Signal Detection Theory Flashcards
what does signal detection theory allow us to predict
how the observer’s criterion, motivation, boredom, etc affect the detection of a stimulus
what is the assumption of the signal detection theory
that there is inherent noise in the visual system
-noise is the randomly fluctuating neuronal activity in the absence of a stimulus (background noise)
signal detection theory assumes that the stimulus a subject is trying to detect is….
always being detected in the presence of noise
-noise is not a problem most of the time but conditions near threshold (ex. weak stimulus) can severely affect the detectability of a stimulus
what is detectability (d’)
the difference btwn the mean values of noise and noise+signal divided by the common std deviation of the 2 distributions
-if d’ is large there is no difficulty detecting a stimulus
by setting a low criterion, your hit rate is ____, miss rate is ______, false alarm rate is _____, and correct rejections _______
hit rate high
miss rate low
false alarm high
correct rejection in btwn
setting a high criterion, your correct rejections is ____, miss rate is ______, false alarm rate ______, hit rate ______
correct rejection is high
high miss rate
low false alarm rate
hit rate in btwn
in the very lax, lax, strict, and very strict criteriorn: how are hit rate and false alarm rate related
very lax: hit rate is very high, false alarm rate veyr hihg
lax: hit rate is very hihg, fals alarm rate is high
strict: hit rate is high, false alarm rate is low
very strict: hit rate is very low, false alarm rate is very low
you can caluclate the hit and false alarm rate for different criterion levels and create a _____ curve
receiver operating characteristic curve (ROC)
the ROC curve is specific to what
a certain d’ value
ex. depends on the detectabilty btwn the noise and noise+signal distributions
what do we have if d’=0
how is the ROC curve
how are false alarm and hit rate related
very weak stimulus and N+S overlap completely
- gives a linear ROC curve
- false alarm rate and hit rate increase linearly, are equal for all criterion values
if the d’ is small but > 0
substatial overlap btwn distri but there is enough neural activity to separate them slightly, stimulus produces a signal just above noise level
- for strict crit: both the hit rate and false alarm rate are bow but hit rate is lsightly higher
- for lax: hit rate and false alarm rate high
if d’ is large what happens
no overlap btwn N and N+S
-hit rate increases substantially before the false alarm rate increases (low)
how do prey animals se ttheir criterion levels
very lax
-regonize a predator but have a very high false alarm rate
what is the likelihood ratio
beta=stimulus freq/1-stimulus freq x $CR-$FA/$H-$M
by showing the same payoff matrix to the subjects you know….
- you have the same bias throughout the trials
- all the subjects will try and set the same criterion levle and therefore you will have a consistence bias amont subjects
signal detection theory and ROC are used int he medical field to…
- asses the validity of a clincal diagnostic procedure
2. for rational decision making
it is important to know the sensitivty and specificity of diagnostic tests. what are these things
sensitivty: the %age of pts correctly identified having the disease you are testing for (hits)
specificity: the %age of pts who are healthy and identified as healthy (correct reject)
for example, if you have a machine that tests for glauc, what will a machine with strict criterion give for sensitivty and specificty? how about a lax criterion machine
strict: sensitivty (correctly id disease) falls but specificity incresaes (id healthy)
lax: sensitivty increaes but speciicity decreases
if the disease is mild, self healing, and won’t result in vision loss, how will the clinician’s criterion be? what about if the diease is vision threatening
mild: criterion will be strict-no big deal to miss some cases
vision threat: lax-to ensure not misdiagnosing
if the treatment for dieases is really expensive but the disease has mild consequcenes, what criterion?
if the treatment is cheap and the disease is serious, what criterion?
expensive: strict to reduce false alarms and avoid cost
cheap: lax to avoid untreated cases
