Surveys may be classified in at least two ways based on how data is collected and
when data is collected. These are not the only ways to classify survey researeh but they
are convenient ones. From a methodological standpoint survey data may be collected by
questionnaire, interview or observation. As well as method, the time when data is
collected may be used to distinguish among survey type. In a cross-sectional study the
sample would consist of different subjects representing all relevant subgroups in a
population measured at one moment in time. In a longitudinal study the sample may
consist of one group of subjects who are studied over time and each subject may be
measured more than once.
Longitudinal surveys can be subdivided into studies which look backward in time,
retrospective studies and those which look forward, prospective studies. Retrospective
studies determine the effect of exposure, retrospectively, on an outcome of interest. For
example, a researcher may set up a retrospective longitudinal survey to investigate
bullying in school. A sample of school children would be selected and then divided into
two groups on the outcome of interest. Here one group would consist of children who
report that they have been bullied in school and the other group would comprise those
children who say they have not been bullied. A potential exposure effect such as type of
school attended, public or state school, is then determined retrospectively by examining
the proportions of children in each school type who were bullied. This design is also
called a causal comparative survey design because it attempts to determine the causes
or reasons for existing differences between groups of individuals, in this case those
children bullied and those not. The inference in this example would be that school type
leads to the observed differences in bullying. Both the outcome, bullying, and the
suggested causal factor, school type, have already occurred and are studied
retrospectively. Causal comparative research is distinguished from experimental research
because variables are not under the control of the researcher and effects or outcomes are
observed and possible causes sought. In experimental research causal factors are
deliberately manipulated with the intention of observing particular outcomes. In causal
comparative research suggested causal factors, such as the type of school, could not be
manipulated because they have already occurred. Caution should be used when
interpreting causal relationships from causal comparative studies. Such studies may lead
to the setting up of ‘true’ experimental designs.
In prospective studies, subjects would be grouped according to exposure or some other
factor and individuals would be followed up and outcomes of interest observed. In the
bullying example, two groups of children, those attending public schools and those
attending state schools, would be followed up over a number of years and the proportion
of those who report being bullied would be observed at different points in time in both
groups. The National Child Development Study (NCDS) is an example of a longitudinal
prospective survey. This study is intended to follow the lives of individuals living in
Great Britain who were born between the 3rd and the 9th of March 1958. Obstetric data
was collected from the first sample of approximately 17,000 children. The first follow-up
of these children took place in 1965 and this gave the study its name, the NCDS. A
second follow-up was in 1969 when the children were aged 11, and futher observations
occurred in 1974, 1981 and 1991.
An alternative approach to describing surveys is to focus on purpose. Surveys may be
primarily descriptive or explanatory. In a descriptive survey a sample would be drawn
Statistics and research design 7