Sensation and Perception / Signals

Sensation and Perception Book

• Do our sensations/senses convey reality?
• most of the time,  but what about tricks or illusions?
• ex. matching lines, constantly descending tones
• So, we cant always sense exactly what's going on.
• Strictly speaking, we'll have to go with no, our perceptions do not always convey reality. 
• People often contrive physical and psychological responses to stimuli (i.e. optical illusions (or visual illusions may be more appropriate.
• But our senses serve us very well, they don't often mislead us
• there should be an important distinction between the physics and psychological of a stimuli.
• What evidence is there that perception is not just physical? 
• Perception involves cognition and psychological context (ex. of 13 or a B? And old lady or a young woman?)
• top -down perception - things from "high" in our mind that influence our context, as well as things that come from "low" in our minds (like our perceptions).

ANOVA clip:

• maybe the parthenon was built to trick the eye into thinking it was perfect. They compensated for our natural visual distortion to make it more perceptually perfect.
• Gestalt Psychology- principles that organized perceptual wholes.
• law of proximity
• law of similarity
• law of closure
• continuation
• common fate
• Helmholtz: "Contructivists theory" - we add information to what is provided in the stimulus to draw inferences with our conscious knowledge. "unconscious inferences" - way of processing sensory information.
• Bev Dottle's art is a good example (with the horses and mountains, etc). 
• Phiphenomena: 2 red dots flashing - we tend to see movement, even though there is none. This is an example of our perception tricking us.
• Method of limits
• methods of constant stimuli
• methods of adjustment
• discrimination threshold
• all of these were difficult to measure when the people might be bias or participants may be dishonest, so, they came up with measures of biases:
• forced choice methods: not so much a yes or no answer, but when or where? - This way there would not be people saying yes 50% of the time (Percentage correct against the percentages of chance)
• signal detection theory: separate participants sensitivity to the signal vs. biases of the patient. This method takes into account the role of decision making in these experiments.
• There were a series of trials were a stimuli either occurs or doesn't occur (and the task should be challenging) and then the person says yes or no.
• How might expectations effect the decision to say yes or no?
• you may try really hard to see something... maybe you'd have a bias or motivation to say yes.
• context of previous experiments may matter too, or social contexts.
• Signal detection theory can detect these biases and separate out our sensitivity to the stimulus from the bias. (4 categories, hit, miss, false alarm, correct rejection).
• There is always ambiguity in the experiments with backgrounds or other stimuli occurring at the same time, random firing neurons, etc.
• Stimuli is a signal to be tested against a background of noise.
• when hits = false alarms, then there is no ability to detect really anything.
• ex. distribution of effects inside the brain. When the distributions sit on top of each other, you aren't really detecting anything.
• d prime - measure of sensitivity; ability to detect.
• criterion - shows bias (yes more or no more). If you say yes more, your signal detection goes up, but so does your false alarms  (low criterion)
• Receiver operating characteristics: plots hits vs. false alarms. If hits = false alarms, the line will be perfectly diagonal. So, the more sensitive you are, the more the curve will be bent up to the left. 

Signal Detection Analysis

• You can't just look at hits of just look at misses, because you get 100% if you say yes all the time, but also 100% on false alarms.
• d prime - differences between distributions of hits and signal absent (sensitivity to signal). 
• A smaller d prime is harder to differentiate
• percentages correlate with the ares of each
• The more separate the two distributions of false alarms and hits are, the more curved away the ROC will be, the greater the d prime is, and vise versa.
• The degree of bend in the ROC is an indication of our sensitivity to the signal.
• ROC should be generated from multiple points from the same person.
• "noise" is background ongoing elements.
• how far apart the two distributions are depends on the actual strength of the signal and our sensitivity to it (d prime).
• You have to take into account the cognitive business that interferes with the actual sensory perception.
• This helps us get the bias out of motivation or setting or expectations.
• SDT separates what the person can truly detect and what is just biases.
• ex. a radiologist in detecting cancer
• This method casts objective light on performance.
• Its a lot like statistical tests with Type 1 (false alarms) vs. Type II errors (misses)