- Количество слайдов: 22
Methodological review of the VOA’s Private Rental Market Statistics GSS Methodology Symposium - 06/07/2016 Neville de Souza Stephanie Astley Valuation Office Agency
What will we cover? • • • Background to VOA and PRMS Overview of the methodological review Sampling biases Investigation of post-stratification weighting Conclusions and further work
What is The Valuation Office Agency? • Purpose: The VOA provides government with valuations and property advice. • Rich source of data collected for administrative purposes – but this makes it a challenge for statistics.
Background to the statistics • The National Statistician’s review of housing market statistics in 2012 called for the rationalisation of the House Price Index and highlighted a gap in information on the PRM. • PRMS have been published since 2011 and its users encompass central and local government, tenants, landlords and agents, property investors, academics. • It is the only official data on rental price levels.
Measuring the Private Rental Market • Landlords and letting agents provide VOA Rent Officers with data pertaining to the properties they let, primarily to calculate Local Housing Allowance (statutory function). • Participation is voluntary and results in a purposive sample of properties, rather than a random sample • Information captured electronically includes the type of property, number of bedrooms and rent charged. • Rent officers only include achieved rents (the price after negotiation between tenant and landlord). • Data is for both rent renewals and new lets.
Private Rental Market Statistics • The data is broken down by boroughs and by bedroom/room category and statistics are calculated on the total monthly rents (which may include some service charges such as fuel and water). • For Local Housing Allowance purposes, some service charges are not eligible and are excluded. • CPIH uses only the rent excluding services.
Median and interquartile range of monthly rents by English region • All bedroom categories between 1 April 2015 and 31 March 2016 Source: Private Rental Market Statistics (PRMS) Table 1. 7
The PRMS sample and known biases • Rents are collected from letting agents and landlords who are willing to provide data on their rents (participation is voluntary). No available sampling frame. • There is an aim to get 10% coverage across the whole country (based on Census 2011 population), but this is not constant across regions and property types within regions. • Properties in receipt of Housing Benefit (HB) are excluded (1. 4 million claimants).
Differences compared to CPIH sample • Price indices data provided to ONS for CPIH/IPHRP undergoes a complex process to ensure there is a matched sample of comparable properties month-to-month during the year. • These are weighted to give regional and national level results that represent the actual property market (DCLG data).
Why review the PRMS methodology? • User feedback suggests interest in low level geographies and a historic series. • Composition of the dataset underpinning PRMS can change from one month to the next, so results at different points in time cannot be reliably compared to infer trends. • In addition, results are not adjusted to produce statistics which are representative of the private rental property market mix in England.
Data sources on the target population • • • a. b. c. d. e. f. Census 2011 English Housing Survey (EHS) Labour Force Survey (LFS) Family Resources Survey (FRS) Council Tax (CT) DWP data on housing benefit claims.
Issues with using Census 2011 • Census 2011 is the most complete data source for the target population, but the rental market has changed considerably since 2011. • Comparison of Census property counts to the English Housing Survey show significant changes at regional level. • Census includes properties in receipt of Housing Benefit. • There is a discrepancy between Census counts for room only rents and PRMS data.
Chi-squared test to compare proportions • A chi-squared test was used to compare the frequencies observed in the 2011 PRM sample by region and type of property with the expected frequencies if the PRM sample had shown the same proportions as Census. • The null hypothesis assumes there is no difference between frequencies by property type in the Census and the PRMS; alternative hypothesis is that frequencies are different.
PRMS sample vs. Census 2011 – London Observed frequencies (PRMS) Census (expected) proportions Expected frequencies Difference squared/expected (Chisquare value) Bungalow/House: Semi-detached Total 3, 429 61, 528 ~10% 6, 313 8 8, 3 19, 708 29, 419, 518 1, 318 6
Interpreting chi-squared values Chi-square distribution with 3 degrees of freedom showing χ2 on the x-axis. Probability = 1% χ2=11. 34 χ2=1, 318
Chi-square test conclusions • The Chi-square test showed that there is less than 1% chance that we would expect to observe sample frequencies that differed this much from the hypothesized frequencies. • So we rejected the null hypothesis and concluded that the PRMS frequencies are different to the Census 2011 frequencies. • This could be corrected with the right poststratification design.
Post-stratification method •
Calculating weights •
Weights example Leicester (UA) Property type Bungalow/House: Detached Flat/Maisonette Leicester Total (inc semi and terraced) PRMS sample (cell) PRMS per cent cell to LA 65 2. 3 1, 999 6. 7 2. 959 1, 409 49. 0 11, 551 38. 6 0. 789 2, 878 100 29, 911 100 Census 2011 per cent (cell) cell to LA weight Source: VOA Lettings information database and ONS Census 2011
Weighted and unweighted means and median monthly rents by half years • Four or more bedrooms, Leicester Source: VOA Lettings information database
Conclusions and further work • There are still a number of questions that need to be answered before VOA can proceed with a new methodology for PRMS. • None of the data sources available for poststratification could be used as benchmarking the target population. • There is no measure of the target population (total PRM less HB claimants). • Modelling EHS trends could be used to update populations at regional level. 21