Happiness and European Parliament – Adventures in European Social Survey Data

Open data opens fascinating views into recent history and near present day in Europe.

This blog post investigates European Social Study 2002-2016 data and explores subjective happiness, trust in European Parliament and views on immigration and refugees by using statistical analysis and modelling. Separate data from World Bank and UNHCR is used as reference.

R-studio workbook used for the investigation along with other material – is available in GitHub (there is a separate copy of the workbook where most graphics are interactive)

 

cntry name 2002 2014 2016 d0216 eff cil 0216 eff ciu 0216 eff m 0216 p 0216 sig 0216 d1416 eff cil 1416 eff ciu 1416 eff m 1416 p 1416 sig 1416
Ireland 7.863 7.331 7.550 -0.312 0.071 0.075 negligible 0.000 *** 0.219 0.071 0.075 negligible 0.000 ***
Sweden 7.874 7.896 7.850 -0.024 0.079 0.083 negligible 0.742 -0.047 0.079 0.083 negligible 0.430
France 7.410 7.351 7.396 -0.014 0.010 0.014 negligible 0.832 0.045 0.010 0.014 negligible 0.418
Belgium 7.695 7.743 7.727 0.033 -0.014 -0.010 negligible 0.538 -0.016 -0.014 -0.010 negligible 0.808
United Kingdom 7.602 7.583 7.649 0.047 0.365 0.369 small 0.410 0.065 0.365 0.369 small 0.292
Netherlands 7.847 7.865 7.937 0.090 -0.021 -0.017 negligible 0.040 * 0.072 -0.021 -0.017 negligible 0.100 +
Finland 8.031 8.038 8.122 0.091 -0.043 -0.039 negligible 0.066 + 0.084 -0.043 -0.039 negligible 0.070 +
Switzerland 8.043 8.088 8.168 0.125 -0.039 -0.035 negligible 0.020 * 0.080 -0.039 -0.035 negligible 0.132
Norway 7.902 7.957 8.087 0.184 -0.055 -0.051 negligible 0.000 *** 0.129 -0.055 -0.051 negligible 0.028 *
Spain 7.457 7.437 7.747 0.290 -0.079 -0.075 negligible 0.000 *** 0.310 -0.079 -0.075 negligible 0.000 ***
Portugal 6.953 6.973 7.437 0.484 -0.075 -0.070 negligible 0.000 *** 0.465 -0.075 -0.070 negligible 0.000 ***
Germany 7.189 7.586 7.757 0.568 -0.077 -0.074 negligible 0.000 *** 0.171 -0.077 -0.074 negligible 0.000 ***
Slovenia 6.906 7.116 7.477 0.571 -0.230 -0.225 small 0.000 *** 0.361 -0.230 -0.225 small 0.000 ***
Hungary 6.325 6.384 6.901 0.577 -0.067 -0.062 negligible 0.000 *** 0.518 -0.067 -0.062 negligible 0.000 ***
Poland 6.422 7.270 7.475 1.053 -0.436 -0.432 small 0.000 *** 0.204 -0.436 -0.432 small 0.002 **

Columns

  • name of country
  • 2002, 2014, 2016 – average subjective happiness in ESS that year
  • d0216, d1416 – difference between two surveys, 2002 vs. 2016 and 2014 vs. 2016
  • eff cil 0216, eff ciu 0216, eff cil 1416, eff ciu 1416 – confidence intervals for effect size with cil denotes lower and ciu upper end of 95% confidence interval
  • eff m 0216, eff m 1416 – assessment on the meaning of the difference
  • p 0216, p1416 – statistical significance of the difference, numeric p-value
  • sig 0216, sig 1416 – statistical significance levels: + 0.1, * 0.05, ** 0.01, *** 0.001

Effect was estimated using Cohen’s d metric. Statistical significance with weighted t-test.

Factors associated with subjective happiness

There was a linkage between subjective happiness and trust in European Parliament in the first version of the blog, but not very clear. In the updated analysis, now using multiple linear regression model, trust in European Parliament did not end up as a factor by the algorithm used.

Most important factor (and positive one) was willingness of people to help others.

Second largest factor (but a negative one) was trust in United Nations. This provides a link to trust in European Parliament which turned out to be the most important and positive factor in the linear model on trust in United Nations.

There seems to be linkage between subjective happiness and these international institutions.

Other key factors identified (in the end of the blog post there is a tabulation of the factors along with explanation of abbreviations used)

  • stfhlth – State of health services in country nowadays (+)
  • stfedu – State of education in country nowadays (+)
  • eisced2 – Level of education (-)
  • stfeco – Satisfaction with present state of economy in country (+)

Education level having a negative impact is curious and would need further investigation to understand it better.

unnamed-chunk-46-3

The model was trained with data from 2002 to 2014 and tested against 2016 results. It performed decently even though it underestimates the real score in Poland and is almost missing that of Spain.

Then again, other types of models were not tried since a linear model often provides fairly easy way to get insight into associations.

unnamed-chunk-46-2

In other models created for this reboot of the blog post, I decided that subjective happiness is not a plausible explanatory factor. It is more naturally as an ultimate outcome and depends on other factors. Therefore, it did not show up in any models as independent variable. But somebody else might have decided otherwise.

Trust in European Parliament

Most important – and positive – factors in this model were trust in United Nations, satisfaction with government and increasing of integration of European Union.

Interestingly, the most important negative factors were willingness of people to help others, question on whether immigrants are good for economy and satisfaction with economy.

unnamed-chunk-51-3

The model was tested against 2016 results and it performed again relatively well – even though the Portuguese were not giving European Parliament as much trust as the model estimated.

unnamed-chunk-51-2

Overall average from ESS surveys gives an impression that trust in European Parliament is highest in the Nordic countries and a few smaller countries. But on the scale of zero to 10 the scores are modest everywhere and there are differences. Some easier and other not so obvious to understand.

trustep
Interactive chart here

Recent history can be seen in the timeline results below easily for Greece which has had its economic crisis and challenges with aid from European Union. Low scores from Russia could be linked to the crisis in Crimea.  Overall, the graph shows intriguing trajectories, each with a story of its own to tell.

unnamed-chunk-40-12
Click here for interactive version

Sample cases: Greece and Ukraine

Greece has not been in the survey since 2010. That was the year when the first economic support package from EU was agreed. As one can see, most indicators were sliding down towards 2010. And it is worth noting that initially trust in European Parliament was quite high – but dropped low. Trust in legal system, trust in police and – no surprise – trust in country’s economy went down as well.

unnamed-chunk-41-31
Click here for interactive version

Recent history has been quite difficult for Ukraine starting with various political crisis to still ongoing dispute on Crimea (which started 2014, i.e. after the last survey Ukraine participated in).

Happiness score increased while trust in European Parliament decreased, even though not much. However, trust in police and legal system seems to have taken big hit -and there was very low level of interest in politics reported in 2008.

unnamed-chunk-41-79
Click here for interactive version

A division of Europe

A k-means clustering exercise was done with a selection of indicators in ESS and it chose to divide Europe into two groups: one consisting of Nordic countries together with Belgium, The Netherlands, Ireland, Switzerland and Austria. And the other with all the rest.

clust
Interactive chart here

The selected indicators show quite clear differentiation between the groups. Here we have converted absolute numerical ESS scores into uniform range to make it easier to compare them on one graph.

In most of the indicators countries in cluster 2 give higher scores which roughly translate to “more” or “better” — with a couple of noteworthy exceptions

  • health – low value is good and high value is poor health
  • impcntrl – low value means more immigration and high less immigration from poor countries

unnamed-chunk-60-3

No formal clustering exercise was done with World Bank Development indices or UNHCR data which were used for context

However, visual box plot comparisons of selected indicators seem to support the clustering.

  • Cluster 2 countries appear more urbanized than cluster 1
  • Proportion of population born in another country is higher in cluster 2 even though there is quite some overlap in distributions
  • There is higher percentage of people active in workforce in cluster 2
  • Government health and education expenditures vs. GDP are higher in cluster 2
  • GDP growth (and per capita growth) is higher in cluster 1 countries but there is significant overlap with cluster 2 distribution
  • significantly higher share of population is using Internet in cluster 2 compared to cluster 1
  • tax revenues as % of GDP are higher in cluster 2
  • Personal remittances received (% of GDP) is higher in cluster 1

unnamed-chunk-60-4

Since there are big differences between countries and the indicators, a logarithmic scale was chosen. And to make that work all scores were shifted by one. Real scores are one percentage unit lower than shown below.

Here we can see clear difference regarding percentage of refugees and asylum seekers in population, being significantly higher in the countries of cluster 2.

unnamed-chunk-60-5

Migration and refugees

Refugees, asylum seekers and immigration has been a hot topic in recent history – and still is.

unnamed-chunk-34-1
Interactive chart here

There are quite different views on the topic of immigration from poor countries outside of Europe as seen in the timeline below. On one end of the spectrum is Hungary (cluster 1) with strongly restrictive views and on the other Sweden (cluster 2) with more open approach.

unnamed-chunk-40-4
Click here for interactive version

The views on benefits of immigration to country’s economy vary a lot as well. However, this is not necessarily the same as views on immigration from poor countries, at least not for all the countries covered in ESS.

But for some it could since the countries with restrictive views on immigration like Hungary and Cyprus are getting low scores while Sweden is on the high end.

indicator name min max source lblmin lblmax
BX.TRF.PWKR.DT.GD.ZS Personal remittances, received (% of GDP) 0.070 8.154 WB NA NA
GC.TAX.TOTL.GD.ZS Tax revenue (% of GDP) 1.230 48.344 WB NA NA
IT.NET.USER.ZS Individuals using the Internet (% of population) 1.874 97.298 WB NA NA
NY.GDP.MKTP.KD.ZG GDP growth (annual %) -14.814 25.557 WB NA NA
NY.GDP.PCAP.KD.ZG GDP per capita growth (annual %) -14.560 24.765 WB NA NA
REFG.PCT Refugees (% of population) – calculated 0.000 2.319 WB NA NA
SE.XPD.TOTL.GD.ZS Government expenditure on education, total (% of GDP) 2.326 8.560 WB NA NA
SH.XPD.GHED.GD.ZS Domestic general government health expenditure (% of GDP) 0.000 9.420 WB NA NA
SL.TLF.CACT.ZS Labor force participation rate, total (% of total population ages 15+) (modeled ILO estimate) 47.853 68.565 WB NA NA
SM.POP.TOTL.ZS International migrant stock (% of population) 0.795 29.387 WB NA NA
SP.RUR.TOTL.ZS Rural population (% of total population) 2.103 50.373 WB NA NA
SP.URB.TOTL.IN.ZS Urban population (% of total) 49.627 97.897 WB NA NA
UNHCR.Refugee.PCT Refugees (incl. refugee-like situations)(% of population) – calculated 0.000 2.319 UNHCR NA NA
UNHCR.Asylum-.PCT Asylum-seekers(% of population) – calculated 0.000 1.603 UNHCR NA NA
UNHCR.Interna.PCT Internally displaced persons(% of population) – calculated 0.000 4.000 UNHCR NA NA
UNHCR.Statele.PCT Stateless(% of population) – calculated 0.000 11.048 UNHCR NA NA
UNHCR.Other.PCT Other population types together(% of population) – calculated 0.000 6.276 UNHCR NA NA

Updated 2018-06-20 – a few misspellings corrected and title of one graph corrected
Updated 2018-06-21 – fixed formatting issues in two tables
Updated 2018-08-14 – added Github link https://github.com/juhariis/ess to the R-markdown document for this blog entry exists, updated conclusions regarding the change of correlation (some less happy countries went down in their trust – reality bites?)
Updated 2018-09-02 – Sometimes there is a need to revisit work that one has done earlier. This blog post from June 2018 is one such item. When returning to this I realised that it very much needed an update. A much longer  version – actually the R-studio workbook used for the investigation along with other material – is available in GitHub
Updated 2018-09-11 – added link to separate web application providing access to indicators by countries (and vice versa). The application is hosted at a free site with certain usage limits so it may stop working when the limits are reached. Sorry about that.
Updated 2018-09-12 – Added link to the full javascript version of the workbook along with links to the active graphs referred to here.
Updated 2018-09-13 – Replaced old maps with new versions, which are interactive in the copy of the workbook in Azure

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.