Title: An Associative Memory Framework for Addiction and Prevention
1An Associative Memory Framework for Addiction and
Prevention
- Alan Stacy
- Presented to the Research Institute on
Addictions, Buffalo, NY, May 2004 - http//tprc.usc.edu/
2- Elements of Framework
- Measurement possibilities examples of findings
in drug use - Improvement within a new center
3I. Three Elements of Associative Memory Framework
- Associative memory connectionism
- Implicit processes
- Neurobiological plausibility
4Examples of Theories Modeling Associative Memory
- Connectionist approaches that model conceptual
associations - Multiple-trace (Hintzman)
- Hopfield Networks (e.g., Hopfield Tank Masson)
- Recurrent Networks (McClelland and Rumelhart
Smith DeCoster) - Others (eg., Nelson Murdock Eich, etc).
- Mixed models (Symbolic models, with some elements
of connectionism ACT J.R. Anderson)
5Connectionist Approaches Basic Idea
- Cognition is
- Very dynamic, can change continuously
- Reflected in a pattern of activation in memory
- This pattern is dependent on
- The input or sequence (e.g., situation, preceding
cognitive state) - Associations in memory
- Symbolic concepts are not necessary (e.g., norms,
attitudes, beliefs, etc). - But they can be re-defined in terms of very
NON-stable concepts - Or combine into mixed models
6Examples of Evidence for Associative Memory
Processes
- Cued-recall Strongly associated memory cues lead
to best memory - False memoryFalse memories are predicted by
strongest associates of words in a study list - Semantic Priming Strong associates are often
best primes
- Research in Amnesics Amnesics remember previous
word pairs (on implicit tests of memory) if the
words are strongly associated in memory. - Neuroimaging associated cues (pictures, images,
objects) trigger specific responses in the brain - Other paradigms Also show importance of strong
associations in memory (humans and animals)
7History of Related Ideas
Association has been a CENTRAL concept in all of
these approaches but is usually ignored in
prevention
8Two examples of specific theories of associative
memory
- Multiple trace theory (Hintzman)
- Hopfield Network (Hopfield Tank)
9Multiple Memory Traces Encode Associations Among
Emotion, Situations, and Behavior
10Memory Activation Depends on the Overlap of
Cognitive State (Input) and Trace Features
Input
Traces
The cognitive state activates memory traces
having the same features. The activated traces
may have additional features, which can influence
behavior
11Hopfield Neural Network for Drug Concepts or
Situations in Adolescence
12Why not expectancy?
- Semantic priming literature expectancy is a
postaccess decision process not an automatic
memory process - Memory literature Not used anywhere to measure
or conceptualize associative memory - All cognitive literatures It is not as general a
concept as associative memory - Other literatures Occasionally used to describe
a predictive relationship in some associative
paradigms
- If the concept helps your research or
intervention, use it! - Expectancy can be called a relationship or
association - Alcohol literature Often used synonymously with
- Belief
- Subjective probability
- An outcome (e.g., have fun)
- A network
- If-then rule
- Typically NOT linked to triggering cues (an
essential concept from Tolman,etc)
13Another Critical Element Implicit Cognition
- Many theories of associative memory do not
postulate control processes or explicit
recollection as central to memory activation - Instead, they imply an implicit process
- Some of these theories have been used formally to
model implicit or automatic processes - PIER (Nelson et al)
- Parallel Distributing Processing (PDP Smith
Mclelland Rumelhardt) - Spreading activation (Anderson)
14Implicit Memory and Cognition
- Revealed on tests that do not require or foster
- Deliberate or conscious recollections
- Introspections about the causes of ones
behavior - Affects behavior through an automatic or at least
relatively spontaneous process (e.g., memory
activation), no conscious weighing of pros and
cons. - The framework does not require ANY traditional
health behavior concept
15People engage in rational decisions regarding
substance use (weigh pros and cons)A sort of
cognitive algebra is implied, consistent with
symbolic approachesbut Addiction is almost
prototypical example of irrational behavior
- A CONTRAST
- Primary Assumption of Most Traditional Approaches
to Drug Use Prevention
16Examples of theories and research areas that rely
on rational processing assumptions
-
- Health Belief Model
- Utility Theory
- Theory of Reasoned Action its Derivatives
- Most Attitude Theories
- Most Prevention Research
17Areas underscoring importance of implicit
processes
- Associative memory
- Decision theory Kahneman (2003)
- The automatic or implicit system is classified as
System 1 the default system for making
decisions. - Social cognition
- Several approaches underscore importance of
implicit cognition - Neuroscience
- Many learning memory processes involve brain
regions not involving conscious processing
(Squire White Knowlton, etc.)
18Three memory systems (based on White, 1996)
involved in associative memory and drug use
AMYGDALA (incentive)
CAUDATE-PUTAMEN (habit)
HIPPOCAMPUS (declarative)
19A Broader Associative Memory Framework of
Alternative Hypotheses
- White suggests the direct (dotted-line) link on
top (path W). - Shultz (1998) suggests pattern matching occurs
via association cortex, also implicated in
semantic priming Gabrielli, 1998 paths A-B.) - Reward signal is broadcast to frontal cortex
(Shultz, path C)
From Stacy, Ames, Knowlton, in press cf.
Franken, in press.
20A Cartoon Showing the General Idea
21Amnesic data provide some of the best evidence
for distinct neural basis of different types of
memory Implicit and Explicit
22Examples Shimamura Squire (1984, JEP Exp. 4)
- Study Trial. Korsakoff patients and alcoholic
controls rated single words one-at-a-time for how
much they liked the word (e.g., BABY). - Test. After the study trial, participants were
asked to perform a different task-- write the
first word that comes to mind in response to a
presented word (e.g., CHILD).
23Vaidya, Gabrieli, Keane, Monti (1995)
Similar Results
Note Visual presentation
- Overall, controls had significantly better recall
(F11.1) - Controls did not differ significantly in implicit
memory (F - Implicit associative memory test detected effect
of study trial (F 6.24).
24II. Some Measurement Possibilities in Research on
Alcohol and Other Drugs
- Tests from basic cognitive research used in drug
use - Some of our tests of associative memory
25Examples of Cognitive Paradigms from Basic
Research
- Lexical decision in semantic priming
- Word naming in semantic priming
- Flicker paradigm
- Implicit association test (IAT)
- Affective priming
- Cue-effectiveness in extralist cued-recall
- False memory
- Process dissociation
- Stroop
- Word and picture association
Implicit cognition, associative memory framework
important new development in 10th Special Report
to US Congress on Alcohol and Health No test is
the gold standard
26Associative Memory Predicts Alcohol Marijuana
Use
Previous Alcohol Use
- Short-term prospective study of 340 college
students (R2.82) - Predictive effects replicate in cross-sectional
studies of marijuana use, HIV-risk, and in
diverse samples including - Adult drug offenders
- High-risk (continuation high school) adolescents
- Adult driving-under-influence offenders
- Hispanic and White ethnicities
.54
Association Responses
.33
Beliefs About Alcohol Use
.09
Later Alcohol Use
Acculturation
.12
Sensation Seeking
Gender
Stacy, 1997, J.Ab.Psych
27Experimental Evidence Revealing Implications of
Word Association (WA) for Implicit Processes
- Research in Amnesics People with impaired
conscious memory show normal memory on word
assocation tests of implicit memory word
association detects memory - such tests do not require deliberate
recollection. - Primed free-association in normal subjects
Strong associates are best cues word association
detects memory. - Extralist Cued-recall Strong associates are best
cues - False memory false memories are predicted by
strongest associates of words in a study list - Semantic Priming Strong associates revealed on
association tests are often best primes
28Example of outcome-behavior association measure
29Some Pros Cons
- PROS
- Unique in measuring relative access compared to
all competitors - A consistent predictor of memory responses in
diverse cognitive paradigms - Similar technique shows predictive utility in
advertising research (top-of-mind awareness) - No demonstrably better measure of association in
memory - Can measure memory even in amnesics
- CONS
- Cannot rule out filtering
- Must code responses
- 1st response may not be enough
- Does not sound as rigorous as RT tests
- Bound to be noise cannot be more than a rough
measure of association
30Example of other, similar association measures
- Continued association (Szalay et al)
- Multiple responses are obtained (very useful)
- Top-of-mind awareness (e.g., 1st brand comes to
mind) - Media research (predicts purchasing)
- Other associations, e.g., between alcohol and
- Program material or skills
- Images/nonverbal memory
31Example in advertising effects nonverbal memory
- Participants are asked to
- Think of a television commercial, the very first
that pops to mind. - Form an image of that commercial
- Sketch their image
- Average Kappa for coder agreement .80 (p .001 Stacy, Pearce et al., 2004, Psych
Marketing) - Why sketch code? Verify that it is a real ad, a
real memory (proven recall in ad literature)
32Example of a students nonverbal response in
open-ended item illustrates that nonverbal
memory can be assessed
33The test assesses relative memory
Example 2 from same sample test
34Other example of associative memory measure
semantic priming in word naming
Alcohol-related prime-target pairs
- She felt good after she had the
drunk
They were more friendly when they were
Neutral pairs, with alcohol targets
drunk
They said they were
beer
They said it was the
Weingardt, Stacy, Leigh (1996 college students)
35Drinking habits predicted facilitation score
(faster responses to alcohol primes)
36A less consistent finding
37Effect sizes as correlations(reversed signs)
Small to moderate correlations
38Bottom line on measures?
- There are feasible measures, with basic research
support and some evidence in drug abuse area - Other research teams also show support
- These measures can be improved further
validated - New measures need to be developed in a parallel
effort - Such measures seem capable of assessing mediators
of behavior AND program effects
39III. Attempts at Improvement Prevention
Applications in a New Center
40Transdisciplinary Drug Abuse Prevention Research
Center (USC TPRC)
- Co-investigators/researchers
- Susan Ames (multiple roles)
- Stan Azen
- Lourdes Baezconde-Garbanati
- Chih-Ping Chou (stat core leader)
- Roger Drake
- Peggy Gallaher
- Elahe Nezami
- Paula Palmer (multiple roles)
- Anamara Ritt-Olson
- Ping Sun
- Jennifer Unger (multiple roles)
- Tom Valente (project PI)
- Director
- Alan Stacy
- Co-directors
- Andy Johnson (also training core leader)
- Steve Sussman (also a project PI)
-
- Staff administrator
- James Pike
- http//tprc.usc.edu/
-
41Most General Statement in Request for
Applications (RFA)NIDA Sought To
- Expand the knowledge base of basic research with
strong translational potential for informing the
design of drug abuse prevention interventions - Accelerate the development of new prevention
interventions arising from transdisciplinary
research interactions - Test research-based interventions
42Goals Are Addressed by Primary Components of
Center
- Two main projects
- Associative memory social networks.
- Developmental (pilot) projects and supplement
project - Training core
- postdoc awards funds for pilot projects student
training - Statistical core
- experts in structural modeling, biostatistics,
psychometrics, assessment - Advisory board consultants
- Basic research and prevention in the community
43Examples of General Questions in Drug Abuse and
Prevention
- How does prevention REALLY work?
- How can prevention effects be improved and
maintained over a long period? - How do drug habits get started and why do they
often escalate? - How can we better understand the mechanisms of
habit formation and escalation? - How do social influences REALLY operate in drug
abuse, and how can they best be tapped in
prevention?
44Reach Out to Basic Researchers in Multiple Areas
Memory
Social Networks
Culture
TPRC Reaches Out Foster Collab, Innovations
Training
Traits Emotion
Neuroscience
Basic to Applied- Translation
Other Areas?
45One example 3 Areas of Basic Cognitive Research
Stacy, Ames, Sussman
Sussman, Valente, Johnson
Knowlton
De Houwer?
Tprc team
Nelson
Chou Tprc team
Sussman, Ames, Stacy
Stacy, Ames
46Examples of other efforts of center
- Assessing executive functions in high-risk
adolescents - A possible moderator of effects of implicit
cognition on behavior - Social networks
47Summary
- TRPC Center Tries to apply some of the most
promising findings from multiple areas of basic
research. - Associative Memory Framework Our work fosters
use and development of this framework - Can be seen as a general framework not owned by
any single team other specifics are viable. - Measures We want to help foster development of
new methods of assessing drug-related associative
memories. - Association measures should be taken to their
limit. - Other measures need to be refined/developed
48Questions or Discussion?
- Moderator slide for exec functions
- Moderator slide for associative effects
- Intervention slide (prevention)
- Specific findings on nonverbal, associative
memory test
49Influence of other variables?Examples of
interaction hypotheses
- Cognitions may be implicitly activated, but
individual differences in executive functioning
may influence whether these cognitions are
applied to behavior - Individual differences in neural systems of
reward may influence development of connections
in memory
50Hypoth 1. Indicators of Executive Functioning May
Moderate the Effects of Implicit Cognition on
Drug Use
51Hypoth 2 Individual differences may moderate
growth of associations involving drugs
- Mechanism?
- Possibly genetic effect on dopamine (reward)
system (e.g., Sobol et al., 1999)
Growth of Associative Strength among Strong
Positive Affect, Cues, Drug Use
Experimental Drug Use
Growth of Addiction, Habit
52Intervention Framework for Prevention
- Focus on input before behavior takes place
- Focus on connections of prevention material or
skills, especially with the input - Goal is to increase prevention-related memory
activation
- Consistent with
- Connectionist models
- Traditional expectancy theories of Tolman
Bolles, etc. - Not addressed by most alcohol expectancy research
or prevention - Again, cues connections not a focus
53Intervention Framework
54General Characteristics of Center Projects
- Relatively high risk youth
- Both projects use a previously documented
effective prevention program (TND Steve Sussman,
PI) as a springboard - Associative memory assessment and processes in
drug use habit and prevention - Social networks used to study and improve program
effects - The Center will go beyond the focus of the 2 main
projects!
55Project Towards No Drug Abuse (TND) Awards
-
- 1. NIDA model program
- 2. CSAP/SAMSHA model program
- 3. Health Canada exemplary program
- 4. Sociometrics, Inc. model program
- 5. U.S. Department of Education model program for
drug abuse prevention grantees - 6. Model program for numerous state Departments
of Health and Education
56The Motivation-Skills-Decision Making Model
(MSD) of Problem
Behavior
Motivation/ Cognitive Misperceptions
Behavior Social and -Violence-
related Self-control Skills -Drug
use Decision Making
57 Project TND Curriculum
Outline Session 1 Introduction Communication
and Active Listening Session 2 Stereotyping Sessio
n 3 Myths and Denial Session 4 Chemical
Dependency Session 5 Talk Show Session
6 Marijuana Panel Session 7 Tobacco Basketball
and Cessation Session 8 Stress, Health, and
Goals Session 9 Self-control Session 10 Positive
and Negative Thought and Behavior
Loops Session 11 Perspectives Session
12 Decision-making and Commitment Note
Three new lessons added to the TND-I 9 sessions
curriculum that are constituents of the newer
TND-II 12-session curriculum.
58Associative memory measures (spontaneous
cognition) predicted HIV risk behavior
59Specific results on DET
60(No Transcript)
61Some convergence of DET with other tests no
confounding with false positives or other
variablesaAll factor loadings p this loading
62Shultz (1998)
63From Shultz (1998) Broadcast from midbrain to
striatum and frontal cortex
- VTA ventral tegmental region of midbrain
- SN substantia nigra region of midbrain
- Midbrain top of brain stem
- Caudate-Putamen region of striatum in basal
ganglia
64(No Transcript)
65Survey measures of expectancy Our research
- Outcome expectancies predict alcohol and other
drug use, as long as 9 years (S,N,B, 1991) - Expectancies about positive and negative outcomes
are clearly separate constructs (S,W,M, 1990 S,
M, P, 1993 L S, 1993 others) - Positive expectancies are often, but not always,
better predictors (LS, 2004 S,W,M, 1990 etc) - Others research often shows two factors, even
they they rotate or model more (do a scree test
on their eigenvalues!) - Something about survey measures of expectancy are
important - But what are they?
- Self-perceptions or other inferences
- Another measure of behavior
- Memories (no evidence anywhere if so, please
show me!) - If-then rules
66What about arousal/sedation?
- In about 5 studies on self-generated outcomes, we
do not find more than 1 clearly arousal-related
words - The norms on word association show that some
arousal words clearly have a meaning of
POSITIVE AFFECT - D. Nelson norms Excited happy
- Happy is by FAR the strongest associate (.37 vs
.03) - Associative memory techniques may derive
COMPLETELY different answers compared to factor
analysis or MDS (not just 1st associations in any
method) - MDS has some very subjective steps fitting words
to different dimensions (is it really arousal, or
positive?). - Overall, I dont think there is substantial
evidence from multiple methods. Know of any? Case
may not be closed
67What about multiple expectancy dimensions?
- Many early factor analyses really showed TWO
factors in a scree test, but more factors were
analyzed (pos, neg). - Some words in expectancy scales are NOT
self-generated spontaneously by subjects - Words used to describe alcohol effects are verbal
labels for SOMETHING. Not necessarily isomorphic
with that something - Positive affect? (fun, happy, talkative, relaxed,
etc) - Researchers read too much into the words or use
procedures that do NOT evaluate semantic
associations. The bottom line involves the words
associative structure in memory, not a factor
analysis excited happy
68Is word association truly an implicit test?
- In the proper experimental paradigm, probably so
(e.g., amnesics vs. non manipulate encoding) - In a study without an encoding trial, we do not
know - There is nothing to stop explicit retrieval or
executive processing - There is no comparison with an explicit retrieval
condition - It does take time to write down a wordthere is
time for other processes to be engaged - Yet, word association PREDICTS implicit memory
processes in other paradigms (false memory,
extralist cued recall, primed free association,
semantic priming, at least) - If word association PREDICTS later implicit
processing, then - It is very useful even though it may sometimes
involve other processes
69Lexical versus Semantic Associations
Suppose you never saw these words before and did
not know their meanings
Now suppose you experienced these word pairs
repeatedly over time, but still learning nothing
else about them
- Later, if you see the letters Cervesa, you may
access an association to the letters in Corona
(their look or sound) - But a picture or a word for beer would not lead
you to think Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
Cervesa Corona
70 Who are the five BEST LEADERS in this class?
Think about the five people in this class who
would make the best leaders for working on group
projects. Write up to 5 names on the lines below,
starting with the best leader on the first line.
After you write their name look at the list of
names on the roster that has been provided. Match
the name to the number and write the number in
the boxes. If you cant think of five names in
this class, then leave the extra lines blank. You
can name yourself if you want.
Also asked who are your five best friends
71(No Transcript)
72Providing dates..
73Many Associative Memory Models are Connectionist
Approaches
74- Implicit cognition (IC)
- General Project Aims
-
- Examines drug abuse prevention program effects
- from an implicit cognition perspective.
- Proposes that behavior is governed by the
current - pattern of activation in memory activation
is often - primarily an implicit or relatively
spontaneous process. - Assumes drug use is not primarily governed by
- reflective decisions in which alternative
consequences - are weighed.
75Neuroscience basic research perspectives
support a distinction between systems
76Implicit Cognition Project Specific Aims
- Study 1
- Evaluate and refine assessments of associative
memory processes and drug abuse - Evaluate three-wave predictive effects of
associative memory on drug abuse
77Example of Continued Association Methods for
Cultural Research
(Szalay and colleagues)
78Cultural Differences in Association
(Szalay and
colleagues)
79Study 1General Predictionsof Drug Abuse by
AssociativeMemoryMeasures
80Study 1 3-Wave Prediction Model
T1
T2
T3
Drug Abuse
Drug Abuse
Drug Abuse
T1
T2
T3
Associative
Associative
Associative
Memory
Memory
Memory
C
T1
T2
T3
Outcome
Outcome
Outcome
Expectancy
Expectancy
Expectancy
T1
T2
T3
Sensation
Sensation
Sensation
Seeking
Seeking
Seeking
81Implicit Cognition Project Specific Aims
- Study 2
- Evaluate effects of TND curriculum on memory
and implicit cognition processes - Apply the best assessments from the preceding
study
82Implicit Cognition Project Specific Aims
continued.
- Communicate findings to other Center projects so
cognitive processes can be better addressed in
theory assessment of mechanisms. - Develop a proposal for a future large scale test
of a memory enhancement component for prevention.
83Implications of IC Project
- Advances in understanding the causes of drug
abuse - Advances in assessment of cognitive antecedents
of - drug abuse cognitive consequences of
prevention - Increase understanding of processes through
which - prevention programs operate
- Develop procedures that make intervention
content - more readily accessible in high-risk settings
84Example of Compound Cue measure For each phrase
below, write the first word it makes you think of
85Project 2 TND-Networked Using Network Theory
to Tailor Project TND
Thomas W. Valente, PhD
86Abstract
- Study designed to use network analysis theory to
structure an existing, evidence based, substance
abuse prevention program - Motivation comes from prior studies that show
- Social networks are important influences on
behavior - Social networks have been used to tailor programs
in the past TTURC study
87The Effects of a Social Network Method for Group
Assignment Strategies on Peer Led Tobacco
Prevention Programs in Schools
Thomas W. Valente, PhD Beth R. Hoffman,
MPH Anamara Ritt-Olson, MA Kara Lichtman, MA C.
Anderson Johnson, PhD In press Am. J. of Public
Health Funded by NCI/NIDA, Transdisciplinary
Tobacco Use Research Center
88Objectives
- Evaluate the feasibility of a network method for
identifying leaders and creating workgroups for
school-based tobacco prevention curriculum. - Nested within a study of FLAVOR, a culturally
tailored program, being compared to CHIPS! a
standard social influences curriculum. - Determine whether more effective than random
groups and teacher defined ones.
89Comparison of 3 Conditions
90Study Design
91Data
92 Who are the five BEST LEADERS in this class?
Think about the five people in this class who
would make the best leaders for working on group
projects. Write up to 5 names on the lines below,
starting with the best leader on the first line.
After you write their name look at the list of
names on the roster that has been provided. Match
the name to the number and write the number in
the boxes. If you cant think of five names in
this class, then leave the extra lines blank. You
can name yourself if you want.
Also asked who are your five best friends
93(No Transcript)
94Group Assignments for One Network Condition Class
95Regression Results on Post Program Appeal (Lower
Scores Better) (N1960 k87 Beta Coefficients)
96Regression Results on Post Program Attitudes
(Lower Scores Better, Beta Coefficients)
97Intention to Smoke
98Classroom Level Analysis (Nk87 Beta
Coefficients)
99Conclusions
- Network method was most effective at changing
short term outcomes - First network-based intervention
- Can use network information to improve existing
programs in schools, organizations, communities,
and other sites - The approaches and algorithms can be improved
(location science) - Networks influence behavior
100Advantages of Network Methods
- Capitalizes on existing interpersonal
relationships - Uses community input
- Establishes a learning organization /community
- Builds social capital
- Is Empowering
10110 Methods Used to Identify Peer Opinion Leaders
102TPRC Training Core and Developmental Grants
103Training Core
- Purpose
- To foster transdisciplinarity
- Training opportunities for students
- Mentorship
- Research experiences
- Offered at BA, MPH, Ph.D., and postdoctoral
levels - BA level Directed Research units
- MPH and Ph.D. level rotation plan for working
with different faculty - Training opportunities for faculty
- Seminars
- Developmental grants
104Developmental Grant Program
- Purpose
- To assist investigators in integrating strategies
and theories from multiple scientific disciplines - To gain a more comprehensive understanding of the
causes of drug abuse and its prevention among
adolescents
105Award Categories
- Established researchers
- transitioning to drug use research from another
research area or seeking seed funding for new
research ideas - up to 20,000/year
- New investigators
- (Ph.D.-level) at the beginning of their research
careers who have not yet been Principal
Investigator on a major grant - up to 15,000/year
- Graduate students
- dissertation research
- up to 5,000/year
106What Were Looking For
- We are interested in projects that
- Are transdisciplinary
- Involve multiple levels of analysis
- Apply basic research to the prevention of drug
abuse - Letters of intent accepted once or twice annually
107(No Transcript)
108(No Transcript)