OFFICIAL DISSEMINATION AND FUEL POVERTY STATISTICS Presentation to

OFFICIAL DISSEMINATION AND FUEL POVERTY STATISTICS Presentation to the Oslo Group Andrew Ray Department for Business, Energy and Industrial Strategy (UK central government ministry) 10 May 2017

Dissemination in the UK: Recent Progress • Shorter publications, better infographics • Starting to move away from spreadsheet tables: developing an energy statistics API, data extraction and tabulator tools • A user consultation event on electricity statistics to find out more about what people would like Focus of talk today is on fuel poverty, to show what we publish is shaped by policy and delivery. 2

What is Fuel Poverty? Households In England we consider a household to be fuel poor if: in fuel poverty face both lower incomes and higher energy needs compared to • they have required fuel costs that are above the national typical households. average • were they to spend that amount, they would be left with a residual income below the official poverty line Annual Fuel Poverty Statistics for England 2014 published 30 th June 2016 Based upon English Housing Survey data 3

The International Context No internal consensus on definitions of energy poverty. Examples of other ‘objective measures’ from the EU: • Households spending more than 10% of their income on energy – Ireland, the rest of Britain. An unofficial definition in France. • Where average monthly expenditure on energy represents a ‘substantial’ share of average monthly income of the household Slovakia. Subjective measures have also been used Sources: EU Energy Poverty Handbook launched in 2016 and ‘Manual for Statistics on Energy Consumption in Households’ (Eurostat, 2013 Chapter 7) 4

A relative measure… 14. 7% 10. 6% 35. 3% 39. 4% Always half below the median but income changes can alter the % in the FP quadrant. Depth of FP would reduce if fewer with very high fuel costs. 5

Fuel poverty in England 2003 -2014 6

But we target those with highest energy costs Aim to eradicate these. FPEER F/G = most nonenergy efficient properties after factoring in additional government support Majority of FP HH now classified as band D rather than Band E 7

So this year’s publication needs to focus more on the target than on the overall number in fuel poverty… FUEL POVERTY TARGET Raise as many fuel poor homes as reasonably practicable to a minimum energy efficiency rating of Band C by 2030; with interim targets of: • Band E by 2020 • Band D by 2025 8

To achieve targets we need to know where the fuel poor live LAs with highest levels of FP concentrated in South West, West Midlands and North. In general LAs in SE & E have lowest levels We have been publishing these figures each year down to quite small areas BUT we are reviewing this methodology.

However, we only have estimates based on a survey The English Housing Survey is a continuous national survey of people's housing circumstances and the condition and energy efficiency of housing in England. It is the only survey in England to collect information on both the occupants and physical characteristics of dwellings. But it only covers about 12, 000 out of 22. 5 million households English Housing Survey (EHS) 10

The new published figures won’t be detailed enough If we change to publishing more robust figures for Local Authorities with confidence intervals we will please statisticians but not policy makes. Is there a way to target fuel poverty interventions in the absence of detailed figures? Two approaches: • Data science methods to try to classify the fuel poor using existing available data • Legislation to enable additional datasets to be used. English Housing Survey (EHS) 11

Data science pilot • Take the ‘known’ fuel poverty sample in the survey data • Link in a wide range of additional data available nationally on household characteristics • Run machine learning algorithms to try to classify fuel poverty – used TPOT • In theory can then use same approach model to classify all households http: //rhiever. github. io/tpot/ - Performs better than simple regression models for fuel poverty - But still not as helpful as more relevant real data. . . 12

Can we obtain more data? Data Within government the data needed to do accurate LIHC targeting exists at address-level, but it is sensitive. Low Income (equivalised, after bills) High Costs • • Property age Property size Property type Central heating Fuel type Boiler info Etc… Tax Credits ESA JSA Universal Credit Pension credits Housing benefit Etc… 13 Household income Household size Housing costs • • Valuation Office NEED Ordnance Survey EPC Register

Data sharing legislation • New primary legislation required before tax credit and VOA data can be used. • Proposals were consulted on last year and legislation has just gone through • Aiming for new targeting model in 2018 • Will be able to provide more accurate maps and data to energy companies 14

Summary • Fuel Poverty annual publication has been improved and made more relevant to the targtes we actually use • But this has raised questions about the local level statistics – are they robust? • In the absence of published statistics that can be used for targeting we will be able to use other means (data linking) to pursue delivery of fuel poverty policies. • However legislation doesn’t necessarily allow us to publish these more detailed numbers 15