Human Information Processing in Visual Sampling
Tarald O. Kvålseth · Ergonomics · 1978
Information gain, entropy and redundancy measures were used to analyse human monitoring of a Gaussian autoregressive process. Particular emphasis was placed on the values of these measures when sampling occurred, yielding the following results: (i) the subjects appeared to ‘ tolerate ’ a relatively high degree of uncertainty (large entropy) before they decided to take a sample, the entropy asymptotes ranging from 2[sdot]12 to 4[sdot]24 bits sample−1: (ii) the entropy at sampling instants increased significantly as the variance of the monitored process increased; (iii) the ratio of the cost of the process exceeding certain threshold limits to the cost of sampling had no significant effect on the sampling entropy, while the interaction of this cost ratio with the process variance had a significant effect on the absolute scatter of the sampling entropy: (iv) there was a rapid loss of information following a sample; (v) sampling redundancy ranged from 1% to 18% with an overall mean of 10%; (vi) neither the process variance nor the cost ratio had any significant effect on the mean sampling information or on the mean sampling redundancy.