Download Recognition Memory: Types, Measuring, and Signal Detection Theory and more Papers Psychology in PDF only on Docsity!
Recognition memory^ PSY 373, Human Memory
Housekeeping March 24, 2009
-^ Next experiment Implicit Learning (completeTuesday) •^ Exams are graded, let’s go over it.
What is recognition memory? • In recall tasks, you have to generate the s • In recognition, you are given the stimulushave to assess your memory for it. • As a control, you’re asked for your memo lures, items that were not presented. • Memory is assessed by the ability to Measuring recognition memory • Types of recognition test. • Hit rate, false alarm rate • Signal detection theory • ROC curves and zROC curves.
distinguish
old from new.
Where are we in the memory systems
taxonomy?
Relationship between recognition •^ Episodic memory •^ Semantic memory?
recall
-^ Fig 9.3 •^ Fig 9.4 •^ Table 9.
Types of recognition tests
-^ Two alternative forced choice (2-AFC)
vs
recognition. • Confidence levels in yes/no recognition. • Item recognition
vs^ associative recognition
2-AFC recognition
vs^ yes/no
recognition
Try to remember the following words...
summer blessing
flavor^ canoe
rattle^ duty
motion garment
building
Yes
carbon
Yes
linen
Yes
tennis
Yes
rattle
Things about yes/no recognition • Each test probe gets a yes or a no response. • Some probes are lures. • You can manipulate proportion of lures. Yes
image Now let’s do 2-AFC recognitio
motion
which was on the list?
duty
senate
which was on the list?
summer
tunnel
which was on the list?
autumn
powder
which was on the list?
flavor
diet
which was on the list?
leader
blessing
which was on the list?
Things to notice about 2-AFC
recognition
-^ The number of lures is always the samenumber of old items. •^ Hard to tell if a correct response is a consequenceof good memory for an item or an easy-to-rejectlure:
oven
Promethe
which was on the list?
An obvious, but very important p
about recognition memory
The relationship of the lure to the old itemtremendous effect on our ability to tell them
Assessing recognition memory
-^ Is^ P
(hit)^
sufficient?
Measures of recognition performance • Probability correct:(hits + correct rejections)/(number of prob • Measures based on signal detection theobias. • What if you say yes to everything? • Need a measure that takes hits and falseinto account.
SDT citation
Figs taken fromhttp://white.stanford.edu/ heeger/sdt/sdt.html
Signal detection theory
-^ Let’s say we have to detect a signal in the pof noise. •^ You have only a strength to go on. •^ You set some criterion to guess whether thewas present or not.
A goofy analogy
-^ Let’s take a group of people who vary in height. •^ Half of the people, chosen at random, getstilts. •^ Try to guess whether a person’s wearing stiltsnot, just given their height.
SDT, underlying assumptions
-^ Usually assumed that noise is Gaussian. •^ Old item distribution simply shifted.
Criterion
Criterion, hit rate and false alarm Given disributions and a criterion, you can cahit and fa rates •^ To guess whether an item’s old or not, ysome criterion. •^ If the item’s “strength,” or “internal respogreater than the criterion, you say yes, otyou say no.
Bias terms
-^ Where the criterion is placed... called
bias
-^ A^ conservative
bias means you avoid saying
-^ A^ liberal
Bias affects hit and fa rate^ bias means you avoid saying no.
Discriminability
-^ How far apart the distributions are, in unitsthe standard deviation, called
′ d.
′^ • dis a measure of
discriminability
Discriminability indpendent of bias
Estimating
′^ dfrom data
Given a hit rate
H^ and a false alarm rate
F
can estimate
′^ das ′^ d= z(H)
−^ z(F A
where
z(x)^
is the
z-transform.
z-transform refresher
Given
a^ normal
distribution,
how
many standard deviations from themean do you have to go to makethe area under the curve
x?^
(It
sometimes helps if you rememberthat .68 of the curve is between μ^ −^ σ
and^ μ
+^ σ.)
x (^0) .1590.5.841 1
d’^ and yes/no recognition
′^ • dcalculated in this way is probably thcommonly
used
measure
of^ discriminab
Reciever operating characteris yes/no recognition.′ • dshould be insensitive to bias. • (Assumes that old noise and new noisesame)
curves
-^ ROC curves plot
P^ (hit
)^ as a function of
P
-^ Allow a way to assess memory at different c •^ No^
discriminability
means
no^
memory
P^ (hit
) =^ P
(fa)
-^ If SDT applies, you should see a specialcurve.
Two processes in recognition memo^ Recollection:^ •^ Controlled.^ •^ Semantic (“deep”).^ •^ Remember.^ •^ Hippocampal.
Familiarity: •^ Automatic. •^ Perceptual •^ Know. •^ Cortical.
Techniques for studying two-pro
recognition
-^ Remember/know. •^ Associative recognition. •^ Process dissociation procedure. •^ Source memory •^ ROC analysis.
Remember/know
Lets try some Remember/Kno • How do you know if people used episosemantic memory? • How about we just ask them? • Remember responses are dissociable fromones.
Results from last semester
Response Type
Probe type
Remember
Know
Not on
Synonym
0.^
0.^
Rhyme
0.^
0.^
Lure^
0.^
0.^
Were “remember” responsesqualitatively different from “kno
responses?
Is there a qualitative difference oR and K just two separate thresh^ Issues that affect remember/kn •^ Depth of processing, Stimulus type (pix vsdivided attention, benzodiazepenes •^ Problem with R/K—how do you knowjust a continuum of strength?
Associative recognition and tw
process theory
What is associative recognition? • Notice that all of the items should be familia • Most^
people
would
assert
that
asso
recognition relies mostly on recollection.
What is source memory?
-^ Asked to answer questions beyond yes/no. •^ Details
of^ experience
supposed
to^ depend
recollection.
Conjoint item and source judgments^ So, are there two processes or not?
Words 0.0 0.2 0.4^ 0.6^ 0.^
1.0 0.8 0.6 hr0.4 0.2 0.0^ 0.0^ 0.2^ 0. 1.01.01.01.01.01.01.01.0 0.80.80.80.80.80.80.80.8 0.60.60.60.60.60.60.60.6 0.40.40.40.40.40.40.40.4 0.20.20.20.20.20.20.20.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ 0.0^ 0.2^ 0.4^ 0.6^ 0.8^ 1.0^ Items: Words
Pictures 0.0 0.2 0.4^ 0.6^ 0.8^ far
1.0 0.8 0.6 hr0.4 0.2 0.0^ 0.0^ 0.2^ 0. 0.6^ 0.^
1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ 0.0^ 0.2^ 0.^
0.6^ 0.8^ 1. 1.0 0.8 0.6 0.4 0.2 0.0^ Items: Pictures 1.0++++^ +++ hr0.5^ + 0.0^ 0.0^ 0.5^ 1.0^ far
1.0+++ +++^ + hr0.5^ +^ IM data^ IM fit+ 0.0^ 0.0^ 0.5^ 1.0^ far IM9 data^ IM8 data IM9 fit^ IM8 fit B
IM7 data^ IM6 data^ IM5 data IM7 fit^ IM6 fit
IM4 data^ IM3 data IM5 fit IM4 fit^ IM3 fit
IM2 data^ IM1 data^ IM2 fit^ IM1 fit
A
far^ IM data^ IM fit+
Conjoint item and source judgments
-2^ -1^0 1 2 z(far) (^210) z(hr)-1 -
Responses 1-9 Responses 1-8 Responses 1-7 Responses 1-6 Responses 1-5 Responses 1-4 Responses 1- (^210) z(hr)^ Responses 1-9^ Responses 1-8^ Responses 1-7-1^ Responses 1-6^ Responses 1-5^ Responses 1-4 -2^ Responses 1-3^ -2^ -1^0 1 2 z(far)
Conjoint item and source judgments
0.0^ 0.5^ Item recognition z-ROC intercept (± CI95%)
1.0^ 1. 1.5^ Words1.0 0.5 0.0 Source recognition d-prime (± CI95%)
Item ratings 1-9 Item ratings 1-
0.0^ 0.5^ Item recognition z-ROC intercept (± CI95%) Pictures^ Item ratings 1-9 Item ratings 1-8Item ratings 1-7 1.0^ 1.
Conclusions
-^ ROC curves (sometimes) are well-descriSDT if you assume unequal variance •^ This
could
come
from
different
l
increments • Two-process account of recognition: Recoand familiarity
Assignment
-^ Next experiment Implicit Learning •^ Read chapter 9
