3 The Nature of Scientific Meta-Knowledge 57
model, or set of competing models, which allows one to develop a set of theory-
based hypotheses to test. The goal is to test each of the hypotheses to see if the
findings are consistent with its theoretical predictions. This allows one to deter-
mine which models are most consistent with the data and which are not suitable for
explaining the phenomena that have been investigated.
Exploratory Inductive Investigations
Galileo is famous for developing exploratory inductive methods in science. In his
experiments on pendulums and gravity, he systematically varied the elements that
he thought might affect the period of the pendulum and the speed of a ball rolling
down an incline. From these exploratory investigations, he derived equations for the
motion of pendulums and falling bodies. The Framingham Heart Study is a modern
variation on his method using natural variation rather than controlled manipulation.
The investigators in this study collected data from many people in Framingham
Massachusetts on a large number of variables that they thought might influence the
likelihood of getting heart disease. They then followed the people over many years
to see if they developed heart disease and identified a number of variables that were
precursors to heart disease.
The kind of data collected in exploratory investigations has a strong effect on the
types of models that can be constructed from the data. Quantitative data support the
construction of constraint-equation models, as we see with Galileo, or multifactor
models, as we see in the Framingham Heart Study. To construct process models,
such as the agent model described earlier, one needs a richer data stream, such as
observational, protocol, or discourse studies provide. The goal of exploratory studies
is to identify patterns in the data and systematic relationships that allow for the
construction of models. These models can then be evaluated using confirmatory
methods.
Confirmatory Investigations
Confirmatory investigations, which are designed to test theory-based hypotheses,
can take many different forms. The best known is the randomized controlled trial, in
which one or more hypotheses are tested by comparing conditions that correspond
to each of the hypotheses being tested. Often such a test of a hypothesis contrasts
an “experimental” condition, which includes some particular feature, with a control
condition that lacks that feature. In such cases, the competing hypothesis is that
the feature will have no effect, which is known as the “null hypothesis.” In order
to ensure the generality of the findings, the participants, or objects being studied,
are assigned randomly to the different conditions. Often special efforts are made to
control any variables that might affect the results, other than those being deliberately
varied. After the experiment has been carried out, the data are analyzed to see if they
are consistent with what was predicted by any of the hypotheses.
In one example of such a randomized controlled trial, we tested hypotheses about
the impact of self-assessment on students’ learning (White & Frederiksen, 1998).