An Introduction to Modeling Causality with Repeated Measurement Designs: Guest Lecture for Experimental Psychology Tulane University 11/21/13 James Gerhart, Ph.D. Rush University Medical Center
Dec 17, 2015
An Introduction to Modeling Causality with Repeated Measurement Designs:
Guest Lecture for Experimental Psychology Tulane University
11/21/13James Gerhart, Ph.D.
Rush University Medical Center
Agenda
• Discuss methods for assessing statistical causality1 and mechanisms of change2 in longitudinal designs – Ecological Momentary Assessment (Diary Data)– Treatment Outcome Research
Statistical Causality and Mechanisms of Change
• Psychologists are interested in explaining behavior and promoting socially valid changes in behavior.
• Thus, we are often interested in identifying causes of behavior.
• Philosophers continue to argue about causality.
• Today we’ll focus on a few suggested methods for estimating causality.
John Stuart Mill’s Criteria
• Establish temporal precedence– Cause precedes effect
• Establish co-variation-– The cause and effect are related
• Rule out alternative explanations
Statistical Causality and Mechanisms of Change
• We can attempt to meet these criteria with Ecological Momentary Assessment (diary data), and treatment outcome data.
• We can attempt behavioral change more efficiently.
Benefits of Diary Data
• Diary studies assess multiple behaviors in the context of day-to-day life (ecologically valid)
• Repeated measurements throughout the day allows us to study changes
• Easy to use with smart phone/tablets• The structure of the data allow us to test for
temporal precedence (1), co-variation (2), and rule out some potential confounds (3).
Structure of Diary DataTime 1 Time 2 Time 3 Time 4 Time 5
Anger Anger Anger Anger Anger…
Pain Pain Pain Pain Pain…
Note. All data are illustrative and not from research subjects
Structure of Diary Data-Measuring Confounds
Time 1 Time 2 Time 3 Time 4 Time 5
Anger Anger Anger Anger Anger…
Pain Pain Pain Pain Pain…
Depression Depression Depression Depression Depression
Medication Medication Medication Medication Medication
Missing Data
• Data replacement techniques have been developed so that data can be estimated and used
• Mixed Models• Data Imputation• Data should be missingat random• Data could be biased if angrier folks skip more
Variance Components
Between subject varianceWithin subject variance
Who is angriest? How does anger fluctuate?
Setting up the Diary Model
How does change in anger relate to change in later pain?
Pain = Prior Anger+Prior Pain+Time+Covariates
Reverse the Model
Anger =Prior Pain+Prior Anger+Time+Covariates
Mechanisms of Change
• For years psychologists have argued back and forth about theoretical orientations
• Meta-analytic research shows that for many disorders (anxiety, anger, depression, PTSD) treatments intended to work, usually do
• But psychodynamic, cognitive, and behavioral models differ drastically
Mechanisms of Change
• There are many explanations for the similarity of effects. – Behaviors are multiply determined
• Maybe psychodynamic treatment changes one part of distress, and behavioral changes another
– Treatments overlap on key processes• Hope, expectation of change, normalization, behavioral
activation, insight, placebo effects, unconscious learning.
• We need to measure these possibilities regularly throughout treatment.
Revisiting John Stuart Mill’s Criteria
• Establish temporal precedence– The mechanism should change before outcome
changes• Establish co-variation– Change in mechanism should correlate with
change in outcome• Rule out alternative explanations– Measure and analyze other potential mechanisms
of change
Structure of Treatment DataTime 1 Time 2 Time 3 Time 4 Time 5
Anger Anger Anger Anger Anger…
Pain Pain Pain Pain Pain…
Depression Depression Depression Depression Depression
Medication Medication Medication Medication Medication
Pain Attitudes Pain Attitudes Pain Attitudes Pain Attitudes Pain Attitudes
Hope Hope Hope Hope Hope
Mindfulness Mindfulness Mindfulness Mindfulness Mindfulness
Distraction Distraction Distraction Distraction Distraction
Setting up a Mechanism Models
1. Pain = Prior Pain Attitude+Prior Pain+Time+Covariates
2. Pain = Prior Hope+Prior Pain+Time+Covariates3. Pain = Prior Mindfulness+Prior Pain+Time+Covariates4. …
• Contact info:James Gerhart, Ph.D.Rush University Medical [email protected]
References
• 1. Duckworth, A.L., Tsukayama, E. & May, H. (2010). Establishing causality using longitudinal hierarchical linear modeling: An illustration predicting achievement from self-control. Social psychological and personality science, 1, 311-317.
• 2. Kazdin, A.E. (2007). Mediators and mechanisms of change in psychotherapy research. Annual review of clinical psychology, 3, 1-27.