Page 138 - Hojnik, Jana. 2017. In Persuit of Eco-innovation. Drivers and Consequences of Eco-innovation at Firm Level. Koper: University of Primorska Press
P. 138
In Pursuit of Eco-innovation
item measures. The size of factor loadings is one important consideration.
In the case of high convergent validity, high loadings on a factor would
indicate that they converge on some common point. At a minimum, all
factor loadings should be statistically significant, and the standardized
loading estimates should be 0.50 or higher, and ideally 0.70 or higher.
The rationale behind this rule can be understood in the context of an
item’s communality; the square of standardized factor loading represents
how much variation in an item is explained by the latent factor (mean-
ing that a loading of 0.71 squared equals 0.50; the factor explains half the
variation in the item, with the other half being error variance). The sec-
ond indicator of convergence is variance extracted; with CFA, the aver-
age percentage of variance extracted among a set of construct items is a
summary indicator of convergence. A value of variance extracted of 0.50
138 or higher is a good indicator of adequate convergence. Moreover, reliabil-
ity is also an indicator of convergent validity; coefficient alpha remains a
commonly applied estimate, although it may understate reliability. The
rule of thumb for either reliability estimate is that 0.70 or higher suggests
good reliability, while reliability between 0.60 and 0.70 may be accept-
able. High construct reliability indicates that internal consistency exists,
meaning that the measures all consistently represent the same latent con-
struct (construct reliability should be 0.70 or higher to indicate adequate
convergence or internal consistency) (Hair et al. 2009).
The eco-innovation construct in our study is composed of three di-
mensions – product, process and organizational eco-innovation – and is
thus a second-order latent factor. Items were grouped together in the ex-
pected grouping by dimension. Poorly fitting items – those that had low
communalities, or had low correlations with other items pertaining to
the same dimension or loaded onto two dimensions – have been exclud-
ed. The convergence and divergence of dimensions were checked by as-
sessing the fit of confirmatory models and inter-dimension correlations.
Furthermore, the contributions of the eco-innovation dimensions-only
model versus contributions of the overall factor-only model were exam-
ined by comparing nested models (dimensions-only and one factor-on-
ly) with an overall model that included both dimension factors and the
overall eco-innovation factor, by employing confirmatory factor analy-
sis. These contributions were analyzed using a test of significant improve-
ments in the model fit (the NFI for the two model differences, computed
with a formula from Bentler 1990).
For testing the proposed hypotheses, we used structural equation
modeling (SEM). The typical application of SEM is to a system of rela-
item measures. The size of factor loadings is one important consideration.
In the case of high convergent validity, high loadings on a factor would
indicate that they converge on some common point. At a minimum, all
factor loadings should be statistically significant, and the standardized
loading estimates should be 0.50 or higher, and ideally 0.70 or higher.
The rationale behind this rule can be understood in the context of an
item’s communality; the square of standardized factor loading represents
how much variation in an item is explained by the latent factor (mean-
ing that a loading of 0.71 squared equals 0.50; the factor explains half the
variation in the item, with the other half being error variance). The sec-
ond indicator of convergence is variance extracted; with CFA, the aver-
age percentage of variance extracted among a set of construct items is a
summary indicator of convergence. A value of variance extracted of 0.50
138 or higher is a good indicator of adequate convergence. Moreover, reliabil-
ity is also an indicator of convergent validity; coefficient alpha remains a
commonly applied estimate, although it may understate reliability. The
rule of thumb for either reliability estimate is that 0.70 or higher suggests
good reliability, while reliability between 0.60 and 0.70 may be accept-
able. High construct reliability indicates that internal consistency exists,
meaning that the measures all consistently represent the same latent con-
struct (construct reliability should be 0.70 or higher to indicate adequate
convergence or internal consistency) (Hair et al. 2009).
The eco-innovation construct in our study is composed of three di-
mensions – product, process and organizational eco-innovation – and is
thus a second-order latent factor. Items were grouped together in the ex-
pected grouping by dimension. Poorly fitting items – those that had low
communalities, or had low correlations with other items pertaining to
the same dimension or loaded onto two dimensions – have been exclud-
ed. The convergence and divergence of dimensions were checked by as-
sessing the fit of confirmatory models and inter-dimension correlations.
Furthermore, the contributions of the eco-innovation dimensions-only
model versus contributions of the overall factor-only model were exam-
ined by comparing nested models (dimensions-only and one factor-on-
ly) with an overall model that included both dimension factors and the
overall eco-innovation factor, by employing confirmatory factor analy-
sis. These contributions were analyzed using a test of significant improve-
ments in the model fit (the NFI for the two model differences, computed
with a formula from Bentler 1990).
For testing the proposed hypotheses, we used structural equation
modeling (SEM). The typical application of SEM is to a system of rela-
