Data generation and analysis using spreadsheets

Abstract

The pages ofBehaviorResearch Methods, Instruments, & Computers contain many references to and descriptions of instructional programs that generate results (quantita­ tive data) resembling data that might be collected under a variety of experimental manipulations (e.g., Eamon, 1980; Fischler, 1980; Keenan & Keller, 1980; Levy & Durnin, 1981; Stout, 1974). Virtually all of these pro­ grams generate random data from a distribution (usually normal) with a mean and variance determined by some theoretical model. These programs were among the earli­ est applications of computers in teaching experimental and cognitive psychology, and they have proven to be valuable instructional devices over the years (see Castellan, 1983). Instructors who wish to use such generated data primar­ ily to demonstrate a statistical procedure or to show the effects of the manipulation of various statistical parame­ ters, however, will encounter difficulties with most of these programs. There are a variety of reasons: the level of measurement may be inappropriate, the models used may be especially complex or contain hidden variables (e.g., some of the EXPERSIM models), it may be diffi­ cult to create the appropriate design within the model, finding a statistically significant difference may require too many data points for a classroom computational dem­ onstration, the programs may take a long time to run, and so on. Furthermore, although most ofthese programs al­ low the instructor to manipulate the means and variances ofthe data generator so that they will generate usable data, doing so frequently involves a complicated series of steps­ often, the creation of an entirely new model. Even when a new model can be created on the fly, other problems remain. A particularly important consideration in creating problems for use as classroom examples in teaching sta­ tistics is that the instructor may want to simplify compu