Page 142 - Hojnik, Jana. 2017. In Persuit of Eco-innovation. Drivers and Consequences of Eco-innovation at Firm Level. Koper: University of Primorska Press
P. 142
In Pursuit of Eco-innovation
(Klem 2000); a model can be either under-identified (this happens when
a structural model has a negative number of degrees of freedom, mean-
ing that we aim to estimate more parameters than is possible with the
input matrix) or over-identified (this happens when a structural mod-
el has a positive number of degrees of freedom and thus indicates that
some level generalizability may be possible). The researchers’ objective is
always steered towards achievement of maximum model fit, with the larg-
est number of degrees of freedom (Ruzzier 2005). The statistical reasona-
bleness of the parameters concerns the second major statistical criterion.
A model with negative variances and correlations greater than one is mis-
specified and can further result in improper results (Klem 2000).
When it comes to determining the adequacy of a structural equation
model, various measures of model fit are available. The two most popular
142 ways of evaluating model fit are those that involve the chi-square good-
ness-of-fit statistic (χ2 test) and the so-called fit indexes that have been
offered in order to supplement the χ2 test (Hu and Bentler 1995). The χ2
test enjoyed substantial popularity at first, while the problems associat-
ed with the goodness-of-fit χ2 tests were recognized quite early. One of
the concerns has centered on the sample size issue. The statistical theo-
ry for T is asymptotic; that is, it holds as sample size gets arbitrarily large.
Therefore, T may not be χ2 distributed in a small sample; therefore, it may
not be correct for model evaluation in practical situations. Furthermore,
T may not be χ2 distributed when the typical underlying assumption of
multivariate normality is violated. Therefore, the standard χ2 test may not
be a sufficient guide to model adequacy, because a significant goodness-
of-fit χ2 value may be a reflection of model misspecification, the power of
the test, or a violation of some technical assumption underlying the esti-
mation method (Hu and Betler 1995).
When an SEM model that looks theoretically sensible is identified
and there are no signs of statistically improper estimates, we check wheth-
er the data fit the model using various goodness-of-fit measures (Ruzzi-
er 2005). With the measurement model specified, sufficient data collect-
ed, and key decisions such as the estimation technique already made, the
researcher comes to the most fundamental question in SEM testing: “Is
the measurement model valid?” Measurement model validity depends on
goodness-of-fit for the measurement model and specific evidence of con-
struct validity (Hair et al. 2009).
Among the fit indexes, we can distinguish three types of fit meas-
ures: 1) absolute fit measures, 2) incremental fit measures and 3) parsi-
monious fit measures. An absolute fit index directly assesses how well an
(Klem 2000); a model can be either under-identified (this happens when
a structural model has a negative number of degrees of freedom, mean-
ing that we aim to estimate more parameters than is possible with the
input matrix) or over-identified (this happens when a structural mod-
el has a positive number of degrees of freedom and thus indicates that
some level generalizability may be possible). The researchers’ objective is
always steered towards achievement of maximum model fit, with the larg-
est number of degrees of freedom (Ruzzier 2005). The statistical reasona-
bleness of the parameters concerns the second major statistical criterion.
A model with negative variances and correlations greater than one is mis-
specified and can further result in improper results (Klem 2000).
When it comes to determining the adequacy of a structural equation
model, various measures of model fit are available. The two most popular
142 ways of evaluating model fit are those that involve the chi-square good-
ness-of-fit statistic (χ2 test) and the so-called fit indexes that have been
offered in order to supplement the χ2 test (Hu and Bentler 1995). The χ2
test enjoyed substantial popularity at first, while the problems associat-
ed with the goodness-of-fit χ2 tests were recognized quite early. One of
the concerns has centered on the sample size issue. The statistical theo-
ry for T is asymptotic; that is, it holds as sample size gets arbitrarily large.
Therefore, T may not be χ2 distributed in a small sample; therefore, it may
not be correct for model evaluation in practical situations. Furthermore,
T may not be χ2 distributed when the typical underlying assumption of
multivariate normality is violated. Therefore, the standard χ2 test may not
be a sufficient guide to model adequacy, because a significant goodness-
of-fit χ2 value may be a reflection of model misspecification, the power of
the test, or a violation of some technical assumption underlying the esti-
mation method (Hu and Betler 1995).
When an SEM model that looks theoretically sensible is identified
and there are no signs of statistically improper estimates, we check wheth-
er the data fit the model using various goodness-of-fit measures (Ruzzi-
er 2005). With the measurement model specified, sufficient data collect-
ed, and key decisions such as the estimation technique already made, the
researcher comes to the most fundamental question in SEM testing: “Is
the measurement model valid?” Measurement model validity depends on
goodness-of-fit for the measurement model and specific evidence of con-
struct validity (Hair et al. 2009).
Among the fit indexes, we can distinguish three types of fit meas-
ures: 1) absolute fit measures, 2) incremental fit measures and 3) parsi-
monious fit measures. An absolute fit index directly assesses how well an
