Page 87 - Ana Kozina and Nora Wiium, eds. ▪︎ Positive Youth Development in Contexts. Ljubljana: Educational Research Institute, 2021. Digital Library, Dissertationes (Scientific Monographs), 42.
P. 87
measuring positive youth development in slovenia
limited, and they were supervised by the school coordinator (teacher or
school counsellor) who answered any questions if they had them.
Data analysis
After examining descriptive statistics, correlations and reliabilities using
IBM SPSS Statistics 27, we considered the ESEM (Exploratory Structural
Equation Modelling) model of DA and the CFA model of the 5Cs using
Mplus (Version 8.6; Muthén & Muthén, 1998–2021) to examine the pro-
posed models’ structural validity. The full information maximum likeli-
hood (FIML) algorithm was used to handle missing data and assess pa-
rameters in the model. Separate ESEM or CFA was conducted for each
construct. If indicated by modification indices and justified by the content
of the items, correlated errors were allowed between these items. ESEM was
carried out for DA and CFA for the 5Cs. We applied ESEM to DA since the
construct can be organised in two different ways (i.e. internal and external
assets or as asset contexts: personal (self), social, family, school and com-
munity; Scales, 2011). This means that the DA factors are intercorrelated
and thus ESEM can provide a better solution since it allows the pre-speci-
fication of target and non-target loadings, while all non-target loadings are
close to 0 and are not fixed at 0 like with the CFA (Morin et al., 2015). The
two possible solutions for DA (internal and external assets and asset con-
texts) will be included in the analysis to compare both models. In the ini-
tial CFA for the 5Cs, 14 pairs of the same-facet items were allowed to cor-
relate (see Tirrell et al., 2019). Item loadings were interpreted according to
Tabachnick and Fidell (2006), who suggested cut-off values ranging from
0.32 (poor), 0.45 (fair), 0.55 (good), 0.63 (very good) to 0.71 (excellent). Model
fit was assessed with chi-squares, comparative fit indices (CFI), root-mean-
square error of approximation (RMSEA), and standardised root-mean-
square residual (SRMR), following recommendations by Hu and Bentler
(1999) for a good fit: CFI > .95, RMSEA < .06 and the SRMR < .08. For an
adequate fit, the following cut-off values were applied: CFI > .90, RMSEA <
.08 and the SRMR < .08 (Hair et al., 1998).
After considering the psychometric properties of the two PYD mea-
sures, Multigroup Confirmatory Factor Analysis (MGCFA) was applied us-
ing Mplus (version 8.6; Muthén & Muthén, 1998–2021) to estimate mea-
surement invariance by gender and school level (i.e. construct, metric and
scalar invariance) for each developmental asset and C of PYD separate-
ly. The series of multi-group models was compared to assess if the same
87
limited, and they were supervised by the school coordinator (teacher or
school counsellor) who answered any questions if they had them.
Data analysis
After examining descriptive statistics, correlations and reliabilities using
IBM SPSS Statistics 27, we considered the ESEM (Exploratory Structural
Equation Modelling) model of DA and the CFA model of the 5Cs using
Mplus (Version 8.6; Muthén & Muthén, 1998–2021) to examine the pro-
posed models’ structural validity. The full information maximum likeli-
hood (FIML) algorithm was used to handle missing data and assess pa-
rameters in the model. Separate ESEM or CFA was conducted for each
construct. If indicated by modification indices and justified by the content
of the items, correlated errors were allowed between these items. ESEM was
carried out for DA and CFA for the 5Cs. We applied ESEM to DA since the
construct can be organised in two different ways (i.e. internal and external
assets or as asset contexts: personal (self), social, family, school and com-
munity; Scales, 2011). This means that the DA factors are intercorrelated
and thus ESEM can provide a better solution since it allows the pre-speci-
fication of target and non-target loadings, while all non-target loadings are
close to 0 and are not fixed at 0 like with the CFA (Morin et al., 2015). The
two possible solutions for DA (internal and external assets and asset con-
texts) will be included in the analysis to compare both models. In the ini-
tial CFA for the 5Cs, 14 pairs of the same-facet items were allowed to cor-
relate (see Tirrell et al., 2019). Item loadings were interpreted according to
Tabachnick and Fidell (2006), who suggested cut-off values ranging from
0.32 (poor), 0.45 (fair), 0.55 (good), 0.63 (very good) to 0.71 (excellent). Model
fit was assessed with chi-squares, comparative fit indices (CFI), root-mean-
square error of approximation (RMSEA), and standardised root-mean-
square residual (SRMR), following recommendations by Hu and Bentler
(1999) for a good fit: CFI > .95, RMSEA < .06 and the SRMR < .08. For an
adequate fit, the following cut-off values were applied: CFI > .90, RMSEA <
.08 and the SRMR < .08 (Hair et al., 1998).
After considering the psychometric properties of the two PYD mea-
sures, Multigroup Confirmatory Factor Analysis (MGCFA) was applied us-
ing Mplus (version 8.6; Muthén & Muthén, 1998–2021) to estimate mea-
surement invariance by gender and school level (i.e. construct, metric and
scalar invariance) for each developmental asset and C of PYD separate-
ly. The series of multi-group models was compared to assess if the same
87
