Queen Mary, University of London
School of Mathematical Sciences
EPSRC Project PhD Studentship
Designs for Subjects Observed in Groups
Applications are invited for a PhD studentship funded by the Engineering
and Physical Sciences Research Council, as part of a high-profile research
grant on Unifying Approaches to Design of Experiments. Applicants must be
EU residents and should have a good first degree in a subject containing a
substantial amount of statistics and/or an MSc in Statistics. Some
knowledge of the design of experiments would be an advantage, but
appropriate training will be given as necessary.
The project is on the design and analysis of experiments when observations
are made on groups of subjects. The student will work mainly under the
supervision of Professor R.A. Bailey (http://www.maths.qmul.ac.uk/~rab),
but will also work closely with other members of the project team,
Professor Steven Gilmour, Dr Barbara Bogacka and three postdoctoral
research assistants.
The Statistics Group at Queen Mary is known internationally for its
research in the design of experiments and has extensive collaborations
with other departments in the College and with industry. The Statistics
Group has expanded over recent years and we now have a thriving group of
PhD students in statistics (currently 5 full-time). We are in a large and
diverse School of Mathematical Sciences, with over 50 academic staff,
about 30 research staff and over 40 research students. We also have a
number of joint projects with statisticians in other parts of the College.
A description of the project is given below. The studentship covers
tuition fees for 3 years and, for UK residents only, living expenses of
approximately £12,500 per annum (2004-05 rate). Application forms can be
obtained online at http://www.qmul.ac.uk/postgraduate/apply/index.html
and informal enquiries can be made to Steven Gilmour (contact details
below).
Project Description
-------------------
Members of the Department of Computer Science have been involved in
several projects to see how people interact with each other. Communication
tools, such as chat tools, whiteboards and conferencing systems, provide
an unprecedented degree of experimental control over the mechanisms of
interaction available to users, and over the history and topology of
interactions across communities as a whole. This provides a powerful way
of testing hypotheses about the basic organisation of human interaction
and tracking the evolution of communicative practices and 'dialects'.
For example, we can create artificial sub-communities who play a
collaborative game together and then compare the kinds of communicative
convention they develop. This design has already been used to demonstrate
the emergence of group specific sub-languages. We can also manipulate the
kinds of participation that are possible (addressee, overhearer,
bystander) and the kinds of interaction mechanism (synchronous vs.
asynchronous, edit own vs. edit partner's and own contributions). We can
also study trust and deception: student projects have investigated a game
which involves the use of deception in groups of three people who can
communicate privately in pairs.
In all of these experimental situations, a group of people has to work
together, and the group usually has a strong influence on individual
behaviour. Typically the experimental unit (in the sense of what an
experimental condition can be applied to) is a group of people for a
certain time period. In the experiments done so far, each group of people
has remained together throughout. Sometimes the response is measured on
the whole group for that time period (eg length of time to complete the
game); other times there is a response on each person in that time period
(eg number of lies told). In both cases the analysis should use the data
at the group level. It is known that statistics become problematic when
the pair, or group, is the unit of analysis, but this is often not
realised when computer scientists report their results.
Data from a pilot study showed that there was considerably more
variability between groups than there was between people within groups.
This suggests that there might be some advantage in designing the
experiment so that people do not stay in the same groups throughout. This
is a novel experimental structure so there are statistical issues to
explore in both design and analysis, and these should be resolved before
we can consider whether such a new design would be suitable in practice.
Steven Gilmour
--
Professor Steven G Gilmour
School of Mathematical Sciences
Queen Mary, University of London
Mile End Road
London E1 4NS
United Kingdom
Tel: +44 (0)20 7882 7833
Fax: +44 (0)20 8981 9587 (department fax, not private)
Web page: http://www.maths.qmul.ac.uk/~sgg
|