International Journal for Research in Engineering Application & Management (IJREAM)

ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

Comparison of Influence of the Financial Crisis

of 2008 and the Political Events of 2014 on
Political Trust in Belgium
Suvobroto Banerjee, Assistant Professor, Amity University Kolkata, India ,
suvobroto.banerjee2019@gmail.com
Shourya Bhattacharya, Graduate Student , KU Leuven , Leuven , Belgium ,
bhattacharyashourya@gmail.com
Abstract : The objective of this study is to understand the effect of the great recession of 2008 on political trust in
Belgium as well as near term micro events on the political trust of Belgium. This study aims to compare and contrast
the effect of the great recession of 2008 and the near term micro events on the political trust of Belgium. The political
trust in Belgium for 2008 and 2014 has been compared with the political trust in Belgium for 2018 and influences of
known factors on political trust of Belgium has been studied.

Keywords — European social survey, Federal election , Financial Crisis , Political Trust , MGCFA , Structural Equation

I. INTRODUCTION (2016)[7] which have explored the effect of macro world

events on political trust and compare the same with effect of
Armingeon et al. (2014)[1]and Batels et al. (2014)[2] has
micro world events on political trust in Belgium. Data for
argued that the financial crisis of 2007-2008 has shown
significant effect on political trust in Europe. Just like most each of the years have been collected from European Social
other countries in Europe Belgium’s political landscape Survey. This study has used data from 2008, 2014 and
had also faced turmoil during 2007- 2008[3]. Just as 2018. All data used in this study has been taken from
landmark events like the financial crisis of 2007- 2008 had European Social Survey Website and details about the
left a mark on political trust, this study aims to understand
source of data can be seen in the references [8].
if similar influence can be observed on political trust in
Belgium due to near term political events of
Belgium(2014-2018)[4]. After the elections of 2014,
II.TEST FOR THE THEORETICAL
governments at all levels were established but the year MODEL
2018 was marked by some more political uncertainty. This Breustedt W. et al. (2017)[5]has primarily used different
study aims to measure and compare the effects of the variants of the political trust model as tested for fit on
political difficulties of 2018 with that of the great European Social Survey Data by Ariely et al.[9] (2011)
recession and aims to contrast the same with the political from different time periods. The same model for political
trust in Belgium in the year 2014. trust has also been used by Poznyak et al. (2014)[10]for
The study is divided into five parts. The first part of the studying political trust in the United States Of America.
study aims to establish the research question and it’s The variables used to construct confirmatory factor
relevance. The second parts aims to show a theoretical basis models by Ariely et al. (2011)[9] and Poznyak et al.
of political trust as studied in Europe. The study primarily (2014)[10] have differences but all models of political trust
have defined the trust in representative political
considers the models of political trust tested by Breustedt institutions and trust in implementing political institutions
W. et al. (2017)[5]. The third part of the study tests the as separate latent variables. The model for political trust
theoretically established models on European Social Survey tested in the works of Breustedt W. et al. (2017)[5]is as
data from 2008, 2014 and 2018. The deviations found has follows:
been reported in the third section and alternatives have been
proposed. The fourth section of the study shows
measurement invariance tests and compares the political
trust of the three time periods. Multi-group confirmatory
factor analysis has been used The fifth section comprises of
discussion and conclusion of the tests for measurement
invariance and comparison of political trust from section
four. The objective of this research work is to extend the
works of Arpino B. et.al. (2020)[6], Erkel Van et. al.

53 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

(2011) as one of the variables required to measure trust in

political institutions. As such in this case
interchangeability of trust in courts and trust in the legal
system is assumed to be justified. All models in this study
except for model 1 This study aims to develop upon the
established model of political trust as adopted by Poznyak
et al. (2014) [10] and the following alternative models are
proposed to estimate political trust:

Figure 1[10]: Model for political trust adopted by

Poznyak et al. (2014)[10] on European Social Survey
data. (Model 1)
Breustedt W. et al. (2017) [10] has considered ordinal
nature of the variables while estimating model 1. All the
observed variables stated in model 1 have not been found in
European Social Survey data of 2008,2014 and 2018 and
this the following models have been considered in Figure 3: Proposed alternative model as created from
accordance to the concept of political trust in Europe as modification of model 2 shown in figure 2 (Model 3)
defined by model 1.

Figure 2: Proposed bi-factor model of political trust, in

accordance to available observed variables in European Figure 4: Proposed alternative model as created from
Social Survey. (Model 2) modification of model 2 shown in figure 2 (Model 4)

As model 2 has been constructed in accordance to the

theoretical establishment of model 1, the observed
variables have been considered as ordinal. Model 2
considers the variables available in European Social
Survey of 2008,2014 and 2018. The variables “Trust in
government” and “Trust in civil services” as used in model
1 has not be used in any other model proposed in this
study. Though a clear evidence of measurement invariance
between using trust in the courts and the trust in legal
system could not be found for Europe, it must be noted
that trust in legal system has been used by Allum et al.[12]

54 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

Model 4 2018 0.991 0.968 754.5 0.000 4 1713

26

Model 4 2008 0.984 0.946 1852. 0.000 4 1666

989

Model 5 2014 0.906 0.754 8915. 0.000 8 1682

068

Model 5 2018 0.916 0.780 9557. 0.000 8 1713

263

Model 5 2008 0.875 0.672 2010 0.000 8 1666

2.039

Table 1: Fit statistics of all tested models

The variables used to estimate model 2 and model 3 have
been considered as categorical variables and as such in
accordance to works of Suh Y [15] (2015) , weighted least
Figure 5: Proposed alternative model as created from square mean and variance adjusted estimators have been
modification of model 2 shown in figure 2 (Model 5) used to estimate parameters for model 2 and model 3.
Figure 4 and 5 have been proposed in line with established The same derives from the original proposed model by
model for political trust where the trust in representative Poznyak et al. (2014) [10] where variables used to estimate
political institutions and trust in implementing political the bi-factor model shown in figure 1 has been considered to
institutions have been shown as separate latent variables. be ordinal. It is to note that in the bi factor models 2 and 3
Model 4 and model 5 attempts to simplify the bi-factor the covariance matrix of latent variables has been found not
models shown in figure 2 and figure 3. Nardis Y.[13] (2014) to be positive definite and thus estimates for parameters
has shown from his research about influence of media on may not be reliable. Model 2 and 3 have AFGI values
trust in the European parliament, that trust in European less than
parliament forms one of the components of trust in 0.90 for the year 2008 which would lead to elimination of
representing political institutions and the same has also been the hierarchical factor models in further analysis. In the
reflected in by Allum et al.[12](2011). simpler models 4 and 5, the variables have been
Torgler B. et al.[14 ] (2008) has shown the importance of considered to be numerical. Hu W. et al. (2017) [16] has
united nations in political trust and the same has also been shown through simulation that 11 point scales can be
included in model 3 and model 5 to test for fit in European treated as numerical variables, which fits the cases shown
Social Survey data from Belgium. in model 4 and 5. It can be seen from table 1 that model 5
has GFI and AFGI that do not comply with respective
TEST MODEL FOR FIT AND EXPERIMENTAL criteria of being greater than 0.95 and 0.90 in all the cases
RESULTS and thus it must be rejected for further studies as well.
The results for model fit of each of the previously tested Model 4 meets all the criteria for model fit and thus it
models are as follows: would be considered as acceptable in this study. It must be
noted that number of observations in each case is much
Model Year GFI AGFI Chi P Degre es Numb greater than the theoretically recommended number of
Sq. value of freedo er of
m obser observations of 200 and thus though all the p values for
vation s tested models are significant which would indicate a bad
Model 2 2014 0.985 0.975 2595 0.000 13 1682 fit, they may be ignored in this case.
.735
The factor loadings for each observed variable
Model 2 2018 0.984 0.974 4078 0.000 13 1713 (description of the variables used can be found in
.995 appendix I) for model 4 can be studied in the following
Model 2 2008 0.92 0.876 3444 0.00 13 1666
table:
3 9.821 0 Year Variable Loading Standard Z P
Error Value Value
Model 3 2014 0.97 0.969 4751. 0.000 19 1682
9 068 2014 trstprl 1.0

Model 3 2018 0.977 0.966 7400. 0.000 19 1713 2014 trstprt 0.94 0.022 42.04 0.000
332 7

Model 3 2008 0.924 0.888 4138 0.000 19 1666 2014 trstep 1.0 0.026 40.37 0.000
5.367 0

Model 4 2014 0.995 0.980 424.4 0.000 4 1682 2014 trstlgl 1.0
36

55 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

2014 trstplc 0.686 0.032 21.17 0.000 From table 2 it can be inferred that trust in political parties
0 and trust in European parliament have very high influence
on defining trust in representative political institutions in
2018 trstprl 1.0
Belgium if the three individual years under study
2018 trstprt 0.890 0.028 31.36 0.000 (2014,2018 and 2008) are considered separately. From table
7 2 it can also be inferred that trust in police has very high
influence on trust in implementing political institutions if
2018 trstep 1.0 0.026 40.34 0.000
the three individual years under study (2014,2018 and
7
2008) are considered separately. As measurement
2018 trstlgl 1.0 invariance has not been established at this stage,
comparison across years can not be conducted. In
2018 trstplc 0.727 0.030 24.06 0.000 2014,2018 and 2008 it can be concluded that trust in
9
political parties have a greater influence on trust in
2008 trstprl 1.0 representative political institutions than trust in police has
on trust in implementing political institutions if the
2008 trstprt 0.945 0.030 31.53 0.000 individual years are considered. From table 3 it can be
0 concluded that trust in representative political institutions
and trust in implementing political institutions are closely
2008 trstep 0.996 0.033 29.91 0.000
1 linked for Belgium in the years under study.

2008 trstlgl 1.0 III. MULTI-GROUP CONFIRMATORY

FACTOR ANALYSIS
2008 trstplc 0.728 0.032 23.07 0.000
2 Model 4 as shown in figure 4 has been fitted as a multi
group confirmatory factor model to compare the constructs
Table 2: Factor loadings of observed variables (Model 4) of political trust in two cases. The two cases of comparison
and the fit indices can be observed in the following table:
Year Latent Covaria Standar d Z Value P Value
factors nce Error Model Years being GFI AFGI Chi Sq. P Degree s
compared of Freedo
Value m
2014 Trust in 3.00 0.136 21.277 0.000
represen
tative
political Model 2014- 0.998 0.991 73.878 0.000 8
instituti
ons & 4 2018
Trust in
impleme Model 2008- 0.999 0.995 38.537 0.000 8
nting
political 4 2018
instituti
ons
2018 Trust in 3.415 0.153 22.362 0.000
represen Table 4: Fit statistics for multi group confirmatory factor
tative models
political
instituti From table 4 it can be observed that GFI and AFGI fit
ons & statistics prove that both multi group confirmatory factor
Trust in models fit well. The Chi-squared statistic points towards a
impleme
nting
improper fit but as the number of observations (table 1) in
political each case is much greater than the traditionally used
instituti number of 200 and thus the conclusion of the chi-squared
ons test can be ignored in this case. The model fit statistics of
2008 Trust in 2.889 0.136 21.277 0.000 all models with restrictions can be found in the appendix
represen for reference.
tative
political A. Measurement invariance test for fitted model for
instituti comparing political trust of 2014 and 2018
ons &
Trust in In order to prove weak invariance model 4 has been fitted
implem without any constraint and with equal loadings across the
nting
political two groups, in this case which is data from 2014 and 2018.
instituti Anova has been used to compare the two model fits which
ons can be observed as follows:

Table 3: Covariance between latent factors (Model 4)

56 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

Mode l Restri Degre AIC BIC Chi Chi Sq. P Table 8: CFI comparison of configurational and metric
ctions es of Sq. Differ Value model for strong invariance (2014-2018)
Freed ence
om Table 7 shows that adding restrictions on intercepts in
addition to loadings significantly changes the model fit.
Confi No 8 67273 67470 73.87 Similar conclusion can be drawn from table 8 as there is
gurati Restri 7 non-zero change in CFI between the two models. Thus in
onal ctions this case the assumption of strong measurement invariance
Metri Equal 11 67272 67450 78.68 4.241 0.236 has to rejected. As strong measurement invariance could
not be proved, In accordance to Meredith W. (1993)[17],
c Loadi 9 9 5
difference between modification indices across groups
ngs
have been studied to find if the effect of freeing a
parameter is substantially different across groups (details
Table 5: Anova comparison of configurational and metric can be found in appendix IV). In accordance to greatest
model (2014-2018) for weak invariance changes in modification indices, addition restrictions have
been added to attempt to prove partial strong invariance.
Model Restrictions Difference in CFI Difference in Greatest differences in modification indices have been
scaled CFI
prioritized. All comparisons have been made between the
Configuratio nal No model with equal loadings across groups, equal loadings
Restrictions and intercepts with increasing additional restrictions. The
comparisons can be found in the following tables :
Metric Equal 0 0
Loadings Additi Degree s AIC BIC Chi Sq. Chi Sq. P
onal of Freedo Differe Value
Table 6: CFI comparison of configurational and metric Restric m nce
tions
model for weak invariance
From table 5 it can be observed that adding constraints for Equal 11 67272 67450 78.689
equal loadings across groups does not significantly change loadin gs
the model fit. The same conclusion can be drawn from
table 6 as well, as there is no difference in CFI between
trstep 14 67301 67460 112.98 35.108 1.156x
the two models. Thus weak measurement invariance ~1 5 10-07
between the two models can be confirmed in this case.
trstep 14 67301 67460 112.98 35.108 1.156x
In order to prove strong invariance additional restrictions ~ 5 10-07
has been added in which both loadings and intercepts are
equal across groups. The comparison of the model with 1,trstlg
equal loadings across groups and equal loadings , l~1
intercepts across groups can be observed as follows:
trstep 14 67301 67460 112.98 35.108 1.156x
Mode l Restri Degre AIC BIC Chi Chi Sq. P
~ 5 10-07
ctions es of Sq. Differ Value
Freed ence 1,trstlg
om l~
1,trstpl
Mode Equal 11 67272 67450 78.68 c~1
l4 Loadi 9
ngs
14 65899 66058 129.7 49.16
Table 9: Comparison of models with additional
Mode l Equal 1.202
35 7 restrictions
4 loadin x10-10
gs Additional Difference in CFI Difference in scaled
Restrictions CFI
and
interc trstep ~ 1 4x10-03 3x10-03
epts
trstep ~ 1,trstlgl ~ 1 4x10-03 3x10-03
Table 7: Anova comparison of configurational and metric
model for strong invariance (2014-2018)
trstep ~ 1,trstlgl ~ 4x10-03 3x10-03
Model Restrictions Difference in CFI Difference in 1,trstplc ~ 1
scaled CFI

Model 4 Equal
Loadings Table 10: CFI comparison for models with additional
Model 4 Equal loadings 0.007 0.008 restrictions with model with equal loadings across groups.
and intercepts
It is to be noted that all additional restrictions shown on
table 9 and 10 are in additional to equal loadings and
intercepts if not explicitly mentioned. From table 9 and 10

57 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

it can be observed that additional restrictions imposed, onal ngs

significantly changes the model fit from the base model Metri Equal 14 67787 67946 134.6 94.99 2.2x1
which in this case has only equal loadings across the c Loadi 27 7 0-16
groups. Thus partial measurement invariance can not be ngs
established as well. It must be concluded that the concept and
of political trust in Belgium as defined by model 4 (figure interc
4) is not clearly comparable for the year 2014 and 2018. epts
B. Measurement invariance test for fitted model Table 13: Anova comparison of configurational and
for comparing political trust of 2008 and 2018 metric model for strong invariance (2008-2018)
In order to prove weak invariance between the construct of Model Restrictions Difference in CFI Difference in
political trust in Belgium for the year 2008 and 2018, scaled CFI
model 4 (figure 4) has been fitted without any restrictions
Configuratio nal Equal
and with equal loadings across groups. The difference Loadings
model fits has been studied using anova and the details can
be found in the following table: Metric Equal Loadings 0.011 0.012
and Intercepts
Mode l Restri Degre AIC BIC Chi Chi Sq. P
ctions es of Sq. Differ Value Table 14: CFI comparison of configurational and metric
Freed ence
om
model for strong invariance (2008-2018)
From table 13 it can be inferred that addition of equality
Confi No 8 67703 67899 38.53 constraints for intercepts in addition with equality
gurati Restri 7 constraints for loadings across the groups introduces
onal ctions significant change to the model fit. Similar conclusion can
Metri Equal 11 67704 67881 45.32 5.927 0.115 be drawn from table 14 as difference between CFI
between the two models is non-zero. Thus it can be
c Loadi 7 9 2
concluded that the assumption strong measurement
ngs
invariance can be rejected in this case.
Table 11: Anova comparison of configurational and
metric model (2008-2018) for weak invariance In accordance to partial measurement invariance proof
shown by Meredith W. (1993)[17], modification indices
Model Restrictions Difference in CFI Difference in have been studied and the restrictions that show changes
scaled CFI
modification indices across the groups have been added to
Configuratio nal No the model in which already constraints for equal loadings
Restrictions and intercepts across groups have been applied. The
model with additional restrictions have been compared
Metric Equal 0 0
Loadings
with the model in which only the loadings are constrained.
The details about modification indices can be found in the
Table 12: CFI comparison of configurational and metric appendix. The difference in model fits has been conducted
model for weak invariance (2008-2018) using anova and the details can be observed in the
following table:
From table 11 it can be observed that addition of equal
loadings constraints across the groups does not change the Additi Degree s AIC BIC Chi Chi Sq. P Value
onal of Freedo Sq. Differe
model fit significantly as the p value for difference in chi- Restric m nce
squared statistic is not significant. The same conclusion tions
can be drawn from table 12 as there is no difference in
CFI. Thus weak measurement invariance between the Equal 11 67704 67881 45.327
political construct in Belgium for the year 2008 and 2018 loadin
gs
can be proven in this case.
trstep 14 67787 67946 134.62 94.997 2.2x10-16
Additional restrictions has been added in order to attempt ~1 7
to prove strong measurement invariance. Additional trstep 14 67787 67946 134.62 94.997 2.2x10-16
restrictions of equal intercepts across groups have been ~ 7
added with the equality constraints across groups. Anova 1,trstlg l
has been used to compare the two models and it can be ~1
trstep 14 67787 67946 134.62 94.997 2.2x10-16
observed in the following table:
~ 7
Mode l Restri Degre AIC BIC Chi Chi Sq. P 1,trstlg l
ctions es of Sq. Differ Value ~ 1,trstpl
Freed ence c~1
om Table 15: Comparison of models with additional
restrictions
Confi Equal 11 67704 67881 45.32
Additional Difference in CFI Difference in scaled
gurati Loadi 7 Restrictions CFI

58 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

trstep ~ 1 0.011 0.012 construct of political trust, it has been concluded that it is
trstep ~ 1,trstlgl ~ 1 0.011 0.012 not possible to understand if the political opinion of the
general population has improved or deteriorated in this
trstep ~ 1,trstlgl ~ 0.011 0.012 time period but it can be inferred that the diverse set of
1,trstplc ~ 1 political events in Belgium has ensured that the general
population has constantly re-evaluated their understanding
Table 16: CFI comparison for models with additional
and outlook of politics in the period under study.
restrictions with model with equal loadings across groups.
It is to be noted that in table 15 and table 16, the
REFERENCES
restrictions are in addition to loadings and intercepts being [1] Armingeon, K., & Ceka, B. (2014). The loss of trust in the
constrained across the group unless explicitly mentioned. European Union during the great recession since 2007: The role of
heuristics from the national political system. European Union
It can be inferred from table 15 that additional restrictions
Politics,15(1), 82– 107.
change the model fit significantly. Similarly, from table 16
[2] Bartels, L. (2014). Ideology and retrospection in electoral
it can be inferred that the fits of the models with additional
responses to the Great Recession. In L. Bartels & N. Bermeo
restrictions are different from the model with just equal (Eds.), Mass politics in tough times: Opinions, votes and protest in
loadings and intercepts across the groups. Thus strong the great recession. New York: Oxford University Press.
invariance could not be proven in this case and it has to be [3] Dave Sinardet (2008) Belgian Federalism Put to the Test: The
concluded that the construct of political trust in Belgium 2007 Belgian Federal Elections and their Aftermath, West
(as estimated using model 4) between the years 2008 and European Politics, 31:5, 1016-1032, DOI:
2018 are not clearly comparable. 10.1080/01402380802234706

[4] "Two formateurs to form new centre-right Belgian Government".

IV. CONCLUSIONS deredactie.be. 22 July 2014

The objective of this study was to understand the structure [5] Wiebke Breustedt (2017). Testing the Measurement Invariance of
of political trust in Belgium and to compare the effect of Political Trust across the Globe. A Multiple Group Confirmatory
Factor Analysis. Methoden, Daten, Analysen. 1864-6956 (Print);
micro and macro events on political trust. Macro in this 2190-4936 (Online). GESIS - Leibniz-Institute for the Social
case refers to the financial crisis of 2008 and micro in this Sciences, Mannheim
case refers to the difficulties observed in formation of
[6] Arpino, B., Obydenkova, A.V. Democracy and Political Trust
Belgian federal government post the election of 2014. Before and After the Great Recession 2008: The European Union
From model 4 (figure 4) it can be concluded that the and the United Nations. Soc Indic Res 148, 395–415 (2020).
construct of political trust in Belgium can be expressed [7] Van Erkel, Patrick F.A, and Van Der Meer, Tom W.G.
using separate constructs of trust in representative political "Macroeconomic Performance, Political Trust and the Great
institutions and trust in implementing political institutions. Recession: A Multilevel Analysis of the Effects of Within-country
Fluctuations in Macroeconomic Performance on Political Trust in
In each year under study (i.e. 2014,2018 and 2008) trust in
15 EU Countries, 1999-2011." European Journal of Political
political parties and trust in European parliament have Research 55.1 (2016): 177-97. Web.
strong influence on trust in representative political
[8] Source of data: The data used in this research work can be
institutions. In each year under study, trust in police has downloaded here:
strong influence on trust in implementing political
https://www.europeansocialsurvey.org/data/themes.html?t=politics.
institutions.
The data files from 2014 and 2018 from the political theme of
The influence that trust in police has on implementing european social survey has been used in this project.

political institutions is lower than the influence of trust in [9] Ariely, G., & Davidov, E. (2012). Assessment of measurement
European union and trust in political parties on equivalence with cross-national and longitudinal surveys in
political science. European Political Science, 11, 363–377.
representative political institutions for each year under
study. From section IV of the study it can be concluded [10] Poznyak, D., Meulemann, B., Abts, K., & Bishop, G. F. (2014).
Trust in American government. Longitudinal measurement
that political trust in Belgium are not clearly comparable
equivalence in the ANES, 1964-2008. Social Indicators Research,
constructs when comparing the political trust between 118, 741–758
2008-2018 and 2014-2018 and thus the differences in
[11] Image Source: The image has been taken from [5]
political trust between 2008-2018 and 2014-2018 can not
be stated in this case as the same may lead to comparison [12] Allum, N., Read, S., & Sturgis, P. (2011). Evaluating change in
social and political trust in Europe using multiple group
of constructs that are not the same across the years under
confirmatory factor analysis with structured means. In E. Davidov,
study. J. Billiet, & P.Schmidt (Eds.), Cross cultural analysis: Methods and
applications. New York, NY: Routledge.
It can be concluded from this study that the construct of
political trust in Belgium has been a changing entity from [13] Nardis, Yioryos. "News, Trust in the European Parliament, and EP
Election Voting: Moderated-Mediation Model Investigating
2008 to 2018 and due to the changing nature of the
Voting in Established and New Member States." The International

59 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

Journal of Press/Politics 20.1 (2015): 45-66. Web. test for multivariate normality for the variables in each
[14] Torgler, Benno. "Trust in International Organizations: An model is warranted. Variables used in model 4 and model
Empirical Investigation Focusing on the United Nations." The 5 have been tested. Results of multivariate normality test
Review of International Organizations 3.1 (2008): 65-93. Web. can be found in the following table:
[15] Youngsuk Suh (2015) The Performance of Maximum Likelihood
Year Model Test Statistic P Multiva
and Weighted Least Square Mean and Variance Adjusted
statistic Values Value riate
Estimators in Testing Differential Item Functioning With
Normali ty
Nonnormal Trait Distributions, Structural Equation Modeling: A
Multidisciplinary Journal, 22:4, 568-580, DOI:
10.1080/10705511.2014.937669 2014 Model 4 Henze- 9.261834 0 No
[16] Wu, Huiping, and Leung, Shing-On. "Can Likert Scales Be Treated Zirkler
as Interval Scales?—A Simulation Study." Journal of Social 2014 Model 4 E 24.64456 0 No
Service Research 43.4 (2017): 527-32. Web. Statistic
[17] Meredith, W. (1993). Measurement invariance, factor analysis and
2014 Model 5 Henze- 9.332276 0 No
factorial invariance. Psychometrika 58, 525–543. doi:
Zirkler
10.1007/BF02294825 Journal of Social Service Research 43.4
(2017): 527-32. Web. 2014 Model 5 E 28.82509 0 No
Statistic
APPENDIX
Year Model Test Statistic P Multiva
I. DESCRIPTION OF VARIABLES
Statistic Values Value riate
Name of variable Description Scale Normali ty
2018 Model 4 Henze- 8.340047 0 No
trstprl Trust in country’s parliament 0-10
Zirkler
trstprt Trust in political parties 0-10 2018 Model 4 E 20.87165 0 No
Statistic
trstlgl Trust in legal system 0-10
2018 Model 5 Henze- 8.197119 0 No
trstplc Trust in the police 0-10 Zirkler
2018 Model 5 E 24.59548 0 No
trstun Trust in the United Nations 0-10 Statistic
trstep Trust in the European parliament 0-10 2008 Model 4 Henze- 16.76151 0 No
Zirkler
2008 Model 4 E 6.689054 0 No
Table 17: Description of variables used in the study Statistic
2008 Model 5 Henze- 6.887859 0 No
Each variable used in this study has values from 0 to 10, all Zirkler
other values available has been removed for this study. The 2008 Model 5 E 20.31953 0 No
following categories has been removed : Statistic
Table 19: Test for multivariate normality
Response value Description
MVN package has been used to test for multivariate
77 Refusal
normality and the same can be seen in the implementation
88 Don’t know of the project. It can be observed from table 19 that
99 No answer multivariate normality could not be proved in any case but
as the number of observations are greater than 1000 in
Table 18: Removed values from all variables
each case, it can be assumed that deviations from
In the variables seen in table 17 the value 0 represents the multivariate normality are not quite substantial. To
lowest agreement to a trust element and 10 represents mitigate the effect of deviations from normality, robust
absolute agreement to a trust element. The agreement to a maximum likelihood estimators (MLM) of lavaan package
trust construct is structred in increasing order from 0 to 10. in R has been used for estimating model 4 and model 5 for
II. ESTIMATORS USED AND TEST FOR NORMALITY each year.

In model 2 and model 3 the variables have been treated as III. FIT STATISTICS FOR MODELS WITH ADDITIONAL
RESTRICTIONS
ordinal and as such weighted least square mean and
variance adjusted estimator has been used. In model 4 and In section IV of the study, in-order to prove measurement
model 5 the variables have been treated as numerical. As invariance model 4 has been fitted with additional
per assumption of confirmatory factor analysis, restrictions. The fit statistics of models with additional
multivariate normality should exist in the variables used to restrictions can be found in the following table:
estimate the models. As the four models tested in this
study use two different combination of variables, separate

60 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.

 International Journal for Research in Engineering Application & Management (IJREAM)
ISSN : 2454-9150 Vol-07, Issue-03, JUNE 2021

Restricti ons Yea GFI AFGI Chi Sq. P Value Loading 2018
r Statistic s s&
Intercep
Equal 201 0.998 0.993 78.689 0.000 ts, trstep
4-
~ 1,
Loading 201
trstlgl ~
8
1, trstplc
s
~1
Equal 201 0.997 0.992 112.985 0.000
4- Table 20: Fit statistics for all models with restrictions
Loading 201
Table 20 shows the fit statistics or all models that have been
8
s& tested with additional restrictions. In each case GFI and
Intercep AGFI show a good model fit. chi-squared test does not
ts accept any of the models but as the number of observations
Equal 201 0.997 0.992 112.985 0.000 are quite greater than 200, the chi-squared test result in this
4- case may not be reliable.
Loading 201
8 V. MODIFICATION INDICES
s,
Intercep Modification indeces have been used in section IV of the
ts, trstep study in order to attempt to establish partial
~1 invariance. The modification indices are of model 4
Equal 201 0.997 0.992 112.985 0.000 with equal loadings and intercepts across groups . The
4-
greatest changes in across groups modification indices
Loading s & 2018
Intercep ts, can be studied in the following tables :
trstep
Relation Group Modification Indices
~ 1, trstlgl ~ 1
f =~ trstlgl 1 3.044
f =~ trstlgl 2 0.240
Equal 2014- 0.997 0.992 112.985 0.000
f =~ trstplc 1 3.044
Loading 2018 f =~ trstplc 2 0.240
s& f1 =~ trstep 1 3.136
Intercep f1 =~ trstep 2 1.258
ts, trstep
~ 1,
trstlgl ~ Table 21 : Modification Indices for additionally
1, trstplc constrained factors in the partial invariance test for
~1
2014 - 2018
Equal 2008- 0.999 0.996 45.327 0.000
Relation Group Modification Indices
Loading 2018
f =~ trstlgl 1 1.135
s
f =~ trstlgl 2 0.632
Equal 2008- 0.997 0.991 134.627 0.000 f =~ trstplc 1 1.135
Loading 2018 f =~ trstplc 2 0.632
s& f1 =~ trstep 1 13.967
Intercep f1 =~ trstep 2 11.938
ts
Equal 2008- 0.997 0.991 134.627 0.000
Table 22 : Modification Indices for additionally
Loading 2018
constrained factors in the partial invariance test for
s&
Intercep 2008 - 2018
ts, trstep
In table 21 and 22 , the latent factors f and f1 represent trust
~1
in the representative political institutions and trust in
Equal 2008- 0.997 0.991 134.627 0.000
implementing political institutions respectively. Additional
Loading 2018
modification indices have not been reported as the
s&
Intercep
difference of the other modifications indices across groups
ts, trstep are not substantially high.
~ 1,
trstlgl ~
1
Equal 2008- 0.997 0.991 134.627 0.000

61 | IJREAMV07I0375032 DOI : 10.35291/2454-9150.2021.0315 © 2021, IJREAM All Rights Reserved.