All pages

Chapter 5: Happiness in Europe

The purpose of this chapter is to use data from the European Social Survey 2010 to develop a variance component model for happiness with explanatory variables at the individual level as well as the country level, and to interpret the findings.

Introduction

A common conclusion in quality of life research is that subjective well-being is only very weakly related to material living conditions (for reviews, see Arthaud-Day, Marne and Near 2005, Diener and Biswas-Diener 2002). Whereas the weak effects of material living conditions are commonly stressed in quality of life research, the situation is quite different in research on health, including subjectively assessed health. In this field, income, social class and other indicators of material living conditions are generally considered to be very important determinants.

In light of the highly divergent views of the importance of material living conditions to health and happiness, a recent study by Subramanian, Kim and Kawachi (2005) is very interesting. They performed parallel analyses of happiness and self-assessed health in the United States and found that these variables were equally strongly related to income and education. In addition, they found that happiness was much more strongly related to marital status than health. Eikemo and Ringdal (2008) replicated the American study based on data from the European Social Surveys for 2002 and 2004, with parallel findings. In the example, we will restrict the analysis to individual and contextual determinants of happiness. For a better overview of the field of well-being in general, see the Well-being module.

Page 1

Download data for Chapters 5 and 6

ESS Multilevel Data is an online service that makes it possible to download ESS survey data together with data for countries and regions. You are now going to visit this site and create and download the data you need to do the exercises in Chapters 5 and 6.

We recommend you to keep the instructions on this page open in one browser window, and to open the ESS MD in another window. This enables you to switch between windows, and it will definitely be useful for those of you who are not able to memorize a whole page at one glance.

It will be useful for your later searches for data to spend some time on ESS MD and to familiarize yourself with the resource. In order to keep the pace up, however, you can also start with these stepwise instructions:

  1. Click this link and open ESS MD in a new window.
  2. There are two gateways into the data. Please select the one called ‘ESS Multilevel Download’.
  3. If you are a registered user of ESS data, please enter your e-mail address. If you are not already registered, you will have to fill in a short form. Registration is free and fast.
  4. You will need data from Round 5 of the ESS, and you will need to download respondents from all the countries. Please tick for all the Round 5 countries and click ‘Next’.
  5. The tree shows all the available levels: a) country b) region (ess region, nuts1, nuts2 and nuts3) and c) individual (the ESS respondents). You will need data for individuals and countries for Chapters 5 and 6.
  6. Start with the individual level variables. Click the ‘+’ in front of ‘ESS5-2010, ed. 3.2 - Individual level’.
    1. Click the ‘+’ in front of the variable group ‘Media and social trust’ and tick the variable:
      - ‘Most people can be trusted or you can’t be too careful’
    2. Click the ‘+’ in front of the variable group ‘Politics’ and tick the variables:
      - ‘Trust in country’s parliament’
      - ‘Trust in the legal system’
      - ‘Trust in the police’
      - ‘Trust in politicians’
      - ‘Trust in political parties’
      - ‘How satisfied with present state of economy in country’
      - ‘How satisfied with the national government’
      - ‘How satisfied with the way democracy works in country’
    3. Click the ‘+’ in front of the variable group ‘Subjective well-being, social exclusion, religion...’ and tick the variables:
      - ‘How happy are you’
      - ‘How often people meet socially with friends, relatives or colleagues’
    4. Click the ‘+’ in front of the variable group ‘Gender, Year of birth and Household grid’ and tick the variables:
      - ‘Gender’
      - ‘Age of respondent’
    5. Click the ‘+’ in front of the variable group ‘Socio-demographics’ and tick the variables:
      - ‘Interviewer code, lives with husband/wife/partner’
      - ‘Interviewer code, lives with husband/wife/partner’ (there are two variables with similar labels)
      - ‘Highest level of education, ES - ISCED’
      - ‘Years of full-time education completed’
      - ‘Household’s total net income, all sources’
      - ‘Feeling about household’s current net income’
  7. Proceed with the country level variables. Click the ‘+’ in front of ‘Country level’.
    1. Click the ‘+’ in front of the variable group ‘Demography’, the ‘+’ in front of the group ‘Population size’ and tick the variable
      - ‘Population size 2008’
    2. Click the ‘+’ in front of the variable group ‘Economy’, the ‘+’ in front of the group ‘National economic accounts’ and tick the variable
      - ‘GDP per capita 2008’
    3. Click the ‘+’ in front of the variable group ‘Economy’, the ‘+’ in front of the group ‘Gini coefficient’ and tick the variable
      - ‘Gini coefficient after taxes - Total population 2005’
    4. Click the ‘+’ in front of the variable group ‘Economy’, the ‘+’ in front of the group ‘Social expenditure’ and tick the variable
      - ‘Social expenditure as a percentage of GDP 2005’
    5. Click the ‘+’ in front of the variable group ‘Composite measures’, the ‘+’ in front of the group ‘Human development index’ and tick the variable
      - ‘Human development index, HDR 2007’
    6. Click the ‘+’ in front of the variable group ‘Composite measures’, the ‘+’ in front of the group ‘Gender empowerment measure' and tick the variable
      - ‘Gender empowerment measure 2009’
    7. Click the ‘+’ in front of the variable group ‘Composite measures’, the ‘+’ in front of the group ‘Multidimensional poverty index (MPI)' and tick the variable
      - ‘Multidimensional Poverty Index (MPI) 2011’
    8. Click the ‘+’ in front of the variable group ‘Composite measures’, the ‘+’ in front of the group ‘Freedom in the world’ and tick the variables
      - ‘Freedom in the world - Political rights 2008’
      - ‘Freedom in the world - Civil liberties 2008’
      - ‘Freedom in the world - Status 2008’
    9. Click the ‘+’ in front of the variable group ‘Health’, the ‘+’ in front of the group ‘Life expectancy’ and tick the variable
      - ‘Life expectancy at birth (years) both sexes 2008’
    10. Click the ‘+’ in front of the variable group ‘Health’, the ‘+’ in front of the group ‘Health expenditure’ and tick the variables
      - ‘General government expenditure on health as a % of total government expenditure 2007’
      - ‘Per capita total expenditure on health 2007’
    11. Click the ‘+’ in front of the variable group ‘Crime’, the ‘+’ in front of the group ‘Corruption Perceptions Index’ and tick the variable
      - ‘Transparency International - Corruption Perceptions Index 2008’
  8. Select your preferred data format in the drop-down box at the top right-hand corner of the page and click ‘Download’. Save the file to your computer.
  9. If you find theese steps a bit too tedious, you may follow this link, select the preferred data format and download the file.
Page 2

Preparing the data

Please open the data file you downloaded above. In order to have a better point of departure for the analyses, you need to work a bit more with the variables in this file. If you find it difficult to do the exercises, you may look at and copy from the syntax at the bottom of this page.

  1. Open the file with data.

The dependent variable

In all ESS rounds, happiness is measured by the following question: ‘Taking all things together, how happy would you say you are? Please use this card.’ The card shows a scale from 0, ‘extremely unhappy’ to 10, ‘extremely happy’. This variable can be regarded as having an ordinal scale, but with 11 categories. We will treat the raw scores as having been measured at the interval level.

  1. Perform a frequency analysis of the variable ‘happy’.

The explanatory variables

We want to have the following variables available for the analysis:

Demographic variables: age, age centered, age squared, gender.

  1. Find 'mean age' in the data and compute a centered version of 'Age'.
  2. Use the centered age variable and compute a squared version of it.
  3. Recode ‘Gender’ into a dummy variable with 0 = man and 1 = female.

Socioeconomic variables: education in years, education level, household income, evaluation of current income.

Years of education (eduyrs) can be used directly, although the variable is not the best indicator of educational attainment in the ESS. It also has some problematic high values. The best indicator of education is eisced.

  1. Recode ‘eisced’ into a new variable with three levels: primary (1,2), secondary (3,4) and tertiary (5,6,7).
  2. Create dummy indicators for the three levels of education.
  3. Recode ‘hincfel’ into a new dummy variable called ‘copeinc’. Values 1 and 2 for ‘hincfel’ = 1 for ‘copeinc’, and 3 and 4 = 0.
  4. The variable ‘hinctna’, ‘Household’s net income, all sources’, is a variable with ten categories, in addition to missing. Our suggestion is to simplify this and compute a new variable, 'hinc4', with the following values: low income - medium income - high income - missing. The categories defining for example 'low income', are not the same across all countries, and we suggest to recode one country at the time. Please open the syntax at the bottom of this page, copy the 'Exercise 9' syntax and compute the simplified variable, ‘hinc4’.
  5. Compute dummies for each of the four values on 'hinc4'.

Social support variables: living together with others or alone, meeting with friends.

  1. The variable icpart2 distinguishes between persons who live together with a husband, partner or cohabitant and those who live alone. Recode ‘icpart2’ into a variable called ‘cohab’, where cohabitants are coded 1 and all others are coded 0.
  2. Meeting socially with friends, neighbours and co-workers is measured by the schmeet variable. Recode ‘schmeet’ into a new variable called ‘social’. Set the values 1 - 5 to 0, and 6 and 7 to 1.

Country level variables: Welfare state classification.

Welfare state classification based on Ferrera
Nordic Liberal Continental Southern Eastern
Sweden

Norway

Denmark

Finland
Ireland

United Kingdom

Netherlands

Luxembourg

Germany
Switzerland

Belgium

Austria

France
Spain

Israel

Italy

Greece

Turkey

Portugal

Cyprus
Russia

Estonia

Czech Republic

Poland

Croatia

Hungary

Latvia

Rumenia

Slovenia

Slovakia

Ukraine

The welfare state classification is made by recoding the country variable (cntry). The classification is inspired by Esping-Andersen (1990), but has been expanded by adding a Southern and an Eastern welfare regime (Arts, W. and Gelissen, J. 2002).

  1. Recode 'cntry' into a new variable 'welstate'.
Users of SPSS: Please click here, select the syntax and copy it into your syntax.

* This syntax prepares data for chapters 5 and 6 in the EduNet module 'Multilevel models'.
* Data downloaded from ESS MD (edition 2 of the ESS 5, http://ess.nsd.uib.no/ess/essmd/).

*Exercise 1.

*Please remember to change the path to the location where your dataset is and, later, to write the path to the location where you would like to save your work.

GET FILE='c:\data\ESSMDw5e2.sav'.

*Exercise 2.

fre happy.

*Exercise 3.

* Descriptive analysis - in order to find the mean value of age.
* Compute age centred.

desc agea.
compute agec = agea - 48.
var labels agec 'Age centred: age- 48'.

* Exercise 4.

* Compute centered age squared.
* Descriptive analysis of the three age variables.

compute agec2 = agec*agec.
var labels agec2 'Age centred squared'.
desc agea agec agec2.

* Exercise 5.

* Compute dummy variable, female.
* Check that the frequencies are identical.

compute female = gndr-1.
var label female 'Female gender from gndr'.
fre female gndr.

* Exercise 6.

* Look at the frequencies for the variable ‘years of education’.
* Note that this variable is not considered to be the best indicator of education.
* Note further that there are about 50 persons reporting more than 30 years of education.
* The best alternatives are edulvlb and eisced.
* eisced has seven values, recode to three levels.
*Check the frequencies of the variables.

fre eduyrs.
recode eisced (1,2=1)(3,4=2)(5,6,7=3)into edlev3.
var labels edlev3 'Education in three level from eisced'.
value labels edlev3 1 'Primary' 2 'Secondary' 3 'Tertiary'.
fre eisced edlev3.

* Exercise 7.

* Create dummy indicators for the three levels of education.
* Check the frequencies of the variables.

recode edlev3 (1=1)(2,3=0)into primed.
recode edlev3 (2=1)(1,3=0)into seced.
recode edlev3 (3=1)(1,2=0)into terted.
var labels primed 'Primary education, Edlev=1'.
var labels seced 'Secondary education, Edlev=2'.
var labels terted 'Tertiary education, Edlev=3'.
fre edlev3 primed seced terted.

* Exercise 8.

* Recode hincfel - Feeling about household's current income.
* Check the frequencies of these two variables.

recode hincfel (1,2=1)(3,4=0)into copeinc.
var labels copeinc 'Living comfortably or coping on present income'.
fre copeinc hincfel.

* Exercise 9.

* Note that Portugal lacks household income.
* Recode hinctnta into hinc4.
* Check the frequencies of these two variables.

do if (cntry='BE').
recode hinctnta (1 thru 5=1)(6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='BG').
recode hinctnta (1=1)(2 thru 6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='CH').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='CZ').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7 thru 10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='CY').
recode hinctnta (1 thru 2=1)(3,4,5=2)(6 thru 10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='DE').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7 thru 10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='DK').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='EE').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='ES').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7 thru 10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='FI').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='FR').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='GB').
recode hinctnta (1 thru 3=1)(4,5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='GR').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='HR').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='HU').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='IE').
recode hinctnta (1,2=1)(3,4,5=2)(6,7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='IL').
recode hinctnta (1 thru 4=1)(5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='NL').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='NO').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='PL').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='RU').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='SE').
recode hinctnta (1 thru 4=1)(5,6,7,8=2)(9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='SI').
recode hinctnta (1 thru 3=1)(4,5,6=2)(7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='SK').
recode hinctnta (1 thru 4=1)(5,6,7=2)(8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='TR').
recode hinctnta (1,2=1)(3,4,5=2)(6,7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
do if (cntry='UA').
recode hinctnta (1,2=1)(3,4,5=2)(6,7,8,9,10=3)(77,88,99=4)into hinc4.
end if.
var labels hinc4 'Household income in 3 cat + missing from hinctnta'.
value labels hinc4 1 'Low' 2 'Medium' 3 'High' 4 'Missing'.
fre hinctnta hinc4.

* Exersice 10.

* Recode hinc4 into dummies.
* Check frequencies.

recode hinc4 (1=1)(2,3,4=0)into lowinc.
recode hinc4 (2=1)(1,3,4=0)into medinc.
recode hinc4 (3=1)(1,2,4=0)into highinc.
recode hinc4 (4=1)(1,2,3=0)into missinc.
var labels lowinc 'Low household income, hinc4=1'.
var labels medinc 'Medium householdincome, hinc4=2'.
var labels highinc 'High household income, hinc4=3'.
var labels missinc 'Missing income, hinc4=4'.
fre hinc4 lowinc to missinc.

* Exercise 11.

* Recode iscpart2 into cohab, living with husband, wife, partner or cohabiting.
* Check frequencies.

recode icpart2 (1=1)(2=0)into cohab.
var labels cohab 'Living with husband wife partner or cohabiting'.
fre cohab.

* Exercise 12.

* Recode schmeet into social, meeting socially.
* Check frequencies.

recode sclmeet (1,2,3,4,5=0)(6,7=1)into social.
var labels social 'Meet sdeveral times a week with friends, relatives collegues'.
fre sclmeet social.

* Exercise 13.

* Recode cntry into welstate.
* Check frequencies.

RECODE cntry ('SE', 'NO', 'DK', 'FI' =1) ('IE', 'GB'=2) ('NL','LU', 'DE'=3)('CH', 'BE','AT', 'FR'=3)('ES','IL','IT', 'GR', 'TR','PT', 'CY'=4) ('RU', 'EE', 'BG','CZ','PL','HR', 'HU','LV','RO','SI','SK','UA'=5) INTO welstate .
var labels welstate 'Welfare state classification based on Ferrera'.
value labels welstate 1 'Nordic' 2 'Liberal' 3 'Continental' 4 'Southern' 5 'Eastern'.
fre welstate.

*Save the changes.

SAVE OUTFILE='c:\data\Multilevel.sav'
/COMPRESSED.
Users of Stata: Please click here, select the syntax and copy it into your syntax.

* This syntax prepares data for chapters 5 and 6 in the EduNet module 'Multilevel models'*
* Data downloaded from ESS MD (edition 2 of the ESS 5, http://ess.nsd.uib.no/ess/essmd/)*

*Exercise 1*

*Please remember to change the path to the location where your dataset is and, later, to write the path to the location where you would like to save your work.

use "c:\data\ESSMDw5e2.dta", clear

*Exercise 2*

tabulate happy

*Exercise 3*

* Descriptive analysis - in order to find the mean value of age*
* Compute age centred*

summarize agea
generate agec = agea - 48
label variable agec "Age centered_ age- 48"

* Exercise 4*

* Compute centered age squared*
* Descriptive analysis of the three age variables*

generate agec2 = agec*agec
label variable agec2 "Age centered squared"
sum agea agec agec2

* Exercise 5*

* Compute dummy variable, female*
* Check that the frequencies are identical*

gen female = gndr-1
label variable female "Female gnder from gndr"
sum female gndr

* Exercise 6*

* Look at the frequencies for the variable ‘years of education’*
* Note that this variable is not considered to be the best indicator of education*
* Note further that there are about 50 persons reporting more than 30 years of education*
* The best alternatives are edulvlb and eisced*
* eisced has seven values, recode to three levels*
*Check the frequencies of the variables*

sum eduyrs
recode eisced (1/2=1) (3/4=2) (5/7=3), gen(edlev3)
replace edlev3 = .n if eisced == 55
label define edlev 1 "Primary" 2 "Secondary" 3 "Tertiary"
label values edlev3 edlev
tab1 eisced edlev3

* Exercise 7*

* Create dummy indicators for the three levels of education*
* Check the frequencies of the variables*

recode edlev3 (1=1) (2/3=0), gen(primed)
recode edlev3 (2=1) (1/3=0), gen(seced)
recode edlev3 (3=1) (1/2=0), gen(terted)
label variable primed "Primary education, Edlev=1"
label variable seced "Secondary education, Edlev=2"
label variable terted "Tertiary education, Edlev=3"
tab1 edlev3 primed seced terted

* Exercise 8*

* Recode hincfel - Feeling about household's current income*
* Check the frequencies of these two variables*

recode hincfel (1/2=1) (3/4=0), gen(copeinc)
label var copeinc "Living comfortably or coping on present income"
tab1 copeinc hincfel

* Exercise 9*

* Note that Portugal lacks household income*
* Recode hinctnta into hinc4*
* Check the frequencies of these two variables*

recode hinctnta (1/5=1) (6/7=2) (8/10=3) (.a .b .c . =4) if cntry == "BE", gen (hincBE)
gen hinc4 = hincBE
drop hincBE
***
recode hinctnta (1=1) (2/6=2) (7/10=3) (.a .b .c . =4) if cntry == "BG", gen (hincBG)
replace hinc4 = hincBG if hinc4 ==.
drop hincBG
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "CH", gen (hincCH)
replace hinc4 = hincCH if hinc4 ==.
drop hincCH
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "CZ", gen (hincCZ)
replace hinc4 = hincCZ if hinc4 ==.
drop hincCZ
***
recode hinctnta (1/2=1) (3/5=2) (6/10=3) (.a .b .c . =4) if cntry == "CY", gen (hincCY)
replace hinc4 = hincCY if hinc4 ==.
drop hincCY
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "DE", gen (hincDE)
replace hinc4 = hincDE if hinc4 ==.
drop hincDE
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "DK", gen (hincDK)
replace hinc4 = hincDK if hinc4 ==.
drop hincDK
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "EE", gen (hincEE)
replace hinc4 = hincEE if hinc4 ==.
drop hincEE
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "ES", gen (hincES)
replace hinc4 = hincES if hinc4 ==.
drop hincES
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "FI", gen (hincFI)
replace hinc4 = hincFI if hinc4 ==.
drop hincFI
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "FR", gen (hincFR)
replace hinc4 = hincFR if hinc4 ==.
drop hincFR
***
recode hinctnta (1/3=1) (4/7=2) (8/10=3) (.a .b .c . =4) if cntry == "GB", gen (hincGB)
replace hinc4 = hincGB if hinc4 ==.
drop hincGB
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "GR", gen (hincGR)
replace hinc4 = hincGR if hinc4 ==.
drop hincGR
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "HR", gen (hincHR)
replace hinc4 = hincHR if hinc4 ==.
drop hincHR
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "HU", gen (hincHU)
replace hinc4 = hincHU if hinc4 ==.
drop hincHU
***
recode hinctnta (1/2=1) (3/5=2) (6/10=3) (.a .b .c . =4) if cntry == "IE", gen (hincIE)
replace hinc4 = hincIE if hinc4 ==.
drop hincIE
***
recode hinctnta (1/4=1) (5/6=2) (7/10=3) (.a .b .c . =4) if cntry == "IL", gen (hincIL)
replace hinc4 = hincIL if hinc4 ==.
drop hincIL
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "NL", gen (hincNL)
replace hinc4 = hincNL if hinc4 ==.
drop hincNL
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "NO", gen (hincNO)
replace hinc4 = hincNO if hinc4 ==.
drop hincNO
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "PL", gen (hincPL)
replace hinc4 = hincPL if hinc4 ==.
drop hincPL
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "RU", gen (hincRU)
replace hinc4 = hincRU if hinc4 ==.
drop hincRU
***
recode hinctnta (1/4=1) (5/8=2) (9/10=3) (.a .b .c . =4) if cntry == "SE", gen (hincSE)
replace hinc4 = hincSE if hinc4 ==.
drop hincSE
***
recode hinctnta (1/3=1) (4/6=2) (7/10=3) (.a .b .c . =4) if cntry == "SI", gen (hincSI)
replace hinc4 = hincSI if hinc4 ==.
drop hincSI
***
recode hinctnta (1/4=1) (5/7=2) (8/10=3) (.a .b .c . =4) if cntry == "SK", gen (hincSK)
replace hinc4 = hincSK if hinc4 ==.
drop hincSK
***
*** Turkey not in data***
***recode hinctnta (1/2=1) (3/5=2) (6/10=3) (.a .b .c . =4) if cntry == "TR", gen (hincTR)***
***replace hinc4 = hincTR if hinc4 ==.***
***drop hincTR***
***
recode hinctnta (1/2=1) (3/5=2) (6/10=3) (.a .b .c . =4) if cntry == "UA", gen (hincUA)
replace hinc4 = hincUA if hinc4 ==.
drop hincUA
***
label var hinc4 "Household income in 3 cat + missing from hinctnta"
label define hinc4 1 'Low' 2 'Medium' 3 'High' 4 'Missing'
label values hinc4 hinc4
tab1 hinctnta hinc4

* Exersice 10*

* Recode hinc4 into dummies*
* Check frequencies*

recode hinc4 (1=1) (2/4=0), gen (lowinc)
recode hinc4 (2=1) (1 3/4=0), gen (medinc)
recode hinc4 (3=1) (1/2 4=0), gen (highinc)
recode hinc4 (4=1) (1/3=0), gen (missinc)
lab var lowinc "Low household income, hinc4=1"
lab var medinc "Medium household income, hinc4=2"
lab var highinc "High household income, hinc4=3"
lab var missinc "Missing income, hinc4=4"
tab1 hinc4 low med high miss

* Exercise 11*

* Recode iscpart2 into cohab, living with husband, wife, partner or cohabiting*
* Check frequencies*

recode icpart2 (1=1) (2=0), gen (cohab)
lab var cohab "Living with husband wife partner or cohabiting"
tab cohab

* Exercise 12*

* Recode schmeet into social, meeting socially*
* Check frequencies*

recode sclmeet (1/5=0) (6/7=1), gen (social)
lab var social "Meet sdeveral times a week with friends, relatives collegues"
tab1 sclmeet social

* Exercise 13*

* Recode cntry into welstate*
* Check frequencies*
* First create numeric country variable*

encode cntry, gen (cntry_num)
recode cntry_num (23 19 7 10 = 1)(16 12=2) (18 6=3)(3 1 11=3)(9 17 13 21 4 = 4)(22 8 2 5 20 14 15 24 25 26=5), gen(welstate)
lab var welstate "Welfare state classification based on Ferrera"
* Define a label group*
label define welstate 1 'Nordic' 2 'Liberal' 3 'Continental' 4 'Southern' 5 'Eastern'
*Assign the label group to the variable*
label value welstate welstate

*Save the changes*

save "c:\data\Multilevel.dta", replace

Page 3

The null model

Load the file you downloaded and prepared in the exercises on the previous pages.

Estimate the null model and calculate the intraclass correlation (ICC). How much of the variation in happiness stems from the variation among countries? Is multilevel analysis to be recommended or will an individual level analysis suffice? How do you interpret the intercept?

SPSS
mixed happy with agec agec2 female seced terted missinc medinc highinc copeinc cohab social
/fixed =
/method = ml
/random = intercept | subject(cntry)
/print = solution.

Note that all the explanatory variables are included first to ensure that all models use the same list-wise deletion criterion in SPSS, i.e. that they are estimated on the basis of the same number of respondents.

Table 5.1. Information Criteriaa

Table 5.2. Estimates of Fixed Effectsa

Table 5.3. Estimates of Covariance Parametersa

Table 5.4. Information criteria table for OLS modela

Answers to the questions

Intraclass correlation ICC=0.5855/(3.7717+0.5855)=0.134.

This means that more than 13 per cent of the variation in happiness scores is due to between-country variation.

Is a multilevel analysis necessary? The between-country variation of 13 per cent is far from trivial, and the estimate of the variance in the intercept residuals is about three times larger than its standard error. The definitive answer as regards the statistical significance of the between-country variation is found in the LR test, where the -2LL of the null model is compared to the -2LL for the one-level model:

Chi square=203 026.467 - 196 165.706 = 6860.76, with 1 df, p=0.000. The outcome is highly significant and indicates that a two-level model is necessary.

The intercept in the null model (7.11) is the weighted average of the country mean scores for happiness. It is slightly higher than the overall mean from the one-level model (7.04).

Stata

To estimate the multilevel models with the same number of respondents, we need to create a filter variable to count the number of missing values for the maximum set of variables:

egen nmiss = rowmiss (happy agec agec2 female seced terted missinc medinc highinc copeinc cohab social)

xtmixed happy || cntry: , mle variance, if nmiss==0

Table 5.5. Stata output 1

Answers to the questions

Intraclass correlation ICC=0.5855/(3.7717+0.5855)=0.134.

This means that more than 13 per cent of the variation in happiness scores is due to between-country variation.

Is a multilevel analysis necessary? The between-country variation of 13 per cent is far from trivial, and the estimate of the variance in the intercept residuals is about three times larger than its standard error. The definitive answer as regards the statistical significance of the between-country variation is found in the LR test, where the -2LL of the null model is compared to the -2LL for the one-level model:

The chi square for the LR test comparing the null model with two levels to the one-level equivalent is found at the bottom of the Stata output: 6860.76, df=1, p=0.000. The outcome is highly significant and indicates that a two-level model is necessary.

The intercept in the null model (7.11) is the weighted average of the country mean scores for happiness. It is slightly higher than the overall mean from the one-level model (7.04).

Page 4

The individual level model

Add all candidates and eliminate variables with coefficients that do not attain statistical significance or add variables in groups - demographic, socioeconomic, social support - to find out which group of variables can explain some of the differences between countries.

Follow the latter strategy and add variables in blocks. Eliminate variables that do not appear to have any statistically significant effect on happiness. Create a table with the pseudo R squares for the two levels for each block with the null model as the baseline.

SPSS

Run the commands to see the output.

mixed happy with agec agec2 female seced terted missinc medinc highinc copeinc cohab social
/fixed = agec agec2 female
/method = ml
/random = intercept | subject(cntry)
/print = solution.

mixed happy with agec agec2 female seced terted missinc medinc highinc copeinc cohab social
/fixed = agec agec2 female seced terted missinc medinc highinc copeinc
/method = ml
/random = intercept | subject(cntry)
/print = solution.

mixed happy with agec agec2 female seced terted missinc medinc highinc copeinc cohab social
/fixed = agec agec2 female seced terted missinc medinc highinc copeinc cohab social
/method = ml
/random = intercept | subject(cntry)
/print = solution.
Stata

Run the commands to see the output.

xtmixed happy agec agec2 female || cntry: , mle variance, if nmiss==0

xtmixed happy agec agec2 female seced terted missinc medinc highinc copeinc || cntry: , mle variance, if nmiss==0

xtmixed happy agec agec2 female seced terted missinc medinc highinc copeinc cohab social || cntry: , mle variance, if nmiss==0

Answer
Table 5.6. Explained variance models formed by adding blocks of variables
Null model + Demographic variables + Socioeconomic variables + Social support variables
Individual level variance 3.772 3.701 3.398 3.268
Between-country variance 0.586 0.578 0.345 0.292
Explained individual level variance 0.000 0.019 0.099 0.133
Explained country level variance 0.000 0.013 0.410 0.502

Run the models to see the output and compute the pseudo R squares in the table. We see that the demographic variables, age, age squared and gender, explain very little of the variation at any level. This changes when the socioeconomic variables are added. They explain about 10 per cent of the individual level variance and, more importantly, about 40 per cent of the between-country variation in happiness, which must be the result of compositional effects. The fixed coefficients indicate that the differences among countries in relation to coping on present income are the most important factor in terms of explaining this variation. Adding the two social support indicators increases the explained variances to 13 and 50 per cent for the individual and country level, respectively.

All fixed coefficients are statistically significant and there are no statistical reasons for excluding any explanatory variables from the model.

Page 5

The final model with level 2 explanatory variables

We have to deal with two problems here: the low number of countries, which limits the number of variables that can added to the model, and the problem that many variables are rather strongly correlated. It is therefore advisable to base the choice on theory and test the model with one or a few country level variables at a time. For happiness, we could reason that people in rich countries or countries with good living conditions will be happier than people in poor countries. This point at Gross Domestic Product (GDP) per capita and the Human Development Index (HDI). We can only use one of them at a time, since GDP is a component of HDI. We could also examine the hypothesis inspired by Wilkinson and Pickett (2010) that inequality lowers levels of happiness and include the Gini coefficient as a measure of inequality. We could also argue that gender equality promotes happiness and include the gender empowerment measure from the Human Development Report. Life expectancy at birth can be seen as an objective quality-of-life measure. Does it also lead to happiness?

Country classifications can also be added as contextual variables. They can build upon a complex set of variables and be an alternative to the more specific continuous variables reviewed above. We can also test the welfare state classification that is already in the data file. Looking at the categories, we see that the classification is correlated to a country’s wealth, especially the difference between the Nordic, Continental and Liberal regimes, on the one hand, and the Southern and Eastern European regimes, on the other.

Page 6

Summary of findings

The main estimates are summarized in three tables. The first one analyses explained individual level variance and country level variance that can be explained by compositional effects. The second table documents how much of the country level variance is explained by adding five continuous country variables and a country classification. If the results were to be published, the main results in these two tables could be presented in a single table. The third table presents the estimates of the fixed (regression coefficients) and random parameters (variance components) for two selected models. With the exception of the column with the variable names, the table could be published as it is.

Together, the three groups of variables - demographic, socioeconomic and social support indicators - explain about 13 per cent of the individual level (within-country) variation in reported happiness scores. The socioeconomic variables, especially ‘coping well on present income’, appear to be the most important ones in relation to individual differences in happiness. About half of the between-country variation in happiness scores is explained by the individual level explanatory variables. This can be interpreted as a compositional effect. Again, the socioeconomic variables are the most important ones, and especially coping on present income.

The second table shows that the gender empowerment measure works best in terms of further explaining the between-country variation in happiness. With this variable added to the full individual level model, around 80 per cent of the between-country variation is explained. Since the country characteristics are rather strongly correlated, only one of them can be added at a time. This also means that, on average, countries that are strong in relation to gender empowerment are also rich, have long life expectancy and low income inequality.

Table 5.6a. Variance components for model with individual level explanatory variables
Null model + Demographic variables + Socioeconomic variables + Social support variables
Individual level variance 3.772 3.701 3.398 3.268
Between-country variance 0.586 0.578 0.345 0.292
Explained individual level variance 0.000 0.019 0.099 0.133
Explained country level variance 0.000 0.013 0.410 0.502

Table 5.7. Explaining the country level variation in hapiness scores
Null Full inda GDP Gini Lifeb Hdi Empowc Welstated
Remaining country level variance 0.586 0.292 0.131 0.166 0.128 0.109 0.103 0.131
Explained country level variance 0.000 0.502 0.777 0.717 0.782 0.814 0.824 0.776
a. Full ind = Full individual level model b. Life = Life expectancy at birth c. Empow = Gender empowerment measure d Welstate = The welfare state classification

Table 5.8. The final model and the model with the welfare state classification
Variable names Estimates of fixed parameters B S.e. t Sig.
Intercept 2.704 0.380 7.123 0.000
Agec Age in years, centred -0.011 0.000 -22.008 0.000
agec2 Age centred squared 0.001 0.000 23.588 0.000
Female Female gender 0.159 0.017 9.433 0.000
Primary education (reference category) 0.000
seced Secondary education 0.126 0.022 5.595 0.000
Terted Tertiary education 0.219 0.024 9.137 0.000
Low household income (reference category) 0.000
Missinc Missing household income 0.287 0.026 11.134 0.000
Medinc Medium household income 0.229 0.024 9.608 0.000
Highinc High household income 0.420 0.027 15.692 0.000
Copeinc Coping on present income 0.920 0.021 43.853 0.000
Cohab Living with a partner 0.713 0.020 36.504 0.000
Social Meeting often socially 0.496 0.018 27.141 0.000
c_gem_2009 Gender empowerment measure 3.423 0.511 6.697 0.000
Estimates of random parameters
Residual variance, individual level 3.268
Residual variance, country level 0.103
Estimates of fixed parameters
(estimates for age to social omitted)
Welstate = 1 Nordic 1.065 0.216 4.928 0.000
Welstate = 2 Liberal 0.530 0.282 1.879 0.072
Welstate = 3 Continental 0.755 0.200 3.777 0.001
Welstate = 4 Southern 0.498 0.216 2.309 0.029
Welstate = 5 Eastern (reference category) 0.000
Estimates of random parameters
Residual variance, individual level 3.268
Residual variance, country level 0.131

The model in which the gender empowerment measure is replaced by the welfare state classification seems to be inferior to the final model with three more fixed parameters. They show that the Eastern countries fare worst and have the lowest level of happiness after controlling for all individual level explanatory variables. The Southern and Liberal welfare regime countries score about half a point higher (on a scale from 0 to 10) than the Eastern regime. In the Nordic regime, the average score is about one point higher than in Eastern Europe.

Finally, some comments on the individual level explanatory variables. Age is the most difficult to interpret since its effects are represented by two variables: age centred and age centred squared. The linear coefficient t is negative and the coefficient for the square term is positive. This indicates that young people are most happy and that the level of happiness decreases until about the age of 53.5 years, when the bottom is reached, before again increasing. The high point is about 5.96 for persons in the reference categories for all dummy variables and in countries with an average level of gender empowerment. The low point is about 5.12, a maximum difference of about 0.84 points. The graph describes the relationship between happiness and age, controlling for all other variables:

Figure 5.1. Relationship between happiness and age controlling for all other variables

Women report being slightly happier than men. The expected gradients in happiness by education and income are observed, the steepest one for income. Note that education could well have indirect effects on happiness through the remaining variables in the table, where only the direct effects are reported. The subjective measure of whether one is coping (well) on present household income has a rather strong effect, even when controlling for income. Those who are coping well report about one point higher happiness than those who have difficulties coping on their present income. Finally, both living together with a partner and meeting often socially with friends is positively related to reporting happiness.

Final words

The analysis of happiness in Europe rests on an important and perhaps unrealistic assumption that all individual level explanatory variables are similarly related to happiness in all countries. There are at least two ways of testing this implicit assumption: firstly, by testing whether one or more of the explanatory variables can be defined as random and, secondly, by estimating one-level models for groups of countries. I leave it as an open exercise to examine this assumption.

Go to next chapter >>

Page 7