Fitting Statistical Models to Complex Survey Data
One-day Workshop organized by
Southampton Statistical Sciences Research Institute – S3RI
Location: Royal Statistical Society, London – 10 MARCH, 2011
Workshop coordinators:
Prof. Danny Pfeffermann and Prof. Patrick Sturgis, University of Southampton.
Sponsored by the Economic & Social Research Council (ESRC)
Motivation
Survey data are frequently analysed for making inference on statistical models assumed to hold for the corresponding population. However, complex survey data are typically hard to handle because of three main reasons: 1- the data are often clustered due to the use of multi-stage stratified cluster samples or longitudinal survey designs, implying that observations within the same cluster (repeated observations for the same unit) are correlated; 2- sample units are often selected with unequal probabilities; when these probabilities are related to the outcome variable, the sampling becomes informative and the model holding for the sample is then different from the model holding in the population; 3- survey data are almost inevitably subject to nonresponse, often of considerable magnitude, which again may affect the model holding for the responding units if the response propensity is correlated with the outcome (not missing at random nonresponse).
Several approaches have been proposed for modelling and analysing complex survey data. These approaches differ in the conditions underlying their use, data requirements, the inference objectives that they accommodate, statistical efficiency, computational demands and the skills required for their implementation. This heterogeneity means that no single approach can be considered as best, or even operational, in all situations.
Overview
The workshop aims to bring together sample survey experts and quantitative social scientists to discuss their data sets and research problems, and how they are currently analysed. Participants are encouraged to contribute to the discussion by describing an interesting data set and the analysis that they have performed, and commenting on problems they have encountered. In addition, there will be three sessions discussing key features of some of the available approaches for modelling complex survey data.
The workshop is designed to foster interaction between practitioners and statistical experts in the analysis of complex survey data, by which participants can get advice on the suitability of different approaches for their intended analysis, and the statistical experts can get feedback on what is the ’current practice’ regarding the analysis of complex survey data within the UK quantitative social science community, including information on issues and barriers for successful analysis of such data.
Provisional Programme
10:00 – 10:30 Arrival and registration
10:30 – 10:45 Welcome and Introduction
10:45 – 12:30 Presentations and discussion of design problems affecting modelling
of complex survey data:
Dr Nick Allum (Essex), Dr Franz Buscha (Westminster), Dr James Carpenter (LSHTM) and Dr Man Yee Kan (Oxford).
12:30 – 13:30 Lunch
13:30 – 15:00 Methods for fitting models to complex survey data:
· The sample model approach for handling informative sampling and nonignorable
nonresponse – Danny Pfeffermann (Southampton)
· Weighting and pseudo- maximum likelihood estimation for fitting models to
complex survey data – Chris Skinner (Southampton)
15:00 – 15:30 Coffee break
15:30 – 16:15 Methods for fitting models to complex survey data (Cont.):
· Using design and response related variables as covariates when fitting models to
complex survey data – Moshe Feder (Southampton)
16:15 – 17:30 General discussion, Questions and Answers
Registration
· Registration is free but limited to 40 participants and will close on 25 February, 2011.
· If you like to register, send an email to the workshop secretariat with name and affiliation.
E-mail: T.Lipscomb@soton.ac.uk Phone: 44 23 80594548.
· Please state “Fitting Models Workshop” in the subject of your message.
For further information on registration for the workshop, please contact the workshop administrator. A workshop website will be established shortly.