Lisa A. Best

Date of Award


Level of Access Assigned by Author

Open-Access Dissertation

Degree Name

Doctor of Philosophy (PhD)




D. Alan Stubbs

Second Committee Member

Laurence D. Smith

Third Committee Member

Roger B. Frey


This dissertation investigated several factors involved in the perception of nonlinear relationships in time series graphs. To model real-world data sets, the graphed data included different sample sizes and levels of variability, and represented different underlying trends. Graph format was also varied. The purpose of the experiments was to determine how these factors affect both trend discrimination and extrapolation accuracy, with the overall goal of determining what types of graphs are optimal in different situations. In Experiment 1, subjects viewed time series graphs on a computer screen and had to identify the type of trend that was present. Six trends (exponential increasing, asymptotic increasing, linear increasing, exponential decreasing, asymptotic decreasing, and linear decreasing) were presented on four graph types (histogram, line graph, scatter plot, and suspended bar graph). The same stimuli were presented in Experiment 2 and subjects extrapolated future data points by adjusting the position of points on the screen. In Experiment 3, subjects were given feedback on their extrapolations in order to determine if this information would improve their forecasts. Experiment 4 examined discrimination and extrapolation accuracy with dynamic displays that included motion. In all experiments, accuracy was higher when variability was lower and sample size was higher. On discrimination tasks, choice accuracy was higher for nonlinear trends than for linear trends. On extrapolation tasks, in contrast, accuracy was lower when exponential trends were presented, due largely to subjects overestimating the rate of change. In regard to graph format, discrimination accuracy was highest when line graphs were used, but extrapolation accuracy was lowest with line graphs. Thus, the optimal graph format depends on the graphical perception task. Line graphs are optimal for discrimination but other graphical formats lead to higher extrapolation accuracy. Neither feedback not dynamic displays improved accuracy.