![]() ![]() An investigator can use inducements (say, by altering the total number of signals presented or providing financial incentives) to encourage an observer to say yes more or less often on different blocks of trials. If, on the other hand, the observers were very liberal about saying they detected a signal, then the proportion of hits will be high and the proportion of false alarms will also be high. Then, the probability of a hit will be low and the proportion of false alarms also will be low. Suppose an observer's attitude was very conservative about saying when a signal was presented. Accurately estimating both the proportion of affirmative responses (yes) on these no-signal trials (called false alarms), as well as the proportion of yes responses when the signal was actually presented (called hits), provides a much more useful data set than just estimating the yes responses on signal trials. To determine if, in fact, the observer was following these instructions, occasional “catch” trials were offered where no signal was presented with the hope that a correct response of “no detection” would occur.Īn early and important contribution of signal detection theory was to suggest that not a few but many trials without signals should be presented. What shall be done when the observer says he/she detects a signal when no signal had been presented? Historically, these responses were called false positives or false alarms, and the general advice to the observers was to avoid making such incorrect responses. ![]() The objective fact, a datum that we all can agree upon, is whether an observer says that he/she detects a signal or does not detect a signal. The only objective fact was the observer's response on that particular trial. The sensations produced by the stimuli were subjective they were private or covert. The responses did not indicate whether or not the signal was actually detected. Such responses were private evaluations, and there was no means of counting such actions as anything more than individual opinions. In determining how human observers detect weak signals, historically, investigators often simply asked them whether they heard, saw, or sensed a given signal. I also must disclose that I had nothing to do with the advertisements for that conference nor, for that matter, the choice of the honoree. Rather, we should be focusing on more general measurements of a signal's detectability and how those different measurements are related. I will argue that, while the equal-variance Gaussian icon is commonly used, it is inappropriate for illustrating the essential contribution of the theory. Therefore, I composed this brief homily on the topic to illustrate what I believe is the critical contribution of this theory, the true gospel if you will. 1 My complaint is that similar logos have become the main message of the theory, rather than only one particular embodiment of it. For a recent conference on signal detection theory, held at Northwestern University and sponsored by the Hugh Knowles Foundation, the cover on the program listing the participants contained a logo that has long been associated with signal detection theory, namely, two equal-variance Gaussian distributions separated by about one standard deviation (see Fig.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |