Welsh Index of Multiple Deprivation (WIMD): identifying groups of small areas based on deprivation indicators
Detecting Deprivation Patterns in Wales. Insights from WIMD 2019 clustering analysis.
This file may not be fully accessible.
In this page
Introduction
The Welsh Index of Multiple Deprivation (WIMD) is the Welsh Government’s official measure of relative deprivation in Wales. WIMD ranks all small areas (known as Lower Super Output Areas, or LSOAs) in Wales from 1 (most deprived) to 1,909 (least deprived) across eight domains (for example: income, education, health) using a weighted sum of 47 different underlying indicators; these are measurable quantities which capture the concept of deprivation for each domain. A lookup table for each indicator description and corresponding abbreviated code can be found at the end of the report. These domains are also used to calculate an overall deprivation rank. For more information on how WIMD was constructed, please see the WIMD section on the Welsh Government website.
In this project we applied clustering techniques to the underlying WIMD indicator data to segment groups of small areas in Wales that share common deprivation characteristics. This approach allows us to reveal and explore patterns in the WIMD indicator data that reflect the different combinations of deprivation across Wales.
Report outline
In this project we clustered the 1,909 areas first into two main clusters, before further splitting those clusters into 3 sub-clusters each. This resulted in 6 sub-clusters representing different deprivation profiles across Wales.
This report will briefly describe the two main clusters before providing more in-depth descriptions and analysis of the six sub-clusters. For each sub-cluster, we will describe:
- general patterns of deprivation ranking
- performance on key WIMD indicators that distinguish it from other sub-clusters
- the rural-urban split
In addition, each sub-cluster is supplemented by three case studies, small areas that are representative of their sub-cluster. By exploring these case studies this report will attempt to draw out the different deprivation profiles of each subcluster and how they vary. We have provided tables showing the relative difference across key indicators that distinguish the sub-cluster from other sub-clusters, including the difference from the national median and which decile the small area falls under for that indicator.
Data quality
Please note that all figures, tables, and appendices are sourced from Welsh Government unless otherwise indicated.
WIMD indicators are produced using a wide range of datasets that were collected in different years (see the WIMD technical report for more detailed information). Datasets underlying WIMD indicators may have been updated with more recent data since the publication of WIMD 2019. Additionally, the LSOAs referred to in this analysis are those that were put in place in 2011. This is because these were the LSOAs used in the WIMD 2019 analysis. In 2021 new LSOAs were defined by the Office for National Statistics (ONS) due to population and household shifts. This resulted in the merging/splitting of some 2011 LSOAs to ensure they met population and household thresholds.
Indicator limitations
The 47 underlying indicators that make up the eight domains of WIMD are referred to through-out the report. In figure titles and when referred to in the text we use the indicator descriptions, however in some figures we have used the abbreviated indicator codes in the interest of readability. A full indicator description to indicator code lookup can be found in the appendix at the end of the report.
The indicator that is most likely to have changed since data was compiled for the 2019 index is the % unavailability of broadband at 30mb/s. In the last few years, the roll-out of faster broadband across Wales has been facilitated by Welsh Government-backed schemes, such as the Superfast Cymru successor project and the Local Broadband Fund. The latest Ofcom Connected Nations report shows that superfast broadband coverage for residences in Wales has increased to 95%, a 2% increase from 2019. However, when split by rural-urban classification, this figure is 98% for urban areas and 85% for rural areas.
The updated broadband information could notably impact our analysis, particularly for Main Cluster 2 where the “% Unavailability of broadband at 30Mb/s” indicator ranks as the 9th most important feature out of 25 indicators used. If the changes in broadband availability are significant then the identified patterns of deprivation could be affected.
Rurality
The classification of urban and rural areas is based on the 2011 ONS Urban Rural Classification dataset that uses the 2011 based output areas. In this dataset, areas with population >10,000 designated as ‘urban’. All other areas are categorised as rural. The urban and rural domains are then subdivided into six settlement types which are explored for each sub-cluster, as well as the distribution of Built-Up-Areas (BUAs) (ONS), which can give a better idea of the average population density within each of the sub-clusters. While it’s important to note that the data about the number of areas in different urban-rural classifications and sizes of BUAs is from 2011 and may not reflect the most recent changes, it remains the latest available information we have.
The two primary clusters
In the initial phase of data analysis, the application of the clustering algorithm tests to our dataset revealed the presence of two distinct groups within the data. These initial groups were identified based on their intrinsic properties which set them apart from each other. The identification of these two primary clusters was a crucial step as it laid the foundation for a more granular exploration of the data.
The unique attributes of the two primary clusters were explored to better understand their defining characteristics and to establish a baseline for the more detailed segmentation that follows later in the report.
Primary Cluster 1
Number of LSOAs: 1017
Population: 1,690,000
Area (km2): 18,840
Primary Cluster 2
Number of LSOAs: 892
Population: 1,463,000
Area (km2): 1,909
Primary Cluster 1 is, on average, less deprived than Primary Cluster 2 across all WIMD domains. The median rank for overall deprivation is 1,400 for Primary Cluster 1 and 447 for Primary Cluster 2 (note: rank 1 = most deprived, rank 1,909 = least deprived).
Figure 1: Distribution of Primary Cluster 1 and Primary Cluster 2 across Wales at LSOA level
Description of Figure 1: A map depicting a visual representation of the distribution of Primary Clusters 1 and 2 across Wales at the LSOA level. The map features two distinct colours to represent both primary clusters. Primary Cluster 2, represented as orange geographies, is mostly found in pockets in the South Wales cities and valleys, and in some North Wales coastal and border towns. Primary Cluster 1, shown in dark blue geographies, represents the rest of Wales and is the dominant cluster throughout central, northern and western-Wales.
Feature importance
Feature importance is a technique used in machine learning to identify features that have the most significant influence on the predictions of a model. It helps us to understand the underlying structure of the data and the model’s behaviour by highlighting which features contribute the most to the model’s decision-making process.
The features that are most important to determining which small areas go into Primary Cluster 1 are: People in income deprivation (INC) and Adults aged 25 to 64 with no qualifications (NOQU).
Figure 2: Importance Scores of Deprivation Variables for Primary Cluster 1
Description of Figure 2: A bar chart showing the 10 most important deprivation variables used to sort data into Primary Cluster 1 or Primary Cluster 2. The chart shows that a few variables have significantly higher importance scores, with the longest bars representing the most influential variables. This means that these variables play a bigger role in defining the characteristics of the primary clusters. The variable INC has the highest importance score, followed by NOQU, KS4, and CHRON. Other variables have shorter bars, representing lesser importance in the clustering process, with KS2 being the lowest importance score for this Primary Cluster.
Rurality distribution
Primary Cluster 1 is proportionally more rural than Primary Cluster 2. 44% of small areas in Primary Cluster 1 are classed as rural, compared to 17% in Primary Cluster 2. Primary Cluster 1 covers approximately 18,833km2 of land area, substantially larger than Primary Cluster 2 which covers 1,902km2. The population of Primary Cluster 1 is approximately 1.7 million and the population of Primary Cluster 2 is approximately 1.5 million.
The table below shows the disaggregated six-fold rurality classification (The National Archives) breakdown aggregated at the primary cluster level – Primary Cluster 1 is comprised of sub-clusters 1-1, 1-2, and 1-3. Primary Cluster 2 is comprised of sub-clusters 2-1, 2-2, and 2-3.
| Classification | Primary Cluster 1 | Primary Cluster 2 |
|---|---|---|
| Urban city and town | 54.1% | 80.5% |
| Urban city and town in a sparse setting | 1.6% | 2.1% |
| Rural town and fringe | 13.1% | 13.3% |
| Rural town and fringe in a sparse setting | 5.8% | 2.1% |
| Rural village and dispersed | 11.1% | 1.8% |
| Rural village and dispersed in a sparse setting | 14.4% | 0.1% |
Description of Table 1: This table presents the percentage distribution of the 6-fold rurality classification (source: ONS) across Primary Cluster 1 and Primary Cluster 2, highlighting significant differences in the urban and rural compositions of the two clusters.
The "Urban city and town" category is the largest in both clusters, with a relatively higher proportion in Primary Cluster 2 (80.5%) compared to Primary Cluster 1 (54.1%). The "Urban city and town in a sparse setting" category shows a minimal presence in both clusters, at 2.1% for Cluster 2 and 1.6% for Cluster 1, reflecting limited representation of sparse urban areas.
Rural classifications are relatively more prevalent in Primary Cluster 1. The "Rural town and fringe" category is similar in both clusters, with 13.1% in Cluster 1 and 13.3% in Cluster 2, suggesting comparable proportions of small rural towns. However, "Rural town and fringe in a sparse setting" is more common in Cluster 1 (5.8%) than in Cluster 2 (2.1%), highlighting a greater presence of sparsely populated rural towns in Cluster 1.
The "Rural village and dispersed" category underscores the relatively rural character of Cluster 1, where it accounts for 11.1% of areas, compared to just 1.8% in Cluster 2. Sparse rural villages and dispersed settings are particularly distinctive in Cluster 1, with 14.4% of areas falling into this category, while they are virtually absent in Cluster 2 (0.1%).
The table below shows the BUA breakdown aggregated at the main cluster level – they include areas of built-up land with a minimum of 20 hectares (200,000m2).
| BUA Category | Primary Cluster 1 | Primary Cluster 2 |
|---|---|---|
| Largest | 39.6% | 45.2% |
| Large | 18.7% | 26.5% |
| Medium | 14.7% | 23.9% |
| Small | 31.1% | 4.5% |
Description of Table 2: This table presents the percentage distribution of BUA categories across Primary Cluster 1 and Primary Cluster 2, highlighting notable differences in the built-up characteristics of the two clusters. The "Largest" category, representing the most built-up areas, is the dominant category in both clusters. Primary Cluster 2 has a higher proportion in this category (45.2%) compared to Primary Cluster 1 (39.6%), reflecting the relative prevalence of urbanised areas in Cluster 2. The "Large" category shows a more pronounced difference, with Primary Cluster 2 having a higher proportion (26.5%) compared to 18.7% in Primary Cluster 1. This indicates a greater presence of moderately built-up areas in Cluster 2. The "Medium" category follows a similar trend, with Primary Cluster 2 (23.9%) again outpacing Primary Cluster 1 (14.7%). This consistent pattern across the "Large" and "Medium" categories highlights the greater diversity and concentration of urban areas in Cluster 2. The "Small" category shows the most contrast, being more prevalent in Primary Cluster 1 (31.1%) than in Primary Cluster 2 (4.5%).
Primary Cluster 1 shows a wide distribution of BUAs. ‘Largest’ (39.6%) is the predominant category, however, the notable portion of the ‘Small’ (31.1%) category suggests a substantial rural component to this primary cluster. ‘Large’ (18.7%) and ‘Medium’ (14.7%) categories are less prominent but help to underscore the diversity of Primary Cluster 1 in terms of settlement sizes and densities.
Primary Cluster 2 shows a stronger tendency towards larger built-up areas. The dominance of the ‘Largest’ category (45.2%) in this primary cluster is more pronounced than in Primary Cluster 1. The ‘Large’ (26.5%) and ‘Medium’ (23.9%) categories also have considerable share and reinforce the urban component of Primary Cluster 2. The relatively small percentage of the ‘Small’ category points to a reduced presence of rural or sparsely populated areas.
Sub-clusters
The division of primary clusters 1 and 2 into six sub-clusters was informed by analysis of the data, we sought to capture the nuances within each primary cluster at a deeper level. By applying another layer of clustering to the narrower subsets of data, we were able to identify distinct patterns and relationships that were not evident at the higher level of clustering. Each of the six sub-clusters represents a more homogenous grouping of data points, characterised by specific attributes, which we can explore further.
In the subsequent sections of this report, we will explore the properties and characteristics of these six sub-clusters in detail. The analysis will focus on the unique aspects of each sub-cluster. Characteristics of these six sub-clusters are summarised in Table 3 below.
| Sub-cluster | Sub -cluster description | Number of small areas | Estimated population | Land area (km-squared) | % urban areas |
|---|---|---|---|---|---|
| 1-1 | Areas with moderate deprivation, ranking towards the middle across all domains | 453 | 766,000 | 2,160 | 66.7 |
| 1-2 | Rural and remote areas with low deprivation, except for relatively poor access to services and housing | 218 | 360,000 | 15,260 | 1.4 |
| 1-3 | Sub-urban and rural areas with low deprivation | 346 | 564,000 | 1,420 | 75.4 |
| 2-1 | Areas with fairly high levels of deprivation, but moderately good access to services | 357 | 580,000 | 810 | 78.2 |
| 2-2 | Urban areas with moderate deprivation | 337 | 565,000 | 930 | 83.7 |
| 2-3 | Urban areas with high deprivation, except for relatively low housing deprivation | 198 | 318,000 | 160 | 88.9 |
Description of Table 3: This table summarises six sub-clusters, detailing their characteristics, number of small areas, estimated population, land area, and proportion of urban areas. It highlights the diversity in geographic, demographic, and deprivation profiles across the sub-clusters.
Sub-cluster 1-1 represents areas with moderate deprivation and a balanced ranking across all domains. It has the highest number of small areas (453) and a population of 766,000, with 66.7% urban areas. Its land area of 2,160 km² indicates a moderate level of geographic spread compared to the other sub-clusters.
Sub-cluster 1-2 is distinctly rural and remote, with low deprivation except for poor access to services and housing. Covering the largest land area (15,260 km²), it includes only 218 small areas, highlighting its sparse population density. The estimated population is 360,000, and urban areas make up just 1.4%, reflecting its rural nature.
Sub-cluster 1-3 features suburban and rural areas with low deprivation. It covers 346 small areas and has an estimated population of 564,000. With a land area of 1,420 km², it is more compact than Sub-cluster 1-2 but has a higher proportion of urban areas (75.4%).
Sub-cluster 2-1, characterised by relatively high deprivation but good access to services, includes 357 small areas and a population of 580,000. It spans 810 km², making it relatively compact, with 78.2% urban areas.
Sub-cluster 2-2 represents urban areas with relatively moderate deprivation, encompassing 337 small areas and a population of 565,000. With a land area of 930 km², it is geographically similar to Sub-cluster 2-1 but has a slightly higher proportion of urban areas (83.7%).
Sub-cluster 2-3 is the most urbanised, with 88.9% of its 198 small areas classified as urban. Despite having the smallest land area (160 km²), it accommodates a population of 318,000, showing high population density. This sub-cluster also exhibits high levels of deprivation, with relatively low housing deprivation as an exception.
| BUA Category | 1-1 | 1-2 | 1-3 | 2-1 | 2-2 | 2-3 |
|---|---|---|---|---|---|---|
| Largest | 16.8% | 4.1% | 52.0% | 41.5% | 44.2% | 53.5% |
| Large | 47.2% | 0.0% | 21.1% | 25.8% | 26.1% | 28.3% |
| Medium | 19.7% | 0.5% | 17.1% | 28.6% | 23.2% | 16.7% |
| Small | 16.3% | 95.4% | 9.8% | 4.2% | 6.5% | 1.5% |
Description of Table 4: This table shows the percentage distribution of BUA categories across six sub-clusters, highlighting variability in area sizes.
In the above table the "Largest" category dominates Sub-clusters 1-3 (52%) and 2-3 (53.5%), with Sub-clusters 2-2 (44.2%) and 2-1 (41.5%) also showing strong urban characteristics. In contrast, Sub-cluster 1-2 has the smallest representation in this category (4.1%), reflecting its rural nature. The "Large" category is most prominent in Sub-cluster 1-1 (47.2%), while other sub-clusters, excluding 1-2 (0%), show more moderate proportions ranging from 21.1% to 28.3%.
The "Medium" category is highest in Sub-cluster 2-1 (28.6%) and 2-2 (23.2%), with smaller shares in Sub-clusters 1-1 (19.7%) and 1-3 (17.1%). Sub-cluster 1-2 again has minimal representation (0.5%). The "Small" category is dominant in Sub-cluster 1-2 (95.4%), followed by Sub-cluster 1-1 (16.3%). Other sub-clusters have minimal representation in this category, ranging from 9.8% to just 1.5%.
Sub-clusters 1-2 and 1-1 display more rural characteristics with higher proportions of smaller areas, while Sub-clusters 1-3, 2-1, 2-2, and 2-3 are characterized by larger and more built-up areas, particularly in the "Largest" and "Large" categories. This distribution underscores the diversity in built-up area types across the sub-clusters.
| Classification | 1-1 | 1-2 | 1-3 | 2-1 | 2-2 | 2-3 |
|---|---|---|---|---|---|---|
| Urban city and town | 64.9% | 1.4% | 73.1% | 75.6% | 81.0% | 88.4% |
| Urban city and town in a sparse setting | 1.8% | 0.0% | 2.3% | 2.5% | 2.7% | 0.5% |
| Rural town and fringe | 15.7% | 3.2% | 15.9% | 16.8% | 11.3% | 10.6% |
| Rural town and fringe in a sparse setting | 10.4% | 3.7% | 1.2% | 3.1% | 2.4% | 0.0% |
| Rural village and dispersed | 4.6% | 31.7% | 6.7% | 1.7% | 2.7% | 0.5% |
| Rural village and dispersed in a sparse setting | 2.7% | 60.1% | 0.9% | 0.3% | 0.0% | 0.0% |
Description of Table 5: table displays the percentage distribution of the 6-fold rurality classification across the six sub-clusters. Each row represents a specific rurality classification, and the columns correspond to the sub-clusters. Urban classifications dominate most sub-clusters, with Sub-cluster 2-3 having the highest percentage of "Urban city and town" areas at 88.4%, followed by Sub-cluster 2-2 at 81%. In contrast, Sub-cluster 1-2 is the most rural, with 60.1% of its areas classified as "Rural village and dispersed in a sparse setting" and an additional 31.7% as "Rural village and dispersed."
Sub-cluster 1-1 shows a more balanced distribution, with 64.9% of its areas in "Urban city and town," 15.7% in "Rural town and fringe," and 10.4% in "Rural town and fringe in a sparse setting." Sub-cluster 1-3 also leans urban, with 73.1% of its areas in "Urban city and town," but still maintains a rural presence, with 6.7% in "Rural village and dispersed." Sub-cluster 2-1 and Sub-cluster 2-2 have similar patterns, with high urban proportions (75.6% and 81%, respectively) but lower rural classifications compared to Sub-cluster 1-1 and 1-3.
Sparse settings are most prominent in Sub-cluster 1-2 and Sub-cluster 1-1, with notable proportions in "Rural village and dispersed in a sparse setting" (60.1% and 2.7%, respectively) and "Urban city and town in a sparse setting" (1.8% in 1-1). Sub-cluster 2-3 has negligible rural representation, highlighting its predominantly urban nature.
Sub-cluster 1-1: Areas with moderate deprivation, ranking towards the middle across all domains
Number of LSOAs: 453
Population: 766,000
Area (km2): 2160
Figure 3: Proportion of LSOAs in Sub-cluster 1-1 at Local Authority Level
Description of Figure 3: A map showing the proportion of LSOAs in sub-cluster 1-1 across different Local Authorities. Lighter areas indicate fewer LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. Two Local Authorities are significantly darker than the others, indicating a higher concentration of LSOAs from sub-cluster 1-1. These darker Local Authorities are Carmarthenshire and Conwy. The lightest Local Authorities on the map are Denbighshire and Vale of Glamorgan.
Description
Sub-cluster 1-1 is the largest sub-cluster in terms of number of small areas comprising the sub-cluster as well as the total sub-cluster population. It is a densely populated, moderately deprived cluster hosting many small areas near rural towns. These areas tend to have an average ranking in the ‘access to services” domain when compared to other sub-clusters, as well as higher employment deprivation relative to other sub-clusters. It faces challenges in income and education, having the highest income deprivation relative to other Primary Cluster 1 sub-clusters, as well as a higher than average proportion of adults without qualifications.
The most prominent six-fold rurality classification in sub-cluster 1-1 is ‘Urban city and town’ (64.9%) indicating a strong urban character. This is further reinforced in the BUA distribution for this sub-cluster, which shows that the ‘Large’ BUA category is the most prevalent (47.2%). Despite the urban dominance, there are still significant rural components to sub-cluster 1-1, with ‘Rural town and fringe’ (15.7%) and ‘Rural town and fringe in a sparse setting’ (10.4%) comprising about 26% of LSOAs in this sub-cluster. This indicates a notable presence of rural and semi-rural areas within the sub-cluster.
Evidence
Small areas in this cluster tend to land nearer the middle of the deprivation rankings in the income (median = 1160), employment (median = 1156), education (median = 1149), and health (median = 1126) domains, whereas the other sub-clusters in Cluster 1 tend to rank as less deprived on average. However, small areas in this sub-cluster tend to rank as much less deprived in the Access to Services domain (median = 1273) compared to the more rural sub-cluster 1-2.
Sub-cluster 1-1 has the highest median percentage of people in income deprivation (11%, see Figure 6) and percentage of adults aged 25 to 64 with no qualifications (15.4%) in Primary Cluster 1.
Small areas in this sub-cluster can be found in all Local Authorities across Wales. Conwy and Carmarthenshire have proportionally more small areas in sub-cluster 1-1 (each with approximately one third of their small areas in this sub-cluster) than other Local Authorities.
Figure 4: Distribution of Sub-cluster 1-1 within Conwy at LSOA Level
Sub-cluster 1-2: Rural and remote areas with low deprivation, except for relatively poor access to services and housing
Number of LSOAs: 218
Population: 360,000
Area (km2) : 15,260
Figure 5: Proportion of LSOAs in Sub-cluster 1-2 at local authority Level
Description of Figure 5: A map showing the proportion of LSOAs in sub-cluster 1-2 across different Local Authorities. Lighter areas indicate fewer LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. Three Local Authorities are significantly darker than the others, indicating a higher concentration of LSOAs from sub-cluster 1-2. These darker Local Authorities are Powys, Ceredigion and Gwynedd. The lightest Local Authorities on the map, indicating a lower concentration of LSOAs from sub-cluster 1-2, are Cardiff, Bridgend, and Caerphilly.
Description
Sub-cluster 1-2 covers the largest land area of the sub-clusters and is predominantly rural, comprising of around 99% rural and remote areas. This sub-cluster includes numerous rural villages and dispersed settlements in Wales, but has difficulties in access to services, housing (housing hazards are more common here), and longer travel times to amenities.
This sub-cluster has a high percentage of LSOAs in the ‘Rural village and dispersed in a sparse setting’ category (60.1%). The ‘Small’ Built-up Area category is overwhelmingly dominant in the BUA category distribution (95.4%). This data points to sub-cluster 1-2 having a strong focus on sparsely populated and dispersed rural communities, contrasting with the other more urban-centric rurality profiles of other sub-clusters.
Evidence
LSOAs in this sub-cluster tend to rank as less deprived than average across most domains, with the exceptions being Access to Services and Housing. Median ranks for this sub-cluster were 114 for Access to Services and 473 for Housing.
Small areas in this sub-cluster performed well on air-quality indicators, for example, the median Population Weighted Average Concentration Value for Nitrogen Dioxide was 3.5, far below the national median of 8.3.
The higher levels of deprivation in the Access to Services domain for sub-cluster 1-2 reflect higher unavailability of broadband at 30mb/s (median = 28%) and longer travel times to essential services and amenities (e.g. median average public return travel time to a sports facility = 139 minutes). LSOAs in this sub-cluster also tend to have a higher likelihood of housing containing serious hazards (median = 31.7%, see Figure 12) compared to all other sub-clusters.
Many of the small areas in Powys (54%), Ceredigion (50%), and Gwynedd (48%) are in sub-cluster 1-2. Conversely, there are no small areas from sub-cluster 1-2 in the Blaenau Gwent, Bridgend, Caerphilly, Cardiff, Merthyr Tydfil, Neath Port-Talbot, Newport, Rhondda Cynon Taf, or Torfaen Local Authorities.
Figure 6: Distribution of Sub-cluster 1-2 within Gwynedd at LSOA Level
Description of Figure 6: A map illustrating the distribution of sub-cluster 1-2 within Gwynedd at the LSOA level. LSOAs that belong to sub-cluster 1-2 are highlighted in blue, while other sub-clusters are shaded in grey. Approximately one half of the LSOAs on the map are blue, indicating that they are part of sub-cluster 1-2.
Sub-cluster 1-3: Sub-urban and rural areas with low deprivation
Number of LSOAs: 346
Population: 564,000
Area (km2) : 1,420
Figure 7: Proportion of LSOAs in Sub-cluster 1-3 at Local Authority Level
Description of Figure 7: This figure is a map showing the proportion of LSOAs in sub-cluster 1-3 across different Local Authorities. Lighter areas indicate fewer LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. Two Local Authorities are significantly darker than the others, indicating a higher concentration of LSOAs from sub-cluster 1-3. These darker Local Authorities are Vale of Glamorgan and Monmouthshire. The lightest Local Authorities on the map, indicating a lower concentration of LSOAs from sub-cluster 1-2, are Blaenau Gwent and Powys.
Description
Sub-cluster 1-3 represents many of the least deprived LSOAs in Wales. These areas are mostly suburban or rural areas close to urban centres with relatively low income and employment deprivation. Although the LSOAs in this sub-cluster are likely to be commuting hubs with good transport links to areas of employment and easy access to services, the proximity to urban centres also leads to lower air quality.
In sub-cluster 1-3 the ‘Urban city and town’ classification is the most prevalent (73%), suggesting a relatively strong urban component. The ‘Largest’ Built-up Area category is the most significant, accounting for more than half of the areas (52%). The ‘Large’ (21.1%) and ‘Medium’ (17%) BUA categories indicate some diversity in urban scale. The rural aspect of the sub-cluster is less dominant but still represented, particularly within the ‘Rural town and fringe’ rural classification (15.9%). Sub-cluster 1-3 is the most urban-centric out of the three Primary Cluster 1 sub-clusters.
Evidence
On average, this sub-cluster ranks as the least deprived sub-cluster across many of the WIMD domains - except community safety, where it is marginally more deprived than sub-cluster 1-2, and physical environment, where its median rank is 933 – small areas in this sub-cluster tend to perform well across many deprivation indicators, for example, people in income deprivation (median = 6%), adults aged 25 to 64 without qualifications (8.7%), and unavailability of broadband at 30mb/s (1.3%). However, these small areas tend to have higher levels of air pollution, for example, the median Population Weighted Average Concentration Value for Nitrogen Dioxide was 9.8, which is on par with sub-clusters that are relatively more deprived.
Examples of sub-cluster 1-3 LSOAs can be found in all local authorities. Proportionally more LSOAs in the Vale of Glamorgan (48%), Monmouthshire (34%), and Cardiff (32%) are in this sub-cluster.
Figure 8: Distribution of Sub-cluster 1-3 within Vale of Glamorgan at LSOA Level
Description of Figure 8: A map illustrating the distribution of sub-cluster 1-3 within Vale of Glamorgan at the LSOA level. LSOAs that belong to sub-cluster 1-3 are highlighted in blue, while other sub-clusters are shaded in grey. Approximately one half of the LSOAs on the map are part of sub-cluster 1-3.
Sub-cluster 2-1: Areas with fairly high levels of deprivation, but moderately good access to services
Number of LSOAs: 357
Population: 580,000
Area (km2) : 810
Figure 9: Proportion of LSOAs in Sub-cluster 2-1 at local authority Level
Description of Figure 9: A map showing the proportion of LSOAs in sub-cluster 2-1 across different Local Authorities. Lighter areas indicate fewer LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. Three Local Authorities are significantly darker than the others, indicating a higher concentration of LSOAs from sub-cluster 2-1. These darker Local Authorities are Rhondda Cynon Taf, Merthyr Tydfil, and Blaenau Gwent. The lightest Local Authorities on the map, indicating a lower concentration of LSOAs from sub-cluster 2-1, are Monmouthshire and Flintshire.
Description
The small areas in sub-cluster 2-1 have significant levels of deprivation, though they have good access to services due to proximity to economic hubs. They face economic deprivation, with housing often in disrepair and potential hazards in the home above the national average, as well as higher than average income deprivation compared to other sub-clusters.
Sub-cluster 2-1 has a pronounced urban inclination, underscored by its high number of small areas in the ‘Largest’ BUA classification (41.5%), followed closely by ‘Large’ and ‘Medium’ categories which account for 25.8% and 28.6% respectively. This distribution highlights the diversity of the urban framework in this sub-cluster, with a clear leaning towards larger and more developed urban settlements. This urban inclination is further evidenced in the six-fold rural classification distribution in this sub-cluster, where the ‘Urban city and town’ class represents 75.6% of this sub-cluster. The ‘rural town and fringe’ rural classification, albeit smaller compared to the urban categories, comprises 16.8% of this sub-cluster, suggesting the inclusion of areas that are on the cusp of urban and rural settings – more remote areas that could possibly serve as transitional zones between highly urban and rural landscapes.
Evidence
On average, small areas in sub-cluster 2-1 rank as relatively more deprived across several of the WIMD domains, such as income (median = 476), housing (median = 447), and employment (median = 444). However, they tend to rank as less deprived than average on access to services (median = 1020). The most notable difference between this sub-cluster and other sub-clusters within Primary Cluster 2 is its increased likelihood of housing being in disrepair (median = 4.7%) and housing having serious hazards (22.7%) which are both above the national average. The median rate of people in income deprivation in this sub-cluster is 22%, higher than the national median of 14%, however this figure is smaller compared to sub-cluster 2-3.
The Local Authorities with proportionally more of their LSOAs in this sub-cluster include Rhondda Cynon Taf (48%), Blaenau Gwent (45%), and Merthyr Tydfil (44%).
Figure 10: Distribution of Sub-cluster 2-1 within Rhondda Cynon Taf at LSOA Level
Description of Figure 10: This figure is a map illustrating the distribution of sub-cluster 2-1 within Rhondda Cynon Taf at the LSOA level. LSOAs that belong to sub-cluster 2-1 are highlighted in blue, while other sub-clusters are shaded in grey. Approximately one half of the LSOAs on the map are part of sub-cluster 2-1.
Sub-cluster 2-2: Urban areas with moderate deprivation
Number of LSOAs: 337
Population: 565,000
Area (km2) : 930
Figure 11: Proportion of LSOAs in Sub-cluster 2-2 at local authority Level
Description of Figure 11: A map showing the proportion of LSOAs in sub-cluster 2-2 across different Local Authorities. Lighter areas indicate a lower proportion of LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. One Local Authority is significantly darker than the others, indicating a higher concentration of LSOAs from sub-cluster 2-2. This darker Local Authority is Torfaen. The lightest Local Authority on the map, indicating a lower concentration of LSOAs from sub-cluster 2-2, is Gwynedd.
Description
The small areas in sub-cluster 2-2 are typically located close to economic hubs with good access to services. Compared to sub-cluster 2-1, these areas tend to experience less economic deprivation and better-quality housing. Many of these small areas can be found in the southern valleys, in inner city areas, and on the northern coastline.
In sub-cluster 2-2 the data again reveals a strongly urbanised profile, with a prominent ‘Urban city and town’ classification which constitutes 81% of the small areas in this sub-cluster. Whilst the ‘Largest’ BUA category represents 44% of the small areas in this sub-cluster, the ‘Large’ and ‘Medium’ BUA categories represent 26.1% and 23.2%, respectively. Whilst both this sub-cluster and sub-cluster 2-1 are predominantly urban, sub-cluster 2-2 shows less of a rural presence indicating less diversity in its rural-urban mix than sub-cluster 2-1.
Evidence
Sub-cluster 2-2 represents some of the less deprived small areas within Primary Cluster 2. On average, LSOAs in this sub-cluster fall around the middle of the rankings in the housing domain (median = 1025) and access to services (median = 926). The median proportion of people in income deprivation in this sub-cluster is 19%, which is 5% above the national average. This sub-cluster is more income deprived than average when compared to sub-clusters 1-1, 1-2, and 1-3, however it is less income deprived than average than sub-clusters 2-1 and 2-3. Compared to sub-cluster 2-1, this sub-cluster has a lower median likelihood of housing being in disrepair (2.3%) and having serious hazards (12.4%).
Sub-cluster 2-2 can be found in all local authorities across Wales. As a proportion of its small areas, Torfaen has the most sub-cluster 2-2 LSOAs (43%). Meanwhile, Gwynedd has proportionally the lowest number of sub-cluster 2-2 LSOAs (1%).
Figure 12: Distribution of Sub-cluster 2-2 within Torfaen at LSOA Level
Description of Figure 12: A map illustrating the distribution of sub-cluster 2-2 within Torfaen at the LSOA level. LSOAs that belong to sub-cluster 2-2 are highlighted in blue, while other sub-clusters are shaded in grey. Approximately one half of the LSOAs on the map are part of sub-cluster 2-2.
Sub-cluster 2-3: Urban areas with high deprivation, except for relatively low housing deprivation
Number of LSOAs: 198
Population: 318,000
Area (km2): 160
Figure 13: Proportion of LSOAs in Sub-cluster 1-1 at local authority Level
Description of Figure 13: A map showing the proportion of LSOAs in sub-cluster 1-1 across different Local Authorities. Lighter areas indicate fewer LSOAs in that sub-cluster within the Local Authority, while darker shades represent a higher proportion of LSOAs comprising the Local Authority. Two Local Authorities are darker than the others, indicating a higher concentration of LSOAs from sub-cluster 1-1. These darker Local Authorities are Blaenau Gwent and Merthyr Tydfil. The lightest Local Authority on the map, indicating a lower concentration of LSOAs from sub-cluster 1-1, is Ceredigion.
Description
Sub-cluster 2-3 is the smallest sub-cluster in terms of land area, population, and number of LSOAs. This small, highly deprived urban sub-cluster generally presents good housing conditions but high economic deprivation. Though it has the highest proportion of income deprivation and adults without qualifications, it has the lowest chance of containing housing with hazards among all six sub-clusters.
Sub-cluster 2-3 strongly favours the ‘Urban city and town’ rural classification (88.4%), and the ‘Largest’ BUA category (53.4%). The minimal presence of small rural areas shows a sub-cluster landscape dominated by larger, more developed urban settings.
Evidence
On average, small areas in this sub-cluster experience the most deprivation across domains like education (median = 119), income (median = 108), and health (median = 103). However, average ranking in the housing domain is less deprived than some other sub-clusters (median = 810). The median proportion of people in income deprivation is 34% and the median proportion of adults aged 25 to 64 with no qualifications is 37%, which are the highest across all 6 sub-clusters. Meanwhile, the median likelihood of housing having serious hazards is 10.1%, the lowest of all 6 sub-clusters.
LSOAs in sub-cluster 2-3 can be found across most Local Authorities, with the exception of Monmouthshire and Ceredigion. The local authorities with proportionally more of their LSOAs in this sub-cluster are Blaenau Gwent (23%) and Merthyr Tydfil (22%).
Figure 14: Distribution of Sub-cluster 2-3 within Blaenau Gwent at LSOA Level
Description of Figure 14: This figure is a map illustrating the distribution of sub-cluster 2-3 within Blaenau Gwent at the LSOA level. LSOAs that belong to sub-cluster 2-3 are highlighted in blue, while other sub-clusters are shaded in grey. Approximately one third of the LSOAs on the map are part of sub-cluster 2-3.
Conclusion
In this report, unsupervised machine learning has been utilised to uncover distinct patterns of deprivation across small areas in Wales. By analysing the WIMD 2019 data, we have segmented Wales into two primary clusters, further divided into six sub-clusters based on similarities across multiple deprivation indicators. This approach facilitates a detailed understanding of the varying degrees and dimensions of deprivation across the 1,909 small areas examined. Each sub-cluster's unique characteristics are explored, highlighting a broad range of deprivation levels across domains such as income, education, health, housing quality and access to services.
The application of the clustering algorithm to the multiple deprivation indicators not only substantiates the identified patterns of deprivation but also demonstrates the potential of machine learning techniques in enhancing our understanding of deprivation across Wales. The successful application of these techniques underscores the possibility for ongoing analysis to continuously refine and update our understanding of deprivation profiles across Wales.
Appendix 1: WIMD Indicator description to indicator code lookup
Police recorded theft (rate per 100)
Indicator code: THEF
Police recorded violent crime (rate per 100)
Indicator code: VIOC
Repeat Absenteeism (%)
Indicator code: REAB
Average public return travel time to a secondary school (minutes)
Indicator code: PUSS
Average public return travel time to a sports facility (minutes)
Indicator code: PUSF
Average public return travel time to a primary school (minutes)
Indicator code: PUPS
Average public return travel time to a post office (minutes)
Indicator code: PUPO
Average public return travel time to a pharmacy (minutes)
Indicator code: PUPH
Average public return travel time to a public library (minutes)
Indicator code: PULI
Average public return travel time to a GP surgery (minutes)
Indicator code: PUGP
Average public return travel time to a food shop (minutes)
Indicator code: PUFS
Average private return travel time to a secondary school (minutes)
Indicator code: PRSS
Average private return travel time to a sports facility (minutes)
Indicator code: PRSF
Average private return travel time to a primary school (minutes)
Indicator code: PRPS
Average private return travel time to a post office (minutes)
Indicator code: PRPO
Average private return travel time to a pharmacy (minutes)
Indicator code: PRPH
Average private return travel time to a petrol station (minutes)
Indicator code: PRPE
Average private return travel time to a public library (minutes)
Indicator code: PRLI
Average private return travel time to a GP surgery (minutes)
Indicator code: PRGP
Average private return travel time to a food shop (minutes)
Indicator code: PRFS
People in overcrowded households (%)
Indicator code: OVCR
Children aged 4 to 5 who are obese (%)
Indicator code: OBCH
Adults aged 25 to 64 with no qualifications (%)
Indicator code: NOQU
Key Stage 4 leavers entering Higher Education (%)
Indicator code: NEHE
GP-recorded mental health condition (rate per 100)
Indicator code: MENH
Limiting long-term illness (rate per 100)
Indicator code: LLTI
Low birth weight (live single births less than 2.5kg) (%)
Indicator code: LBW
Key Stage 4 average point score
Indicator code: KS4
Key Stage 2 average point score
Indicator code: KS2
People in income deprivation (%)
Indicator code: INC
Likelihood of housing containing serious hazards (%)
Indicator code: HQUAH
Likelihood of housing being in disrepair (%)
Indicator code: HQUAD
Likelihood of poor-quality housing (%)
Indicator code: HQUA
Foundation Phase Average Point Score
Indicator code: FOUN
Households at risk of flooding score
Indicator code: FLRS
Fire incidences (rate per 100)
Indicator code: FIRE
Working-age people in employment deprivation (%)
Indicator code: EMP
% Unavailability of broadband at 30Mb/s
Indicator code: DIG
Premature death (rate per 100,000)
Indicator code: DEAT
Police recorded criminal damage (rate per 100)
Indicator code: CRDG
GP-recorded chronic condition (rate per 100)
Indicator code: CHRON
Cancer incidence (rate per 100,000)
Indicator code: CANC
Police recorded burglary
Indicator code: BURG
Anti-Social Behaviour (rate per 100)
Indicator code: ASB
Ambient Green Space Score
Indicator code: AMBGS
Population Weighted Average Concentration Value for Particulates < 2.5 µm
Indicator code: AIQP2
Appendix 2: built-up area BUA size classification
| Population range (Usual resident population) | BUA size classification | Approximate settlement type |
|---|---|---|
| 0 to 4,999 | Minor | Hamlet or village |
| 5,000 to 19,999 | Small | Larger village / small town |
| 20,000 to 74,999 | Medium | Medium towns |
| 75,000 to 199,999 | Large | Large towns / smaller cities |
| 200,000+ | Major | Cities |
Description of Table 5: table displays the percentage distribution of the 6-fold rurality classification across the six sub-clusters. Each row represents a specific rurality classification, and the columns correspond to the sub-clusters. Urban classifications dominate most sub-clusters, with Sub-cluster 2-3 having the highest percentage of "Urban city and town" areas at 88.4%, followed by Sub-cluster 2-2 at 81%. In contrast, Sub-cluster 1-2 is the most rural, with 60.1% of its areas classified as "Rural village and dispersed in a sparse setting" and an additional 31.7% as "Rural village and dispersed."
Sub-cluster 1-1 shows a more balanced distribution, with 64.9% of its areas in "Urban city and town," 15.7% in "Rural town and fringe," and 10.4% in "Rural town and fringe in a sparse setting." Sub-cluster 1-3 also leans urban, with 73.1% of its areas in "Urban city and town," but still maintains a rural presence, with 6.7% in "Rural village and dispersed." Sub-cluster 2-1 and Sub-cluster 2-2 have similar patterns, with high urban proportions (75.6% and 81%, respectively) but lower rural classifications compared to Sub-cluster 1-1 and 1-3.
Sparse settings are most prominent in Sub-cluster 1-2 and Sub-cluster 1-1, with notable proportions in "Rural village and dispersed in a sparse setting" (60.1% and 2.7%, respectively) and "Urban city and town in a sparse setting" (1.8% in 1-1).
Sub-cluster 2-3 has negligible rural representation, highlighting its predominantly urban nature.
Contact details
Statistician: David Basch
Email: stats.inclusion@gov.wales
Media: 0300 025 8099