Вы находитесь на странице: 1из 16

Applied Research Project and Report Applications for Economics & Management Course 6064

THE DETERMINANTS OF HAPPINESS


( Word Count: XXXX)

Declaration of Original Work In submitting this assignment 1. We declare that this written assignment is our own work and does not include (i) material from published sources used without proper acknowledgment or (ii) material copied from the work of other students. 2. We declare that this assignment has not been submitted for assessment in any other course at any university. 3. We have a photocopy or electronic version of this assignment in our possession.

___________________________

___________________________

___________________________

___________________________

The Determinants of Happiness I. SUMMARY It is the purpose of our research project to examine the determinants and influences of happiness. In order to prove our hypothesis that the level of happiness is mainly affected by the four major factors of family, health, job, and religion, we selected a range of indicator variables from the European Social Survey that we believed would model the influences above. Data screening, however, led to the situation that a significant fraction of our indicator variables describing the influence of job & work did not prove appropriate for factor analysis. As a result, we ended up with three latent factors indicating the determinants of family, health, and religion. These three factors were then implemented in a regression analysis in order to determine their impact on the level of happiness. In the end, our factors were complemented by the remaining independent variables to gain a broader picture and more solid results. Concerning the results of our research, we can state that age turned out to be insignificant in the determination of happiness while family and religion only showed little correlation with our dependent variable. This is surprising since we especially predicted family to be a major impact on the level of happiness. Finally, we concluded that happiness mainly depends on the status of health and job satisfaction where the latter one actually evolved as our strongest determinant of happiness. Summing up, we have to state, however, that our regression model is far from being reliable, accounting for roughly 18 percent of variation only. II. INTRODUCTION In the pursuit of happiness, the difficulty lies in knowing when you have caught up.1 (R. H. Grenville) The aim of this research paper is the investigation of the determinants of happiness. The drive to achieve happiness characterises most humans life. Nevertheless it can take totally different forms such as a general objective in life, a fundamental concept of our religious belief, or even part of our political system2. That is why we decided to make the determinants of happiness the focus of our project. Our detailed research question can be read as Which aspects in life determine happiness?. We identified four basic categories of variables that, in our opinion, exert a major influence on the level of happiness. These are job, family, health, and religion, which will all be underlined by a range of indicator variables describing them in greater detail. It is our hypothesis that we will end up with four latent variables along the line of our established categories, all of them having a significant influence on the level of happiness. In order to reach a conclusion we plan to set up a regression modelling the level of happiness. For the dependent variable, we decided on the question Taking all things together, how happy would you say you are? (the variable HAPPY) of the European Social Survey. Certain indicator factors assumed to be relevant in the determination of happiness will be identified and linked to our four established categories in order to model the behaviour of HAPPY. Factor analysis will be applied to produce latent variables from the range of indicator
1 http://www.inspirednow.com/goals--happiness.html 2 The pursuit of happiness is included as one of the undeniable rights of men in the United States Declaration of

Independence

The Determinants of Happiness variables. Finally, we hope to have gained a range of meaningful results in order to draw a number of solid conclusions from our analysis and come up with a rough explanation of the determination of happiness. It must be noted, however, that the complexity of the concept of happiness does not allow us to come up with a fundamental and comprehensive answer but rather an indication of the major trends and impacts. III. DATA DESCRIPTION The project will be based on the third round of the European Social Survey conducted in 2006 and 2007. Therefore we will deal with secondary data only. We narrowed down the data set to five countries, that is, Austria, Ireland, Norway, Portugal, and Ukraine. The study, founded by the European Commission in association with the European Science Foundation, is of crosssectional nature and can be considered a representative sample. The fact that it is funded by government and EU agencies allows the achievement of a high degree of data quality. Therefore, our concerns for accuracy, precision, and reliability are negligible. On the contrary, however, the apparent disadvantage is that it was designed for a purpose differing from our research question so validity problems may be faced.

3.1 The Dependent Variable As mentioned in the introduction, we have chosen to use HAPPY (Taking all things together, how happy would you say you are?) as the dependent variable. According to the ESS questionnaire, it is measured on a scale from zero to ten with zero being extremely unhappy and ten extremely happy. We will use this variable for our regression analysis in order to see how the range of chosen factors models the level of happiness. This allows us to check whether our hypothesis that happiness is mainly affected by the categories of job, family, health, and religion is correct. As shown below the variable is approximately normally distributed, proven by the fact that the Q-Q plot closely follows a straight line with a slope of one. According to our histogram, however, it seems that the distribution is actually slightly left skewed. This shape indicates that the majority of people interviewed are in general rather happy. Nevertheless, the Central Limit Theorem (which will be more deeply explained below) allows us to disregard the slight skewness.

The Determinants of Happiness 3.2 The Independent Variables Considering our four categories of influence, we came up with a number of indicator variables that we believed would add explanatory power for our independent factors. As a result, we identified a total of 15 indicating variables from the ESS questionnaire that allow us to draw conclusions about the four basic factors described above. Since we plan to use factor analysis to extract latent variables we made sure that each category contains at least three indicator variables. On the other hand, we tried to keep the number of variables in each category low in order to avoid overfitting, the situation in which a model is uncovered that cannot be observed in the population. If our plan turns out to be correct, we would end up with four different latent variables from the extraction process that represent the four categories illustrated in detail below.
STFJB STFECO JOB How satisfied are you with your job? How satisfied are you with the present state of the economy in your country? Are you satisfied with the balance between time on your job & time on other aspects? Do you feel that you get paid appropriately, considering your efforts & achievement? HEALTH How is your health in general? EVMAR PARTNER FAMILY Are or have you ever been married? Dou you live with your husband/wife/partner? Number of children ever given birth to/fathered Number of people living regularly as members o household

STFJBOT

NBTHCLD

PDAPRP

HHMMB

HEALTH

RLGBLG

HLTHHMP

AGE STFHLTH

Are you hampered in daily activities by illness/disability/infirmary/mental problems? Age of respondent, calculated State of Health Services in your country nowadays

RLGDGR

RELIGION Have you ever considered yourself as belonging to any particular religion or denomination? How religious are you?

PRAY

How often do you pray apart from religious services?

In terms of measurement scale and level of measurement, our independent variables show a wide range of different categories. These are summarized in the table below.
NUMERICAL Interval STFJB STFECO STFJBOT STFHLTH RLGDGR Ratio NBTHCLD AGE Nominal EVMAR PARTNER RLGBLG CATEGORICAL Ordinal PDAPRP HEALTH HLTHHMP PRAY

3.3 Data Screening of Independent Variables 3

The Determinants of Happiness

Since the underlying data originates from a survey, a few points have to be kept in mind. First of all, social desirability bias may cause questioned individuals to state answers that differ from their true opinions, feelings, or thoughts. Furthermore, random responses, accidental errors, and sabotage may lead to answers different from the truth. Lastly, due to the large number of interviews conducted there is a high chance of data entry errors. These thoughts have to be kept in mind throughout this project and especially highlight the need of preliminary data screening.
Descriptive Statistics N

Min

Max

Mean

St. Dev.

Skewness Statistic St. Error -.980 -.403 -.385 .308 .706 .381 4.288 .963 .610 -1.844 .257 -.332 .562 -.220 -.269 .033 .025 .033 .033 .025 .025 .030 .025 .025 .025 .025 .025 .025 .025 .025

Kurtosis Statistic St. Error .972 -.349 -.354 -.953 -1.502 -1.856 51.173 1.612 .310 2.423 -.774 -.642 -1.684 -.948 -1.643 .066 .050 .066 .066 .049 .050 .061 .049 .049 .050 .050 .050 .050 .050 .050

STFJB STFECO STFJBOT PDAPRP EVMAR PARTNER NBTHCLD HHMMB HEALTH HLTHHMP AGE STFHLTH RLGBLG RLGDGR PRAY Valid N

5571 9594 5561 5540 9828 9549 6477 9827 9820 9743 9671 9706 9760 9770 9671 3275

0 0 0 1 1 1 1 1 1 1 14.83 0 1 0 1

10 10 10 5 2 2 31 15 5 3 97.58 10 2 10 7

7.16 5.43 6.21 2.83 1.33 1.41 2.41 2.82 2.11 2.72 46.2597 5.44 1.36 4.97 4.37

2.101 2.335 2.274 1.137 .472 .491 1.423 1.531 .858 .547 17.96510 2.509 .481 2.971 2.484

Our sample size of 9,846 interviewed people is large enough to allow reference to the Central Limit Theorem. It states that if the sample size is sufficiently large, normality can be assumed. Therefore we can disregard the rule of thumb that non-normality exists if the absolute value of skewness and/or kurtosis is above 1. Especially NBTHCLD does not seem to have a normal distribution according to skewness and kurtosis figures, showing the highest values in both cases for all the variables. Also HLTHHMP shows critical values for both skewness and kurtosis. Furthermore, AGE, PARTNER, RLGBLG, and PRAY show kurtosis levels above 1. Nevertheless, the Central Limit Theorem allows us to disregard these concerns. In addition, a large sample size guarantees correlation reliability that will be required for the application of factor analysis. When analysing extreme values, we found a maximum value for NBTHCLD the number of children ever given birth to/fathered of 31 that can be clearly marked as impossible. For HHMMB, our first thoughts considered a maximum of 15 to be an improbable value as well. However, simple calculations showed that three generations of a family living together in the same household can easily reach a two-digit number. Hence, including a small buffer, we figured that 15 is a possible value. The screening for outliers will give us a more accurate picture and understanding of the questioned values. We defined outliers as values outside the 4

The Determinants of Happiness interval of +/- 3.5 standard deviations from the mean. According to calculations based on this rule of thumb and as predicted above, variables NBTHCLD and HHMMB show outliers. This is especially illustrated in the Box-and-Whisker Plots below.

In the case of HHMMB, the situation of outliers can be explained by the focus of the ESS survey; among other requirements, the ESS states that potential candidates for interviews must be members of a private household. The vast majority of private households in Europe consist of nuclear families. As a result, large households become outliers. As explained above, however, we believe that the maximum of 15 household members found in the data set is possible and therefore relevant in our investigation. As a result, outliers for HHMMB will be included in further analysis. In contrast, a maximum of 31 for NBTHCLD is considered to be impossible. Responses around the higher end of the interval cannot be serious or are due to errors and mistakes. As a result, we agreed to exclude any values for NBTHCLD that are above 10. Even with this restriction we keep several responses that are marked as outliers in the Box-and-Whisker Plot since the mean is slightly above 2 (or 2.41 to be exact). 3.4 Missing Value Analysis
Univariate Statistics: Missing Value Anaylsis Missing N nbthcld age stfjb stfeco stfjbot pdaprp evmar partner health hlthhmp stfhlth rlgblg rlgdgr pray 6463 9671 5571 9594 5561 5540 9828 9549 9820 9743 9706 9760 9770 9671 Mean 2.38 46.2597 Std. Deviation 1.240 17.96510 Count 3383 175 4275 252 4285 4306 18 297 26 103 140 86 76 175 Percent 34.4 1.8 43.4 2.6 43.5 43.7 .2 3.0 .3 1.0 1.4 .9 .8 1.8 No. of Extremesa Low 0 0 High 349 0

The Determinants of Happiness


Univariate Statistics: Missing Value Anaylsis Missing N nbthcld age stfjb stfeco stfjbot pdaprp evmar partner health hlthhmp stfhlth rlgblg rlgdgr pray 6463 9671 5571 9594 5561 5540 9828 9549 9820 9743 9706 9760 9770 9671 Mean 2.38 46.2597 Std. Deviation 1.240 17.96510 Count 3383 175 4275 252 4285 4306 18 297 26 103 140 86 76 175 Percent 34.4 1.8 43.4 2.6 43.5 43.7 .2 3.0 .3 1.0 1.4 .9 .8 1.8 No. of Extremesa Low 0 0 High 349 0

The data estimations show evidence on high proportion of missing values in some variables. In the variable PDAPRP (Do you get paid appropriately considering effort and achievement?) there is a proportion of 43.7 percent of missing values. We believe that this is explained by the fact that only people who currently have work were able to answer this question. Furthermore, some people could have found this topic sensible so they are unwilling to provide an answer. In our opinion, the same reasons also apply for STFJBOT and STFJB. We therefore decide to accept the high amount of missing values for these variables, keeping in mind that the proper responses given are only valid for people that have a job. For the variable NBTHCLD (Number of children ever given birth/fathered to?) the percentage of missing values was 34.4 percent. This could be explained by the fact that people who have never given birth to/fathered a child may have answered not applicable, no answer or simply left the question blank instead of stating 0, thereby being recorded as a missing value in the data set. As a consequence, there are a number of ways to deal with missing values. We have decided to use pairwise deletion. This means that we omit the case from just those parts of the analysis that use the variables for which values are missing. Pairwise deletion is possible for all of our variables and we think that using this method will provide us with the most accurate result. IV. FACTOR ANALYSIS As discussed above, it is our plan to end up with four different latent variables, each of them representing one identified category of indicator variables. To start off, however, we will run tests to check data suitability and sufficient correlation for data analysis. The Kayser-MayerOlkin Measure of Sampling Adequacy (KMO MSA), Bartletts Test of Sphericity, and the determinant of the correlation matrix will be used to obtain appropriate sets of indicator variables.

The Determinants of Happiness The KMO MSA applies to the data set as a whole and Kaiser-Meyer-Olkin Measure of Sampling Adequacy. .638 can be defined as the ratio of Bartlett's Test of Sphericity Approx. Chi-Square 13074.876 the sum of all squared df 105 correlations in the sample to the sum or all squared Sig. .000 correlations plus the sum of all squared partial correlations. Our obtained result of 0.638 is of medium quality. There would definitely be room for improvement but it is a value we can work with.
KMO and Bartlett's Test

Moving on, the Bartletts Test of Sphericity is a test statistic used to examine the hypothesis that the variables are uncorrelated in the population. In essence, it checks the null hypothesis that the matrix R is significantly different from the identity matrix that is, no correlation among the variables. Since the level of significance is 0 (and thus below 0.05), we can reject the null hypothesis and conclude that there is correlation between at least two variables. Furthermore, we have to check for the absence of multicollinearity and for a sufficient common variance. For the multicollinearity test, we produced a correlation matrix. Here, our point of interest is the determinant. A determinant equal to 0 would indicate a problem of multicollinearity in the correlation matrix. Even though it shows a low value with 0.025, we can proof absence of multicollinearity since its value is greater than zero. In order to check that our data shows a sufficient level of common variance we apply the measures of sampling adequacy for each variable (MSA) and the particular communalities. The concept of MSA is the preliminary measure of common variance relative to other sources of shared variance. In a good factor model, the proportion of common variance should be higher than the sum of the squared partial correlations. The relevant values are obtained from the anti-image matrix that is not fully included in the report but summarized below for the benefit of simplicity and clarity.
Summary: Anti-Image Matrix Variable Anti-Image Correlation STFJB 0.723 STFECO 0.730 STFJBOT 0.679 PDAPRP 0.778 EVMAR 0.617 PARTNER 0.502 NBTHCLD 0.502 HHMMB 0.402 Variable HEALTH HLTHHMP AGE STFHLTH RLGBLG RLGDGR PRAY Anti-Image Correlation 0.661 0.651 0.570 0.611 0.801 0.713 0.719

As it can be seen, THE MSA results are strong. The majority of variables are satisfactory, almost all of them being above the critical level of acceptance at 0.5. The only exception is HHMMB the number of household members - with 0.402. Also PARTNER and NBTHCLD, show low values, both at 0.502; however, both of them are above 0.5 and can therefore be included in further analysis. HHMMB, on the other hand, is considered inadequate according to the rule of thumb. Therefore we decided to exclude it from further multivariate analysis.
Extraction Communalities Initial STFJB .268 Extraction .428

The Determinants of Happiness In order to gain further insight the estimates of communalities are evaluated. Communality is the proportion of common variance given the extracted factor model. Hence, we can check whether the extracted factors explain an adequate amount of common variance
STFECO STFJBOT PDAPRP EVMAR PARTNER NBTHCLD .173 .236 .150 .508 .374 .224 .198 .381 .229 .648 .566 .101

Here, the results are not as strong as we HHMMB .408 .453 hoped. We will use only those variables for HEALTH .397 .713 which the extracted factors explain an .327 .443 HLTHHMP adequate amount of variance given the .605 .957 extracted factor model. Hence, any variables AGE .103 .087 that show extraction communality lower than STFHLTH .386 .473 0.4 are excluded. The value of 0.4 indicates RLGBLG that 40% of common variance is modelled. RLGDGR .513 .663 According to the table above, STFECO, PRAY .515 .675 STFJBOT, PDAPRP, NBTHCLD, and STFHLTH do not meet these criteria. This Extraction Method: Principal Axis Factoring. contradicts our hypothesis. We believed that the job factors as well as the number of children would play an essential role in our model. Now, only one variable from the job category is identified as appropriate; in theoretical terms, it is the only variable from the range of job indicators for which the extracted factors explain an adequate amount of common variance. Hence, we have to adjust the rest of the procedure as well as our hypothesis. For the Job Factor, there are too little indicators available now to guarantee a satisfactory modelling. Therefore we decided to treat STFJB as an independent variable and remove it from the factor analysis as well. However, it will be added again to the regression during the final phase of our multivariate analysis. As a result, we are left with the three categories Family, Health and Religion, hopefully leading to three extracted factors of significance. In addition, at this point of multivariate analysis, we also agreed to exclude the variable AGE from factor extraction. This decision is based on the fact that this variable shows a different scale set-up in comparison to the rest. All other variables are measured on an anchored scale. Finally, we run KMO MSA, Bartletts Test of Sphericity, individual MSA and communality analyses again to make sure that the requirements of suitable data, the absence of multicollinearity and a sufficient level of correlation or common variance, are still met in our adjusted set of indicator variables. We observed that KMO MSA has decreased slightly from 0.638 to 0.634 while the Bartletts Test of Sphericity still shows a significance level of 0 but at a much lower number of degrees of freedom. Hence, we can conclude that we deal with sufficient correlation in the data set in order to extract factors. The determinant of the correlation matrix has increased significantly to 0.164. Thereby problems of multicollinarity can be ruled out. MSA for each of the remaining variables from the anti-image show negligible variations compared to previous values, ranging from 0.502 to 0.792. Hence, we obtained satisfactory results from our final MSA values so that we can conclude a sufficient level of common variance. Negligible changes in communalities confirmed that finding. The Final Extraction Communalities Table shows the new communalities for the adjusted range of indicators after the inappropriate variables have been excluded.
Summary: Final Anti-Image Matrix Variable Anti-Image Correlation EVMAR 0.562 PARTNER 0.478

The Determinants of Happiness Initial


HEALTH HLTHHMP RLGBLG RLGDGR PRAY 0.525 0.531 0.783 0.689 0.698

Final Extraction Communalities Extraction EVMAR .507 0.555 PARTNER .349 0.496 HEALTH .351 .606 HLTHHMP .325 .514 RLGBLG .384 .474 RLGDGR .507 .688 PRAY .500 .653

When it comes to factor extraction, we have decided on the common factor model since our research project is of social nature. The common factor model assumes that each variable has some unique variance and some variance that is common with the other variables. The used method is principal axis factoring. In order to define the number of factors to extract from an initial solution, we implemented two restrictions: 1. According to the Kaisers rule (or latent root criterion), the minimum value of the eigenvalues is set to 1. This is explained by the fact that a developed factor makes sense only as long as it gives more information than single variables. 2. A minimum of 50 percent of the variance explained across all the observed variables.
Total Variance Explained Factor Initial Eigenvalues Total d1 i2 m 3 e 4 n 5 s 6 i o7 n 0
Extraction Method: Principal Axis Factoring.

Extraction Sums of Squared Loadings Cumulative % 33.267 55.540 75.509 82.543 89.092 95.285 100.000 Total 1.935 1.124 .926 % of Variance 27.644 16.060 13.223 Cumulative % 27.644 43.704 56.927

% of Variance 33.267 22.273 19.970 7.033 6.549 6.193 4.715

2.329 1.559 1.398 .492 .458 .434 .330

As can be seen from the graph above, three factors satisfy our two requirements, explaining a satisfactory amount of 56.927 percent of variance. The Scree Test, a plot displaying the eigenvalues in decreasing order over the number of factors, confirms this result: Three extracted factors show an eigenvalue greater than 1. After the third factor, the graph indicates a deep fall in the level of eigenvalue to a point below 0.5, which is clearly out of our predetermined range. Hence, it seems like our modified hypothesis has become true. To aid interpretation, we applied a rotation factor matrix. Rotation is a geometric transformation of the factor matrix so that the 9

The Determinants of Happiness relationship between the variables is still the same but orientation of the factors is different. In particular, we used Kaisers Varimax Rotation that minimises the complexity of factors by maximising the variance of the loadings on each factor.

Rotated Factor Matrixa Factor 1 are or ever been married lives husband/wife/partner at f4 subjective general health by illness/disability/infirmary/ mental problem belonging to particular .683 -.827 -.004 -.053 .062 .086 -.036 .051 religion or denomination how religious are you at religious services
Extraction Method: Principal Axis Factoring. Rotation Method: Varimax with Kaiser Normalization. a. Rotation converged in 4 iterations.

2 .162 -.081 -.776 .715

3 .710 .699 -.060 -.005

.158

with -.011 -.027

hampered in daily activities .043

how often pray apart from .804

Factor 1 is closely correlated to the indicator factors related to religion. These are Belonging to a particular religion or denomination, How religious are you, and How often do you pray apart from religious services. Therefore, we can summarize this group under the factor of religion. Secondly, Are or ever married and Living with a partner both show high correlation with factor 2, renamed as the factor of family. Finally, we can identify the factor of health from Subjective general health and hampered in daily activities by illness in form of the third factor. We will refer to it as the factor of wellbeing in order to avoid confusion with the indicator variable HEALTH.

10

The Determinants of Happiness Now we can move on to factor creation. Here fore we used factor scores through the regression method. The factor score for each factor extracted is a linear combination of all the variables used in factor analysis, weighted by their correlation with the factor. V. REGRESSION ANALYSIS As an introduction to the regression analysis, let us quickly go back to our research question. We planned to model the level of happiness by multiple independent variables. Hence, we try to build a multiple linear regression of the variable HAPPY over our three latent variables created by factor analysis and the independent variables of STFJB and AGE. Summarising this algebraically, we would obtain the following regression equation: HAPPY = 0 + 1 Religion Factor + 2 Family Factor + 3 Wellbeing Factor + 4 Age&Household Factor + 5 STFJB + 6 AGE + Ei where 0 is the constant, n the coefficents, and Ei the error term (for n = 1,2,..,5) To start off, we will build a regression based solely on the latent factors extracted from factor analysis and then add the indicating variables STFJB and AGE at a later point in order to avoid complexity of the model in the present stage of the analysis.

5.1 The Basic Model


Model Summary: Basic Model Model R
1 .307
a

R Square
.094

Ad. R Square
.094

Std. Error of the Estimate


1.780

DurbinWatson
1.817

ANOVAb Model 1 Regression Residual Total Sum of Squares df 3006.342 28836.221 31842.563 3 9106 9109 Mean Square 1002.114 3.167 F 316.451 Sig. .000a

a. Predictors: (Constant), REGR factor score 4 for analysis 1, REGR factor score 3 for analysis 1, REGR factor score 2 for analysis 1, REGR factor score 1 for analysis 1 b. Dependent Variable: how happy are you

Coefficients: Complete Model Unstandardized Coefficients B St. Error (Constant) 7.308 .019 Religion Family Wellbeing -.130 .650 -.179 .020 .022 .023

Standardized Coefficients Beta

Sig.

95% Confidence Interval for Lower Upper Bound Bound 7.272 -.170 .606 -.224 7.345 -.090 .693 -.135

Collinarity Statistics Tolerance FIV

391.892 -.063 .293 -.078 -6.345 29.335 -7.845

.000 .000 .000 .000

.998 .999 .998

1.002 1.001 1.002

11

The Determinants of Happiness We gain the regression for HAPPY as the dependent variable over our three extracted factors: HAPPY = 7.309 - 0.063 Religion Factor + 0.293 Family Factor 0.078 Wellbeing Factor R squared, the percentage of HAPPY modelled by our regression, of 0.094 is very low. This indicates a poor goodness-of-fit of the model and means that our three factors extracted account for only 9 percent (rounded to next percentage) of the total level of happiness. Using a significance level of 0.05, all of our extracted factors are nevertheless significant since t-test significance is 0 in all cases and none of the 95% confidence levels includes 0. Referring to the F-Test for overall significance from the ANOVA table, we can also state that, in general, our basic model is significant. In conclusion, except for the very poor fit, our model seems to be all right. Now, adding STFJB and AGE will complete our model.

5.2 The Complete Model


Model Summary: Complete Model Model R R Square 1 ANOVAb Model 1 Regression Residual Total Sum of Squares df 3233.189 15052.950 18286.140 5 5226 5231 Mean Square 646.638 2.880 F 224.496 Sig. .000a .420
a

Ad. R Square .176

Std. Error of the Estimate 1.697

DurbinWatson 1.877

.177

a. Predictors: (Constant), how satisfied with job, REGR factor score 3 for analysis 1, REGR factor score 1 for analysis 1, REGR factor score 2 for analysis 1, REGR factor score 4 for analysis 1 b. Dependent Variable: how happy are you

Coefficients: Complete Model Unstandardized Coefficients B St. Error (Constant) 5.515 .106 Religion Family Wellbeing STJFB AGE -.085 -.144 .534 .261 -.002 .026 .031 .031 .011 .002

Standardized Coefficients Beta

Sig.

95% Confidence Interval for Lower Upper Bound Bound 5.307 -.137 -.206 .473 .239 -.005 5.722 -.033 -.083 .595 .284 .001

Collinarity Statistics Tolerance FIV

52.096 -.041 -.063 .241 .294 -.016 -3.219 -4.590 17.170 22.806 -1.047

.000 .001 .000 .000 .000 .295

.950 .838 .802 .950 .676

1.053 1.193 1.248 1.053 1.480

Summarized, our regression line can now be expressed as: HAPPY = 5.515 - 0.041 Religion Factor 0.063 Family Factor + 0.241 Wellbeing Factor + 0.294 STFJB 0.016 AGE 12

The Determinants of Happiness

Adding the independent variables STFJB and AGE to our regression increases the reliability to 0.177 in comparison to our poorly fitting basic model. Hence, it roughly explains 18 percent of the variance in the level of happiness. Even though this is a substantial increase, the value is still too low to speak of a reliable model. In case of overall significance of the model, it is also confirmed this time by a F-test significance level of 0, which is below our reference value of 0.05. Referring to the coefficients table of the complete model, it can be concluded that STFJB and our extracted factors are significant. The significance test for AGE, in contrast, indicates a value of 0.295 > 0.05. Therefore we can conclude that it is not significant in our regression. Finally, low Variance Inflator Factor Indices and high tolerance levels allow us to rule out the concern of multicollinarity. Examining the single variables more closely, we realise that two of the variables indicate a positive influence on the level of happiness while the other three exert a negative one. The first factor, Religion, is negatively correlated. The majority of its indicator factors, PRAY and RLGBLG, have an anchored scale attached that relates a high degree of religious belief to low numbers and vice-versa. Hence the higher the belief in religion, the lower the negative influence on the level of happiness. Therefore the negative correlation makes sense since religion is an essential factor in the life of many people that provides them with strength and hope. Its coefficient, however, is rather small (0.041), indicating only a small impact on the level of happiness. Another factor of negative influence on happiness is the one of family. Here the indicator variables EVMAR and PARTNER answer the questions for marriage and living status with partner first with Yes and then with No. Hence, the lower the factor of family, the closer and more intimate the relationship, and the greater the level of happiness. Also this factor shows a rather limited influence on happiness due to its coefficient of 0.063. The factor of wellbeing, on the other hand, shows a much greater correlation with the level of happiness; here the influence is of positive nature so the higher the factor, the higher the level of happiness. Also STJB is of positive influence as it can be easily explained: the more satisfaction a person receives from his job, the happier he is. This variable actually shows the highest level of influence out of all our dependent variables. Finally, AGE has a negative influence of low relevance. This would indicate that the older a person gets, the unhappier he becomes. Finally, the fact that we use Ordinary Least Squares method to obtain our coefficient estimates requires us to check the assumptions for standard multiple regression in order to make sure that our model is actually valid. These assumptions are: 1. Our factors and the error terms Ei are not correlated 2. The expected value of Y, the level of happiness, is a linear function of our independent variables 3. The error terms are random variables with mean 0 4. In addition, the error terms have a constant variance to insure homoschedasticity 5. The random error terms are not correlated with one another 6. The error terms are normally distributed If these assumptions hold true then our Ordinary Least Square estimators are, according to the Gauss-Markov Theorem, best linear unbiased estimators; that is, they are a linear function of our dependent variable, have an expected value that equals the coefficients, and show the smallest variance in the class of linear unbiased estimators, so that they are considered the most efficient.

13

The Determinants of Happiness

We can conclude from the distribution of the standardized residuals and from the Normal Probability plot that the error term is approximately normal. This is evident from the fact that the observed residuals in our P-P plot approximately follow the straight line that would represents a normal distribution. However we observe that the distribution is slightly left skewed. This is not a problem since the number of observation is high (N=9846), and therefore our analysis is statistically reliable due to the Central Limit Theorem. In addition, the histogram indicates that error terms show a mean close to 0. We also need to assert that we have homoschedasticity in order to satisfy assumption number four. We do this by checking the scatter plot. Our variable should look like a random array of dots evenly dispersed around zero in order to meet the assumption of homoschedasticity. This applies to the case shown in our scatter plot below.

V. CONCLUSION Returning to our final regression model, we can draw certain conclusions from our research project. According to our investigation, job satisfaction and the status of health are the two essential determinants of the level of happiness in life. Religious belief and family relationships also play a role, even though a significantly weaker one. Surprisingly, the age has been proven to be insignificant in the determination of happiness. A reason for that may be that the benefits and costs of a higher age, such as children, marriage, higher income 14

The Determinants of Happiness and/or responsibility in job but weaker health, are already represented in other categories of variables, such as in the factors of family or health. During factor analysis, we modified our hypothesis in such a way that we now expected three major factors representing the areas of health, family, and religion. In so far, this has been proven correct since we were able to extract exactly these three factors. However, our hypothesis is contradicted by the finding that health and job satisfaction represent such a major share of the determinants of happiness in life. We rather expected roughly equally influence from all three, or originally four, areas of focus. Summing up, it is needless to say the model and our results are far from being perfect. Therefore we want to indicate a few additional points that we believe would lead to a substantial result. First of all, an extended range of independent variables would significantly enhance our model and, especially, increase our reliability. Due to the limited scope of this project, we only focused on four major areas, and then, on a few indicator variables only. The list of possibly significant variables in endless and varies for every individual human being. Second, we have to recall the limitations on our sample. Our study only focuses on five countries that, since they are all European, show relatively common cultural, social, and moral standards and values. A broader picture would provide us with more reliable and comprehensive results. Third, we were forced to exclude a high number of indicator variables during multivariate analysis. Therefore we ended up with two factors that were based on only two indicator variables and thus on a number that is below three which is generally considered as sufficient and satisfactory. A broader range of indicator variables would provide us with more accurate and precise factors.

VI. LIST OF REFERENCES 6.1 Data Source European Social Survey; Third Round, 2006/2007; London; UK 6.2 Text Books Carlson, Newbold, Thorne; Statistics for Business and Economics; 7th Edition, XXX; Pearson Education; Upper Saddle River, NJ, USA Norusis, Marija; SPSS Statistics 17.0, Statistcal Procedures Companion; 2008; Prentice Hall; Upper Saddle River, NJ, USA 6.3 Webpages http://www.inspirednow.com/goals--happiness.html (last assessed 31st May, 2010)

15

Вам также может понравиться