You are here:

Innovations in small area estimation methodologies

Reliable statistics are crucial for targeted policy planning. For greater effectiveness, however, users of statistics across a range of policy research areas (e.g. health, crime, employment, income) and governmental scales (national, regional, local) need detailed information that extends beyond estimates at the national level. Small Area Estimation (SAE) is concerned with the development of methodologies for performing estimation at fine spatial scales for which survey data are either non-existent or too sparse to provide direct estimates of acceptable precision.

The last decade has seen a rapid increase in the use of SAE methods. Reflecting the increasing importance of SAE for policy purposes, a variety of statistical agencies and Governmental organisations are actively developing their own suites of SAE. In the UK the Office for National Statistics (ONS) has responded to user demands by producing estimates of average household income for wards and unemployment rates for Local Authority Districts, which are now accredited as National or experimental Statistics. The Welsh Assembly Government (WAG) is actively seeking both to gain new spatial information through SAE as well as develop in-house SAE capacity. Public Health England feature small area health statistics on their local health website. The initial demand for small area statistics has been followed by user requests for small area statistics more complex than the traditional SAE reliance on means and proportions. In response to user requests, ONS is exploring the estimation of the income distribution and the incidence of income poverty for Middle Super Output Areas. ONS is also exploring suitable SAE methods for replacing the Census in the production of health indicators, population estimates by ethnicity and for adjusting census results for net undercount. In Mexico the National Council for the Evaluation of Social Development Policy (CONEVAL) is required by law to publish estimates of the incidence of income poverty, inequality and multidimensional deprivation-jointly encompassing income and social-deprivation e.g. lack of access to education, food and health- for municipalities. The National Statistics Office of Brazil (IBGE) has similar SAE targets. Despite recent methodological advances, such as the ones proposed by European Commission framework programmes in SAE including EURAREA (FP5), SAMPLE (FP7) and AMELI (FP7), the demand for complex small area statistics has created significant methodological and applied real-world challenges that remain unresolved.

This project will develop novel SAE methodologies that overcome key SAE challenges whilst simultaneously tackling real-world applied policy priorities in the partner organisations (ONS, CONEVAL, WAG and IBGE). The applications to substantive problems will include SAE of selected attributes of interest to users, including income, inequality, deprivation, health and ethnicity. These are the foci of Work Strands 1 & 2. An additional challenge is that at present the different SAE methodological approaches remain largely unconnected, locked in disciplinary silos. This project will bridge these currently fragmented SAE methods and then exploit their linkages both for the enhancement of the methods and for an unprecedented performance comparison. This is the focus of Work Strand 3. Finally, the frequently technical presentation of SAE methods and the lack of accessible learning resources are key factors that impede the widespread use of these methods by social scientists. Through our commitment to NCRM’s training and capacity building activities, this project will deliver new learning resources that will ensure both the mainstreaming of SAE across the academic and policy social science communities and a step-change in SAE methodological capacity across these user groups.

Work Strands

Strand 1 – Innovations in SAE statistical methodologies for distributions and complex indicators

1. Model specification and data transformations in SAE

2. Robust, semi-parametric and non-parametric methodologies for continuous and discrete outcomes
2.1 Semi-parametric estimation of distribution functions
2.2 Robust prediction of random effects via discrete mixtures and non-parametric estimation
2.3 Robust SAE methods for discrete outcomes

3. Robust measures of precision in SAE

Strand 2 – SAE using Indirect Survey Calibration (ISC) / Spatial Microsimulation

1. Model specification
1.1 Model (Benchmark) selection.
1.2 Donor pool selection.

2. ISC algorithms
2.1 Impact of weight range restrictions on estimate quality.
2.2 Benchmark relaxation strategies.
2.3 Integerisation.

3. Estimating precision

Strand 3 - Bridging the gaps between Strands 1 and 2

1. Spatial variability of Census covariates
2. Statistical theorisation and translation
3. Empirical performance evaluation of the methods across Strands 1 and 2

Highlights:

Small Area Estimation to be read at the Royal Statistical Society on 9th May 2018

Report published by the Welsh Government on estimation of discontinuities in national surveys based on methods

Open sources R software for estimating and mapping disaggregated indicators


People

University of Southampton

Professor Nikos Tzavidis (PI)

Professor Li-Chun Zhang

Dr Yves Berger

Professor Graham Moon

 

University of Liverpool

Dr Paul Williamson

 

University of Sheffield

Dr Adam Whitworth

 

University of Exeter

Dr Karyn Morrissey

 

University of Portsmouth

Professor Liz Twigg

 

Freie Universitat Berlin

Professor Dr Timo Schmid

 

University of Wollongong

Professor Ray Chambers

 

Australian National University

Professor Stephen Haslett

 

University Technology Sydney

Professor James Brown

 

Partner Organisations

UK Office for National Statistics

 

Welsh Assembly Government

Mexican National Council for the Evaluation of Social Development Policy (CONEVAL)

National Statistics Office of Brazil

 

Activities