Skip to content

Exploring Personal Well-being and Place

Released: 10 June 2014 Download PDF

Abstract

This article presents the findings of new analysis by ONS on the connections between personal well-being and place. Using data from the APS Personal Well-being dataset (2012/13), regression analysis is used to look at the contributions of the place and the people who live there in shaping the level of personal well-being across different areas of the UK.

2. Introduction

Previous publications by ONS (October 2013) have shown  how people rate aspects of their personal well-being, such as life satisfaction, a sense that daily activities are worthwhile, and their levels of happiness and anxiety varies across different areas of the UK. Personal well-being also appears to vary in relation to the specific characteristics of where we live. For example, personal well-being is higher on average in rural areas than in urban areas and in the least deprived areas compared to the most deprived (ONS, May 2013). However, it was noted that these results only reflected the average personal well-being ratings between relatively large geographical areas and this may mask important differences between smaller or local areas within them.

Just looking at the aggregated average personal well-being ratings in different areas, it may be tempting to conclude that differences between places (for example, differences in physical, environmental, social and cultural factors) are the drivers of individual differences in personal well-being outcomes. However, where an individual lives is just one aspect that may influence an individual’s response to the personal well being questions and the differences between people living within these areas could be larger than the differences between areas.

The main purpose of this working paper is to use regression analysis to investigate the relationship between personal well-being and where we live, in particular, the degree to which regression analysis of the Annual Population Survey (APS) data can help us understand the relationship between personal well-being and the ‘place’ where an individual lives. In doing so, the paper is a follow up to the previous ONS “What matters most to personal well-being” article (ONS, May 2013), that used similar regression techniques to examine which factors matter most to well-being but that only gave limited consideration to the effect of place. This article particularly focuses on the following questions:

  • Are some parts of the Great Britain associated with higher personal well-being than others?

  • Are some types of areas associated with higher personal well-being than others?

  • To what extent the place where we live is associated with our personal well-being?

  • To what extent the differences in average personal well-being between the areas are driven by the characteristics of people who live in these areas compared with the other characteristics of the areas themselves?

The knowledge that to fully understand differences in average personal wellbeing across places it is necessary to consider the characteristics of the people living there in addition to the other (e.g., physical) characteristics of the place itself is important when determining policy. The final section of this article discusses this issue.

Overall, the paper is aimed mostly at an audience of researches in academia and policy researchers in government. The paper aims to stimulate further debate on the subject of the relationship between personal well-being and place among these audiences. However, the results and findings are also likely to be of interest to those engaged in personal well-being issues at the local level.

2.1 Explaining differences in personal well-being between areas

The differences in observed average personal well-being ratings between areas can be due to:

  • the effects relating to people with different characteristics living in different areas (individual effects).

  • the effects relating to differences between the areas themselves (rather than the people who live there). This is called area effects. 

It is important to note that people may choose to live in a particular area for many different reasons, but their reasons may be similar to those of other people who have opted to live in the same area. For example, they may have similar preferences for the types of schools in the area, access to the countryside and to local amenities. They may also have similar levels of household income which influence their choices of neighbourhood and home. All of this would lead to people with similar characteristics or incomes clustering together in particular areas, which can be described as a ‘sorting’ process. Because people tend to ‘sort’ into areas alongside people of similar characteristics, we may find that the average characteristics of people in one area may differ quite substantially from the average characteristics of people in a different area. Meanwhile, a wide range of studies have established that peoples’ individual characteristics are related to how they rate their personal well-being (ONS May 2013). For example, people who are unemployed rate their well-being much lower than people in employment and those who rate their health as poor also tend to rate their well-being this way. Taken together, these issues may suggest that personal well-being is really about the people in the place rather than the place itself.

However, it is also likely that personal well-being can be influenced by the characteristics of places where people live. The types of local features which could influence personal well-being are likely to be those things which relate to perceived quality of life, such as the local environment1, social and cultural life, the local labour market and economic conditions, access to health care and schools and geographical features such as green space. If differences in personal well-being levels across areas relates primarily to characteristics of the place, then people with similar individual characteristics would be expected to rate their personal well-being differently across areas because of different features of the places where they live. 
 
This article explores the differences in personal well-being between local areas and types of areas and investigates using regression analysis to what extent these differences occur because of differences in the types of people in different areas (individual effects) versus different outcomes for the similar types of people in different areas (area effects). The article also describes some important methodological and conceptual challenges associated with examining this topic that need to be borne in mind when reviewing the results. The article concludes by placing the results of the regression work in context.

2.2 Methodological challenges

It is important to be aware that several methodological challenges arise in attempting to understand the relationship between personal well-being and where we live. Some of these challenges are summarised below. 

  • Difficulty in disentangling the individual and area effects. The two explanations proposed above (i.e. individual versus area effects) suggest that the effects of each can be isolated to provide a clear understanding of their separate contributions to personal well-being. However, as many researchers such as Cummins et al (2007) have noted, there may be a ‘mutually reinforcing and reciprocal relationship between people and place’.  For example, the place may contribute to or undermine the health of those living there- so place may have an effect on health which in turn has an effect on personal well-being (Macintyre et al 2002). Indeed a recent research from Imperial College London found that environmental factors, such as air pollution, accounted for roughly 5-10% of a person’s risk of disease, which was significant in terms of a population. As such, in this example the ‘area effect’ is influencing wellbeing via its impact on an individual effect (health). In other words, there is not always a clear separation between area and individual effects (see also the discussion of sorting below).

  • Lack of clarity about the most appropriate geographical level at which to capture any ‘area’ effects on personal well-being.  As people vary significantly in their choices of where to live and why, it is not clear at what geographical level (e.g., region, local authority or neighbourhood), differences between areas matter most to personal well-being. Also, some aspects of where we live, such as sense of community, may not coincide neatly with established geographical boundaries so may be difficult to detect in analysis using standard geographical definitions.

  • Difficulty in establishing which aspects of the ‘place’ are most important to personal well-being. There are often strong correlations between different factors related to a ‘place’. For example, previous analysis (ONS, May 2013) found a positive relationship between living in a rural area and personal well-being after holding equal individual characteristics, but many factors that differ between rural and urban areas are correlated to each other. Rural areas are likely to have more green space and less air pollution and several studies (White et al 2013, Douglas 2014 or Ferreira et al 2013) have found a significant relationship between access to green spaces or pollution and personal well-being.  Isolating the separate effects of each on personal well-being is very challenging.

2.2.1 Effect of ‘Sorting’

It is worth focusing on how sorting links to individual or area effects.  It can be observed that different types of people are clustered into different areas. The fact that this has occurred makes it important that individual effects are examined when seeking to explain differences in average personal wellbeing across areas, (for example, when an area is home to many people who possess individual characteristics usually associated with high wellbeing, that area will likely have a high average wellbeing rating). However, it should be noted that area effects can also be influenced by the sorting of people, because the characteristics of your neighbours can act as an area effect that influences personal well-being.

There is also a separate, but related, question of why this sorting occurs in the first place, and the role of area effects to influence it. This question is not examined in this report. Evidence on this would be found in literature on factors influencing the location choice of households, these factors may include area effects such as the characteristics of neighbours, physical environment, pollution levels etc. As mentioned, this is not within the scope of this article. This article takes the observed distribution of households and individuals across areas as a starting point to the analysis, and examines to what extent the observed differences in average personal well-being between the areas are driven by the characteristics of people who live in these areas compared with the other characteristics of the areas themselves.

Notes for 2. Introduction

  1. The characteristics of the people living in the area are also part of the local environment.

3. Methodology and Data

3.1 How personal well-being is measured

The research is based on data from the Annual Population Survey (APS) from April 2012 to March 2013. The survey provides a representative sample of people living in residential households in the UK and includes about 165,000 respondents.

Since April 2011, ONS have included four questions on the APS which focus on different aspects of personal wellbeing: how we assess our life satisfaction; whether we feel our lives have meaning and purpose; and our recent experiences of positive and negative emotions. These are measured by asking people aged 16 and over four questions:

  • Overall, how satisfied are you with your life nowadays?

  • Overall, to what extent do you feel the things you do in your life are worthwhile?

  • Overall, how happy did you feel yesterday?

  • Overall, how anxious did you feel yesterday?

Those taking part in the survey are asked to give their answers on a scale of 0 to 10 where 10 is ‘completely’ and 0 is ‘not at all’. For the first three questions which are about positive aspects of personal well-being, a higher score indicates higher personal well-being. However, the fourth question asks about anxiety which is a negative emotion so a higher score here indicates lower personal well-being. In this report the abbreviations ‘life satisfaction’, ‘worthwhile’ ‘happiness’ and ‘anxiety’ are used to refer to the findings in relation to these four questions.

3.2 ‘Area’ data used in the analysis

The APS dataset provides detailed geographical information on individuals which potentially allows analysis at lower geographical levels. Due to sample sizes, the smallest spatial units at which personal well-being data in the APS can be analysed robustly are local authority districts.  However, ‘place’ may have an effect on personal well-being at a more local level than this. In order to overcome sample size problems and to explore the relationship between personal well-being and neighbourhoods, smaller area groupings were used in this analysis as well. These areas were classified together on the basis of similar environmental or socio-economic characteristics and are based on Census output areas. The data used for grouping areas together were derived from various sources such as Census, Index of Multiple Deprivation and small area income estimates.  These datasets were all merged into a single data file and then used in the analysis.

A limitation of this approach is that, although the groups are based on smaller areas, they are ‘types’ of small areas and as such are spread out across Great Britain. As a result they may not capture differences associated with a specific location in the country. Also, an important but untested assumption is that all of the characteristics of these areas are identical.

Note that some of the place related variables, such as rural/urban identification could not be created for all of the UK. As such, the analysis reported in this article does not include respondents from Northern Ireland.  Additionally, in some cases it excludes respondents from Scotland; these cases are referred to clearly in the text.

Areas, area groupings and the geographical levels included in the analysis:

  • Local authority counties

  • Local authority districts

  • Weekly average net household income after housing costs in the area - equivalised for household composition (MSOA)

  • Weekly average net household income in the area (MSOA)

  • Index of multiple deprivation (LSOA)

  • Green space (LSOA)

  • Rural urban areas (OA)

  • Built up areas (OA)

  • Population density (OA)

  • Output area classification (OA)

Information on these areas and the datasets used can be found in the Technical Annex.

3.3 Regression analysis

First, the differences between individuals were ignored and regressions were run for each of the four personal well-being questions using only a set of area indictors (dummy variables that indicate where the person lives) for each area or area type considered.

The coefficients (b) represent the difference between the average personal well-being rating in a given area and the average personal well-being in the reference area. These regressions, referred to as ‘the basic model’, show how different areas were related to personal well-being without differentiating between area and individual effects.

Then, a second set of regressions were run including both the area-related and the individual-related variables in the model. These regressions show both the effect of individual characteristics and the additional effects of areas on personal well-being.

     PWBi = a + bPi + ei

     where

     PWB = individual i’s rating for one of the personal well-being outcomes such as life satisfaction (on a scale of 0 to 10)

     P = variables (dummy) that indicates where the individual i lives

     b = unstandardised coefficients showing area effects

     e = an error term that represents unobserved factors associated with personal well-being but that are uncorrelated with area effects.

In these regressions, the coefficients (b) show the effect of areas on personal well-being after accounting for individual characteristics and circumstances. Similarly, the coefficients (g) show the impact of individual characteristics and circumstances on their personal well-being after controlling for the effects of the areas.

Note that, it is not possible to control for all the individual characteristics and circumstances that are associated with personal well-being. If some of these unobserved characteristics also influence sorting into areas, then these factors would be attributed incorrectly to area effects. Because of this as well as the other methodological challenges noted earlier, the estimated area effects in this analysis can best be understood as an approximation of the ‘true’ area effects. Finally, the estimates are based on a sample and there is the possibility that different samples can give different results.

At each stage, the regressions were carried out using the Ordinary Least Squares (OLS) technique.  Although ordered probit is the regression technique which is usually considered best suited to the ordered nature of the responses to the personal well-being questions (i.e., with responses on a scale from 0 to 10), they were not suitable to use for the variance analysis. As in our previous regression analysis (ONS May 2013, ONS February 2014) we also estimated the regressions in ordered probit to test the robustness of OLS results. The statistical significance, the signs and the relative sizes of the regression coefficients  between the two methods were very similar, however, the findings from the ordered probit regressions are not reported in this article. The findings in this report are based on the results of the OLS regressions not only for the sake of simplicity and ease of interpretation, but also to enable the variance analysis to be undertaken. 

3.3.1 Variables included in the regression models

As an extension of previously published regression findings (ONS, May 2013), the same variables were included in this analysis but more variables were added focusing particularly on aspects of the areas where people live. The following is the full list of variables included in the models:

  • Age

  • Sex

  • Ethnicity

  • Migration status

  • Relationship status

  • Economic activity status

  • Housing tenure

  • Self-reported health

  • Self-reported disability

  • Highest qualification held

  • Socio-economic status

  • Presence of dependent and non-dependent children in the household

  • Religious affiliation

  • Mode of interview (telephone or personal interview)

  • Day of the week of the survey

  • Month of the survey

  • Length of time at the current address1

  • A ‘place’ indicator

The development of the regression models is described more fully in ONS (May 2013). 

3.4 Analysis of variance

The extent of the individual or area effects on overall personal well-being, or on the differences in average personal well-being between the areas, can be examined via the coefficients of the regressions before and after accounting for individual characteristics and circumstances. However, to assess the extent to which differences in personal well-being between areas are due to individual or area effects formally and also the extent to which areas matter for individual outcomes of personal well-being , a method called variance decomposition analysis has been used. This involves the use of estimates from the regressions to calculate the variation in personal well-being ratings attributable to either area differences or to differences in the people living in the area.

This technique has also been used in other contexts such as labour economics to explore whether and to what extent sorting on the basis of individual characteristics influences wages (Gibbons et al 2010 or Combes et al 2007). This article used the methods developed by Gibbons et al (2010) to decompose the variation in individual personal well-being ratings into individual and group (or area) specific components. This method gives an estimate of the extent that the observed differences in average personal well-being between the areas reflect differences in the people living in them and/or whether it reflects differences in the areas themselves.
The detailed methodology of the variance decomposition method with an application to area disparities can be found in the paper by Gibbons et al 2010.

In brief, the regression analyses produced three different estimates of the amount of variance which were used to estimate how much of the observed differences in average personal well-being across the areas could be attributed either to the characteristics of the people living in each area (individual effects) or to differences in the areas themselves and the extent to which areas matter for individual outcomes of personal well-being. The following provides a brief description of the estimates which indicate the extent to which areas matter for individual outcomes of personal well-being:

  • The ‘raw variance share’, or RVS which is derived from the basic model2, shows how much of the variation in overall personal well-being ratings can be explained by the area, regardless of whether the differences relate to the area itself or to the people living there.

  • The ‘correlated variance share’, or CVS, shows how much of the variation in overall personal well-being ratings can be explained by the area, after accounting for the characteristics of the people living there. This measure excludes the direct contribution of individual effects on the average personal well-being of the area, however, it includes any indirect effect that the people living in the area may have on the area effects3.

  • The ‘uncorrelated variance share, or UVS, also shows how much of the variation in overall personal well-being ratings can be explained by the area, after accounting for the characteristics of the people living there. However, it focuses only on the contribution of area effects that are uncorrelated with the observed individual characteristics and circumstances of the people living in the area, therefore it can be interpreted as the ‘net’ effects of the area.

These variance estimates are then used to provide some indication on the contribution of individual or area effects to observed differences in average well-being between areas.

3.5 Interpreting the numbers

Two types of results are reported in the sections which follow. These are findings from:

  1. the variance analysis; and

  2. the regression analysis.

3.5.1 Understanding the variance analysis results

The numbers in Table 1 and Table 2 represent:

  • the proportions of overall personal well-being ratings that can be explained by area-related variables: 
    - before individual characteristics and circumstances are taken into account (RVS), and
    - after individual characteristics and circumstances are taken into account, either directly by area effects (UVS), or additionally including the indirect effect that the individuals living in the area may have on the area effects (CVS)

  • the proportion of the area share  of the area differences in observed average personal well-being.  That is the contribution of area effects (versus individual effects) to the differences in average personal well-being between areas (CVS or UVS / RVS).

3.5.2 Understanding the regression analysis results

The numbers included throughout the text and in the tables in section 5 are the unstandardised coefficients for each variable included in the regression models. This shows the size of the effect that the variable has on the specific aspect of personal well-being considered.

In interpreting the findings, it is important to remember that these numbers represent the difference between two groups when all other variables in the model have been held equal. The comparisons are therefore between two people who are otherwise the same in every respect apart from the particular characteristic, circumstance or area being considered. This helps to isolate the effect of any specific characteristic, circumstance or area on personal well-being.

Under the tables, when results are referred to as ‘significant at the 5% level’, this means there is a probability of less than .05 (or less than one in twenty) that the result could have occurred by chance.

Notes for 3. Methodology and Data

  1. This is an individual level variable but it is potentially related with ‘place’
  2. It is the R-squared from the basic model
  3. Because the estimated area effects are potentially correlated with individual characteristics/circumstances

4. Results - Variance Analysis

This section reports the results of the variance analysis. It focuses on the extent to which average differences in personal well-being between areas arise because of differences in the characteristics of people who live in these areas (individual effects) or differences in characteristics of the areas (area effects). Variance analysis also provides an estimate on the extent to which areas matter for individual outcomes of personal well-being.

To assess the role of individual versus area effects in understanding the differences in observed average personal well-being between areas, first, the differences between individuals were ignored and separate regressions were run for each of the four personal well-being questions and for each set of area indictors (dummies). These regressions, referred to as ‘the basic model’, showed how different areas were related to personal well-being without differentiating between area and individual effects. The results showed that there are statistically significant and occasionally sizeable differences in average personal well-being between areas1. These are represented by RVS in Table 1 which show that:

  • Life satisfaction had strongest association with area.

  • The weakest associations were found between daily emotions such as anxiety and happiness and area.

  • Among the area variables included in the analysis, areas grouped on the basis of the output area classification contributed most to the explanation of the differences in overall personal well-being ratings;

  • Local authorities contributed most to the explanation of the differences in overall anxiety yesterday ratings.

The analysis also found areas based on one common characteristic such as rural-urban, or green space were only able to explain a small amount of the differences in overall personal well-being even though these characteristics are derived from small geographies. However, this should not mean that these characteristics are not important for personal well-being. As mentioned before, it is implicitly assumed that area characteristics or the issues faced by the people living in that particular type of area are identical. While this is likely to be the case for IMDs and OACs it is less likely to be so for the rural-urban or built up areas. For example, among the urban places even within the same city there are large differences in terms of economic wealth, quality of life and character of the area. Similarly, in rural areas there may be differences in terms of access to public transport or differences in living costs relative to incomes.

Table 1: Variance decomposition of areas - proportion of the overall personal well-being ratings explained by the area before and after accounting for the characteristics and circumstances of people

Area Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Local Authority County
RVS 0.56% 0.45% 0.36% 0.46%
CVS 0.20% 0.18% 0.15% 0.37%
UVS 0.19% 0.17% 0.15% 0.35%
Local Authority District
RVS 1.14% 1.08% 0.85% 1.07%
CVS 0.57% 0.59% 0.53% 0.93%
UVS 0.56% 0.57% 0.52% 0.91%
Equivalised net weekly household income after housing costs in the area
RVS 0.60% 0.44% 0.36% 0.03%
CVS 0.01% 0.01% 0.01% 0.04%
UVS 0.01% 0.01% 0.01% 0.04%
Net weekly household income in the area
RVS 0.54% 0.43% 0.37% 0.03%
CVS 0.01% 0.02% 0.01% 0.04%
UVS 0.01% 0.02% 0.01% 0.03%
Index of Multiple Deprivation 
RVS 1.27% 0.87% 0.72% 0.14%
CVS 0.03% 0.01% 0.03% 0.01%
UVS 0.03% 0.00% 0.02% 0.01%
Green space per hectare
RVS 0.46% 0.36% 0.20% 0.11%
CVS 0.06% 0.03% 0.02% 0.03%
UVS 0.05% 0.03% 0.02% 0.03%
Built up areas
RVS 0.42% 0.33% 0.21% 0.15%
CVS 0.07% 0.04% 0.03% 0.06%
UVS 0.06% 0.04% 0.03% 0.05%
Rural-Urban
RVS 0.45% 0.35% 0.21% 0.16%
CVS 0.08% 0.04% 0.04% 0.07%
UVS 0.07% 0.04% 0.03% 0.06%
Population density
RVS 0.63% 0.49% 0.31% 0.17%
CVS 0.06% 0.03% 0.05% 0.04%
UVS 0.06% 0.03% 0.04% 0.04%
Output Area Classification
RVS 1.80% 1.48% 1.01% 0.38%
CVS 0.12% 0.15% 0.08% 0.15%
UVS 0.10% 0.13% 0.07% 0.13%

Table source: Office for National Statistics

Download table

The regressions were then run including both the area-related and the individual-related variables in the model. Again, separate regressions were run for each of the four personal well-being questions for each of set of area indictors. These regressions show how each personal characteristic is associated with personal well-being and the additional effects of areas on personal well-being after controlling for individual characteristics1.

Analysis of variation represented by CVSs and UVSs in Table 1 shows how much each area related variable considered in the analysis contributed to the overall differences in average personal well-being after taking account of individual characteristics and circumstances. It shows that the area related variables were only able to explain a small amount of the differences in personal well-being. For example, only about 0.10% of the variance in life satisfaction can be attributed exclusively to living in different OAC areas. The variable representing local authority districts explained more of the variance in personal well-being than any other area variable considered in the analysis.

Table 1 showed that, in general area effects (as shown by CVS or UVS) only explain a small variation in overall personal well-being after accounting for individual characteristics of the people. For details of the overall explanatory power of the regression models see technical appendix section 1.3).

Table 2 shows how much of the observed differences in average personal well-being between the areas could be attributed to the ‘area’ effects only (as opposed to individual effects).

Table 2 shows that for the positive measures of personal well-being, different attributes of the people living in each place (or sorting effects) accounted for most of the area differences in well-being. For example, individual characteristics and circumstances accounted for around 90% of the area differences in personal well-being between the OAC groups, with the areas accounting for the remaining 10% approximately (i.e. 6%,7%,9% for life satisfaction, happy yesterday and ‘worthwhile’ measures). However, for local authorities, area effects appeared to matter more than the other places considered here. For the local authority districts in particular, individual effects accounted approximately for half of the area differences in average personal well-being leaving area effects to account for the other half.

In contrast, area effects appeared to have a larger influence on the feelings of anxiety. For local authorities particularly, most of the observed variation in anxiety could not be attributed to the individual characteristics and circumstances of the people living in these areas. However, note the variation in ‘anxiety yesterday’ explained by the areas (as shown by RVS) were generally lower than the other personal well-being outcomes (even before accounting for individual characteristics and circumstances). Therefore, the proportions presented in Table 2 refer to an already small variation explained by the areas.

Table 2: Contribution of 'area' effects to area differences in average personal well-being

Area Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Local Authority County        
CVS/RVS 36% 40% 42% 80%
UVS/RVS 34% 38% 42% 76%
Local Authority District
CVS/RVS 50% 55% 63% 87%
UVS/RVS 49% 53% 61% 85%
Equivalised net weekly household income after housing costs in the area
CVS/RVS 2% 2% 3% 100%
UVS/RVS 2% 2% 3% 100%
Net weekly household income in the area
CVS/RVS 2% 5% 3% 100%
UVS/RVS 2% 5% 3% 100%
Index of Multiple Deprivation 
CVS/RVS 2% 1% 4% 7%
UVS/RVS 2% 0% 3% 7%
Green space per hectare 
CVS/RVS 13% 8% 10% 27%
UVS/RVS 11% 8% 10% 27%
Built up areas
CVS/RVS 17% 12% 14% 40%
UVS/RVS 14% 12% 14% 33%
Rural-Urban
CVS/RVS 18% 11% 19% 44%
UVS/RVS 16% 11% 14% 37%
Population density
CVS/RVS 10% 6% 16% 24%
UVS/RVS 10% 6% 13% 24%
Output Area Classification
CVS/RVS 7% 10% 8% 39%
UVS/RVS 6% 9% 7% 34%
 

Table source: Office for National Statistics

Download table

Note that, it is not possible to control for all the individual characteristics and circumstances that are associated with personal well-being. If some of these unobserved characteristics also influence sorting into areas, then these factors would be attributed incorrectly to area effects.

Note also that areas grouped as rural/urban, household income, built-up and population density cover England and Wales, IMD, OAC and local authorities cover Great Britain, and green space covers England only, therefore, the results within Tables 1 and 2 are not directly comparable. However, to test whether the higher variation in average personal well-being in the areas covering Great Britain was influenced by relatively higher well-being ratings in Scotland, we estimated IMD, OAC and local authorities without including the respondents from Scotland. The results were similar.

Notes for 4. Results - Variance Analysis

  1. The findings from these regressions are presented in section 5

5a. Results - Administrative Areas

The previous section on variance analysis provided a formal method of examining the extent to which area effects as opposed to individual effects accounted for differences in average personal wellbeing across areas. It showed that individual effects were generally more important than area effects for explaining the differences in observed average personal well-being between the areas (but that this result was not so strong for local authorities and did not apply for all the regressions explaining anxiety yesterday). Estimates of area effects can also be examined via the estimated coefficients of the regressions before and after controlling for individual characteristics. This section focuses on the size of the regression coefficients for the different area variables included in this analysis to both reconfirm the results from the previous section but also to provide additional detail.

There are three types of tables included in this section. The first of these (for example, table 3) compares the regression results for a) the ‘basic’ regression model without individual effects and b) the regression model after accounting for individual characteristics and circumstances by summarising the spread between the area with the highest and the area with the lowest average personal well-being in each case.

The second type of table shows the unstandardised coefficients for each of the areas considered in that section before controlling for individual characteristics and circumstances (for example, table 7) whilst the third type of table does the same but this time for the regression results after controlling for individual characteristics and circumstances (for example, table 9). Note that for local authorities and for the areas grouped by output area classifications these latter tables are not included in the article due to size considerations. However, the results are available in the reference tables.

The full results of each regression model are provided in the accompanying reference tables. The discussion in this section focuses on the size of identified area effects. However, the full results also include coefficients for each of the individual effects included in the model. In brief the results for the individual effects in each of the regressions were very similar and the key points were that:

  • Amongst the observed individual characteristics and circumstances included in the analysis, people’s self-reported health made the largest contribution to their levels of personal well-being, followed by their work situation and relationship status.

  • Other observed individual characteristics such as age, sex, ethnic group, migration status, religious affiliation, level of qualification, presence of children, reasons for economic inactivity, occupation or housing tenure were also associated in different ways to personal well-being, but none to the same extent as self-reported health, unemployment and relationship status. 

5. 1 ‘Area’ effects among administratively defined areas

5.1.1 Local authority counties (Great Britain)


Differences in average personal well-being levels between local authority counties were reported in the ONS publication “Personal wellbeing across the UK (ONS, October 2013)”. The publication showed that there was some (statistically significant) variation between the average personal well-being levels of the people living in different local authority across the UK.

The results of the regression analysis can be found in Reference tables 1. They show, for example, before controlling for individual effects, the average life satisfaction in Stoke-on-Trent and in Inner London were 0.43 and 0.42 points respectively lower than average life satisfaction in York which is the reference region in the analysis. At the high end, the average life satisfaction was 0.36 points higher in the county of Eilean Siar, Orkney and Shetland than in York.

Table 3 shows that before accounting for individual characteristics, the difference between the local authority with the lowest coefficient for average life satisfaction and the local authority with the highest coefficient for life satisfaction is 0.79 points (on a scale of 0–10). The table also shows the spread was largest for the anxiety yesterday question1.

Table 3: Distribution of area effects before and after controlling for individual characteristics: Local Authority Counties

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.79 0.60
Worthwhile 0.91 0.65
Happy yesterday 0.93 0.65
Anxious yesterday 1.21 1.23

Table source: Office for National Statistics

Table notes:

  1. Data for Great Britain from April 2012 to March 2013

Download table

It should be noted that whilst Table 3 is illustrating the maximum extent of the spread between the highest and lowest values, the full results in the accompanying reference tables 1 show that average well-being ratings between most of the local authorities were similar to each other. For example, results for 77 of the 139 local authorities were not statistically significantly different from the reference local authority of York.

After accounting for individual effects, with the exception of anxiety yesterday, the spread between the local authorities with the lowest and highest average well-being was less than the one found in the basic model (with no individual characteristics). For example, the difference in the observed average life satisfaction between the local authority with the lowest average life satisfaction and the local authority with the highest life satisfaction is now 0.60 points (on a scale of 0-10) as shown in Table 3. Additionally, the results in reference tables 1 show that the number of local authorities with statistically significantly different results to the reference local authority of York was much lower at just 13 out of 139.

These findings imply that some of the differences in average personal well-being between the local authorities can be mainly explained by the individual characteristics of the people living in them. For example, after accounting for individual effects, the average life satisfaction in inner London was no longer significantly different to average life satisfaction in York. By contrast, living in the county of Eilean Siar, Orkney and Shetland was still associated with higher average life satisfaction than living in York (0.30 points higher on average) suggesting the possible existence of positive area effects within this local authority.

Note that for anxiety, the results and implications were different. As shown by the variance shares previously in Table 2, for local authorities, individual effects accounted for a smaller proportion in the observed average anxiety ratings between the areas than the area effects.

Note that the estimates are based on a sample, therefore, there is a possibility that different samples give different results.

5.1.2 Local authority districts (Great Britain)

To examine the association between where we live and our personal well-being at a more localised level a similar analysis has also been carried out for the 407 local authority districts of Great Britain.

Table 4 shows the distribution of area effects. For example, before controlling for the individual characteristics the difference in the observed average life satisfaction between the local authority with the lowest average life satisfaction to the local authority with the highest life satisfaction is 1.35 points (on a scale of 0-10). The largest difference was observed for anxiety yesterday outcomes. The distribution of area effects after accounting for individual characteristics indicate that individual effects only accounted for some of the differences in average personal well-being between the local authority districts. The spread between the lowest and highest areas remained relatively large.

Table 4: Distribution of area effects before and after controlling for individual characteristics: Local Authority Districts

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 1.35 1.12
Worthwhile 1.52 1.08
Happy yesterday 1.53 1.21
Anxious yesterday 2.20 2.42

Table source: Office for National Statistics

Table notes:

  1. Data for Great Britain from April 2012 to March 2013

Download table

The full results of the regressions are presented in Reference tables 2. For example, in the basic models when taking the district of York as the reference area, London district of Harlow had the lowest average ratings for life satisfaction, happiness yesterday and feelings doings things that are worthwhile (0.84, 0.79 and 0.87 point lower than the average ratings observed in York, respectively), while the respondents living in the Horsham district reported the highest anxiety yesterday levels compared to the residents of York (1.14 points higher than York).

After accounting for individual effects, the average ratings for the positive personal well-being measures in Harlow district were still one of the lowest among the local authorities and statistically different from York. They were now 0.45, 0.50 and 0.56 points lower than the average ratings observed in York for life satisfaction, happy yesterday and ‘worthwhile’, respectively. The respondents living in the district of Horsham still had the highest average anxiety yesterday rating compared to the average anxiety levels observed in the district of York.

The tables also show after accounting for individual effects a larger number of local authorities had average well-being ratings which are not statistically significant from the reference local authority of York (only 41 local authorities were statistically significantly different from York after taking into account individual effects compared to 124 local authorities before taking individual effects into account). Again, these findings imply that some of the differences in average personal well-being between the local authorities can be explained by the individual characteristics of the people living in them. Similarly, for anxiety levels, the results and implications were different. As shown by the variance shares previously in Table 2, among local authorities, individual effects accounted for a smaller proportion in the observed average anxiety ratings between the areas than the area effects.

Note that the estimates are based on a sample, therefore, there is a possibility that different samples give different results. Also, for local authority districts, sample sizes are relatively small which increases the size of confidence intervals and reduces the chances of finding statistically significant difference between areas.

Notes for 5a. Results - Administrative Areas

  1. The coefficients in the regression as shown in reference tables are relative to a reference area and would change if a different reference area were chosen. However, the minimum to maximum spread is not affected by the choice of reference area.

5b. Results - LSOA/MSOA Area Groups

This section focuses on the findings for small area groupings classified together on the bases of similar environmental or socio-economic characteristics. As noted earlier, although the groupings are derived from smaller geographies, they are spread out across the Great Britain. As a result they may not capture differences associated with a specific location in the country. Also, it is implicitly assumed that the area characteristics including the issues faced by the residents of the same type of area are identical and this is a limitation of using these groupings.

5.2.1 Middle Layer Super Output Area (MSOA) level groups

Average household incomes in the area

Average household income of a neighbourhood1 is one of the contextual factors of an area which is potentially associated with personal well-being. Average income in an area could affect personal well-being in a number of ways. For example, several researchers such as, Diener et.al (1993), Ferrer-i-Carbonell (2005) or Lutmer (2004) found that a person’s or household’s income relative to their neighbours is one of the factors associated with personal well-being. Average income of an area could also be viewed as an indicator of the desirability of the area. For example, high income areas are generally associated with desirable characteristics, such as low crime, good resources and public services and stable communities living in the area which are all potentially related positively to personal well-being.

The MSOAs are the smallest geographical level for which income information is available. In this section we classified MSOAs on the basis of the average household income in the area to explore how living in relatively rich or poor MSOAs was associated with personal well-being.

Two sets of deciles of the average net weekly household income in the MSOAs were created; average net household income and average equivalised2 net household income after housing costs to adjust for differences in household composition. The information on average income estimates were derived from the small area income estimates3 for England and Wales. The income refers to the income a household receives from wages and salaries, self-employment, benefits, pensions, plus any other sources of income. The figures have been produced using a modelling methodology that combines survey, census and administrative data. Note that these estimates generally have large confidence intervals (i.e. they are subject to variability) and the estimates of average income do not reflect the spread of income (i.e. income inequality) in the areas. Indeed a number of deprived neighbourhoods (indicated by IMD) within the MSOAs with relatively high average household incomes in the APS dataset were found.

Looking only at the differences in average personal well-being across areas without controlling for personal characteristics and circumstances, personal well-being was found to be higher on average in the areas with higher incomes and lower in the areas with lower incomes compared to the areas with the lowest average household incomes. People living in the richer MSOAs also reported slightly lower anxiety levels yesterday than people living in the poorest MSOAs. The findings are shown in Tables 5 to 8. 

Table 5: Distribution of area effects before and after controlling for individual characteristics: Deciles of net household weekly income in an area after housing costs

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.41 0.05
Worthwhile 0.35 0.05
Happy yesterday 0.38 0.07
Anxious yesterday 0.19 0.17

Table source: Office for National Statistics

Table notes:

  1. Net household weekly income is equivalised for household composition
  2. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

Table 6: Distribution of area effects before and after controlling for individual characteristics: Deciles of average weekly net household income in an area

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.33 0.06
Worthwhile 0.25 0.05
Happy yesterday 0.31 0.09
Anxious yesterday 0.17 0.17

Table source: Office for National Statistics

Table notes:

  1. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

Table 7: Effects of relative net household weekly income in an area after housing costs on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy  yesterday Anxious yesterday
2nd income decile  0.078* 0.089* 0.021 -0.037
3rd income decile  0.124* 0.115* 0.115* -0.021
4th income decile  0.232* 0.201* 0.196* -0.108*
5th income decile  0.284* 0.227* 0.247* -0.116*
6th income decile  0.325* 0.268* 0.245* -0.071
7th income decile  0.367* 0.315* 0.341* -0.191*
8th income decile  0.391* 0.348* 0.325* -0.116*
9th income decile  0.397* 0.341* 0.378* -0.093*
10th income decile  0.407* 0.317* 0.368* -0.097*

Table source: Office for National Statistics

Table notes:

  1. The reference group for area net household weekly income is the bottom income decile
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

Table 8: Effects of relative net household weekly income in the area on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd income decile  0.163* 0.173* 0.182* -0.139*
3rd income decile  0.276* 0.25* 0.271* -0.129*
4th income decile  0.235* 0.235* 0.233* -0.093*
5th income decile  0.318* 0.27* 0.297* -0.107*
6th income decile  0.322* 0.293* 0.341* -0.16*
7th income decile  0.367* 0.34* 0.37* -0.132*
8th income decile  0.395* 0.382* 0.364* -0.129*
9th income decile  0.409* 0.347* 0.422* -0.132*
10th income decile  0.493* 0.419* 0.496* -0.172*

Table source: Office for National Statistics

Table notes:

  1. The reference group for relative income is the bottom income decile
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

After accounting for individual characteristics and circumstances the average household income in the MSOA did not appear to matter additionally to the life satisfaction, feelings of worthwhile and happiness of the respondents living in these areas. The findings are shown in Tables 9 and 10 where it can be seen there are hardly any results which are statistically significantly different to the reference group (bottom decile). These results are in line with those reported in Table 2 in section 4 which showed that individual characteristics and circumstances were able to explain almost all of the differences in average life satisfaction, and ‘worthwhile’ and ‘happiness yesterday’ between these areas.

Table 9: Effects of relative net household weekly income in an area after housing costs on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd income decile  -0.022 -0.003 -0.051 0.0263
3rd income decile  -0.041 -0.033 -0.016 0.0975*
4th income decile  -0.005 -0.007 0.005 0.0460
5th income decile 0.009 -0.016 0.020 0.0663
6th income decile  -0.018 -0.038 -0.047 0.1550*
7th income decile  -0.012 -0.028 0.019 0.0520
8th income decile  -0.003 -0.009 -0.012 0.1384*
9th income decile  -0.018 -0.032 0.023 0.1701*
10th income decile  -0.011 -0.055 -0.001 0.1548*

Table source: Office for National Statistics

Table notes:

  1. The reference group for area net household weekly income is the bottom income decile
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

Table 10: Effects of relative net household weekly income in the area on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd income decile  0.000 0.029 0.060 -0.031
3rd income decile  0.046 0.053 0.085* 0.031
4th income decile  -0.012 0.014 0.029 0.074
5th income decile  0.034 0.013 0.057 0.084
6th income decile  -0.005 0.004 0.065 0.062
7th income decile  -0.005 0.002 0.047 0.116*
8th income decile  0.009 0.029 0.03 0.111*
9th income decile  -0.012 -0.032 0.061 0.143*
10th income decile  0.028 -0.001 0.094* 0.111*

Table source: Office for National Statistics

Table notes:

  1. The reference group for relative income is the bottom income decile
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013) and the Small Area Income Estimates (2007/08)

Download table

The results were different for the feelings of anxiety yesterday data. After accounting for individual characteristics, living in higher income areas appeared to be associated with higher anxiety yesterday than living in the lower income areas. Those living in MSOAs with the top 30% of income had higher anxiety levels on average than those living in MSOAs with lowest average incomes. For example, respondents living in the MSOAs with the highest equivalised average net household incomes after housing costs reported their anxiety levels yesterday 0.15 points higher than the respondents living in the lowest income areas after controlling for individual effects.

The Reference Tables 3 and 4 provide the results for all the variables in the regression model for the MSOA areas grouped by average household incomes.

5.2.2 Lower Layer Super Output Area (LSOA) level groups


Social theories suggest that people are affected by neighbourhood context in a number of ways. For example, there are empirical studies showing that unemployment, besides having adverse effects on personal well-being of those who actually have no job, can also affect the well-being of other individuals such as their families, colleagues and neighbours. However, Many of the contextual factors that may have a potential influence on personal well-being at an area (or neighbourhood) level are highly correlated. For example, neighbourhoods with poor housing conditions may have high unemployment. This in turn makes it very difficult to identify exactly which contextual factor(s) may have an association with wellbeing. Therefore, instead of looking at neighbourhoods based just on one characteristic, such as average unemployment rate, this article utilised a number of composite indices that condense key information on the characteristics of the areas and its neighbourhood composition, namely the Output Area Classification (OAC) and the Index of Multiple Deprivation (IMD).

Deciles of Index of Multiple Deprivation (IMD) (Great Britain)


Previously published regression analysis by ONS (May 2013) considered how the relative level of deprivation in the area may affect personal well-being. In this section small LSOA level areas were analysed by comparing a standard ranking system based on how deprived they are relative to other areas. The ranking system, called the Index of Multiple Deprivation (IMD), takes into account a number of different aspects of the local area (economic, social, housing and environmental issues) and combines them into a single deprivation score for each area.

Table 11 indicates that individual characteristics and circumstances accounted for most of the differences in observed average well-being between these types of areas (also shown in Table 2).

Table 11: Distribution of area effects before and after controlling for individual characteristics: Indices of Multiple Deprivation (IMD)

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.69 0.08
Worthwhile 0.64 0.13
Happy yesterday 0.55 0.05
Anxious yesterday 0.33 0.09

Table source: Office for National Statistics

Table notes:

  1. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013), English IMD (2010), Scottish IMD (2012) and Welsh IMD (2013)

Download table

Table 12 shows that before accounting for individual characteristics and circumstances the reported average well-being increases as the deprivation of the local area decreases. 

Table 12: Effects of relative deprivation in the area on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd decile  0.157* 0.134 0.171* -0.104*
3rd decile  0.243* 0.212* 0.216* -0.122*
4th decile  0.333* 0.291* 0.332* -0.172*
5th decile  0.372* 0.317* 0.367* -0.187*
6th decile  0.463* 0.361* 0.428* -0.229*
7th decile  0.518* 0.439* 0.5* -0.319*
8th decile  0.586* 0.483* 0.532* -0.330*
9th decile  0.602* 0.479* 0.511* -0.299*
10th decile (least deprived) 0.691* 0.546* 0.641* -0.324*

Table source: Office for National Statistics

Table notes:

  1. The reference group for deprivation is the bottom decile (most deprived)
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013), English IMD (2010), Scottish IMD (2012) and Welsh IMD (2013)

Download table

The individual characteristics and circumstances accounted for most of the differences in the feelings of ‘worthwhile’ between the areas. However, those living in relatively less deprived areas rated their levels of life satisfaction and ‘happiness yesterday’ on average somewhat higher than those living in the most deprived areas after their individual characteristics were taken into account, however, on the average no difference was found in anxiety levels between the respondents living in the most and the least deprived areas. The findings are shown in Table 13.

Table 13: Effects of relative deprivation in the area on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd decile  0.026 0.032 0.063 -0.018
3rd decile  0.004 0.023 0.02 0.03
4th decile  0.015 0.034 0.077* 0.025
5th decile  -0.007 0.008 0.057 0.056
6th decile  0.031 0.005 0.075* 0.039
7th decile  0.039 0.04 0.106* -0.016
8th decile  0.075* 0.053 0.11* -0.008
9th decile  0.056* 0.016 0.062 0.044
10th decile (least deprived) 0.077* 0.026 0.129* 0.068

Table source: Office for National Statistics

Table notes:

  1. The reference group for deprivation is the bottom decile (most deprived)
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from the Annual Population Survey (April 2012 to March 2013), English IMD (2010), Scottish IMD (2012) and Welsh IMD (2013)

Download table

Reference table 5 provides the results for all the variables in this version of the regression model.

Relative concentration of green space in LSOA areas (England)

A considerable body of research (for example, Ambrey et.al (2012 and 2013) suggests that proximity to green space is associated with personal well-being through its effects on health. This may relate either to increased opportunities for physical activity in areas with more green space or to other health related issues that differ between areas with more or less green space, such as air pollution. The accessibility of local green spaces was also found to be valued more by urban residents than by those living in non-urban areas.

The findings in this section show the relationship between the amount of green space in local areas and personal well-being. For this analysis, deciles were created based on the green space per square metre in the LSOA areas.

Table 14 indicates that individual characteristics and circumstances accounted for some of the differences in observed average well-being between these types of areas (also shown in Table 2). 

Table 14: Distribution of area effects before and after controlling for individual characteristics: Deciles of green space per metre squared

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.38 0.15
Happy yesterday 0.32 0.11
Worthwhile 0.32 0.10
Anxious yesterday 0.30 0.15

Table source: Office for National Statistics

Table notes:

  1. Data for England from April 2012 to March 2013

Download table

The results shown in Table 15 indicate that a statistically significant relationship exists between the amount of the green space in the local area where a person lives and his/her personal well-being before accounting for the composition of the area. Residents living in areas where the amount of green space per square metre is relatively large reported higher ratings for all the measures of positive personal well-being (and lower anxiety levels yesterday) than people living in areas with smaller amounts of green space per square metre.

Table 15: Effects of the amount of green space in the area on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd decile  0.01 0.017 -0.033 -0.033
3rd decile  0.024 0.041 -0.053 -0.076
4th decile  0.023 0.051 -0.038 -0.094
5th decile  0.011 0.023 -0.03 -0.075
6th decile  0.064 0.037 -0.025 -0.125*
7th decile  0.094* 0.092* 0.028 -0.175*
8th decile  0.156* 0.125* 0.048 -0.222*
9th decile  0.264* 0.24* 0.138* -0.244*
10th decile  0.376* 0.323* 0.263* -0.296*

Table source: Office for National Statistics

Table notes:

  1. The reference group for green space is the ‘bottom decile’
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England from April 2012 to March 2013

Download table

Table 16 shows after accounting for personal characteristics and circumstances the amount of green space in the area appeared to have some additional association with personal well-being. People living in the areas with the largest amount of green space (top 20%) rated their  life satisfaction and ‘happiness yesterday’ higher on average than people living in the areas with the least amount of green space (the bottom decile). Consistent with literature, our results also found that green space appeared to matter more for ‘anxiety yesterday’ ratings than other well-being ratings. Results show that reported average levels of anxiety decreases as the amount of the green space in the local areas increases and the coefficients showing the association between personal well-being and green space were generally larger than the ones found for the other measures of personal well-being.

Table 16: Effects of the amount of green space in the area on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
2nd decile  0.021 0.024 -0.016 -0.025
3rd decile  0.044 0.057* -0.026 -0.066
4th decile  0.047 0.064* -0.005 -0.085
5th decile  0.012 0.019 -0.012 -0.054
6th decile  0.044 0.013 -0.025 -0.083
7th decile  0.055 0.047 0.014 -0.116*
8th decile  0.062* 0.028 -0.01 -0.146*
9th decile  0.107* 0.086* 0.022 -0.126*
10th decile  0.152* 0.098* 0.083* -0.135*

Table source: Office for National Statistics

Table notes:

  1. The reference group for green space is the ‘bottom decile’
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England from April 2012 to March 2013

Download table

Reference table 6 provides a summary of the results for all the variables in the regression model.

Notes for 5b. Results - LSOA/MSOA Area Groups

  1. This section only examines the relationship between the average household income in the area and personal well-being. For the relationship between household income and personal well-being see the recently published ONS article.
  2. Equivalisation adjusts for differences in household composition in order to give a measure that can be used meaningfully to compare incomes between different sizes and types of households. Further information on definitions of income and equivalisation can be found in the 2012 Family Spending Edition.
  3. The latest available data was 2008 small area income estimates. It is assumed that the average household incomes in these areas did not change significantly since then.

5c. Results - Output Area Groups

5.3 ‘Area’ effects based on Output Area groups

Rural-Urban (England and Wales)

Previously published regression analysis findings (ONS, May 2013) showed that after taking into account individual and household characteristics, the average well-being of people living in rural areas was higher than the average well-being of people living in urban areas. However, it was also shown that the size of the relationship between living in an urban or rural area and personal well-being was generally small. It was also noted that these results only reflected the average personal well-being ratings between relatively large geographical areas and this may mask important differences between smaller or local areas within them.

To explore this further, this analysis used the 2001 census rural urban classification which splits urban areas into three different groups and rural areas into five different groups. Due to its vastly large size, a separate category was created for London.

Table 17 indicates that (with the exception of anxiety yesterday outcomes) individual characteristics and circumstances accounted for most of the differences in observed average well-being between these types of areas (also shown in Table 2).

Table 17: Distribution of area effects before and after controlling for individual characteristics: Rural and urban areas

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.52 0.31
Worthwhile 0.39 0.21
Happy yesterday 0.43 0.11
Anxious yesterday 0.66 0.53

Table source: Office for National Statistics

Table notes:

  1. Data for England and Wales from April 2012 to March 2013

Download table

Table 18 shows that before accounting for individual characteristics, in general rural areas are associated with higher well-being than urban areas. Examining urban areas alone, people living in other types of urban areas also reported higher average life satisfaction and lower average levels of anxiety yesterday than people living in London.

Table 18: Effects of living in types of rural and urban areas on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Urban (Less Sparse) 0.157* 0.095* 0.031 -0.227*
Urban (Sparse) 0.306* 0.174* 0.084 -0.08
Town and Fringe (Sparse) 0.366* 0.166* 0.316* -0.657*
Village (Sparse) 0.519* 0.389* 0.316* -0.376*
Hamlet and Isolated Dwellings (Sparse) 0.49* 0.373* 0.316* -0.405*
Town and Fringe (Less Sparse) 0.318* 0.26* 0.316* -0.341*
Village (Less Sparse) 0.455* 0.372* 0.316* -0.411*
Hamlet and Isolated Dwellings (Less Sparse) 0.479* 0.386* 0.316* -0.457*

Table source: Office for National Statistics

Table notes:

  1. The reference group for the rural and urban type areas is London – a very large urban area
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Table 19 shows that after accounting for individual characteristics and circumstances, generally there remains a positive association between personal well-being and living in rural areas compared with living in London. The coefficients were largest for life satisfaction and ‘happy yesterday’ for sparse villages whilst anxiety yesterday levels were lowest relative to London in the Sparse Town and Fringe category.

Examining urban areas alone, the average life satisfaction ratings in the other urban areas were still higher than the average ratings in London. However, for the ‘happy yesterday’ measure, London was no different from either the urban (sparse) or urban (less sparse) areas, however, London was lower on this measure than the Town and Fringe areas.

Table 19: Effects of living in types of rural and urban areas on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Urban (Less Sparse) 0.088* 0.062* 0.024 -0.178*
Urban (Sparse) 0.213* 0.147 0.061 -0.012
Town and Fringe (Sparse) 0.186* 0.027* 0.21* -0.527*
Village (Sparse) 0.309* 0.211 0.283* -0.249*
Hamlet and Isolated Dwellings (Sparse) 0.234* 0.123 0.246* -0.248*
Town and Fringe (Less Sparse) 0.128* 0.108* 0.057 -0.222*
Village (Less Sparse) 0.186* 0.131* 0.126* -0.252*
Hamlet and Isolated Dwellings (Less Sparse) 0.186* 0.119* 0.107* -0.275*

Table source: Office for National Statistics

Table notes:

  1. The reference group for the rural and urban type areas is London – a very large urban area
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Reference table 7 provides the results for all the variables in the regression model.

Built up areas (England and Wales)

Another way of investigating differences between types of areas is to use the ONS Built-Up Areas categorisation. This data provides information on the villages, towns and cities where people live, and allows comparisons between people living in built-up areas and those living elsewhere. Similar to disaggregated rural and urban analysis, London is included as a separate category.

Table 20 indicates that individual characteristics and circumstances generally accounted for most of the differences in observed average well-being between these types of areas (also shown in Table 2).

Table 20: Distribution of area effects before and after controlling for individual characteristics: Built-up areas

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.45 0.19
Happy yesterday 0.35 0.15
Worthwhile 0.35 0.12
Anxious yesterday 0.43 0.28

Table source: Office for National Statistics

Table notes:

  1. Data for England and Wales from April 2012 to March 2013

Download table

Table 21 suggests that, before accounting for individual effects, average well-being ratings generally increase as the size of the built up area reduces. As these groupings are highly correlated with the green space and rural urban groupings analysed before, the findings were very similar.

Table 21: Effects of living in types of built-up areas on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Major (1 million -3 million) 0.098* 0.04 -0.018 -0.228*
Large (500,000 - 999,999) 0.16* 0.063* -0.011 -0.199*
Medium (100,000-499,999) 0.143* 0.1* 0.009 -0.202*
Small (10,000-99,999) 0.214* 0.15* 0.103* -0.265*
Minor (<10,000) 0.37* 0.299* 0.228* -0.375*
non-Built Up 0.452* 0.346* 0.336* -0.431*

Table source: Office for National Statistics

Table notes:

  1. The reference group for built-up areas is London
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Table 22 shows that there remains an association with personal well-being and living in relatively less built-up areas, compared with living in London, even after controlling for individual characteristics and circumstances. For example, reported average life satisfaction, the feelings of doing things that are worthwhile and happiness yesterday were 0.19, 0.15 and 0.10 points respectively higher for those living in the least built up areas than those living in the most built up areas (London). Similarly, the reported anxiety yesterday levels of the people living in the least built-up areas were 0.28 points lower on average than those living in London.

Table 22: Effects of living in types of built-up areas on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
Major (1 million -3 million) 0.06* 0.03 -0.004 -0.191*
Large (500,000 - 999,999) 0.118* 0.061* 0.011 -0.177*
Medium (100,000-499,999) 0.08* 0.074* 0.01 -0.161*
Small (10,000-99,999) 0.093* 0.068* 0.056 -0.178*
Minor (<10,000) 0.154* 0.118* 0.089* -0.235*
non-Built Up 0.191* 0.099* 0.148* -0.277*

Table source: Office for National Statistics

Table notes:

  1. The reference group for built-up areas is London
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Reference table 8 provides the results for all the variables in the regression model.

Population density (England and Wales)

Research suggests that population density is another potential factor which may influence personal well-being. Studies have generally shown an inverse relationship between the two (i.e. personal well-being increases as population density decreases). A high level of urbanisation (high density) has also been shown by some studies to be associated with higher levels of psychosis and depression (Cramer et. al 2004 or Sundquist et. al 2004).

There are several ways in which population density may be related to personal well-being, including:

  • Air and noise pollution- it increases with density and this in turn leads to health problems both physical and mental.

  • Restricted physical activity.

  • Social stress – for example, Lederborgen et.al (2013) found that brain area activity differences associated with urbanisation may be linked to the higher incidence of schizophrenia in urban areas compared to non-urban areas. They interpret their findings as showing a causal relationship.

  • Lack of trust or fear of crime.

To explore the association between personal well-being and population density in the local area, 13 groups were created based on the 2011 Census population per hectare in the output areas.

Table 23 indicates that individual characteristics and circumstances generally accounted for most of the differences in observed average well-being between these types of areas (also shown in Table 2).

Table 23: Distribution of area effects before and after controlling for individual characteristics: Population Density per Hectare

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.63 0.22
Happy yesterday 0.45 0.18
Worthwhile 0.53 0.12
Anxious yesterday 0.6 0.38

Table source: Office for National Statistics

Table notes:

  1. Data for England and Wales from April 2012 to March 2013

Download table

Table 24 shows before accounting for individual effects, reported average well-being ratings generally reduce as the population density increases.

Table 24: Effects of population density per hectare in the output area on personal well-being before controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
1 person per hect 0.031 0.055 -0.023 -0.138
2-3 people per hect -0.077 -0.061 -0.079 0.063
4-13 people per hect -0.141* -0.118* -0.152* 0.045
14-25 people per hect -0.179* -0.140* -0.207* 0.154*
26-35 people per hect -0.232* -0.191* -0.275* 0.113
36-45 people per hect -0.253* -0.171* -0.25* 0.108
46-54 people per hect -0.322* -0.251* -0.359* 0.216*
55-65 people per hect -0.374* -0.298* -0.396* 0.256*
66-80 people per hect -0.39* -0.328* -0.39* 0.278*
81-112 people per hect -0.403* -0.324* -0.342* 0.273*
113-250 people per hect -0.522* -0.436* -0.454* 0.349*
251+ people per hect -0.596* -0.477* -0.407* 0.463*

Table source: Office for National Statistics

Table notes:

  1. The reference group is an average of zero person per Hectare
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Table 25 shows that population density of the local area has no additional association with a person’s feelings that the things they do in their lives are ‘worthwhile’. However, there was still an association between other measure of personal well-being and living in relatively less dense areas. For example, people living in the most populated output areas rated their life satisfaction on average 0.22 points lower than those living in the least dense areas. Their reported average anxiety yesterday was also 0.27 points higher than those living in the least populated areas. For ‘happy yesterday’ measure, least dense areas were no different from the most populated areas.

Table 25: Effects of population density per hectare in the output area on personal well-being after controlling for individual characteristics

Coefficients

  Life satisfaction Worthwhile Happy yesterday Anxious yesterday
1 person per hect 0.001 0.06 -0.038 -0.108
2-3 people per hect -0.033 0.004 -0.034 0.044
4-13 people per hect -0.08* -0.021 -0.081* 0.015
14-25 people per hect -0.078* -0.003 -0.105* 0.107
26-35 people per hect -0.111* -0.034 -0.155* 0.057
36-45 people per hect -0.116* -0.004 -0.116* 0.038
46-54 people per hect -0.144* -0.053 -0.183* 0.111
55-65 people per hect -0.155* -0.061 -0.178* 0.13*
66-80 people per hect -0.126* -0.05 -0.139* 0.134*
81-112 people per hect -0.096* -0.01 -0.066 0.095
113-250 people per hect -0.17* -0.057 -0.138* 0.131*
251+ people per hect -0.219* -0.062 -0.089 0.268*

Table source: Office for National Statistics

Table notes:

  1. The reference group is an average of zero person per hectare
  2. * shows that the relationship is statistically significant at the 5% level
  3. Data for England and Wales from April 2012 to March 2013

Download table

Again these groupings were highly correlated with the green space, rural urban and built-up area groupings analysed before, therefore the findings were very similar.

Reference Table 9 provides the results for all the variables in the regression models.

Output Area Classifications (Great Britain)

The Output Area Classification (OAC) is another composite index that condenses key information on the characteristics of the areas and its neighbourhood composition: The Output Area Classification (OAC) distils key results from the 2001 Census for the whole of the UK at a fine grain to indicate the character of local areas. The classification groups areas into clusters based on similar characteristics: demographic, household composition, housing, socio-economic, employment and industry sector.

Table 26 shows that there is quite a large difference between the OAC area with the highest and lowest average personal wellbeing before accounting for individual characteristics. However, once individual characteristics are considered, the difference in average personal well-being between OAC areas reduces considerably. This is consistent with the results reported in Table 2.

Table 26: Distribution of area effects before and after controlling for individual characteristics: Output Area Classification

Points on the 0–10 point scale

  Minimum to maximum (before) Minimum to maximum (after)
Life satisfaction 0.93 0.22
Happy yesterday 0.93 0.35
Worthwhile 0.84 0.28
Anxious yesterday 0.77 0.55

Table source: Office for National Statistics

Table notes:

  1. Data for Great Britain from April 2012 to March 2013

Download table

There are 52 OAC areas so it is not possible to display all the coefficients in a table in this article. However, the regression results can be found in the accompanying Reference Tables 10. These show that among the 52 neighbourhood types, the highest average personal well-being ratings were found in the areas located in the countryside and in the prospering suburbs. The lowest average personal well-being was observed in the relatively deprived areas (neighbourhood types 5 and 7).

After controlling for individual characteristics and circumstances, (with the exception of anxiety yesterday levels) individual effects were able to explain most of the differences in average personal well-being between these areas and there was very little additional association with personal well-being and living in most of these areas. For example, only 6 out of 51 OAC area groupings were statistically significantly different from the reference group. Neighbourhoods with a negative association with personal well-being were located in relatively deprived city areas in England, especially in London and some around Birmingham (area types 7). People living in some of these areas (e.g. area types 7b) also reported higher anxiety levels on the average than people living the reference area, however, reported anxiety yesterday levels was also found to be higher in more affluent and educated parts of the cities mainly in London and Scotland (area types 2a2, 2b2) than in some of the deprived neighbourhoods. Similar to the findings in previous sections, neighbourhoods with a positive association with well-being were located in the rural areas.

Note in Reference Table 10, the areas are identified by their OAC codes. Full explanations of each of the OAC subgroups can be found in the OAC summary document.

6. Discussion and Policy Implications

This article has extended earlier ONS research on the relationships between personal well-being and place and offers a further step in understanding the relationship between where we live and personal well-being. The main findings of the regression and the variance analysis are summarised below:

  • Before accounting for individual characteristics and circumstances, the data showed that areas or types of areas have considerable associations with individual personal well-being outcomes.

  • After accounting for individual characteristics and circumstances, the data showed that areas or types of areas have much lower associations with individual personal well-being outcomes.

  • Observed differences in average life satisfaction, happiness yesterday and peoples’ feelings that the things they do in their lives that are ‘worthwhile’ between the areas or area types mainly reflected the different characteristics of people living in them.

  • Local authority districts had the largest amount of variation in average personal well-being unexplained by individual effects.

  • Area effects appeared to matter more for the feelings of anxiety between the areas or area types than for the other measures of personal well-being for some of the areas and area types.

  • Among the individual characteristics and circumstances considered, people’s self-reported health was the most important factor associated with personal well-being, followed by their work situation and then their relationship status. Other factors such as people’s age, sex, ethnic group, migration status, religious affiliation, level of qualification, presence of children, reasons for inactivity, occupation and home ownership were also associated in different ways to personal well-being, but none to the same extent as health, unemployment and relationship status.

It should be noted, however, that the regression technique applied to the personal well-being data cannot be considered to give definitive answers to the issue of the importance of place on personal wellbeing. The results above are important indications of the likely impact of place on personal well-being but they are only part of the story. To understand this, it is necessary to consider both a number of caveats to the regression work and also to look at the subject in a slightly wider context.

A key caveat is that it is more difficult to disentangle the individual effects and the area effects than the above suggests. The two aspects cannot be entirely separated. For example, air pollution may negatively impact on an individual’s health which is a case where the area effect is influencing personal well-being via its impact on an individual effect (health). It is difficult for regression, in particular a cross-sectional type regression analysis, to disentangle these impacts. It is also difficult to know at what geographical level different factors might influence personal well-being and to ensure that all the possibilities are covered. More advanced statistical techniques and additional data (such as panel data) may be required to explore this further.

Similarly, it is difficult for regression work to establish which aspects of place may impact on personal well-being. This is because there are often strong correlations between different factors related to place. For example, rural areas will typically have low noise pollution, low crime, low air pollution, more scenic views etc. relative to urban areas and isolating which of these is actually the important factor(s) in influencing personal well-being is very challenging.

In terms of providing a wider context, it needs to be borne in mind that individuals are not randomly distributed across the country. Individuals have choices about where to live, and in making their choices they will weigh up many issues, some of which will include the characteristics of the local area. Typically, it is found that people of similar characteristics (in terms of income etc.) tend to be clustered together in similar areas. When examining differences in personal well-being across areas it is the existence of this clustering of individuals that explains much of the difference in observed average personal well-being across areas.

Why this sorting and clustering takes place, and the degree to which aspects of the local area influence this process are not in the scope of this article, but are clearly interesting topics for anyone wishing to consider the average personal well-being differences across areas and how they change over time.

So what conclusions should be taken away from this working paper?

  • A key finding was that the analysis largely confirmed that individual characteristics are usually key to personal well-being and also to the average personal well-being differences between areas (subject to the caveats).

  • Area effects were found to exist to some degree by the regression work. However, it is difficult to isolate exactly what the key determinants of area effects may be. Additionally, the process of isolating individual compared to area effects is not clear cut with some overlap likely to exist. Finally, the analysis did not examine the degree to which ‘area effects’ may influence individuals housing location choice – yet it is this clustering that is one of the key influences in why average personal well-being across two different areas will often be different.

  • In terms of policy, the relative importance of individual effects may suggest that people based policies would have greater potential for well-being improvements than area-based policies. However, this finding needs to be balanced against the respective costs of the different policies whilst the uncertainties around the analysis discussed above would also need to be considered. In reality, the mix of appropriate place- versus individual-based policies is likely to differ depending on policy aims.

  • Although this analysis found relatively little direct area effects on personal well-being, further research on area effects should be encouraged. This could involve examination of additional aspects of local areas beyond those included here. More consideration could also be given to neighbourhoods and the factors that encourage people to choose one neighbourhood over another, drawing on the body of literature on ‘residential location choice’. 

Background notes

  1. Details of the policy governing the release of new data are available by visiting www.statisticsauthority.gov.uk/assessment/code-of-practice/index.html or from the Media Relations Office email: media.relations@ons.gsi.gov.uk

References

  1. Ambrey, C.L. and Fleming, C.M. (2012). ‘Valuing Australia's Protected Areas: A Life Satisfaction Approach’, New Zealand Economic Papers, vol. 46, no. 3, 191–209
  2. Ambrey, C.L., Fleming, C.M., (2013). ‘Public Greenspace and Life Satisfaction in Urban Australia’, Urban Studies, vol. 51, no. 6, 1290-1321
  3. Balas, D. (2013). ‘What makes a happy city?’, Cities 32 (2013) S39-S50
  4. Combes, P.P., Duranton, G. and Gobillon, L. (2007). ‘ Spatial wage disparities: Sorting matters!’, Journal of Urban Economics, Volume (Year): 63 (2008), Issue (Month): 2 (March), 723-742
  5. Cramer, V., Torgersen, S. and Kringlen, E. (2004). ‘ Quality of Life in a City: The Effect of Population Density’, Social Indicators Research, Vol. 69, No. 1 (Oct., 2004), pp. 103-116
  6. Cummins, S. Curtis, S. Diez-Roux, A.V. and Macintyre, S. (2007). ‘ Understanding and representing ‘place’ in health research: A relational approach, Social Science & Medicine 65 (2007) 1825–1838
  7. Diener, E., Sandvik, E., Seidlitz, L., & Diener, M. (1993). The relationship between income and subjective well-being: Relative or absolute? Social Indicators Research 28, 195–223
  8. Dorling,D.,  Smith, G.,  Noble, M.,  Wright, G.,  Burrows, R., Bradshaw, J., Joshi, H., Pattie, C., Mitchell, R., Green, A. E., McCulloch, A. (2001). ‘How much does place matter?’, E&PA, Volume 33, No. 8., 1335-1369
  9. Douglas, I. (2004). ‘Urban green space and mental health’. UK MAB Urban Forum
  10. Ferreira, S., Akay, A., Brereton, F., et al. (2013). ‘Life satisfaction and air quality in Europe’, Ecological Economics. 88:1-10
  11. Ferrer-i-Carbonell, A. (2005). ‘ Income and well-being: an empirical analysis of the comparison income effect’, Journal of Public Economics 89 (2005) 997–1019
  12. Ferrer-i Carbonell, A. and  Frijters, P. (2004). ‘How important s methodology for the estimates of the determinants of happiness?’, The Economic Journal, 114 (July), 641-659
  13. Gibbons, S, Overman, H.G. and Pelkonen, P. (2010). ‘Wage disparities in Britain: People or Place?’, SERC Discussion Paper, 60
  14. Greene, W.H. (2000). ‘Econometric Analysis’, Upper Saddle River, NJ: Prentice Hall, fourth edition
  15. Larrabee. B.R. (2009). ‘Ordinary least squares regression of ordered categorical data: Inferential implications for practice’, A Report submitted in partial fulfilment of the requirements for the degree of Master of Science, Kansas State University
  16. Lederbogen, F, Haddad,L. and Meyer-Lindenberg, A (2013). ‘Urban social stress – Risk factor for mental disorders. The case of schizophrenia’, Environmenal Pollution (2013), 1-5
  17. Lutmer, E.F.P. (2004). ‘Neighbours as negatives: Relative earnings and well-being’, NBER Working Paper No. 10667
  18. Macintyre, S, Ellaway, A. and Cummins, S. (2002). ‘Place effects on health: how can we conceptualise, operationalise and measure them?’, Social Science & Medicine 55 (2002) 125–139
  19. Morrison, P.S. (2007). ‘Subjective wellbeing and the city’, Social Policy Journal of New Zealand, 31(July), 74-103
  20. Office for National Statistics - Oguz, S., Merad, S. and Snape, D. (May 2013). ‘Measuring National Well-being – What matters most to personal well-being?
  21. Office for National Statistics (October 2013). ‘Measuring National Well-being – Personal Well-being across the UK, 2012/13’
  22. Osborne, J., and Waters, E.  (2002). ‘Four assumptions of multiple regression that researchers should always test. Practical Assessment, Research & Evaluation, 8(2)
  23. Oswald, A., & Wu, S. (2009). Well-being across America: Evidence from a random sample of one million U.S. Citizens. University of Warwick working paper
  24. Sundquist, K., Frank, G.L. and Sundquist, J. (2004). ‘Urbanisation and incidence of psychosis and depression: Follow-up study of 4.4 million women and men in Sweden’, British Journal of Psychiatry, (2004) 184, 293-298
  25. White,M.P, Alcock, I., Wheeler, B. W.  and Depledge, M. H. (2013). ‘Would You Be Happier Living in a Greener Urban Area? A Fixed-Effects Analysis of Panel Data’, Psychological Science, published online 23 April 2013

7. Technical Annex

7.1 Why undertake a regression analysis?

In analysis which looks at the relationship between two variables, it can be tempting to infer that one variable is directly related to the other. For example, people in one ethnic group may have higher life satisfaction than those in another ethnic group, but can we assume that the differences observed in relation to life satisfaction ratings are primarily about ethnic differences? This conclusion would only be justified if we could show there were no other important differences between the ethnic groups which might affect the findings such as differences in health or employment status.

Regression analysis allows us to do this by holding all the variables in the model equal while measuring the size and strength of the relationship between two specific variables. If the regression results show a significant relationship between ethnicity and life satisfaction, then this means that two people who are identical in every way apart from their ethnicity would indeed rate their life satisfaction differently. This implies a direct relationship between ethnicity and life satisfaction even when the other variables included in the analysis are taken into account. Therefore, the key benefit of regression analysis is that it provides a better method than analysis looking at the relationship between only two variables at a time of indentifying those factors which matter most to personal well-being.

However, every analytical method has its limitations and regression analysis is no exception. The following sections summarise some key considerations which should be borne in mind in terms of the statistical assumptions underlying the techniques used here and the types of inference which can be drawn from the findings.

7.1.1 Using OLS for ordered responses and the robustness of the OLS estimates

A key implicit assumption in OLS regression is that the dependent variable (the outcome we are trying to explain, such as the personal well-being ratings) is continuous. Continuous data is data that can take any value (usually within a range). For example, a person’s height could be any value within the range of human heights or time in a race which could even be measured to fractions of a second. The personal well-being survey responses, however, are discrete, i.e., they can only take on a relatively small number of integer values, such as 6 or 10 with no other values such as halves in between.

OLS regression also assumes that the values of the dependent variable (e.g., personal well-being ratings) are cardinal, i.e. the interval between any pair of categories such as between 2 and 3 is of the same magnitude as the interval between any other similar pair such as between 6 and 7. As the personal well-being responses are only rankings we cannot know whether for example the distance between 2 and 3 is the same as the distance between 6 and 7. For example, it may be the case that it doesn’t take much for people to move from 2 to 3 in life satisfaction ranking, but it may take a lot more for them to jump from 6 to 7. Therefore, the OLS regression approach may not be well suited for modelling this kind of dependent variable.

There are a number of alternatives to OLS for modelling discrete response variables, such as logit or probit regression. In these models the categories of the responses are treated separately (i.e. there is no order to the categories, for example, 4 is not higher than 3). The disadvantage of these methods is that the information contained in the ordering of the personal well-being ratings is lost. However, a way of overcoming this issue is to create two categories, for example rankings below 7 and above 7, but the categories will be artificial.

An alternative method is to treat the response variable as ordinal and use regression techniques, such as ordered logit or ordered probit that are developed to deal with ordinal data. Ordinal data values can be ranked or ordered on a scale such as from 0 to 10 with each higher category representing a higher degree of personal well-being (or lower personal well-being in the case of anxiety) and unlike the OLS method, ordered probit or ordered logit regression does not assume that the differences between the ordinal categories in the personal well-being rankings are equal.

They capture the qualitative differences between different scores. It is important to note that ordinal probit/logistic performs several probit/logistic regressions simultaneously, assuming that the models are identical for all scores. The latter assumption can be relaxed but the interpretation of the results becomes more difficult. The major advantage of such models is that it takes the ordinal nature of the response variable (i.e. personal well-being rankings) into account without assuming equality of distance between the scores. Similarly to OLS, it identifies statistically significant relationships between the explanatory variables (e.g. age, health, etc) and the dependent variable (personal well-being ranking); however, the estimated coefficients have no direct interpretation.

The existing literature also suggests that OLS may still be reasonably implemented when there are more than five levels of the ordered categorical responses, particularly when there is a clear ordering of the categories e.g. levels of happiness with 0 representing the lowest category and 10 representing the highest category (for example see Larrabee, 2009). Indeed, several studies including our regression analysis (May 2013 and February 2014) applied both methods to personal well-being data and found that there is little difference between the OLS and the theoretically preferable methodologies such as ordered probit. For example, see Ferrer-i-Carbonell and Fritjers (2004) for a detailed discussion of this issue. Also, Greene (2000) points out, the reasons for favouring one method over the other (such as ordered probit or ordered logit) is practical and in most applications it seems not to make much difference to the results. The main advantage of OLS is that the interpretation of the regression results is more simple and straightforward than in alternative methods.

Because the ordered probit regressions were not suitable for the variance analysis only the findings from the OLS regressions are reported in this article. However, we have estimated the regressions in ordered probit to test the robustness of the OLS results. As before, the statistical significance, the signs and the relative sizes of the regression coefficients were similar between the two methods were very similar.

7.1.2 Diagnostic checks of the OLS regressions

Post regression diagnostics identified some violations of the OLS regression assumptions such as model specification and the normality of residuals. However, as some studies (for example see Osborne and Waters, 2002), suggest that several assumptions of OLS regression are ‘robust’ to violation such as normal distribution of residuals and others are fulfilled in the proper design of the study such as the independence of observations. In this analysis, using the survey design controlled for the potential dependence of the individual observations with each other (see section 7.2) and applying the survey weights provided some protection against model misspecification.

As there is no formal statistical test that can be used to identify multi-collinearity when the covariates in the model are dummy variables, an informal method of cross-tabulating each pair of dummy variables can be used. When cross-tabulations showed very high correlation between the variables they were not used in the regression. An example where this was the case is between the variables "reported being a Muslim" and "reported being Pakistani"; to get around this problem in this example, the dummy variables for the individual religions from the model were replaced with a single dummy variable "reported a religion".

Stata automatically computes standards errors that are robust to heteroskedasticity when the regressions are estimated incorporating survey design.

Additionally, estimating the models using different specifications as well as two methods (OLS and ordered probit) confirmed that the magnitude and the statistical significance of the parameter estimates did not significantly change and the general inferences from the analysis remained the same.

7.1.3 The explanatory power of the models

It is important to note that the explanatory power of the regression models used here is relatively low. Indeed, the amount of variance that has been explained by the model is similar to that of other reported regression analyses undertaken on personal well-being. For the ‘happy yesterday’, ‘anxious yesterday’ and ‘worthwhile’ questions, around 10% - 14% of the variation between individuals is explained by the variables included in the model. By contrast, a much higher proportion (19%) of the individual variation in ratings for life satisfaction was explained by the model.

The lower explanatory power of the model could be due to leaving out important factors which contribute to personal well-being. For example, genetic and personality factors are thought to account for about half of the variation in personal well-being. It has not been possible to include variables relating to personality or genes in the models as the APS does not include data of this type.

The subjective nature of the outcome variable also means that it is probably measured with some imperfect reliability. The lower the reliability of the outcome variable, the more unclear its correlations with other variables will tend to be.

7.1.4 Omitted variable bias

In an ideal world, a regression model should include all the relevant variables that are associated with the outcome (i.e. variable being analysed such as personal well-being). In reality, however, we either cannot observe all the potential factors affecting well-being (such as personality) or are limited by whatever information is collected in the survey data used in the regression analysis.

If a relevant factor is not included in the model, this may result in the effects of the variables that have been included being mis-estimated. When the omitted variables are correlated with the included variables in the model, the coefficient estimates of those variables will be biased and inconsistent. However, the estimated coefficients are less affected by omitted variables when these are not correlated with the included variables (i.e. the estimates will be unbiased and consistent).

In the latter case, the only problem will be an increase in the estimated standard deviations of the coefficients which are likely to give misleading conclusions about the statistical significance of the estimated parameters.

7.1.5 Causality

Regression analysis based on cross-sectional observational data cannot establish with certainty whether relationships found between the independent and dependent variables are causal. This is particularly the case in psychological contexts where there may be a reciprocal relationship between the independent and the dependent variables. For example, the usual assumption is that individual characteristics or circumstances like health or employment status are independent variables which may affect personal well-being (viewed here as a dependent variable). However, some of the association between health and well-being may be caused by the impact of personal well-being on health.

Furthermore, as the data used in the regression analysis here are collected at one point in time (i.e. cross-sectional), they are not able to capture the effect of changes over time and which event preceded another. For example, it is not possible to tell from this data whether the perception of being in bad health precedes a drop in well-being or whether a drop in well-being precedes the perception that one is in bad health. We can only definitely say that the perception of being in bad health is significantly related to lower levels of well-being compared to people who say they are in good health. Therefore, while the regression analysis here can demonstrate that a relationship between two variables exists even after holding other variables in the model constant, these findings should not be taken to infer causality.

7.1.6 Multi-collinearity- dependence (or correlations) among the variables

If two or more independent variables in the regression model are highly correlated with each other, the reliability of the model as a whole is not reduced but the individual regression coefficients cannot be estimated precisely. This means that the analysis may not give valid results either about individual independent variables, or about which independent variables are redundant with respect to others. This problem becomes increasingly important as the size of correlations between the independent variables (i.e. multi-collinearity) increases.

7.2 Taking the design of the APS sample into account in the analysis

The primary sampling unit in the Annual Population Survey is the household. That is, individuals are grouped into households and the households become units in sample selection.

Regression analysis normally assumes that each observation is independent of all the other observations in the dataset. However, members of the same household are likely to be more similar to each other on some or all of the measures of personal well-being than they are to members of different households. If the analysis ignores this within-household correlation, then the standard errors of the coefficient estimates will be biased, which in turn will make significance tests invalid. Therefore, to correctly analyse the data and to make valid statistical inferences, the regressions are estimated in Stata with the specification of the survey design features. The survey weights were also used in the estimation of the model as these allow for more consistent estimation of the model coefficients and provide some protection against model misspecification.

7.3 Analysis of variance

The detailed methodology of the variance decomposition method with an application to area disparities can be found in the paper by Gibbons et al 2010.

7.4 Areas and area groupings used in the analysis

All the area related variables were constructed using 2001 Census geographies.

7.4.1 Administrative geographies

Administrative geography concerns itself with the hierarchy of areas relating to national and local government in the UK. The analysis considered two types of administrative geographies: Local Authority County and Local Authority District.

7.4.2 Geographies used in the construction of area groupings

Areas are classified together on the basis of similar environmental or socio-economic characteristics and are based mainly on Census 2001 output areas and super output areas.

Output areas (OAs) are created for Census data, specifically for the output of census estimates. 2001 Census OAs were built from clusters of adjacent unit postcodes but as they reflected the characteristics of the actual census data they could not be generated until after data processing. They were designed to have similar population sizes and be as socially homogenous as possible based on tenure of household and dwelling type (homogeneity was not used as a factor in Scotland).

In constructing the areas, urban/rural mixes were avoided where possible; OAs preferably consisted entirely of urban postcodes or entirely of rural postcodes. They had approximately regular shapes and tended to be constrained by obvious boundaries such as major roads. OAs were required to have a specified minimum size to ensure the confidentiality of data.

In England and Wales 2001 Census OAs were based on postcodes as at Census Day and fit within the boundaries of 2003 statistical wards and parishes. If a postcode straddled an electoral ward/division or parish boundary, it was split between two or more OAs.

The minimum OA size was 40 resident households and 100 resident people but the recommended size was rather larger at 125 households. These size thresholds meant that unusually small wards and parishes were incorporated into larger OAs.

In Scotland OAs were based on postcodes as at December 2000 and related to 2001 wards. However, the OAs did not necessarily fit inside ward boundaries where confidentiality issues made it more appropriate to straddle boundaries. The minimum OA size was 20 resident households and 50 resident people, but the target size was 50 households.

Super output areas were designed to improve the reporting of small area statistics and are built up from groups of output areas (OAs). Statistics for lower layer super output areas (LSOAs) and middle layer super output areas (MSOAs) were originally released in 2004 for England and Wales. Scotland also released statistics for data zones (equivalent to LSOAs) in 2004 and intermediate geographies (equivalent to MSOAs) in 2005.

LSOAs in England and Wales have a minimum population of 1,000 and 400 households and a maximum population of 3,000 and 1,200 households. MSOAs in England and Wales have a minimum population of 5,000 and 2,000 households and a maximum population of 15,000 and 6,000 households.

Data zones (DZs) and Intermediate zones (IZs) in Scotland in Scotland are smaller in population size than their LSOA and MSOA counterparts in England and Wales. DZs have a minimum population of 500 and IGs have a minimum population of 2,500.

Further information is available on ONS geographies.

7.4.3 Socio-economic and environmental characteristics of areas

Index of Multiple Deprivation

The Indices of Deprivation measure relative deprivation for small areas (LSOAs). Deprivation is a wider concept than poverty, and so the indices are constructed from a number of different types, or domains, of deprivation. These domains are combined into a single index, the Index of Multiple Deprivation (IMD), which ranks the areas in order of deprivation. A rank of 1 identifies the most deprived area.

The IMD deciles in this analysis are created by treating the most-deprived 10% of these areas as a single (non-adjacent) area, named Decile 1. The next most-deprived 10% are then grouped into a single area, named Decile 2, and so on.

Each of the four Nations of the UK produces its own Index. These Indices are not directly comparable because they use different domains and indicators, reflecting the priorities in the individual countries, and are published on different timescales covering different time periods.

However, the two domains – income and employment- are common to all four national IMDs and they contribute around half the weight of each IMD. They also use similar indicators such as simple percentage of individuals receiving one or more income or employment benefits. This implies they are relatively comparable across countries. For this analysis, the IMD deciles are created separately for each GB nation and then it is matched with the individual data. Although the values of the IMD deciles variable are not comparable across different nations, the variable can be considered a fairly good proxy for the relative deprivation of the area where an individual lives.

The deciles are based on the latest publications of the IMD Indices. These were: the English IMD Index - 2010, Scottish IMD Index - 2012 and the Welsh IMD Index - 2011.

Output Area Classifications

The 2001 Area Classification are used to group together geographic areas according to key characteristics common to the population in that grouping. These groupings are called clusters, and are derived using census data.

The largest cluster is the supergroup, of which there are seven. Each supergroup is further split into groups (21 in total) and further into subgroups (52 in total). This analysis is based on 52 subgroups.

Detailed information on these groups can be found in output area classifications.

For this analysis, output area classifications were matched with the individual data in the APS.

Rural - Urban Classification

The Rural/Urban Definition and LA Classification were developed following the 2001 Census and defines the rurality of very small 2001 Census based geographies.

Four settlement types are identified and assigned to either a 'sparse' or 'less sparse' regional setting to give eight classes of Output Areas.

Information on the rural-urban classification was available in the APS.

For further information please see the Rural-Urban Definition page.

Built-up areas classification

Built-up areas (BUAs) are a new geography, created as part of the 2011 Census outputs.

This data provides information on the villages, towns and cities where people live, and allows comparisons between people living in built-up areas and those living elsewhere. Census data for these areas (previously called urban areas) has been produced every 10 years since 1981.

For this analysis, output area classifications were matched with the individual data in the APS.

For further information please see ONS geography built-up areas.

Average household income in the MSOA areas

For this analysis, we classified MSOAs on the basis of the average household income in the area to explore how living in relatively rich or poor MSOAs was associated with personal well-being. Middle Layer Super Output Areas (MSOAs) are the smallest geographical level for which income information is available.

Small area (Middle Layer Super Output Area) income estimates covers MSOA areas in England and Wales. The estimates are calculated using a model based method to produce four estimates of average weekly income. The income refers to the income a household receives from wages and salaries, self-employment, benefits, pensions, plus any other sources of income. The figures have been produced using a modelling methodology that combines survey, census and administrative data.

Two sets of deciles of the average new weekly household income in the MSOAs were created; average net household income and equivalised net household income after housing costs to adjust for differences in household composition.

The deciles in this analysis are created by treating the areas with the lowest average household income as 10% and grouping these areas as a single (non-adjacent) area, named Decile 1. The next 10% are then grouped into a single area, named Decile 2, and so on.

For this analysis, the average household income deciles were created separately and then it was matched with the individual data in APS.

Further information on small area income estimates is available in small area model-based income estimates.

Green space

Deciles of the amount of green space per square metre in the output areas were created using the data from 'Land Use Statistics (Generalised Land Use Database)'.

The deciles in this analysis are created by treating the areas with the lowest amount of green space per square metre as 10% and grouping these areas as a single (non-adjacent) area, named Decile 1. The next 10% are then grouped into a single area, named Decile 2, and so on.

For this analysis, the green space deciles were created separately and then it was matched with the individual data in APS.

Population density

For this analysis 13 groups were created based on the 2011 population per square metre in the output areas. The groups were created separately and then it was matched with the individual data in APS.

Get all the tables for this publication in the data section of this publication .
Content from the Office for National Statistics.
© Crown Copyright applies unless otherwise stated.