Page 147 - Ebook HTKH 2024

Additionally, there were twelve MNEs with ninety-one international
manufacturing subsidiaries in the sample. Subsequently, the dominant subsidiaries

were again operated in the automotive industry (approximately 85).
The parent MNEs are from top developed countries; the United States is
reasonably contributed to the largest part, with nearly 38 per cent due to the most
significant economy (Dunning, 2008). Besides, the MNE’s international subsidiaries
cover 36 countries with China’s major component (13.5 per cent). This can be explained
by the fact that China is now the most attractive emerging market that gained significant
development in recent years and became the second-largest economy, only after the
United States (Dunning, and Fujita, 2007). China also draws the attention of MNEs to

build manufacturing plants due to the availability of low-priced resources (Dunning,
2008).
4. Results
4.1. Measurement model
A confirmatory factor analysis model was conducted, simultaneously several
model diagnostics such as path estimates (indicators with factor loadings lower than 0.6
were deleted), standardized residuals, and modification indices (freeing the
corresponding error covariance parameters between the error terms for pairs of

measured variables) were used to improve the fitness level of the structural
infrastructure measurement model (Anderson & Gerbing 1988; Arbuckle 2005; Hair et
al. 2006). The final result of CFA revealed that the model fitted the data well. All the
fitness index (c2 = 164.354; DF = 18; p < 0.001; CFI =0.964, NFI = 0.96, GFI = 0.96,
TLI= 0.928, and RMSEA = 0.096) satisfied the good fit thresholds recommended
by Hair et al. (2005) and Hooper et al. (2008). The goodness of fit index CFI, NFI, TLI
were higher than there commended the satisfactory level of 0.9 whereas the root mean

square error of approximation was lower than 0.1.
Additionally, each construct had high composite reliability (ranging from 0.825 to
0.933), exceeding the usual 0.7 benchmarks (Hair et al., 2005). Convergent validity was
considered satisfactory as the standardized loading for each item and the AVE both
exceeded the 0.5 thresholds recommended by Hair et al. (2005). The internal
consistency of the multi-item scales was also judged to be satisfactory as the

Cronbach’s α coefficients each exceeded the 0.7 cut-offs recommended by Santos
(1999), showing that common bias was not a problem. Discriminant validity was also
evident as the squared correlation among the constructs was less than their individual
AVE (Fornell and Larcker, 1981, Hair et al., 2014).
4.2. Hypotheses testing results
Based on the CFA result, the structural model was subsequently specified using
the maximum likelihood estimation method to test our hypotheses. We simultaneously
conducted three models in which MNE subsidiary performance was in turn measured

by ROE, ROC, and a latent variable combined of ROE and ROC. All three models gave
the same conclusion for hypothesis testing; meanwhile, the model with MNE subsidiary
performance measured by the latent variable (combined of ROE and ROC)

139