September 2, 2005
Source: University of Waterloo:
UW Survey Research Centre examines outcomes
WATERLOO, Ont. -- Survey says . . .
When we answer a survey or read the results of one, we assume that
the results tell us something about relationships and casual
For instance, if a survey about reactions to graphic warning labels
on cigarette packages found that after 10 years of nasty smoking
labels fewer people were still smoking, it might be an indication
that the labels work.
Or would it? Prof. Mary Thompson, co-director of the University of
Waterloo Survey Research Centre, devotes some of her research to the
extraction of meaning from surveys.
"There is a great deal of scope for using surveys to observe
processes and to see how one kind of event leads to another, but the
surveys have to be designed appropriately to examine supposed
relationships and connections," she said.
Thompson, of the Department of Statistics and Actuarial Science, is
particularly interested in longitudinal studies -- where observations
of individuals are made several times, typically over the course of
years. A longitudinal study gives the researchers snapshot pictures
of a population as time progresses. But there can be fairly
significant problems in interpreting the results.
"Every study has dropout, but if you are sampling over a long time
the dropout rate may be pronounced. Do you replenish the sample, and
if so, how is this done to make the best use of continuing
respondents and new recruits, who may be statistically quite
different? These decisions need to be made in the design of the
sampling scheme and are key to the success of a longitudinal survey."
Thompson is working on a survey, developed by UW Prof. Geoff Fong of
Psychology and other colleagues, of smokers and their responses to
anti-smoking measures such as graphic labels. She applies
mathematical and statistical theory to the design of this kind of
population health survey.
"The researchers who conceived the study are psychologists and health
scientists concerned with whether this type of action is effective.
My role is to try to ensure that their results can help to answer the
research questions," she said.
The role of the statistician is often to model error or noise so that
it is easier to see the patterns that are sometimes hidden in
mountains of data.
"One fundamental technique is to try to find patterns in
multi-dimensional data by reducing the dimensions," Thompson said.
"Sometimes that is a straightforward process, as in many engineering
applications, but with data from human respondents it can be harder
to find discernible patterns and appropriate reductions of
dimensionality. The role of theory is very important and designing
and administering a survey carefully may shed light on parts of the
structure and allow us to see that changes in one variable might have
tremendous effects on another."
As well, observing systems "without taking proper account of noise
and randomness can lead you astray and make you misinterpret
results," she said. "But, through careful mathematical modeling of
the randomness, you can look at the phenomenon and make sense of the
Thompson's research is the kind that allows science, engineering and
the social sciences to make tremendous breakthroughs.
"I don't think of what I am doing as being very 'gee-whiz.' because
by its very nature it has to be abstract.
"But the statistician's role in science through probabilistic
modelling is crucial. In a sense that role is to develop the
abstractions which support the inferences underlying those remarkable
discoveries," Thompson explained.
(Written by Graeme Stemp, SPARK -- Students Promoting Awareness of
Research Knowledge -- writing program.)
Prof. Mary Thompson, (519) 888-4567, ext. 5543; methomps at uwaterloo.ca
Jim Fox, UW Media Relations, (519) 888-4444; jfox at uwaterloo.ca
Release no. 196 -- September 2, 2005