| Abstract: | Abstract.— Subjects were instructed to extrapolate curvilinear functions with time (trial) as the dependent variable. Two power functions were used in the experiment and they were each presented either numerically or graphically as plots. During each trial (representing one year) the subjects were asked to predict the year when a certain value of the dependent variable or event would be reached, plus a 90% confidence interval around this prediction. The results showed that the more curvilinear of the two functions was the more difficult to use for predictions of a future event but that there were no significant differences between predictions involving numerical and graphical presentations. Initially, the subjects gave too wide confidence intervals around their predictions, but as they received more information the estimated confidence intervals became too narrow in relation to their performance. |
