and the extant literature. This was done by interviewing senior practitioners and pilot-testing the instrument.
For the construct validity, the items were tested for scale reliability. Various reliability test re-sults are shown in Table 5.2, which summarizes the item-to-total correlations and principal component scores for the sample. Item-to-total correlations exceed 0.30 (Dunn et al., 1994) in all cases. The principal component scores meet minimal levels of 0.30 and above in all cases (Hair et al., 1998). Thus, all of the scales reflect unidimensional characteristics. The reliability (internal consistency) of the items comprising each dimension was examined using Cron-bach’s alpha. Following the guideline established by Nunally (1978), an alpha score of higher than .70 is generally considered to be acceptable. The Cronbach alpha ranges from .831 to .966 for the eight constructs, and are thus also satisfactory, as coefficient alphas meet or exceed 0.70 in all instances (Nunnally, 1978), indicating a high internal consistency. Except for one item in the behavioural process integration construct of SCM competencies, all the items were retained. The construct validity is also tested for convergent and discriminant va-lidity. We assessed convergent validity by reviewing the t tests for the factor loadings and by examining composite reliability and average variance extracted from the measures (Hair et al., 1998). Although many studies have used 0.5 as the threshold reliability of the measures, 0.7 is a recommended value for a reliable construct (Chin, 1998). As shown in Table 5.2, our com-posite reliability values range from 0.925 to 0.992. For the average variance extracted by a measure, a score of 0.5 indicates acceptability (Fornell and Larcker, 1981). The average va-riances extracted by our measures range from 0.500 to 0.822, which are above or equal to the acceptability value. In addition, Table 5.3 exhibits the loadings of the measures in our
re-66
search model. As expected, all measures are significant on their path loadings at the level of 0.01. Finally, we verified the discriminant validity of our instrument by comparing the aver-age variance extracted (AVE) (Fornell and Larcker, 1981) from each latent construct to the square of the correlation between this construct and every other construct, which has been used by some IS studies (Segars and Grover, 1998). The result, in Table 5.4, confirms the dis-criminant validity: the square of the average variance extracted for each construct is greater than the levels of correlations involving the construct. The results of the inter-construct corre-lations also show that each construct shares larger variance with its own measures than with other measures.
67 Table 5.3 Loadings of the measures
Construct Items Standard Loading Standard Error t-value Construct Items Standard Loading Standard Error t-value
EOP2 0.903 0.047 21.048 EOG2 0.988 0.065 19.483 EOP3 0.944 0.048 23.065 EOG3 0.983 0.065 19.353 EOP4 0.964 0.049 24.123 EOG4 0.893 0.065 17.027 EOP5 0.845 0.054 18.636 EOG5 0.801 0.072 14.848
EOP6 0.848 EOG6 0.763
Man-agement
EMNG1 0.893 0.065 16.937 Opera-tional
ESTG2 0.628 0.085 14.477 SPCP2 0.748 0.069 13.291 ESTG3 0.922 0.077 10.172 SPCP3 0.749 0.080 13.312 ESTG4 0.826 0.077 14.702 SPCP4 0.782 0.080 14.034 ESTG5 0.783 0.084 13.266 SPCP5 0.807 0.079 14.572 ESTG6 0.700 0.079 12.609 SPCP6 0.810 0.072 14.633
IT
EIT2 0.902 0.052 21.371 SBP2 0.678 0.089 10.563 EIT3 0.813 0.056 17.578 SBP3 0.639 0.089 9.973 EIT4 0.797 0.059 17.002 SBP4 0.729 0.090 11.331 EIT5 0.702 0.081 13.922 SBP5 0.734 0.095 11.407 EIT6 0.870 0.058 19.902
68
Table 5.4 Comparison of AVE and squared roots correlations
Var. EOP EMNG ESTG EIT EOG SOP SPCP SBP
EOP 0.801
EMNG 0.215 0.901
ESTG 0.604 0.390 0.808
EIT 0.524 0.457 0.610 0.880
EOG 0.153 0.520 0.355 0.546 0.828
SOP 0.532 0.235 0.565 0.503 0.324 0.705
SPCP 0.750 0.266 0.692 0.662 0.314 0.526 0.779
SBP 0.547 0.381 0.647 0.565 0.367 0.442 0.561 0.820
* The shaded numbers in the diagonal row are square roots of the average variance ex-tracted.
69
5.3 Results of Multiple Regression Analysis
The first analysis focuses on comparing the impacts of ERP benefits in each region to the SCM competency measure. The five constructs, each composed of six items of ERP bene-fits, are used as independent variables in a series of regression models, with each SCM com-petency individually treated as a dependent variable. To provide a clearer picture of the role of the ERP benefits in affecting SCM competency, Table 5.5 reports the results of all regression models. Standardized beta coefficients are shown in the table for all significant (p < .05) va-riables. Model significance is also reported in the R²column. All R²values are significant at p
< .001.
In the model, R²values show that this model accounts for 45% of the variance in the operational process, 62% of the variance in the planning and control process, and 30% of the variance in the behavioral process. That is, 30% or more of the variation in three constructs of SCM competencies is explained by most of the ERP benefits. Three of five constructs of ERP benefits explained most of 17 measures of SCM competencies. There are operational, mana-gerial, and strategic benefits of ERP. IT infrastructure benefits explained some SCM compe-tencies; but organizational benefits could not predict any of them. That is, most of the results support the corresponding hypotheses. Strategic and managerial benefits are the most com-mon significant predictor variables relative to SCM competencies. Managerial benefit is a significant predictor for the operational process of SCM competencies (Beta coefficient is .32).
Strategic benefit is a significant predictor for the planning and control process and the beha-vioural process (Beta coefficients are .45 and .30.) These findings will be compared with the results of the structural model, and will be discussed below. A summary of the hypotheses test results is provided in Table 5.7.
70
Table 5.5 Multiple regression results
Standardized coefficients Beta
SCM competences ERP Benefits
Items Competences
Operational Management Strategic IT Infrastruc- ture Organizational R2
SOP 0.20 0.32 0.21 0.11 0.45
SOP1 Relevancy 0.23 0.30 0.13 0.31
SOP2 Responsiveness 0.19 0.33 0.17 0.33
SOP3 Cross-functional unification 0.20 0.24 0.18 0.34
SOP4 Standardization 0.21 0.25 0.19 0.29
SOP5 Operational fusion 0.25 0.28 0.13 0.31
SOP6 Supplier management . 0.34 0.17 0.36
SPCP 0.24 0.24 0.45 0.62
PCP1 Information management 0.13 0.26 0.41 0.47
SPCP2 Internal communication 0.22 0.50 0.42
SPCP3 Connectivity 0.19 0.20 0.38 0.42
SPCP4 Collaborative forecasting and planning
0.21 0.18 0.37 0.40
SPCP5 Functional assessment 0.30 0.16 0.31 0.41
SPCP6 Activity-based and total cost methodology
0.26 0.22 0.28 0.41
SBP 0.26 0.30 0.15 0.30
SBP1 Role specificity 0.17 0.21 0.22 0.25
SBP2 Guidelines 0.18 0.19 0.19 0.17
SBP3 Information sharing 0.15 0.22 0.12 0.13
SBP4 Gain/risk sharing 0.20 0.25 0.13 0.19
SBP5 Strategic alignment 0.15 0.19 0.17 0.18
71
5.4 Results of the Structural Model Analysis
The structural model tested in the present study is shown in Figure 2. This model was estimated using AMOS 7.0. The χ statistic of 2.073 is within the acceptable limit (Byrne, 2 1989). Several goodness of fit indices of the measurement model have been widely used in IS research and are presented in Table 5.6. The Tucker-Lewis index, also known as the non-normed fit index (NNFI), and the comparative fit index (CFI) are all above 0.90, gesting a good fit between the structural model and the data. RMSEA is well below the sug-gested threshold value of 0.08 (Browne and Cudeck, 1992). The parsimony-adjusted NFI of the revised model is 0.848, which is significantly above the suggested value of 0.60. Williams and Hazer (1986) indicate highly acceptable levels of parsimony and fit of the overall model.
All of these fit indices are acceptable, suggesting that the overall structural model provides a good fit with the data. The results of estimating the structural model are presented in Figure 5.1.
72
Table 5.6 Fit indices of structural model
χ 2 2091
df 1009
χ /df 2 2.073
Normed fit index (NFI) 0.848
Tucker-Lewis index 0.909
Comparative fit index (CFI) 0.915
GFI 0.756
RMR 0.024
RMSEA 0.061
Lower bound 0.058
Upper bound 0.065
The squared multiple correlation (SMC) values, which are similar to R2 in regression analysis, show that this model accounts for 50% of the variance in the operational process, 70% of the variance in the planning and control process, and 41% of the variance in the beha-vioural process. Most of the paths are significant and positive, supporting the corresponding hypotheses, except for the organizational process and IT infrastructure. These findings are discussed below. A summary of the hypotheses test results is provided in Table 5.7.
Figure 5.1 and Table 5.7 show the results, and illustrate that the SCM competencies in the operational process, the planning and control process and the behavioural process were positively influenced by ERP benefits. These results basically support all of our hypotheses.
73
74
Table 5.7 Summary of the Structural Model
Hypotheses Beta coeffic- ients
Path Coeffi-cients
Results
Ha1: The operational benefits of ERP positively affect SCM competences in operational process.
0.20 0.12 Supported
Hb1: The managerial benefits of ERP positively affect SCM competences in operational process.
0.32 0.34 Supported
Hc1: The strategic benefits of ERP positively affect SCM competences in operational process.
0.21 0.21 Supported
Hd1: The IT infrastructure benefits of ERP positively af-fect SCM competences in operational process.
0.11 0.07 Not sup-ported He1: The organizational benefits of ERP positively affect
SCM competences in operational process.
0.08 Not
sup-ported Ha2: The operational benefits of ERP positively affect
SCM competences in planning and control process.
0.24 0.25 Supported
Hb2: The managerial benefits of ERP positively affect SCM competences in planning and control process.
0.24 0.28 Supported
Hc2: The strategic benefits of ERP positively affect SCM competences in planning and control process.
0.45 0.48 Supported
Hd2: The IT infrastructure benefits of ERP positively af-fect SCM competences in planning and control process.
0.04 Not
sup-ported
He2: The organizational benefits of ERP positively affect SCM competences in planning and control process.
-0.09 Not
sup-ported Ha3: The operational benefits of ERP positively affect
SCM competences in behavioral process.
0.11 Not
sup-ported Hb3: The managerial benefits of ERP positively affect
SCM competences in behavioral process.
0.26 0.19 Supported
Hc3: The strategic benefits of ERP positively affect SCM competences in behavioral process.
0.30 0.26 Supported
Hd3: The IT infrastructure benefits of ERP positively af-fect SCM competences in behavioral process.
0.12 Not
sup-ported He3: The organizational benefits of ERP positively affect
SCM competences in behavioral process.
-0.07 Not
sup-ported
75
CHAPTER 6
DISCUSSION OF STRUCTURAL EQUATION MODEL AND HYPOTHESES TESTING RESULTS
The main objective of this study is to investigate the relationship between ERP benefits and SCM competencies. The findings show how ERP benefits impact on SCM competencies.
Table 5.5 summarizes the results of multiple regression, and Figure 5.1 shows the results of
Table 5.5 summarizes the results of multiple regression, and Figure 5.1 shows the results of