The Best of Creative Computing Volume 1 (published 1976)
Exper Sim: Experimental Simulation (teaching research design through computer simulation)
EXPER SIM: Experimental Simulation
by Dana B. Main
University of Michigan
EXPER SIM (Experimental Simulation) is a system for teaching research design
through computer simulation. It includes not only a set of computer programs and
accompanying written materials for Student and faculty use, but a classroom
pedagogy designed to emphasize the learning of research strategies in the
context of a
simulated scientific community.
EXPER SIM has been developed for the student and instructor who are naive in the
use of computers and who do not know how to program.
It is appropriate for any undergraduate course concerned with research questions
and analyzing empirical results from experiments addressing such questions.
Although the system was conceived and developed within a behavioral science
context, it also claims applicability to biological, physical and political
science.
Written Materials that Accompany the Program
The student receives a written description of a particular problem area and a
list of variables that he can manipulate in experiments. He is given the range
of possible values of the manipulable variables,further informed which may be
numeric or key words. He is informed of the possible dependent variable allowed
in the mode He is the future informed of the default value of each manipulable
variable should he choose to ignore it in any particular experiment. The
default value may be constant from one execution to the next, or it may be
randomly selected from the set of possible values.
Accompanying these materials is a student's guide for using the Michigan
Experiment Simulation Supervisor MESS). All models in the library are controlled
by a large supervisory program that handles communication between the student
and the computer. Once the student learns to use the supervisor commands he can
easily explore other models based on quite different subject matter.
Use of LESS for Smaller Computers
For smaller computers with a minimum of 8K storage the Louisville Experiment
Simulation Supervisor (LESS), written by Arthur Cromer, may be implemented
rather than the Michigan (MESS) programs. They both, in time, will contain the
same library of data-generating models. The difference in the two programs lies
in the size of the supervisor managing the libraries. MESS has more flexibility,
allows names of variables, permits abbreviations and accepts considerable
misspelling. LESS, because it is designed for smaller computers, is more
restrictive but still very easy for students to use.
Input: Experimental Designs
A student designs an experiment by specifying 1)the number of experimental
groups in his design, 2) for each group, the values of the manipulable variables
and the name(s) of the dependent variable(s), and 3) the number of subjects
(within a specified range) for each group. (It is also possible for a model to
contain the
capability of repeated measures. If so, then the number of measures on a subject
can be specified.)
Output: Experimental Results
This information is submitted to the computer and serves as commands to a
data-generating model. The student's output are values of the specified
dependent variable(s) which can be plausibly interpreted as raw data. All
data-generating models in the MESS library are probabilistic; therefore, the
very same design generates different values of the dependent variable.
Depending on his research goal the student generates hypotheses, designs
experiments, summarizes results, and explores relationships between manipulable
variables as well as functional relationships among different dependent
variables. He gets into problems of scaling and is motivated to acquire some
statistical skills in order to make inferences about the underlying model
generating the data. He is encouraged to report his findings in terms of support
or refutation of possible theories.
Operationalizing New Variables
Sometimes a student may be informed of only a subset of the manipulable
variables and of all, some, or none of their possible values. He may also be
informed of only a subset of possible dependent variables utilized by the
data-generating model. He is made aware, either initially or later, of one or
more X-variables which may affect his results. The concept of X-variables in
instructional simulations was first suggested to us by Richard Johnson in 1972
and was developed for classroom use by Cromer and Thurmond at the University of
Louisville (reported in the Proceedings of the 1972 Conference on Computers in
Undergraduate Curricula). In Cromer and Thurmond's instructional models the
student is provided with a complete set of variables, less one: the X-variable.
information gained from runs on the other variables may or may not lead the
student to infer the X-variable. If he does, the instructor provides him with a
computer command that causes the X-variable to contribute to the values of the
data in subsequent runs. If the student has not correctly inferred the
appropriate variable, but some other implausible or
