Researchers

Prof Ray Chambers, Nikos Tzavidis, University of Southampton.
Dr Marie Cruddas,
Office for National Statistics.
1 May 2003 - 30 April 2006

Context

Multilevel hierarchical modelling is a key tool for small area estimation. However, this methodology is model dependent and makes strong Guassian assumptions in terms of random area effects. The project will develop methodology based on multiquantile modelling that can be used in small area estimation.

Aims and Objectives

• To investigate the use of multiquantile models as an alternative to multilevel models for small area estimation;
• To develop appropriate diagnostics for random effects in multilevel models;
• To apply multiquantile methodology to estimation of target variables for small areas;
• To disseminate results via articles and conferences.

Methodological Aspects

We will be developing and applying multiquantile regression models to survey data, focussing on:

• estimating average levels of income and employment in small areas
• building models for the distribution, rather than average values, of the survey data within and across small areas

Research Design

• We will use a multiquantile regression model for a target variable in small areas to identify a unique quantile score for every individual in the sample;
• The relationship between this quantile score and small area indicators will be used to assess whether there are small area effects;
• This relationship will be used to shed light on how contextual information and group structure in the data explain between-area variation in the target variable;
• The fitted quantile model will be used to define an area specific model for small area estimation.

Outputs

In addition to outputs in academic journals, there will be presentations to the official statistics community at conferences in 2004 (Toronto) and in 2005 (Sydney). There will also be two short courses based on research from the project.