For information - please pass on to any interested colleagues
RSS South Wales Local Group Half-Day Meeting
Date/Time 2.30 pm, Wednesday, October 28, 1998
(Tea and biscuits at half-time)
Venue Training Suite, Office for National Statistics,
Cardiff Road, Newport
(Just off Junction 28, M4 - get on-site directions
from the main entry gate)
State space time series modelling with an application
to benchmarking in official statistics: James Durbin
Time Series Multilevel Modelling of Longitudinal Data from Complex
Surveys: Moshe Feder (Southampton University)
Extended abstracts are given below
For further information, please contact
Alan Watkins [[log in to unmask] or 01792 295853]
Paul Smith [[log in to unmask]]
Abstracts
--------------------------
State space time series modelling with an application
to benchmarking in official statistics: James Durbin
The origins of modern time series modelling in the 1950s
will be considered. The emergence of Box-Jenkins ARIMA models and
state space time series models will be described. The utility and
generality of state space models for current applied work in time
series analysis will be compared favourably with that of ARIMA
models. Benchmarking in official statistics is the reconciliation
of monthly or quarterly data derived from sample surveys with
accurate values obtained from censuses or administrative
resources. The use of state space methods will be illustrated by
applying them to the benchmarking of Canadian retail sales data.
Time Series Multilevel Modelling of Longitudinal Data from Complex
Surveys: Moshe Feder (Southampton University)
Longitudinal survey observations define time series, usually of
short length. Combining the measurements for all the units permits the
fitting of low-order time series models, despite the short lengths
of single series. We illustrate this paradigm using data from the
Israel Labour Force Survey, which employs a rotating panel sampling
scheme of two quarters in sample, two quarters out of sample and
then two quarters in again. The model comprises separate two-level
mixed models for each time point, but allows the second level effects
(corresponding to households) and the first level residuals
(corresponding to individuals) to evolve stochastically over time.
In view of the large number of unknown parameters, direct
maximisation of the likelihood yields unstable estimators.
Instead, a two-stage procedure is adopted. At the first stage, a
separate two-level mixed model is fitted for each time point, thus
yielding estimators for the fixed effects and the second level
variances. At the second stage, the time series likelihood is
maximised only with respect to the coefficients of the time series
models. This two-stage procedure has the further advantage of
permitting appropriate first and second level weighting to
account for possible informative sampling effects.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|