This research proposed to determine, in the context of preliminary data analysis, whether one can generate more--and more complex--"insights" (meaningful or possibly relevant relationships suggested by the data) by looking at a graphic (multiple bar chart) representation--as opposed to numeric table--of a large, multivariate quantitative dataset (twenty variables with twenty four observations each), displayed and manipulated h interactively in a personal computer-based system. If the more complex observations made possible by graphic representations can be explored in more detail--with further help from statistical and mathematical techniques-then the probability of achieving truly novel and useful solutions can be increased. The major issue involved is not how to communicate more effectively information to a large audience; it is rather what would stimulate deeper, sharper, and more expeditious analysis of a problem.

An experiment--of a "posttest only control group" design--was conducted, with eighty Subjects. Half of those Subjects were randomly assigned to a treatment group (graphic representation of a quantitative dataset) and the other half, to a control group (multivariate representation of same dataset). Individual experimental sessions took approximately two hours, with an interactive tutorial--designed to give both groups the same level of basic skills for handling the computer program--followed by sixty minutes (maximum) for problem analysis.

The null hypothesis was there would be no differences between the scores of Subjects looking at a graphic versus a numeric representation of data for each of four classes of "insight" generation: 1 "Insights" ignoring complexity levels 2 Multiple-field "insights", exclusive of single-field "insights" 3 Multiple "field-group" (such as age groups) "insights" 4 Number of different complexity levels

A methodology was developed for objective scoring of the raw data (written notes with requested observations and inferences). Observations were eliminated on the basis of repetition, incompleteness, and lack of validation from underlying dataset.

The differences between "insights" produced by the "graphic" and “numeric" groups were statistically significant. The major differences corresponded to the higher levels of "insight" complexity-—those observations relative to a large number of problem variables or to the whole dataset. The "graphic“ group produced a significantly larger number of such observations.