This chapter describes the methodology used to create a poverty typology for Scotland. It covers the source of data, the construction of indicators to capture a range of poverty experiences, and the statistical technique used to identify households with similar combinations of poverty indicators - in other words, with similar experiences of poverty. In places the reader is referred to more detailed information in the appendices.
To be able to create poverty types, a dataset was required that collects information on different aspects of poverty from a large number of households in Scotland. Households in Scotland are included in many of the UK-wide large-scale surveys, such as Understanding Society, the Family Resources Survey and the General Lifestyle Survey. Although these surveys collect information on a range of topics, the number of households in Scotland that they interview is not large.
The Scottish Household Survey ( SHS), funded by the Scottish Government, was set up in 1998 to provide robust evidence on households in Scotland. The large-scale nature of the survey means that it allows disaggregation of information both geographically and in terms of population sub-groups (such as families with children or older people). Interviews are carried out in approximately 3,900 households each quarter. This report uses data from the latest two years of the survey - 2013 and 2014 - which when pooled contains information from approximately 28,000 households in Scotland. This dataset provides a big enough sample size to allow analysis of low-income households in three different life stages:
- Families with children: households with a child aged under 16 or aged 16-17 and in full-time education
- Working age without children: households with no children and all adults aged under pension age (men under 65 and women under 60)
- Older age: households with at least one adult aged over pension age (men 65 and over, and women 60 and over)
Identifying low income households
As in 'Poverty in Perspective', this research looks only at low-income households - that is, households below 70% of median equivalised net household income. The 70% median threshold was selected because it is used in the official child poverty measure that identifies households with combined low income and material deprivation ( DWP, 2015).
Defining income in the SHS
The term net annual household income refers to income from employment, benefits and other sources, after taxation and other deductions, that is brought into the household by all members of the household. Income as collected by the SHS is not the same as that used by other Government surveys that specialise in collecting income information - such as the Family Resources Survey ( FRS), used in the government's Households Below Average Income ( HBAI) series. Income data from the SHS should not, therefore, be compared with other sources without careful consideration of the methods used. There are several reasons why the SHS data on income may not be completely accurate:
- The SHS collects information from, or about, the Highest Income Householder only (and their spouse or partner in couple families).
- In other specialist income surveys, respondents are asked to refer to pay slips, benefit records or bank statements to support their information.
- It is likely that some people may not know the precise income of their partner, so may make a 'best guess'.
- Like with other 'sensitive' information, some respondents may understate their income because they do not want to reveal how much they really earn.
- As is often the case with income information, even in more specialist income surveys, there is a significant amount of missing data from households unwilling or unable to provide income information. Consequently values for some or all of the main components of income have been imputed by the data providers (Scottish Government, 2014).
An independent report that compared the SHS (and the Scottish House Condition Survey ( SHCS)) and FRS/ HBAI data concluded that '…considering the completely different and much less onerous method of collecting income data in the SHS/ SHCS the agreement in the income distributions with the FRS/ HBAI data is remarkably good' (Raab et al, 2004, p1).
Imputing income for 'other adults'
To further increase the quality of the income data, a procedure was undertaken to impute the income of 'other adults' in SHS from the average income of 'other adults' in the FRS/ HBAI data. This procedure mirrored elements of the approach used by the Scottish Government (2010) in a project using earlier years of the SHS. That report found that main activity status was the key predictor of the income of 'other adults'. Consequently the median income of 'other adults' in the FRS/ HBAI 2013 and 2014 surveys was imputed to 'other adults' in the SHS, according to main activity status.
Median income was calculated from the SHS sample. The percentage of households on low income (below 70% of the median income) within each life stage is presented in Table 1. As mentioned earlier, it is likely that the SHS is overestimating the percentage older age households on low income.
Table 1. Number of households by poverty status in each life stage
|Whether equivalised net household income below 70% median||Life stage||All households|
|Working age with children||Working age without children||Older age|
|70% median or above|
|Below 70% median|
Notes: Weighted column percentages
Notes: Unweighted counts
Base: Households in Scotland
Source: Scottish Household Survey, 2013 and 2014 pooled
Deriving poverty indicators
The original 'Poverty in Perspective' methodology selected a range of indicators that helped to capture the lived experience of poverty for households with a low income. How wide ranging those indicators should be was derived from discussions with key stakeholders (including policy makers, local service providers and academics) and by drawing on the Bristol Social Exclusion Matrix - which was designed to help researchers specify an approach for measuring multidimensional disadvantage (Room, 1995; Levitas, 2007).
The SHS covers many topics included in the three dimensions identified in the B-SEM (Resources, Participation and Quality of Life), such as housing, transport, health, employment, income, education and neighbourhood satisfaction. The poverty indicators used in the analysis are listed below.
- Household income decile
- Some or deep financial difficulties
- No savings
- Difficulties paying rent/mortgage in last year
- No car
- Cannot rely on friends/neighbours for help
- Provides regular unpaid care
- Has not done any free cultural activities in past year
- Has not done any paid cultural activities in past year
- No internet access at home
- Feels cannot influence local decisions
Quality of life
- Overcrowded accommodation
- Home never warm or accommodation has serious heating problem
- Ever been homeless
- High number of neighbourhood problems
- Feel unsafe walking alone in neighbourhood or alone at home
- Lives in one of the 20% most deprived locals areas
- Dissatisfied with local council
- Not visited countryside in last 12 months
- Experienced discrimination or harassment
There are also a range of other socio-demographic characteristics of households that help to explain the make-up of the poverty types:
- Household type
- Gender 
- Age 
- Ethnic group 
- Number of children
- Age of youngest child
- Main income source
- Household work status
- Adult and child health
- Highest qualification 
- Housing tenure
Data analysis: Latent Class Analysis
Latent Class Analysis ( LCA) was used to group households together into 'poverty types', based on similarities in their poverty indicator profiles, and to estimate the size of each these 'poverty types'. LCA is a member of the family of latent variable models (which includes factor analysis and structural equation modelling), and is a statistical technique used to find groups or subtypes of cases, 'latent classes', in multivariate categorical data.
In deciding how many latent classes are needed to best represent the variation in experiences of poverty, both model fit diagnostics and substantive interpretation of classes are considered. For the latter, the core output from the latent class models is used: estimated conditional item response probabilities - in other words, the probability of reporting a poverty indicator given membership of a particular poverty type. These are used to assess how commonly or uncommonly particular indicators of poverty are experienced across the poverty types, and to assess how useful the indicators are for distinguishing between types of poverty. Certain poverty indicators can be dropped from the models if they do not help to define distinct subtypes of poverty, which makes interpretation easier. Initial analysis did lead to the dropping of some indicators  , and not all indicators are used in each life stage  .
In arriving at a final set of models, how clearly the latent classes can be differentiated from each other is maximised - in other words, whether the 'ideal types' cover the majority of experiences of poverty, or whether there are many households whose experiences fall in between them. This can be done partly on the basis of the item response probabilities, but also partly from inspecting the posterior probabilities of class memberships: ideally each case to be assigned 'cleanly' (or with a high probability) to one and only one class.
Interpreting the analysis involves describing the sets of poverty indicators that each poverty type has a high probability of experiencing, and the indicators that help distinguish between the poverty types. Factors that can drive poverty - such as work status, education and health - and other socio-demographic characteristics - such as age, family composition and ethnic group - are used to help describe the sorts of households that experience different types of poverty. Graphical representations of the poverty types are used to succinctly describe and compare the typology.
Of course there are limitations with this method, as there is with all research methodology. It is important to emphasise that not all households within a poverty type have exactly the same poverty experience - but they will have similar experiences. In other words, although a poverty type contains households with a similar set of poverty indicators, not all with have exactly the same combination of poverty indicators. When the poverty types are described below, therefore, we talk about the likelihood of an individual household experiencing a certain poverty indicator. For example, one household may be unable to save, and another may have savings, but they fall within the same poverty type because they are similar on enough other indicators to group them together  .
Presenting the poverty types
The poverty types are presented and discussed in the next three chapters - each focusing on a different life stage: families with children, working age households without children, and older age.
Two graphics are used to illustrate each poverty type:
1/ A 'poverty wheel', showing the lived experience of poverty for households in that poverty type. Each segment of the poverty wheel represents a different poverty indicator. The proportion of the segment that is shaded reflects the probability that households in that poverty type have that poverty indicator. The segments are coloured to signal the domain of the B-SEM to which they belong: yellow = resources, blue = participation, green = quality of life. Not all 20 of the poverty indicators used in the analysis are represented in the poverty wheel, just the 13 that best differentiate between the poverty types. 
2/ A chart showing the socio-demographic characteristics of households in that poverty type (including factors that could lead to a household having low income). This chart is coloured red.
As well as presenting the main features of each poverty type, we explore how each group might be identified, how the causes and symptoms of their experience of poverty might be alleviated, and the implications of these findings for local and national poverty strategies.