RELEVANT ISSUES THAT INFLUENCE THE TRUST IN PARLIAMENT Applied research project
BOCCONI UNIVERSITY Academic year 2009/2010
1) we declare that this written assignment is our own work and does not include material from published sources used without proper acknowledgment or material copied from the work of other students 2) we declare that this assignment assignment has not been submitted submitted for assessment in any other course at any university. 3) we have a photocopy or electronic electronic version of this assignment assignment in our possession.
Munteanu Vladimir
Yuliya Pauliukevich
ID 1237551
ID 1377381
___________
Racul Iulian
ID 1325240
___________
________
Summary We found that there is a worldwide problem of low levels of trust in government institutions in general and in parliaments in particular. In this project our team analyzes the essential factors that influence the overall level of people¶s trustiness in country¶s parliament. The study presents and analyses the data trying to draw the main conclusions related to our problem. The main statistical tests used in the project are Factor analysis and Regression analysis. First of all, we made an accurate selection of eight variables, strongly related to the main argument. Then we used the data on these variables in the analysis, using the tests mentioned above. The second step in our project was to define the two main factors, with different covariates, which were used further in the Regression analysis. Thanks to Factor Analysis we found two factors that stand for two fundamental characteristics of the level of people¶s trust in the country¶s legislative institution, which are Security and Wellbeing. It was important for us to find and to select the most relevant variables and to demonstrate the linkage among them. That is why we tried to do a ³logical´ analysis and to select the most appropriate and interesting variables to describe and support better the main idea of our project. We considered the relationship among such broad issues, as health, safety, happiness, and democracy could explain the attitude of the population towards the parliament. Considering people as rational individuals we also took into account the attitudinal difference between males and females. That is why we also included the variable ³Gender´ in order to stress the attention on this important issue. In this way we tried to investigate how gender differences influence the general trust of people in parliament. Important to mention is the fact that after doing the analysis of the missing values, we realized that we can consider further all the variables which were initially selected because the values for all the variables selected seem to be available. The study examined public opinion in depth and detail and several important conclusions emerged from it concerning public opinion and the level of public trust in the legislative institution. Our initial assumption was that Security will play a bigger role in the subject matter, as it is the main concern of people. As it turned out, Wellbeing has a much bigger impact on the dependent variable and it was a bit of a surprising result. Also we hereby wanted to assess the influence of immigration, which is a rather new issue, on the trust on parliament. As it proved to be, its role is indeed an important one and we can assume that it will rise in the future. We consider all these results very of rather great significance and we want to stress their usefulness for any administrative unit that studies the consequences of the reforms of the parliament towards the trust of the population in this legislative body.
Introduction The relevant professional literature deals with the term ³trust´ in various spheres of the Social Science; with explanations on the state and international levels of the phenomenon of low levels of trust in government institutions in general and parliaments
in particular; with the degree to which the phenomenon threatens democracy; and with reforms that can be introduced in order to bring about a change in the trend. In our research we deal with the first issue - of democracy, but leave the floor to politicians in deciding which reforms should be made in order to increase the confidence of people in politicians, or, in particular, the Parliament. Many parliaments have introduced reforms that are designed to improve the connection between themselves and the public, and there is an expectation that the reforms (which we consider as a main tool in gaining population confidence in the legislative body) will have an effect, at least indirectly, on the level of trust in them. In fact, the reforms have had little effect on public trust, because the low level of trust apparently results more from social and cultural developments (the issues we tried to analyze, such as general level of happiness, satisfaction, safety or health) than from objective reasons, connected with the functioning of the parliaments. We found that several international organizations have been dealing in recent years with the decline in public trust in government institutions, and its ramifications. In a report prepared in 2005 for the OECD it was found that many states, which are members of the Organization have started to focus attention on strengthening relations with their citizens as a result of the decline in the rates of participation in the elections, in the rates of membership in political parties, and in the trust that the public expresses in central public institutions, including parliaments. From the information and data gathered from different sources we found out that the decline of the public trust in parliaments in most of the states has taken place in the last 10-15 years. It was interesting to discover that the level of trust which the public has in parliament does not necessarily correspond to that in members of the parliament. There are states in which the members of parliament enjoy greater trust than does the parliament ± especially in countries in which the members are elected in direct elections. The main reason for the difference is that the expectations from them as individuals are different than the expectations from the parliament as a body. In several countries it is the parliament which enjoys greater trust. The conclusion we draw from examining researches and analysis already done is that almost all were analyzing the public opinion on political issues (such as reforms, confidence in the members of parliament etc.), while we are trying to deal with this problem from a social subjective point of view, explaining how actions of parliaments influence people¶s general state. Let¶s define the steps we followed in our Applied Research process: 1. Defining the problem (which we have already done) 2. Collecting the data (from Database ³Life Course and Attitudes; EES Round 3 2006/2007) 3. Processing the data (Screening & Cleaning) 4. Analysing the data (Factor Analysis and Regression Analysis) 5. Drawing conclusions
Data
description
There are many sources for comparative data regarding the level of public trust in government institutions, including parliament. We decided to use the data set drawn from Round 3 of the European Social Survey (ESS), conducted in 2006/2007 by the European Commission, the European Science Foundation and various scientific funding bodies. The ESS3-2006 main questionnaire is made up of the core module and two rotating modules: Timing of life (the organization of the life course in Europe) and Personal and social well-being (creating indicators for a flourishing Europe). ³Timing of life´ module aims at furthering our understanding of the views of European citizens on the organization of the life course and of their strategies to influence and plan their own lives. ³Personal and social wellbeing´ module seeks to evaluate the success of European countries at promoting the personal and social well-being of their citizens. As mentioned above, we tried to include our variables under some more general social issues. So, we divide our variables into 4 groups, each group representing one general social issue: I. Happiness (here we included the variable concerning happiness + the variable about the general satisfaction with standard of living); II. Democracy (the variable concerning immigration (bad or good for country's economy)); III. Health (here we deal with the variable regarding the subjective opinions on general health + the variable on worry about becoming victim of violent crime has effect on quality of life); IV. Safety (here we refer to the variable about how often people worry about their home being burgled + how safety they feel of walking alone in local area after dark); + the variable ³Gender´ which we use for the Regression Analysis. Now, we will try to describe every variable and to demonstrate their relevance to our main problem question: ³trust in county¶s parliament´. It is appropriate to start our data description with the dependent variable, which is, also, our main research question and then to continue, one by one, explaining the independent variables. y
y
y
y
TRSTPRL (dependent variable): measures on a scale from 0 to 10 the extent to which the interviewed person has confidence, or, trust in country¶s parliament. In this case, 0 means ³no trust at all´ up to 10 which means ³complete trust´; CRVCTWR (independent variable): measures on a scale from 1 to 4 the extent to which the interviewed person worries about becoming a victim of violent crime. In this case , 1 means ³all or most of the time´ up to 4 which means ³never´; BRGHMWR (independent variable): measures on a scale from 1 to 4 the extent to which the interviewed person worries about their home being burgled. In this case, 1 means ³all or most of the time´ up to 4 which means ³never´; AESFDRK (independent variable): measures on a scale from 1 to 4 the extent to which the interviewed person worries about feeling of safety of walking alone in
y
y
y
y
y
local area after dark. In this case, 1 means ³very safe´ up to 4 which means ³very unsafe´; HEALTH (independent variable): measures on a scale from 1 to 5 the extent to which the interviewed person worries about subjective general health. In this case, 1 means ³very good´ up to 5 which means ³very bad´; HAPPY (independent variable): measures on a scale from 0 to 10 the extent to which the interviewed person is happy. In this case, 0 means ³extremely unhappy´ up to 10 which means ³extremely happy´; STFSDLV (independent variable): measures on a scale from 0 to 10 the extent to which the interviewed person is satisfied with standard of living. In this case, 0 means ³extremely dissatisfied´ up to 10 which means ³extremely satisfied´; IMBGECO (independent variable): measures on a scale from 0 to 10 the extent to which the interviewed person consider immigration bad or good for country's economy. In this case, 0 means ³bad for the economy´ up to 10 which means ³good for the economy´ And finally, GNDR: identifies if the interviewed person is male or female. In this case this we are dealing with a Dummy variable. Originally the dataset provided us with values 1 for ³male´ and 2 for ³female´. We decided to change the variable¶s values to 0 for ³male´ and 1 for ³female´. We decided to make the change in order to show the distinction between males and females in a more comprehensible way.
Before initiating the main data analysis we considered appropriate to list the main assumptions for the two tests we are using in this research. Assumptions for Regression: I. II. III. IV.
E () =0 so that estimators are unbiased Var () = 0 so that there are no heteroskedasticity problems Cov (i, j) = 0 so that the error are not correlated with one another Cov (i, xi) = 0 so that the errors are independent from the values
Assumptions for Factor Analysis: I. Large enough sample to yield reliable estimates of the correlations among the variables II. Statistical inference is improved if the variables are multivariate normal III. Relationships among the pairs of variables are linear IV. Absence of outliers among the cases V. Some degree of collinearity among the variables but not an extreme degree or singularity among the variables VI. Large ratio of N / k
Screening & Cleaning of Data The goal of data screening is to identify, and remedy, any problems with data prior to beginning the data analyses that will address the research question. This is a necessary step because data problems can produce misleading (invalid or unreliable) results or make it difficult or impossible to run the desired analysis. The most frequent and common problems encountered with the data are: social desirability bias; inability to answer questions, resulting in missing values; random response; sabotage; accidental errors; interviewer effect (intentional or non-intentional influence on responses) etc. First of all, we run a Missing Value analysis to detect if values for one or more variables might be unavailable. The results we concluded from the following table:
Univariate Statistics Missing N
Mean
Std. Deviation
Count
No. of Extremes
Percent
Low
a
High
crvctwr
42457
3.23
.824
543
1.3
1244
0
brghmwr
42571
3.04
.910
429
1.0
2489
0
aesfdrk
42414
2.07
.821
586
1.4
0
2409
imbgeco
40437
4.93
2.494
2563
6.0
0
1340
happy
42615
7.14
2.059
385
.9
2670
0
health
42919
2.29
.931
81
.2
0
755
stfsdlv
42754
6.46
2.400
246
.6
2044
0
gndr
42903
97
.2
a. Number of cases outside the range (Mean - 2*SD, Mean + 2*SD).
When dealing with missing values one should decide whether a variable with a high proportion of missing values should be dropped from the analysis. The advantage is that we use only ³known´ data, while the disadvantage is that we may lose an important variable. So, we encountered this dilemma in our analysis of missing values with the variable IMBGECO which showed a pretty large percentage of missing values (6%). We considered this variable important for our further analysis, so we decided to keep it and to rely on it. One further step after the missing values analysis is to start examining the data for impossible or improbable values or doing univariate data screening: Since we have only variables that are based on a metric scale and a dummy variable, the influence of impossible values is rather minor. Also in each of these variables there is a specific value that is excluded, that represents these cases. The same reasoning applies when comparing the values of the variables with the Mean +/- Standard Deviation in order to find the extreme values or outliers. Since all
variables are measured on metric scale and we have a dummy variable, we exclude this case also. The normality condition is a very important one and we verified it by comparing the module of kurtosis of each distribution of the variable to be less the 1. All variables passed the test with a single exception, the variable GNDR. But, as it is a dummy variable, it is not unusual to have a rather non-normal distribution. But, regarding the big sample size, we continue to use and rely on it because it will be very useful in the regression analysis. Since all the data has been cleaned and screened, we can proceed to the two methods of Multivariate Analysis: Factor Analysis and Regression Analysis.
Multivariate Analysis The first step in the Multivariate Analysis is the Factor Analysis (FA). FA is a method of data reduction in which we use a small set of factors (new variables) in order to try to represent the common behavior of a larger set of observed variables. Our decision was to reduce the 6 variables that represent the feeling of security of the respondents and the general wellbeing of the respondents. These variables are CRVCTWR, BRGHMWR, AESFDRK (representing the security) and HEALTH, HAPPY, STFSDLV (representing the wellbeing). The first thing that we can observe is the sample size which is well above 1000. This is a very good beginning in order to find correlations and appropriate factors. After the screening and cleaning of the data, we will do two tests to verify if there is enough correlation in the data set. The tests are Kaiser-Meyer-Olkin (KMO) and Barlett¶s test of sphericity. We used SPSS in order to obtain them as shown below. KMO and
Bartlett's Test
Kaiser-Meyer-Olkin Measure of Sampling Adequacy. Bartlett's Test of Approx. Chi-Square Sphericity df Sig.
.683 8040.74 7 15 .000
Also we verified that the determinant of the correlation matrix to be different than 0. In our model this Determinant is 0.294 and this means that there is no sign of multicollinearity. The KMO is the ratio of the sum of all squared correlations in the sample to the sum of all squared correlations plus the sum of all squared partial correlations. Although it should be >0.7 in order to be satisfactory, our result 0.683 is
quite close. Barlett¶s test of sphericity compares our correlation matrix with an identity matrix. In our case, Sig=0.000, so the difference between those two is significant. The next step is to continue to finding the actual factors. The fact that all the communalities were above the acceptance level, pointed out the idea that no variables should be excluded. The Total Variance Table can point out the important factors by comparing the Eigenvalues that are greater than 1. The factors that show an Eigenvalues less than 1 are useless and explain no more than a simple variable, so incorporating them into a factor is not desirable. The table was created in SPSS.
Total Variance Explained Component
Extraction Sums of Squared
Rotation Sums of Squared
Loadings
Loadings
Initial Eigenvalues
Total
% of
Cumulative
% of
Variance
%
Total
Variance
Cumulative %
Total
% of
Cumulative
Variance
%
1
2.426
40.432
40.432
2.426
40.432
40.432
1.938
32.305
32.305
2
1.351
22.519
62.951
1.351
22.519
62.951
1.839
30.645
62.951
3
.793
13.217
76.168
4
.618
10.303
86.471
5
.421
7.016
93.487
6
.391
6.513
100.000
Extraction Method: Principal Component Analysis.
As we can see from the table, the first two factors are the reliable ones. The first has an Eigenvalue of 2.426, explains 32.305% of the variance and includes three variables: CRVCTWR, BRGHMWR and AESFDRK. This factor represents the SECURITY. The second reliable factor has an Eigenvalue of 1.351, explains 30.646% (with a cumulative percentage with the first of 62.951%) and includes also three variables: HEALTH, HAPPY, STFSDLV . This factor represents the WELLBEING. A further confirmation of our conclusions is the Scree Plot representing on the Xaxis the number of the factors and on the Y-axis the Eigenvalues. As we can observe from it, after the first two factors, the curve becomes flatter and goes below Eigenvalue 1.
The following and the last table will be the Rotated Factor Matrix. It presents all the details connection between each variable and the Factors. The values that were not significant have been erased. As we can observe, the Rotation method was Varimax with Kaiser Normalization. Rotated Component Matrix
a
Component 1 how often worry about
2 .839
becoming a victim of violent crime how often worry about your
.837
home being burgled feeling of safety of walking
-.701
alone in local area after dark subjective general health
-.574
how happy are you
.862
satisfied with standard of
.838
living Extraction Method: Principal Component Analysis. Rotation Method: Varimax with Kaiser Normalization. a. Rotation converged in 3 iterations.
This test provides the necessary information about the strong correlation between the factors with each of the variables. Although every value is above 0.5 (which is the acceptance level), we observe two relative low values. The first is ³feeling
of safety walking alone in the local area after dark´ with -0.701 and ³subjective general health´ with -0.574. Our assumption about these low values is about the fact that both of them have a negative correlation with respect to the others. This means that they are measured on a reversed scale with respect to the other variables, but nevertheless are important and significant, so they must not be excluded. The final conclusions that can be made out of Factor Analysis are several important ones: the first is that 6 variables were reduced to two factors; the first factor explains SECURITY and the second one explains WELLBEING; and finally that these two factors explain 62.951% of the covariance among variables.
The second step in the Multivariate Analysis is the Regression Analysis (RA). RA is a method that is useful in two ways. Namely, the possibility to predict the value of a dependant variable based on the values of some independent variables and also to predict the impact of changes in the independent variable on the dependant variable. In this case we use as a dependent variable TRSTPRL (trust in country¶s parliament) and as independent variables we take the two factors obtained in the factor analysis and also GNDR and IMBGECO. The final result of the analysis will make us come closer to the answer of the question: ³Which factors influence the trust in the parliament and by what means?´ The first step is to measure the correlation between our two factors (SECURITY and WELLBEING), GNDR and IMBGECO. We can do so by making a Coefficients Table and discuss the significance of each variable.
Coefficients Model
Unstandardized
Standardized
Collinearity
Coefficients
Coefficients
Statistics
B 1 (Constant)
a
Std. Error
Beta
t
3.339
.119
SECURITY
.271
.032
.108
WELLBEING
.710
immigration bad or
Sig.
Tolerance
VIF
27.972
.000
8.371
.000
.942 1.062
.032
.279 22.085
.000
.977 1.024
.131
.012
.136 10.692
.000
.968 1.033
-.122
.064
.057
.948 1.055
good for country's economy gender
a. Dependent Variable: trust in country's parliament
-.024
-1.905
The first thing to be discussed is the comparison between the significance level of each of the variables and 0.05 which is the acceptance level. As we can see, each of the significance variables is 0.00 (which is below 0.05) and the sole exception is gender which is slightly above 0.5 with a value of 0.57. Nevertheless we will continue with this variable and use it in our Regression Analysis because it is an important variable and it might be useful to compare the results from the model with the facts from the real life. The next step is to exclude the possibility of variables with a high level of collinearity. The acceptance level in this case is a VIF <10. Multicollinearity might become an issue in other case. In our case, as we can observe from the table, VIF has values between 1.024 and 1.062 and excludes this situation. When analyzing the obtained coefficients we come to some interesting conclusions. The first conclusion is that besides the equal level of significance between the two factors, we observe that a change in WELLBEING influences the dependent variable much more than a change in any other variable. Even if for every one of us, security might seem the most important thing, our model predicts that the trust in parliament is more influenced by how we feel about life and health. Our assumption for this phenomenon is that the mentality of a great majority of people is that parliament has a much bigger influence in the economic and social life of a person. Security, on the other hand, is more the responsibility of each of us. We must take appropriate measures in order to guarantee the high level of security for us and our families. Another conclusion that we can draw from looking at the coefficients is the ascending role of the emigrants in the collective mentality. The assumption is that an unwise change in the policy of the parliament or government towards the emigrants might affect the collective trust in the parliament rather serious. Finally, as we expected, given the fact that we changed the variable GNDR (³0´ to female and ³1´ to male), there is a minus sign in front of the coefficient. We will explain the role of this dummy variable latter on. Although we have a sample with a very big number of respondents, we have to build in SPSS a Normal Probability Plot in order to compare the observed distribution of the error term with a theoretical normal one. As we can see below, the observed points practically follow the straight line and so, the assumption holds.
Another important step in analyzing the model is to analyze R 2. It is the square of the correlation coefficient and explains how good is the regression at approximating the real data provided in the survey. Its value is showed below in the table. b
Model Summary Model R 1
R Square a
.353
.125
Adjusted R
Std. Error of the
Square
Estimate .124
2.327
Durbin-Watson 1.687
a. Predictors: (Constant), gender, REGR factor score 2 for analysis 18, immigration bad or good for country's economy, REGR factor score 1 for analysis 18 b. Dependent Variable: trust in country's parliament
As we can see, R2 is equal to 0.125, which means that, roughly speaking, with our model we explain approximately 13% from the real data. Even that this number is not so high, it would be rather difficult to predict the values of such a broad variable with such few factors. Trust in parliament and in politics in general is a complex life issue and it is very hard to be predicted. Almost infinitely many things influence the opinion of different people and we selected those that seemed with the highest level of significance. One last thing to be observed is the fact that Durbin-Wattson statistic value is somewhere around 2, precisely 1.687. This is a proof that there is very little autocorrelation between the error terms and there is very little bias of the outcome of our analysis. The final step, after checking all the issues and deviations that might have influenced the result, is to write the Regression Equation that we obtained: Trust in Parliament=3.339+0.271*SECURITY+0.710*WELLBEING+0.131*IMBGECO0.122*GNDR We observe that the only negatively correlated variable is GNDR and depending on what value it takes, it affects the trust in parliament. The only two possible values are ³0´ for female and ³1´ for male so we have two situations: For female: Trust in Parliament=3.339+0.271*SECURITY+0.710*WELLBEING+0.131*IMBGECO For male: Trust in Parliament=3.339+0.271*SECURITY+0.710*WELLBEING+0.131*IMBGECO0.122*1 As we have been expected, the known fact that men are more interested in politics makes them more reluctant in trusting the parliament, which is the way democratic societies manifest legislative powers. Also the rest of the factors and variables are positively correlated and induces the idea that the main method from which the parliament as an institution can gain trust is by promoting policies and laws sustaining security, wellbeing and other vital attributes of modern life. An important
achievement of our project was also to prove a rather big importance of the immigrants in the mentality of the society. Our group also thinks that, as the time passes, and immigrants became a part of our life, the significance and the impact of this ³variable´ will increase. Conclusions In the end there are several conclusions that can be made after executing all these analyses. First of all are the facts that behave on the paper as they do in reality. Security and Wellbeing have a positive influence on our trust in the parliament. We tend to trust politicians in general when we feel satisfied with life, feel no threat for our personal security and commonly when we feel good. From these considerations, any politician should realize that he represents the institution that is the supreme form of democracy and is responsible for what he does or says. An interesting result was the fact that wellbeing has a much more influence on the dependent variable. As we have mentioned already, this is due to the fact that the policies and laws influence mainly the people¶s wellbeing and so the expectations and the influence on these measures are greater. The singles variables that we have chosen, about immigrants and gender, involved two common issues. The first one is a rather recent one and was preceded by globalization and liberating the labor markets and the second one is about the difference in opinion of different sexes. As we stressed before, the policies towards immigrants prove to have an important weight in the common mentality of the people and this influence will rise for sure. The result about the difference in opinion between men and women was predictable. Those who are more interested in politics, namely the men, have less trust in parliament and its reforms. Finally, we hope that with this research project we provided some insight about a general issue such the peoples¶ trust in parliament. It might be very handy to use these results if in the future we¶ll be in the posture of a policymaker or to extend our research in order to get a better function for our dependent variable.
³Politicians are the same all over. They promise to build a bridge even where they is no river.´ Nikita Khrushchev
