dwelling level housing stock modelling and database ... · the ‘simpleco2’ model 29 appendix b...

Dwelling level housing stock modelling and database development for Thurrock Council Prepared for: Louisa Moss Private Housing Service and Adaptation Team Manager Thurrock Council July 2012

Client report number 271-148

2 Dwelling level housing stock modelling

BRE Client report number 271-148 Commercial in confidence

© Building Research Establishment Ltd 2012

Prepared by

Name Rob Flynn

Position Associate Director

Signature

Approved on behalf of BRE

Name John Riley

Position Director, BRE Housing Stock Performance

Date 6th July 2012

Signature

BRE Garston WD25 9XX T + 44 (0) 1923 664000 F + 44 (0) 1923 664010 E [email protected] www.bre.co.uk

This report is made on behalf of BRE. By receiving the report and acting on it, the client - or any third party relying on it - accepts that no individual is personally liable in contract, tort or breach of statutory duty (including negligence).

mailto:[email protected]

http://www.bre.co.uk




Executive Summary

BRE were commissioned to provide information on key housing private sector housing indicators to Thurrock Council. Principal among these indicators is the rate of Category 1 Housing Health and Safety Rating Hazards found in the stock as this is now the minimum standard for housing which every housing authority is obliged to keep under review. The information provided on Category 1 Hazards will also provide the principal inputs for the Health Impact Assessment of private sector housing which follows on from this project.

The estimates for Category 1 Hazards and the other indicators have been provided using a stock modelling methodology. Such approaches have been used for many years by BRE but this particular set of models is the first of its kind to make significant use of the Experian UK Consumer Dynamics Database of dwelling and household indicators as inputs to the models. This and data from the governments English Housing Survey have been the main input data to the models.

The methodology is explained in some detail in the report and appendices but begins with dwelling level inputs provided by Experian and expands on these using inference techniques to provide sufficient information to calculate likely the energy efficiency of each dwelling in the stock. Some of the housing standards can be directly inferred from this data while others use regression techniques to predict whether a dwelling is likely to meet the standard.

A particular feature of this project is the integration of the model estimates with data held by the local authority. The most significant use has been the integration of [ ] records of households receiving means tested benefits which will improve the model estimates of the Vulnerable Households and improve targeting where dwellings occupied by such households are estimated to fail any of the housing standards.

The table below summarises the modelled estimates for Thurrock private sector housing. Modelled data, private sector stock: authority level summary

In general the results are better than for the national stock but there are still an estimated 6,683 dwellings where a Category 1 Hazard is expected to be present. Most of these hazards (4,771) are fall hazards with the next highest incidence being from Excess Cold (1,557).

The Borough level data is of considerable interest and valuable to place Thurrock in a national context and helpful to predict the likely demand for private sector housing services but it is when the database is used to to cross tabulate different variables and the results are mapped that some of the most useful information is provided. The map below illustrates how cross tabulating Category 1 Hazard for Excess Cold with




Vulnerable Households can indicate where concentrations of households living on the lowest incomes and in the hardest dwellings to heat are likely to be found.

Percentage of private sector dwellings with the presence of a HHSRS Category 1 Hazard for Excess Cold and occupied by vulnerable households

Such maps, and the dwelling lists that underly them, are useful when considering how best to target the Council’s efforts to improve private sector housing whether these are focussed on publicity, joint working or use of enforcement powers.

The database that provides these results and maps is the main output of this project and is currently a unique asset as it is the first to use a dwelling based method of modelling (previous models have been based on area based models). One of it’s major advantages is the relative ease with which local data can be used to update it. It is therefore important that knowledge of the methods used by the authority to provide their input should be retained for future use in keeping the database live. It is also important to look for new opportunities to add to this data which may be presented by activities resulting from this project such as increased enforcement activity or improved joint working.




Contents

Introduction 6 Background 6 Regional and local information requirements 7 Coalition Government Policy: Laying the foundations: A Housing Strategy for England 8 New Government Policy: The Green Deal 9

BRE Dwelling Level Housing Stock Models 10 Updating the BRE Dwelling Level Housing Stock Models 11 Results from the BRE Housing Stock Models 14

Database development 25

Applying the dwelling level housing stock model information 26

Appendix A – Dwelling level housing stock modelling methodology The ‘SimpleCO2’ model 29

Appendix B – Use of the local authority data Private dwelling address list 32 Council Tax Benefit claims 32 Renovation grants 33 Energy data 33

Appendix C – Definitions of the modelled indicators




Introduction

BRE were commissioned to provide information on key housing indicators to Thurrock Council. These include Decent Homes, the Housing Health and Safety Rating System (HHSRS) and Fuel Poverty. The information has been derived from a series of models which make use of the Experian UK Consumer Dynamics Database using a range of statistical methods. This data provides officers at Thurrock with detailed information on the likely condition of the stock and the geographical distribution of properties of interest, such as those that might benefit from energy efficiency improvements, or other forms of intervention.

Before considering the stock models themselves we first of all set out the basic legislative, government and local reporting requirements on private sector housing information. We then proceed to the results themselves.

Background

Local housing authorities are required to review housing conditions in their districts in accordance with the Housing Act 2004 s3. This is a wide ranging requirement as it refers to other parts of the Act as well as other legislation, which between them cover:

• dwellings that fail to meet the minimum standard for housing i.e. dwellings with Category 1 hazards under the HHSRS

• HMOs

• selective licensing of other houses

• demolition and slum clearance

• the need for provision of assistance with housing renewal

• the need for assistance with adaptation of dwellings for disabled persons.

While these are the areas covered by the legislation, the latter does not proscribe precisely what information should be collected when an authority reviews its housing.




Regional and local information requirements

The information to be collected is therefore largely for the authority to decide, although it needs to take account of Government policy. The previous Government (through the CLG) took the view that ‘primary responsibility for repairing and improving homes in the private sector lies with the owner’. They recognized, however, that some - particularly the elderly and other vulnerable groups - would not have the resources necessary for this. Authorities were expected, through the powers they have under the Regulatory Reform Order 2002, to have developed policies on providing such assistance. Government funding for such assistance was channelled via the Regional Assemblies, who were expected to prioritise those most in need.

Audit Commission guidance1 further advised on the need and methodology for collecting information and how it should be used.

The Coalition Government, however, has abolished Regional Assemblies, the Regional Development Agencies, the Government Offices for the regions and the Audit Commission.

In their place the DCLG in their Draft Structural Reform Plan has announced among other measures that it will ‘Move from local authorities primarily reporting to central government to local authorities that report to local people’. This has included the abolition of Comprehensive Area Assessment and it is pledged to cut local government inspection. It has also announced its intention to ‘design and implement a new approach with less reporting burdens on local government and greater transparency for local people’. This has been followed up by a letter from the Minister to all local authorities, revoking all Local Area Agreements, and effectively handing over control of Local Agreements to local authorities. The letter specifically states that authorities wishing to drop any targets may do so, and where authorities choose to retain targets DCLG will not monitor them. It is therefore very much for Thurrock to decide on it’s housing priorities.

The strategic housing vision for Thurrock is:

“To provide high quality housing opportunities in safe, clean and green environments. We will continue to care for our existing homes and ensure that new developments are well designed and support our ambition for our communities to flourish alongside business growth. Thurrock will be a place to live, learn and enjoy.”

In order to support this vision, there are some key priorities, many of which can be supported by the information delivered though the BRE stock model. Thurrock’s housing priorities are

• Managing and improving housing supply and choice • Meeting and supporting the needs of vulnerable groups • Investing in the Housing Stock and environmental sustainability • Ensure all services are effective and achieve value for money

Before turning to the modelled results we briefly consider how the most recent government policy is shaping local authority housing information requirements.

1 Updated strategic approach to housing KLOE for use from 12 April 2010 Audit Commission January 2010




Coalition Government Policy: Laying the foundations: A Housing Strategy for England

The Coalition Government has recently published a strategy document 'Laying the Foundations: A Housing Strategy for England'2. The main focus of this document is on areas other than private sector housing; notably on increasing housing supply and reforming social and affordable housing. However, two chapters of the report do focus on traditional areas of private sector housing policy interest; the private rented sector and empty homes.

The growth of the private rented sector in recent years is acknowledged, and they state that they are ‘looking at measures to deal with rogue landlords and encouraging local authorities to make full use of the robust powers they already have to tackle dangerous and poorly maintained homes’. They commend the approach adopted by Liverpool City Council, which works closely with landlords through a Landlord Forum and Panel operating a Landlord Accreditation Scheme, but still operates a robust enforcement regime. They also commend the partnership with Liverpool Primary Care Trust. The ‘Healthy Home Programme’ has identified over 2000 hazards and led to an estimated £3 million investment by landlords into the housing stock, delivering sustainable health improvements and enhancing community wellbeing. The Health Impact Assessment which will be informed by the Category 1 Hazard models is a first stage in the development of such a program.

The document also draws attention to the Green Deal (discussed in more detail below), which will enable all consumers to install energy efficiency measures at no upfront cost. Householders will repay the cost of the measures through their energy bill savings.

From 2016 reasonable requests by tenants for such energy efficiency improvements will not be able to be refused and from 2018 it will be unlawful for landlords to rent out properties that do not reach a minimum standard of energy efficiency. While there will be various caveats to these powers, they will provide a new minimum standard for rented accommodation. Part of this project includes provision of a private rented sector that should assist in identifying such dwellings.

The document does, however, also point out that the Housing Health and Safety Rating can be used effectively to improve rented property and quotes Hull City Council’s PEAL scheme (Proactive housing and Environmental Action Locally) as an example of how training for landlords in the Rating System can result in effective self regulation.

So, while new powers will become available in 2016, there are existing powers that can be used effectively to deal with the hazards identified by the BRE models.

Empty homes brought back into use will qualify for the New Homes Bonus where, for the following six years, the Government is to match funding the Council Tax on long term empty properties brought back into use. The document also draws attention to the Empty Homes Toolkit provided by the Homes and Communities Agency. From 2012-15, £100million capital funding from within the Affordable Homes Programme will be available to tackle long-term empty homes. While the data provided by this project cannot necessarily assist with identification of empty homes, the database provided would be the logical place for such information to be stored could it be gathered from other sources.

2 Laying the Foundations: A Housing Strategy for England CLG 2011

http://www.communities.gov.uk/publications/housing/housingstrategy2011

http://www.communities.gov.uk/publications/housing/housingstrategy2011




New Government Policy: The Green Deal

We referred to the forthcoming enforcement powers under the Green Deal, but this policy is primarily about encouragement to take up energy efficiency measures. Local authorities can, if they choose, become Green Deal Providers. Others, while not intending to be providers (i.e. installing and financing Green Deal measures), are taking major steps to engage with providers; e.g. the Birmingham Energy Savers scheme, which will result in favoured status for a particular provider.

The Green Deal and Energy Company Consultation Document 3 and the associated local authority information note4 outline the various roles an authority can adopt. These can be summarised as to provide, partner or promote, depending on the extent of engagement the authority decides on. For all three levels of engagement, however, they are ‘likely to act as partners, adding value by, for example, providing information on local housing stock, and endorsing and helping market company activity, using their position as a trusted interlocutor with households to increase local acceptance and take-up’.

The Home Energy Conservation Act 1995 is to be retained, and is likely to be used to encourage local authorities by requiring annual reporting on how they plan to engage with the Green Deal and the future Energy Company Obligation (ECO). Guidance on this is due to be issued in Spring 2012. Both the consultation and the information note refer to the local authority role in indentifying street-by-street opportunities for Green Deal roll out.

It is quite clear that the availability of small-area data, on energy efficiency in particular, will be of particular value as authorities rise to the challenge of the Green Deal.

3 The Green Deal and Energy Company Consultation Document DECC 2011

http://www.decc.gov.uk/en/content/cms/consultations/green_deal/green_deal.aspx 4 Local Authorities and the Green Deal DECC 2011 http://www.decc.gov.uk/en/content/cms/consultations/green_deal/green_deal.aspx

http://www.decc.gov.uk/en/content/cms/consultations/green_deal/green_deal.aspx

http://www.decc.gov.uk/en/content/cms/consultations/green_deal/green_deal.aspx




BRE Dwelling Level Housing Stock Models

The original BRE Housing Stock Models, developed in 2003, were the result of an 18-month development program supported by the BRE Trust. These models, until recently, were based on a combination of data from the English Housing Survey (EHS) and nationally available small area datasets such as the census. These models proved to be useful and very popular with authorities, with over 230 of the Local Authorities in England purchasing the models.

Over the last few years BRE have been looking to further improve the quality of this service. The models provided to Thurrock Council are the first of BRE’s new approach which improves upon the previous methodology and is dwelling rather than area based.

This new methodology makes use of a model originally devised for modelling carbon emission which we call ‘SimpleCO2’. The model uses basic dwelling descriptions from the Experian UK Consumer Dynamics Database e.g. dwelling type and age to define broad dwelling archetypes. It then uses knowledge of distributions from the English Housing Survey (EHS) to predict the likely fuel type, boiler type, insulation levels and other key variables which are required to calculate levels of energy efficiency.

Variables such as ‘SimpleSAP’ (an estimate of the Governments energy efficiency rating known as SAP) and Excess Cold are direct outputs of the calculations preformed by the model. For the other variables (the full list is provided later in this section) we have adopted a top down methodology similar in many respects to the approach used by our previous set of models. We use data from the EHS and Experian to establish relationships between the housing standards and various dwelling and social characteristic using logistic forms of regression analysis. Once these relationships are established, they are used to create a regression model which calculates the likelihood of a dwelling failing a given standard.

There are therefore, in effect, two methodologies being used. The first deals with the lack of information required to undertake an energy efficiency calculation by modelling missing input data such as insulation levels. The second models the probability of failing a standard using formula derived from the EHS and applied to the Experian data. It then converts the probability to a pass/fail value based on the expected pass/fail level in the census output area (COA) in which the dwelling is found. These approaches ensure indicative values are provided for each variable and each dwelling i.e. pass/fail for a standard or an actual value for ‘SimpleSAP’. The probabilistic models used to reach these values mean that the individual dwelling data can only be indicative but in our view this makes as full a use as is possible of the dwelling level input data.

Additional information on our methodology is included in Appendix A.




Updating the BRE Dwelling Level Housing Stock Models

As a general rule it is preferable to use data collected by the Council where it is available in preference to modelled data.

BRE suggested a number of possible data sources internal to the Council and Council officers have supplied the following data which has been incorporated into the BRE base models:

• An address list for all private dwellings

• Dwellings where at least one occupant claims Council Tax benefit

• Energy improvement data

• Renovation grant data

This information was put to use in three ways:

• To correct original information from Experian

• To lower the need for imputation by replacing imputed values for energy efficiency measures with collected data where possible

• To replace final modelled outputs with collected data where this was available

The address list for all private dwellings did not allow us to update any information, but did allow us to identify the dwellings that were in the private sector. The Experian data also has a tenure variable, but this is not 100% reliable.

The Council Tax benefits data was used to identify vulnerable households. A household in receipt of means tested benefits is considered vulnerable, so identifying those receiving Council Tax benefit allows us to pinpoint a large proportion of, although not all, vulnerable households. This is an example of replacing final modelled data with real data.

The renovation grant data was used to identify houses that are not in disrepair. It is also used to adjust the overall total of houses in disrepair, on the assumption that a house that received a grant was in disrepair, but is not any more.

The energy data was used to replace imputed data relating to energy performance such as insulation levels. For some of these cases we were also able to obtain data on basic information such as the dwelling type. These are examples of lowering the need for imputation and correcting Experian data respectively.

The energy data was particularly interesting in that it allowed us to make an assessment as to how accurate our imputation methods have been for Thurrock. Table 1 below shows a list of the information that could be used to assess the accuracy of the imputation methods, in that it is data that one would not expect to change, e.g. number of storeys.




Table 1: Local authority data updates

Number of cases Percent of cases

Variable Information

provided Updated In

agreement Updated In

agreement Dwelling type 184 89 95 48% 52%

Number of storeys 27 13 14 48% 52%

Type of wall 145 34 111 23% 77%

Presence of wall insulation 145 100 45 69% 31%

Heating system 305 14 291 5% 95%

Hot water provision 305 18 287 6% 94% It is reassuring to see that the hot decking appears to have been quite successful for many variables, though it should be remembered that this is a biased sample, in that these dwellings were selected for energy improvements works. They may, therefore, be subject to any bias in the original selection process used to identify households that would benefit from energy improvement measures. It should be noted that because differing levels of information were available for each case, the amount of information provided differs for each variable.

It is worth adding an extra word or two on the topic of dwelling types, particularly in light of the low accuracy achieved here. The dwelling type required by BRE for modelling purposes and the dwelling type used by Experian are, for some categories, quite different. For example, Experian do not distinguish between end terraces and mid terraces, while BRE do not distinguish between houses and bungalows. Consequently we are forced to make some quite difficult imputations here, as there is little data available to help determine for example whether a terraced house is end terrace or mid terrace. This is why the accuracy rate for dwelling type appears to be lower than Experian’s quoted figure, because BRE has been forced to modify it quite heavily.

A more detailed explanation of how each data set was used can be found in Appendix B.

The stock totals used in the model come directly from the Experian tenure data. By purchasing the tenure variable from Experian it is possible to break the housing stock model information down by tenure. Before we show the housing stock models results the stock totals are provided by tenure in Table 2 below.

Table 2: Experian stock totals by tenure

Tenure Number of dwellings

% of all stock

Owner Occupied 45,483 71%

Private Rented 6,794 11%

Social 11,896 18%

Total 64,173 -




The stock totals were compared to the reported neighbourhood statistics figures5 shown in Table 3 as a way of validating the Experian information.

Table 3: Neighbourhood statics stock totals by tenure

Tenure Number of dwellings

% of all stock

Owner Occupied and Private Rented Dwelling Stock 52,780 82%

LA Dwelling Stock 10,322 16%

RSL Dwelling Stock 1,347 2%

Total Dwelling Stock 64,450 -

The figures compare favourably, though in both cases certain categories need to be combined in order to make comparisons. If we add together the LA and RSL totals from the neighbourhood statistics we get 11,669 dwellings, very close to the 11,869 figure for social stock from the Experian data, which would reflect a later date and hence possibly incorporate any new build or tenure changes. Similarly if we combine the owner occupied and private rented figures from the Experian data we get 52,277, very close to the figure of 52,780 from the neighbourhood statistics.

6 Important note: while we can provide ‘SimpleSAP’ ratings from the ‘SimpleCO2’ software, under no circumstances must these be referred to as SAP as the input data is insufficient to produce an estimate of SAP or even RdSAP for an individual dwelling that meets the standards required by these methodologies.




Results from the BRE Housing Stock Models

Tables 4 and 5 describe the stock at ward and borough level, providing estimated percentage and stock totals for the following indicators:

• Dwellings which fail the Decent Homes standard

• Dwellings which fail the Decent Homes standard due to:

o inadequate thermal comfort

o the presence of a Category 1 Rating System Hazard

§ the presence of a Category 1 hazard for excess cold (using SAP ratings as a proxy measure in the same manner as the English House Condition Survey)

§ the presence of a Category 1 hazard for falls (including falls associated with baths, falling on the level, falling on stairs and falling between levels)

o disrepair

o non modern facilities and services

• Fuel poverty (this is based on the full fuel poverty measure based on 10% of earnings being spent on heating costs)

• Dwellings occupied by a vulnerable households

• Non decent homes occupied by a vulnerable households

• An estimate of the SAP rating which to emphasise its origin from a reduced set of variables we refer to as ‘SimpleSAP’6.

Definitions for all of the modelled indicators are provided in Appendix C.

Table 4 summarises the key statistics at the authority level by tenure. Table 4: Modelled data by tenure: authority level summary

The differences between the tenures generally reflect the position at national level with the private rented stock in poor condition than the owner occupied stock.





Table 5: Modelled data, private sector stock




Table 5: continued




Summary of results: private sector stock

Table 6 summarises the key statistics at the authority level by tenure. Table 6: Modelled data, private sector stock: authority level summary

Most of the house condition and energy efficiency indicators suggest the housing stock in Thurrock to be slightly better than the national average.

The headline indicators and their comparison with England are given below:

• The percentage of non decent homes at 32% is 3% lower than the average for England in 2009.

• For HHSRS Category 1 hazards the model estimate, 13%, is 9% lower than the national average in 2009.

• The model estimate for dwellings which fail thermal comfort, 17%, is 7% higher than the national average.

• Excess cold, which is the most common Category 1 hazard, at 3% is 6% lower than the national average.

• The one indicator provided solely of socio-economic conditions, vulnerable households, shows these conditions to be better than the national average (18% compared to 22%).

• The percentage of vulnerable households in non decent homes, 6%, is 2% lower than the national average in 2009.

• The model estimate for fuel poverty is 12% which is 6% lower than the national average in 2009 of 18%.

• The ‘SimpleSAP’ rating for Thurrock of 55 is higher i.e. better than the national average SAP rating for 2009 of 51.




Housing Stock Model outputs by census output areas

Outputs from the models are also produced in map form at the level of the COA. These typically comprise around 125 households and usually include whole postcodes, which have populations that are largely similar.

The maps show the percentages of private sector homes in each output area that are, for example, estimated to have a HHSRS Category 1 hazard. The ranges are defined by using the standard deviation of the COA statistics. The outputs in the lightest and darkest colours on the map are the extreme ends of the range, which therefore highlight the best and worst areas.

The COA data in map format is provided in this report for the following indicators:

• The presence of a Category 1 Rating System Hazard

• The presence of a Category 1 Hazard for excess cold (using SAP ratings as a proxy measure in the same manner as the English House Condition Survey)

• An estimate of the SAP rating which to emphasise its origin from a reduced set of variables we refer to as ‘SimpleSAP’7.

• Fuel poverty (this is the full fuel poverty measure based on 10% of earnings being spent on heating costs)

• Efficient fuel (areas with low levels of efficient fuel, in the main have a high prevalence of electric heating systems)

• Dwellings occupied by a vulnerable households





Map 1: Percentage of private sector dwellings with the presence of a HHSRS Category 1 Hazard




Map 2: Percentage of private sector dwellings with the presence of a HHSRS Category 1 Hazard for Excess Cold




Map 3: Average ‘SimpleSAP’ Ratings per dwelling, private sector




Map 4: Percentage of private sector dwellings with efficient fuel types for main heating




Map 5: Percentage of private sector dwellings occupied by households in fuel poverty




Map 6: Percentage of private sector dwellings occupied by vulnerable households




Database development

• The MS access database prepared for Thurrock as part of this project is organised into 3 broad areas:-Data

• Summary tables

• GIS outputs

The data contains all of the input data used in the SimpleCO2 models, the outputs of those models (i.e. direct housing standards) and the indirect housing standards. It also includes the address details of the dwellings, the index of multiple deprivation, and an indicator as to which data was successfully matched as part of the Experian address matching exercise.

Where data from Thurrock has been used to update or improve the BRE models a flag this is indicated by a flag variable in the appropriate file.

In addition to the data itself there are number of tables that simply contain text values for the numeric data employed in the ’SimpleCO2’ model (e.g. for tenure 1=owner occupied, etc).

The summary tables are in fact queries, created from the data. They are referred to as summary tables as these queries produce the basic data for the summary tables seen elsewhere in this report.

The GIS outputs are also queries, though this time the results of these queries are used to generate the maps, also seen elsewhere in this report.

The contents of the database are, principally, the assorted data and queries needed to create most of the tables and maps in this report. There are, however, some additional features such as search facilities for particular streets or postcodes. All useful outputs can be accessed via the form entitled “Interface”.

For a more detailed explanation of the items in the database see the Thurrock database guidance document.




Applying the dwelling level housing stock model information

The analysis provided on the dwelling level housing stock models in this report is displayed at local authority and ward level tables or as COA level maps. All of these figures have been calculated by aggregating dwelling level information from the database which has been provided. The main strengths of this project is the dwelling level database which can be used by the Council officers to manipulate, amalgamate and extract information from the database to aid Thurrock’s Housing Services projects and reporting.

To give one example, if looking to target energy improvement measures one of the indicators which may be useful is the presence of Excess Cold Category 1 hazards. Map 7 below is a copy of a COA map provided earlier in the report of the presence of Excess Cold Category 1 hazards at COA level. This map shows that the presence of Excess Cold Category 1 hazards is spread throughout the local authority.

For targeting purposes the local authority may want to focus on dwellings occupied by vulnerable households as these households are more likely to be eligible for assistance to improve their homes (Map 8). This map shows that vulnerable households are predominantly found in the urban areas within the local authority.

Within the database it is possible to create a variable which combines Excess Cold Category 1 Hazards and dwellings occupied by vulnerable households which can then be mapped in the same way at COA (Map 9). This map provides a potentially more useful representation of areas within the local authority on which officers may wish to target energy saving measures.




Map 7: Percentage of private sector dwellings with the presence of a HHSRS Category 1 Hazard for Excess Cold

Map 8: Percentage of private sector dwellings occupied by vulnerable households




Map 9: Percentage of private sector dwellings with the presence of a HHSRS Category 1 Hazard for Excess Cold and occupied by vulnerable households

This is an example of where the data within the database can be manipulated to provide a useful variable which can be used by the authority, by combining two variables to target resources. We have shown that, by displaying the aggregated information at COA level, maps can quickly and effectively give an idea of where higher levels of the targeted demographic can be found. The dwelling level information can then be used to extract a list of addresses which the models have estimated as being in that targeted demographic, for example, dwellings most likely to have Excess Cold Category 1 hazards and be occupied by vulnerable households. Figure 1 is a screenshot of the database providing this information for the first five addresses in East Tilbury (with street numbers obscured). Figure 1: Extract from database of addresses estimated to be most likely to have Excess Cold Category 1 hazards and be occupied by vulnerable households

Dwelling Level Housing Stock Modelling



Appendix A – Dwelling level housing stock modelling methodology

The ‘SimpleCO2’ model

BRE are the original developers of the Building Research Establishment Domestic Energy Model. The model calculates from measures of building characteristics the energy costs of a dwelling assuming a standard heating and living regime. The model has a number of outputs including an estimate of the SAP rating and carbon emissions.

BRE have developed a variant of the BREDEM software (‘SimpleCO2’) that can calculate outputs that are indicative of the full BREDEM outputs described above from a reduced set of input variables. The minimum set of variables the software can accept is information on:

• Dwelling type

• Dwelling age

• Number of bedrooms

• Heating fuel

• Heating system

• Tenure

BRE are using the Experian UK Consumer Dynamics Database as a source of these variables. BRE take the above variables from this database and convert them into a suitable format for the ‘SimpleCO2’ software.

The above variables are insufficient on their own for the software to calculate a ‘SimpleSAP’ rating or carbon emissions estimate. Additional variables are required and as these values cannot be precisely inferred then a technique known as hot deck imputation is undertaken. This is a process of assigning values in accordance with their known proportions in the stock. One area where this technique is used is for predicting heating fuels as the Experian data can only tell us whether a dwelling is on the gas network or not. We do not know what fuel might be used by dwellings not on the gas network, so in most cases this information will be assigned using probabilistic methods to dwellings known to be off the gas grid. The process is actually far more complex e.g. dwellings with particular characteristics such as larger dwellings are more likely to be assigned with oil as a fuel than smaller dwellings.

The reason for taking this approach is to ensure that the national proportions in the data source are the same as those found in the stock nationally (as predicted by the EHS or other national survey). While there is the possibility that some values assigned will be incorrect for a particular dwelling (as part of the assignment process has to be random) they ensure that examples of some of the more unusual types of dwelling that will be present in the stock are included.




While this approach is an entirely sensible and commonly adopted approach to dealing with missing data in databases intended for strategic use, it raises issues where one of the intended uses is planning implementation measures. Mindful of this all variables where hot deck imputation has been applied by BRE are identified.

It is important to note that some variables have been entirely assigned using hot decking imputation techniques. These include presence of cavity wall insulation and thickness of loft insulation as there is no reliable database with national coverage for these variables.

The ‘SimpleCO2’ software takes the combination of Experian and imputed data and calculates the ‘SimpleSAP’ rating for each dwelling in the national database including those in the Thurrock area. The calculated ‘SimpleSAP’ ratings are the basis of our estimates of SAP and excess cold. We discuss in a later section how we derive the other key variables.

The estimates of ‘SimpleSAP’ etc. cannot be guaranteed due to their being calculated from modelled data which may itself be inaccurate. They do, however, provide the best estimates that we are aware can be achieved from a data source with national coverage and ready availability. The input data could, however, be improved

• in it’s accuracy for example through correcting erroneous values

• in it’s depth of coverage for example by providing more detailed information on age of dwellings or

• in it’s breadth by providing additional input variables such as insulation.

Improving any of these would enhance the accuracy of the output variables. For this reason it is always worth considering utilising additional information sources where they are available.

The approach described above can provide estimates for the variables we are offering to supply where they:

1. are produced by the ‘SimpleCO2’ software i.e. ‘SimpleSAP’

2. can be derived from the outputs i.e. excess cold (using the SAP <31.5 surrogate) or

3. can be derived from the imputed input variables e.g. HHSRS Category 1 Hazard from excess cold.

The other variables we are offering to supply estimates of, however, are not based solely (or in most cases not at all) on the thermal characteristics of the dwelling and these include:

• The presence of a Category 1 Rating System Hazard

• The presence of a Category 1 Fall Hazard

• The presence of a household in Fuel Poverty (this is the full fuel poverty measure based on 10% of earnings being spent on heating costs)

• Disrepair (using the Decent Homes standard definition of disrepair)

• The presence of a vulnerable households (defined as a dwelling likely to be in receipt of a means tested benefit)




These present a different challenge and we have adopted a top down methodology similar in many respects to the approach used by our previous set of models. We use data from the English Housing Survey and statistical techniques such as logistic regression to determine the combination of variables that are most strongly associated with failure of a particular standard. We develop a formula to predict likelihood of failure and then apply that to the variables in the national Experian dataset to provide a likelihood of failure for each dwelling. This is then assigned a failure/compliance to standard value on an area basis. Thus if the aggregate values for a census output area are that 60% of the dwellings in the area fail a particular standard then 60% of the dwellings with the highest failure probabilities will be assigned as failures and the remaining 40% as passes.

The presence of a Category 1 Hazard failure is the only exception to this as it is found by combining Excess Cold, Fall Hazards and Other Hazards such that failure of any one of these hazards leads to failure of the standard.




Appendix B – Use of the local authority data

Private dwelling address list

This address list was used as the means of matching all other Thurrock data to the BRE models. The address list came with a UPRN field, a Unique Property Reference Number that uniquely identifies each dwelling. All other data supplied by the authority also came wil a UPRN field, allowing them to be matched.

The address data and UPRN were passed to Experian UK who were able to use the address to match the UPRN fields to the BRE models. A total of 53570 addresses were sent to Experian for data matching. Experian returned a file of 53570 cases in which 50609 of these cases had been successfully matched to Experian data. There were, however, a small number of cases where the Experian unique ID code was either not unique, or was missing, rendering them unsuitable for matching with the BRE mode. In the end a total of 49,871 addresses were successfully matched to the BRE models.

The third function of the private address list was a as a check on tenure. All dwellings matched to the list should be private sector stock, either owner occupied or private rented tenure. In fact, a total of 4465 cases that had been matched to the file were classified as Social Rented by Experian. This is not a surprising occurrence as the Experian data is not 100% accurate, indeed the number of dwellings apparently misclassified is about what would be expected according to Experian’s own estimates of their accuracy. It does, however present something a of a problem in that the dwellings that Experian have classified as social have been modelled on that basis, and consequently are likely to appear better in terms of many of the housing standards than they would do if they were modelled using their correct tenure. Unfortunately as there is no way of determining the correct tenure, all that can be done is to flag these cases in the database so as to make sure that the authority is aware that the modelled data for these cases may be less accurate than for other cases.

Council Tax Benefit claims

This data was used to update the vulnerability status of 6,044 dwellings in the model. Where a dwelling contained one or more people in receipt of council tax benefit, the dwelling was classified as one containing a vulnerable household.

This form of information is observational, in that the data represents a clarification of something already thought to be true. The BRE models predicted that there would be a certain number of vulnerable occupants, but this information clarifies where they are. Consequently, the information here will not alter the modelled number of vulnerable households.




Renovation grants

Where a dwelling was found to have been in receipt of a renovation grant the disrepair status of the dwelling was changed to a pass.

This form of information represents an intervention, in that the council has altered the status of the dwelling, and changed what the BRE models originally estimated to be true for these dwellings. Consequently, the predicted number of dwellings failing disrepair has been reduced by 860 in light of the information provided by the authority.

Energy data

The energy data provided by the authority was of great value as it allowed BRE to replace a number of imputed energy characteristics with real ones. The following information was available from the energy data:

• 27 cases where information was available on the number of storeys

• 285 cases where information was available on the levels of loft insulation

• 145 cases where information was available on wall type

• 145 cases where information was available on wall insulation

• 305 cases where information was available on the heating type, boiler type and heating controls

• 305 cases where information was available on the method of water heating

• 307 cases where information was available on the hot water tank insulation




Appendix C – Definitions of the modelled indicators

Decent Homes and components

A decent home is one that meets all of the following four criteria:

a) meets the statutory minimum standard for housing. This was the Fitness Standard up to April 2006 when it was replaced by the Housing Health and Safety Rating System (HHSRS). Homes posing a Category 1 hazard under HHSRS are considered non decent under the updated definition of decent homes.

b) provides a reasonable degree of thermal comfort (adequate heating and effective thermal insulation).

c) in a reasonable state of repair (assessed from the age and condition of a range of building components including walls, roofs, windows, doors, electrics and heating systems).

d) has reasonably modern facilities and services (assessed according to the age, size and layout/location of the kitchen, bathroom and WC and any common areas for the blocks of flats, and to noise insulation).

The detailed definition for each of these criteria is included in A Decent Home: Definition and guidance for implementation, Communities and Local Government, June 2006.

Excess Cold (HHSRS Category 1 hazard)

Households living in homes with a threat to health arising from sub-optimal indoor temperatures. The assessment is based on the most vulnerable group, who for this hazard are those aged 65 years or more (the assessment does not require a person of this age to be an occupant). The EHS does not measure the achieved temperatures in the home and therefore this hazard is based on homes with an energy rating of less than 35 based on the SAP 2001 methodology. Under the SAP 2005 methodology, the comparable threshold was recalculated to be 31.5 and the latter is used when providing statistics for the HHSRS Category 1 hazard.

Fall Hazards (HHSRS Category 1 hazard)

The HHSRS Falls model includes the following hazards:

• 19. Falls associated with Baths etc

• 20. Falling on the level etc

• 21. Falling on stairs etc

• 22. Falling between levels etc




Households in Fuel Poverty

A household is said to be in fuel poverty if it spends more than 10% of its income on fuel to maintain an adequate level of warmth (usually defined as 21 degrees for the main living area, and 18 degrees for other occupied rooms). This broad definition of fuel costs also includes modelled spending on water heating, lights, appliances and cooking.

The Fuel Poverty Ratio is defined as:

Fuel Poverty Ratio = Fuel Costs (usage * price) Full Income

If this ratio is greater than 0.1 then the household is counted as being in Fuel Poverty.

Full Income definition is the official headline figure. In addition to the basic income measure, it includes income related directly to housing (i.e. Housing benefit, Income Support for Mortgage Interest (ISMI), Mortgage Payment Protection Insurance (MPPI), Council Tax Benefit (CTB).

Fuel costs are modelled, rather than based on actual spending. They are calculated by combining the fuel requirements of the household with the corresponding fuel prices. The key goal in the modelling is to make sure that the household achieves the adequate level of warmth set out in the definition of fuel poverty while also meeting their other domestic fuel requirements.

Vulnerable households

A household in receipt of at least one of the principle means-tested or disability related benefits.

The EHS definition of a vulnerable household from April 2005 to March 2007 was a household in receipt of: income support, housing benefit, attendance allowance, disability living allowance, industrial injuries disablement benefit, war disablement pension, pension credit, child tax credit or working credit. For child tax credit and working tax credit, the household is only considered vulnerable if it has a relevant income of less than £15,050.

For the 2009 Housing Stock Models the definition also includes households in receipt of council tax benefit and income based job seekers allowance.




‘SimpleSAP’ Rating, an estimate of SAP8 SAP (Standard Assessment Procedure) is the UK Government’s standard methodology for home energy cost ratings. SAP ratings allow comparisons of energy efficiency to be made, and can show the likely improvements to a dwelling in terms of energy use. The Building Regulations require a SAP assessment to be carried out for all new dwellings and conversions. Local authorities, housing associations, and other landlords also use SAP ratings to estimate the energy efficiency of existing housing. The version on which the Average SAP Rating model is based is SAP 2005.

The SAP ratings give a measure of the annual unit energy cost of space and water heating for the dwelling under a standard regime, assuming specific heating patterns and room temperatures. The fuel prices used are averaged over the previous three years across all regions in the UK. The SAP takes into account a range of factors that contribute to energy efficiency, which include:

• thermal insulation of the building fabric;

• the shape and exposed surfaces of the dwelling;

• efficiency and control of the heating system;

• the fuel used for space and water heating;

• ventilation and solar gain characteristics of the dwelling.

SAP is not affected by the individual characteristics of the household occupying the dwelling or by the geographical location.


dwelling level housing stock modelling and database ... · the ‘simpleco2’ model 29 appendix b...

Documents