If money can't buy happiness, what can?

By Katya Vladislavleva and Matt Price

March 25, 2015

The perception of freedom is fundamental to happiness, yet fresh analysis of recently collected UN data yields some surprising facts about what really drives it. The reality is that it’s much more than just financial security. In this article, we reveal the factors that make people feel free – and therefore happy – today.

We should strive to increase not just our GDP, but also the collective level of happiness we experience.

Introducing the Freedom Index

It is often said that money can’t buy you happiness – and it is now formally accepted in many quarters that happiness is not simply a function of economic prosperity. The concept of happiness economics holds that to create satisfied societies, we should strive to increase not just our GDP, but also the collective level of happiness we experience.

Research already undertaken in this field has yielded surprising results. Far from being dominated by rich countries, the Happy Planet Index (happyplanetindex.org) shows Costa Rica, Vietnam and Cambodia leading the world rankings.

Despite worldwide acceptance of the importance of happiness, however, policymakers still focus on well-established economic drivers. It seems that politicians the world over believe that voters still care most about money. We believe this belief persist for two reasons. Firstly, they consider happiness hard to measure and, secondly, they are accustomed to placing their trust in established economic models that are financial in subject and theoretical in nature.

This is misguided. A vast amount of hard data is harvested every year by the United Nations, in countries all over the world. Much of it is based on interviews and surveys, all carefully designed by professionals to yield reliable insights. As a result, data exists for an enormous range of indicators.

The temptation for policy-makers to rely on theoretical models is arguably more insidious. Despite the ideal worlds depicted in such models, there is a tendency to default to established theory throughout the developed world, from academia to the corridors of power.

The truth is out there

Theory is important, of course. But as data scientists, we believe nothing but the truth. We care only for empirical observations; for reality. And reality often has a way of hiding the truth. At DataStories, it is our mission to reveal it, in all its unexpected glory.

We were keen to investigate whether one universally component of happiness can be linked to any economic drivers. Because if it can, the world’s policy makers can turn their attention to improving our happiness. Finally, they will be able to hear the story the data is telling them.

That component is the perception of freedom of choice. It is widely accepted, that freedom of choice is an important aspect of happiness; indeed, it is one of the principles on which most western democracies are founded. Given that a wealth of reliable data collected by the UN and other globally recognised sources can be used to investigate the link between people’s perceived freedom of choice with many other factors, plus its obvious importance to policy-makers as a component of happiness, we decided to examine it in depth.

In Autumn 2014, we set out to see what drives it – and whether there are any objectively measure indicators impacting subjective freedom of choice. The answer – as always – lay in the data.

Globally trusted data sources

This project involved data that everyone cares about – the relative economic, demographic and cultural performance of the world’s nations. We used the following data sets, all published by The United Nations and available in the public domain:

  1. Human Development index data
  2. Inequality Adjusted HDI data
  3. Gender Inequality
  4. Multi-dimensional Poverty Index
  5. Gender Development Index datalink

The above data, downloadable from the UN website, contains 16 tables with data from all the above and additional tables on Health: Children and Youth (Table 7), Adult Health and Health Expenditures (Table 8), Education (Table 9), Command over and Allocation of Resources (Table 10), Social Competencies (Table 11), Countries Personal Insecurities (Table 12), International Integration (Table 13), Environment (Table 14), Population (Table 15), and Supplementary Indicators on Perceptions of Well-being (Table 16).

How data science sees the problem

Our objective was to understand the perception of people in various countries of their freedom to do what they choose with their lives.

The Freedom Index tracks people’s satisfaction with their freedom of choice, defined as the percentage of respondents answering ‘satisfied’ to the Gallup World Poll question, ‘In this country, are you satisfied or dissatisfied with your freedom to choose what you do with your life?’

We took the results of this question as a starting point and combined it with data from several UN surveys to look for causal relationships. In essence, the question we were asking was: Can we see from the data what drives people’s satisfaction with their freedom of choice?

With so many variables, advanced analytic techniques are required to undertake this examination. To answer our question, we pursued a methodology consisting of four steps:

  1. Considering the perception of freedom of choice a key performance indicator (KPI).
  2. Identifying the economic indicators in the data that are related to the perception of freedom of choice in a mathematical sense (and also understanding which ones are unrelated).
  3. Building predictive models of perception of freedom of choice using the drivers discovered in step 2.
  4. Exploring the new models to understand exactly how those drivers influence the KPI (and in which direction) via the use of ‘What if,’ best case and worst case scenarios.

It is worth noting that our learning algorithms undertake steps (2) and (3) automatically, quickly isolating the variables that truly predict the KPI (satisfaction with freedom of choice).

The result was a distilled data set of the six variables most important to people’s perception of freedom of choice – along with interactive models that could be used for ‘what-if’ scenarios.

Without algorithms capable of performing the heavy lifting, the above process would be long and arduous. Results would be error-prone at best; wholly inaccurate at worst.

Looking at the results

Despite the recent popularity of happiness economics, it was important, first of all, to see how human development and GDP per capita impact on the perception of freedom of choice.

Let’s begin with the Human Development Index (HDI) for all listed countries. TAP on the graphs for more info.

Next, we can superimpose the data for GDP per capita on the same data points.

So far, so predictable. With the exception of a few outliers, the countries’ HDI figures are strongly correlated to their respective GDP. (If you’re wondering which country it is that boasts the gigantic GDP of $133,713 per capita, it’s the tiny but oil-rich Middle Eastern state, Qatar.)

Now, here comes the first surprise. When we plot that data against perception of freedom of choice, we see very little correlation.

Check this page for additional data exploration options.

So what really are the drivers of freedom of choice?

A more visual way to look at the data is to use a map, in which the stronger the perception of freedom of choice, the darker the colour of the country. When we do this, we can see that the highest figures are observed in Cambodia (95), Switzerland (94), Sweden (93), Kuwait (93), Australia (93), Norway (92), Denmark (92), Costa Rica (92), and Canada (92). No data was available for the pink areas.

World colored by the freedom of choice satisfaction. For an interactive map see THIS PAGE (loads slowly on mobile devices).

World countries colored by the level of the freedom of choice satisfaction. Data collected by the Gallup poll in 2013. Published by United Nations in 2014.

Clearly, there are some similarities with the findings of Happy Planet. This suggests, anecdotally at least, that the perception of one’s freedom of choice is indeed an important factor in overall happiness.

So what factors really drive the perception of freedom of choice, if wealth and human development are not relevant?

To answer this question, we first had to exclude the human development index and also population growth over the last ten years. The result was a data set comprising 146 indicators for 189 countries. With so many variables, advanced analytic techniques were our only hope of getting to the truth.

By running a deep learning algorithm known at DataStories as ‘KPI Drivers’, we discovered that with a mere six of the 146 indicators we were able to build reliable predictive models*.[Ten independent experiments of our deep learning algorithm (20,000 cross-validations in total) all produced consistent variable importance values as depicted above, with an average correlation coefficient of predicted versus observed perception of freedom of choice satisfaction of 88%.]

Six steps to freedom of choice

At the outset, we had no alternative but to assume that all candidate indicators had the same bearing on our end measure: the perception of freedom of choice. By the time we had run the KPI Data Story on our data set, however, we had distilled the initial mass of data down to six pivotal factors:

  1. Satisfaction with standard of living (accounts for 38.4% importance)
  2. Action taken to preserve the environment (36.3%)
  3. Public health expenditure (14.8%)
  4. Female participation in the labour force (6.3%)
  5. Consumer price index (2.2%)
  6. Forest area % (1.9%).

The models based on these six drivers are 88% accurate when it comes to predicting the perception of freedom of choice, in comparison with the observed data.

So what do the results tell us?

While standard of living is not an unexpected driver, what is arguably most interesting is the importance of environmental preservation. Could this represent a shift in the importance of environmental policies? Quite possibly. Only four years ago, the same analysis on the 2009 UN HDI data produced the following six drivers:

  1. Women treated with respect (35%)
  2. Level of political engagement (21%)
  3. Perception of personal safety (15%)
  4. Presence of female social networks (11%)
  5. Existence of protected/conservation areas (10%)
  6. Number of doctors per 10km (8%)

Analysis of 2014

Drivers for freedom of choice satisfaction measured in 2013. TAP on segments for more info.

Analysis of 2010

Drivers for freedom of choice satisfaction measured in 2010. TAP on segments for more info.

We can see that, while all are different, the older figures almost all pertain to personal happiness or health, especially with regard to the welfare of women. Only the existence of conservation areas, influencing just 10% of the overall figure, betrays any interest in the environment.

Check this page for a simple visualization of freedom drivers.

Easy ways to boost perceived freedom of choice

Having run the KPI Data Story, we are left with a distilled model, comprising only the six most important variables. This distilled model can be explored interactively, in the form of ‘what-if’ scenarios.

A first step in this exploration might be to see how the perception of freedom of choice changes when all but one of the driving inputs have average values.

Only six indicators out of 146 considered emerge as drivers necessary and sufficient to predict satisfaction with the freedom of choice.

These models get more interesting – and actionable – when they are run for individual countries (provided that we know all six variables for that country). By altering the value of a single variable and keeping all the others unchanged, we can see the extent to which changes in that input can lead the KPI in the right direction. Take France, for example. The values of the drivers are depicted below, along with the corresponding value for predicted perception of freedom of choice.

The fastest way to improve people satisfaction with the freedom of choice for France is to improve their satisfaction with government's actions to preserve the environment.

We can see that one way for France to improve the perception of freedom of choice is to improve the perception of its populace on its actions to preserve the environment. If the latter were to increase by 14% (from its current value of 53% to 67%) the former would increase by 5%, assuming the other five indicators remained the same.

In the graphs below, the red dot represents the effect on perception of freedom fo choice when perceptions on action to preserve the environment are boosted by 14%. The blue dot represents the KPI’s current value.

For France, if the people satisfaction with the government actions to preserve the environment were to increase by 14% (from its current value of 53% to 67%) the freedom of choice satisfation would increase by 5%, assuming the other five indicators remained the same.

Is it possible to eliminate subjectivity?

So far, we have used a data set defined specifically to include the maximum number of useful variables. But due to the various collection methodologies employed, some of these variables are subjective in nature. Why did we not simply eliminate these softer values from our model?

The best way to answer this question is to explain what happened when we tried to predict the Freedom Index using only objectively measured indicators (i.e. excluding variables sourced from polls).

We excluded all 18 subjective indicators from the analysis. All indicators from table 16 (Perceptions of Wellbeing) were excluded, as well as several indicators that are, in our opinion, difficult to measure accurately, namely child deaths due to outdoor air pollution, child deaths due to Indoor air pollution, child deaths due to unsafe water, unimproved sanitation or poor hygiene, vulnerable employment, justification of wife beating among women and justification of wife beating among men.

The results changed markedly, in large part due to the absence of perceptions on action to preserve the environment. In the absence of this key variable, others are selected. The resulting models achieve just 77% correlation between the predicted perception of freedom of choice and the observed value, compared to the 88% correlation achieved when we included subjective values.

The full breakdown can be seen in the following chart:

Analysis of 2014 with objective indicators only

Drivers for freedom of choice satisfaction measured in 2013. TAP on segments for more info.

The most important driver now becomes the Gross Enrollment Ratio for Secondary Education (38.2%) – the ratio of people actually enrolled in secondary education to the number of people eligible for it.

The second driver is the employment to population ratio (30.4% importance), followed by the prison population per 100,000 (12.8%), domestic food price index (8.1%), income inequality (only 7.6%), and deaths from natural disasters (3%). A very different picture from our more accurate model and to a degree, validation of the data collected via more subjective methods.

More examples of variation by country

For the UK, the picture is as follows:

For UK to increase perception of freedom of choice, the domestic food price index should decrease, the employment to population ratio should increase or the prison population should be halved

We can observe that for UK to increase perception of freedom of choice, the domestic food price index should decrease, the employment to population ratio should increase or the prison population should be halved.

Spain is interesting, starting with the accuracy of the model: the prediction produced for our KPI is 72.6%, while the observed value is 74%. In the graphs below (where all drivers are set to the value for Spain) we see a remarkably high number of GER Secondary, which means that a lot of people from non-standard age groups are enrolled in secondary education and other kinds education at the same time.

Spain has four ways to improve FoC –increasing the employment to population ratio (the highest increase impact would come from 12% increase in employment to population and would lead to 6% increase in FoC), decreasing inequality in income, decreasing the domestic food price index or decreasing prison population per 100,000.

Based on the graphs, we can see that the Spain has four ways to improve FoC – increasing the employment to population ratio (the highest increase impact would come from 12% increase in employment to population and would lead to 6% increase in FoC), decreasing inequality in income, decreasing the domestic food price index or decreasing prison population per 100,000.

Conclusion

While the importance of standard of living and consumer price indices might not surprise many, especially with much of the world in the grip of recession or austerity measures, there was one surprise, especially when the findings of this project were compared with those from four years earlier: the growing importance of the environment.

Perhaps as a result of the growing frequency of extreme weather events, or maybe simply a media-led effect as the world comes to terms with the reality of human-led climate change, the perception that one’s government is acting positively to protect the environment is now the second most important factor in people’s satisfaction with their freedom of choice.

While each country has its own story and economics will always be important in a capitalist society, this mental shift has serious implications for politicians who want to stay in their jobs. For them, this data story makes it clear what voters really care about, both now and in the immediate future.

Another fascinating data story

At DataStories, we believe that, although data exploration and visualisation are critical first steps in understanding the system behind any data set, informed decision-making involves much more than that.

To truly understand any data-generating system, we need to sharpen the questions we ask about it and introduce additional context. This involves the creation of reliable, accurate models that capture the behaviour of the system – and using those models to run what-if scenarios. The first step is to establish what we truly know and do not know about the system’s key performance indicators (KPIs), based on the data presented to us.

We need to understand which variables (often referred to as columns, factors, features or attributes) are important. This can only be achieved with deep learning algorithms that have been subject to many repeated validations, thereby ensuring we neither over-predict nor under-predict the system, given the data available. Crucially, we need to ascertain whether any variables are missing, due to relationships between the variables provided and the KPIs failing to produce accurate predictions.

This area of study is becoming ever more complex. Requests to analyse data from systems that behave very differently from their component parts are becoming a rule rather than an exception. This means more KPIs and more variables to examine – and ultimately, more data to deal with. But more data does not equate to more information content. Moreover, decisions often need to be made fast. In these circumstances, advanced analytics solutions like ours often represent the only way to make supportable decisions.

To read more data stories as they are published, register for our newsletter at the bottom of the page.