Capabilities

Supply chain agility is referred to as a type of operational capability (Liu et al., 2013) that reflects a high-level routine or a collection of routines that are used to respond to market changes. The dynamic capability that is determined to be a provider in this study is distinguished from operational capability and is regarded as a higher-level concept that is used to adapt operational routines and capabilities to develop new value-creating strategies (Liu et al., 2013). In this study, agility is viewed as a capability that partner enterprises can receive from the successful formation of virtual enterprises for exploiting fast changing opportunities in the market and sustaining competitive advantage. Furthermore, supply chain agility affects a number of significant enterprise business performance variables (Yusuf et al., 2012; Roberts and Grover, 2012). However, Ngai et al. (2011), Liu et al. (2013) and DeGroote and Marx (2013) tested relation between agility and business performance, no evidence is found on how business performance has been impacted by a strategy of joining in virtual enterprise for providing agility in the supply chain. Therefore, we offer the following hypothesis:

H5. Supply chain agility positively influences business performance.

73 Analysis technique

To analyse multiple variables simultaneously, multivariate analysis had been selected, especially the structural equation model (SEM) as a proper method for this study. The structural equation model has the following advantages:

 The structural equation model is powerful technique that can estimate for a series of separate multiple regression equations simultaneously. structural equation model surpasses traditional regression models by including multiple independent and dependent variables to test associated hypotheses about relationships among observed and latent variables (Carvalho and Felix O. Chima, 2014);

 This method can conduct analysis together in the same time that performed separately. For instance, there are many following analyses included in the structural equation model:

causal modelling or path analysis; confirmatory factor analysis; second order factor analysis; covariance structure models; and correlation structure models. structural equation model also enables statistical analysts to handle difficult data including; time series with auto-correlated error, non-normal data and even incomplete data (Alavifar et al., 2012);

 It enables researchers in measurement of direct and indirect effects and performing test models with multiple dependent and independent variables.

Recently, the structural equation model is increasingly seen as a useful quantitative technique for specifying, estimating, and testing hypothesised models describing relationships among a set of meaningful variables (Hair et al., 2010). Therefore, to find the causal relationship among the determining factors shown previously (see Figure 4-1) simultaneously in this study, the structural equation model technique was applied to test the hypotheses proposed in the conceptual model and to indicate the direct and indirect relationships between the proposed factors.

Since the structural equation model is commonly used for the study of virtual enterprise and agility, some empirical studies are compared in Table 4-1 by factors, the used fit indices and the software. However high-order factor analysis is rarely found in the literatures. Cronbach's alpha value, composite reliability, convergent and discriminant validity tests were executed for all the empirical studies.

74

Table 4-1: Papers that adopted the structural equation model

NB: See abbreviation from Figure 4-2

As shown in Figure 4-2, this research uses a methodology with three steps (Hair et al., 2010) as follows; firstly to identify underlying constructs, exploratory factor analysis eliminates variables with weak or negative correlations and then to measure the model fit, confirmatory factor analysis is conducted to specify how latent variables depend upon or are indicated by the observed variables and finally to test the hypotheses, the path analysis is performed for the developed structural model.

Empirical study

The proposed hypothetical conceptual model is shown in Figure 4-2 was tested by the empirical study based on the questionnaire. The survey was targeted at business companies who are responsible for planning, coordinating, control, realising and monitoring all internal and network-wide material and product flows, with the necessary information flow, in industrial and trading sectors along the complete value-added chain for conforming to customer requirements in Ulaanbaatar (the Mongolian capital). From the members’ list of the Mongolian Logistics Association, companies have a direct export and import with companies abroad for last three years¹⁰ were selected as a target group.

Since it is difficult to find physically integrated companies, enterprises with active web pages both in English and Mongolian were selected to narrow down the target group. Internet and web services have emerged as a serious technology to provide the middleware platform to support effectively the operations of a virtual enterprise (Rezgui, 2007) that enable alliances to be agile with quick response with which it can respond to changing market requirements (Gunasekaran and Yusuf, 2002). These criteria indicate companies that are interested in collaborating with potential partners.

10 Data collected from statistics of Mongolian Customs on the web page of http://www.customs.gov.mn

Authors Factors Fit indices^a Software

Cao and Dowlatshahi (2005) Two 1^st orders and one 2^nd order factor χ²/df, SRMR, RMSEA, GFI, AGFI, CFI LISREL

Swafford et al. (2008) Four 1^st order factors χ², RMSEA, SRMR, CFI, GFI COMPUSTAT

Braunscheidel and Suresh (2009) Five 2^nd order, one 1^st order factors χ² PLS-Graph 03.00 Liu et al. (2013) Five 1^st order factors χ², df, RMSEA, CFI, IFI, NFI, NNFI LISREL

DeGroote and Marx (2013) Three 1^st order factor χ²/df, CFI, RMSEA Mplus

75

Measurement model

Exploratory factor analysis (EFA) is conducted for factors reduction

 Principal component analysis with varimax rotation is used for factor extraction

 For modification, variables with not strong or negative correlations are eliminated then extract into suitable number of factors, again

 Reliability is indicated with Cronbach s alpha coefficient

 Statistical package SPSS^® (version 20.0 for Windows) is used

Confirmatory factor analysis (CFA) is conducted to test model fit

 Maximum likelihood method is used for calculation of covariance matrix

 Fit indices calculated to measure model fit with as set of observation

- Absolute fit indices measure how well the model is specified by the observed data.

χ²: df (chi square per degree of freedom) ratios on the order of 3:1 the standardised root mean square residual (SRMR) < 0.09 the root mean square error of approximation (RMSEA) < 0.08

- Incremental fit indices measure how well the estimated model fits relative to some alternative baseline model . the incremental fit index (IFI) > 0.9

the normed fit index (NFI) > 0.9

the Tucker-Lewis index, also known as the non-normed fit index (NNFI) > 0.9 the comparative index (CFI) > 0.9

 For modification of measurement model an additional causal relationship is established.

 Reliability (composite reliability) and validity (average variance extracted and average shared squared variance ) tests were executed to measure consistency and accuracy of measurement models

 The SPSS^® AMOS software (version 20.0 for Windows) is used

Structural model

 Standardised regression coefficient (path coefficient) measures the power of effect from causal variable to an endogenous variable. In statistical significance testing the p-value is used.

 T-value is used for checking the significance level of estimation for the structural model

 The squared multiple correlation (R²) for the regression equation indicates the proportion of variance in the dependent variable that is accounted for by the set of independent variables in the multiple regression equation.

Figure 4-2: Schema for structural equation modelling

Hard and soft copies of questionnaires were conveniently distributed to randomly selected companies. Participants were treated as autonomous agents by informing them about the study and allowing them to voluntarily choose to participate or not. The cover letter for survey explains the purpose of the research, provides concept definitions and the confidential guarantee of participation. Five draft questionnaires with the cover letters were submitted to a focus group of two academics and three practitioners, to check the readability and possible ambiguity of the questionnaire and four of them replied. The interviews were conducted with respondents and minor changes were made such as rewording some questions, removing several unnecessary items and simplifying the language.

In the first round 400 questionnaires were distributed and 179 responses were received. In the second round, another 100 were distributed and 54 were returned. Out of 233 responses, 205 were usable. The other 28 unusable responses did not contain sufficient data for further analysis. Although this response rate (41%) is not unusual it is recognised that 205 responses cannot cover the total business firms in the whole market. The non-response bias (Armstrong and Overton, 1977) was tested by comparing the chi-squares of overall assessments of key

76

factors of the responses from the single-mailing respondents, and the respondents of hard copy of questionnaire. No significant differences were found between these two groups and the result indicated there is not significant non-response bias exist in this study.