* Apologies for any cross-posting *

There are still places available for this meeting, but a registration fee required.



NEW PERSPECTIVES IN MULTILEVEL MODELLING 

Thursday 16th October, 2.30pm - 5.00pm, Royal Statistical Society, 12 Errol Street, London, EC1Y 8LX

(Event jointly presented by the Social Statistics and General Applications sections of the Royal Statistical Society)

Chair - Nick Allum (Department of Sociology, University of Essex) Discussant - Paul Clarke (ISER, University of Essex)

Multilevel modelling has become a mainstream statistical method in quantitative social science. It is used for analysing structured data in which observations on individuals are nested within one or more higher level units, such as schools or geographical areas, and these structures leads to complex patterns of correlation between the observations. With multilevel models now in widespread use, it is always worthwhile to go back to basics and reconsider some of these models' basic tenets, in order to assess how well they are suited to the range of contexts in which they are commonly used. At this meeting, three papers will presented in which current practice is reconsidered and new solutions suggested.

The event will be of interest to academics, graduate students and quantitative researchers in any applied field of research.  

Registration with payment is required - you can book here: http://www.statslife.org.uk/events/events-calendar/eventdetail/279/-/new-perspectives-in-multilevel-modelling 

Please contact Nick Allum ([log in to unmask]) for more information about the programme and [log in to unmask] for venue and ticket enquiries.

*Presentations*

Explaining Fixed Effects: Random Effects modelling of Time-Series Cross-Sectional and Panel Data

Andrew Bell and Kelvyn Jones  (University of Bristol)

This article challenges Fixed Effects (FE) modelling as the 'default' option for time-series-cross-sectional and panel data. Understanding the conceptual differences between within- and between-effects is crucial when choosing modelling strategies. The downside of Random Effects (RE) modelling - correlated lower-level covariates and higher-level residuals - is omitted-variable bias, solvable with Mundlak's (1978) formulation. Consequently, RE can provide everything FE promises and more, and this is confirmed by Monte-Carlo simulations, which additionally show problems with another alternative, Plümper and Troeger's Fixed Effects Vector Decomposition method, when data are unbalanced. As well as being able to model time-invariant variables, RE is readily extendable, with random coefficients, cross-level interactions, and complex variance functions. An empirical example, considering the relationship between democratisation and globalisation, shows that disregarding these extensions can produce impoverished and misleading results. We argue not simply for technical solutions to endogeneity, but for the substantive importance of context and heterogeneity, modelled using RE. The implications extend beyond political science, to all datasets with a multilevel structure.

Regression analysis of country effects using multilevel data: a cautionary tale

Mark L. Bryan (University of Essex)
Stephen P. Jenkins (London School of Economics and Political Science)

Cross-national differences in outcomes are often analysed using regression analysis of multilevel country datasets, examples of which include the ECHP, ESS, EU-SILC, EVS, ISSP, and SHARE. We review the regression methods applicable to this data structure, pointing out problems with the assessment of country-level factors that appear not to be widely appreciated, and illustrate our arguments using Monte-Carlo simulations and analysis of women's employment probabilities and work hours using EU SILC data. With large sample sizes of individuals within each country but a small number of countries, analysts can reliably estimate individual-level effects within each country but estimates of parameters summarising country effects are likely to be unreliable. Multilevel (hierarchical) modelling methods are commonly used in this context but they are no panacea.

Bayesian methods to model spatial and spatio-temporal data

Marta Blangiardo (Imperial College)

In the last few decades the availability of spatial and spatio-temporal data has increased substantially, mainly due to the advances in computational tools which allow to collect real-time data coming from GPS, satellites, etc. This means that nowadays in a wide range of fields, from social science to epidemiology, from ecology to climatology researchers have to deal with geo-referenced data, i.e. including information about space (and possibly also time). The Bayesian approach is particularly effective at modelling large datasets including spatial and temporal information due to its flexibility and ease with which it can formally include correlation and hierarchical structures in the data. In this talk I am going to give an overview of the Bayesian modelling approach commonly used to model spatial data. I am going to present the different types of spatial data and focus on area level data, modelling the spatial dependency through a neighbourhood structure. The conditional autoregressive structure is described and applications on disease mappings and ecological regressions are included. Then I am going to present how these methods can be extended to deal with spatio-temporal applications, using random walks to model the temporal dynamics and discussing several alternatives for space-time interactions.

******************************************************
Please note that if you press the 'Reply' button your
message will go only to the sender of this message.
If you want to reply to the whole list, use your mailer's
'Reply-to-All' button to send your message automatically
to [log in to unmask]
Disclaimer: The messages sent to this list are the views of the sender and cannot be assumed to be representative of the range of views held by subscribers to the Radical Statistics Group. To find out more about Radical Statistics and its aims and activities and read current and past issues of our newsletter you are invited to visit our web site www.radstats.org.uk.
*******************************************************