Trend estimation from graphic displays

The effect of variance of data, horizontal scale size, and the quantity of data points plotted, on trend estimation accuracy was examined using two graph sizes as pictorial displays. A factorial design was employed using two groups of eighteen subjects. The dependent variables were the slope, intercept, and the sum of squared deviations about the least squared error regression line as measured in millimeters difference between the estimated and true lines. The results indicated that no significant difference in estimation accuracy occurred due to graph size except by one measure, the absolute intercept. The implications of this finding are discussed. Estimation accuracy increased significantly as the quantity of data plotted increased. Estimation accuracy decreased significantly as the variance of the plotted data increased. A quantity by variance interaction was found with the absolute intercept measure. The intermediate quantity condition (12 data points) was most sensitive to the detrimental influence of increased variance. Intercepts were underestimated, and slopes were overestimated significantly for both graph sizes. Methodological problems and suggestions for future research are discussed.