Causal Inference and Alternative Explanations S.A. Murphy Univ. of Michigan May, 2004.

Slides:



Advertisements
Similar presentations
Measurement Concepts Operational Definition: is the definition of a variable in terms of the actual procedures used by the researcher to measure and/or.
Advertisements

Significance Testing.  A statistical method that uses sample data to evaluate a hypothesis about a population  1. State a hypothesis  2. Use the hypothesis.
Forecasting Using the Simple Linear Regression Model and Correlation
Chapter 2: The Research Process
Research in Abnormal Psychology
Correlation AND EXPERIMENTAL DESIGN
Research in Psychology Chapter Two
Experimenting to Improve Clinical Practice S.A. Murphy AAAS, 02/15/13 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.:
Journal Club Alcohol, Other Drugs, and Health: Current Evidence May–June 2013.
Research in Abnormal Psychology
An Experimental Paradigm for Developing Dynamic Treatment Regimes S.A. Murphy Univ. of Michigan March, 2004.
Constructing Dynamic Treatment Regimes & STAR*D S.A. Murphy ICSA June 2008.
Experiments and Dynamic Treatment Regimes S.A. Murphy Univ. of Michigan Florida: January, 2006.
Hypothesis Testing and Dynamic Treatment Regimes S.A. Murphy Schering-Plough Workshop May 2007 TexPoint fonts used in EMF. Read the TexPoint manual before.
An Experimental Paradigm for Developing Adaptive Treatment Strategies S.A. Murphy Univ. of Michigan UNC: November, 2003.
Experiments and Dynamic Treatment Regimes S.A. Murphy Univ. of Michigan PSU, October, 2005 In Honor of Clifford C. Clogg.
Hypothesis Testing and Dynamic Treatment Regimes S.A. Murphy, L. Gunter & B. Chakraborty ENAR March 2007.
Clustered or Multilevel Data
An Experimental Paradigm for Developing Adaptive Treatment Strategies S.A. Murphy Univ. of Michigan ACSIR, July, 2003.
An Experimental Paradigm for Developing Adaptive Treatment Strategies S.A. Murphy Univ. of Michigan February, 2004.
1 Possible Roles for Reinforcement Learning in Clinical Research S.A. Murphy November 14, 2007.
Experiments and Dynamic Treatment Regimes S.A. Murphy Univ. of Michigan April, 2006.
Multivariate Analyses & Programmatic Research Re-introduction to Multivariate research Re-introduction to Programmatic research Factorial designs  “It.
SMART Experimental Designs for Developing Adaptive Treatment Strategies S.A. Murphy ISCTM, 2007.
Experiments and Adaptive Treatment Strategies S.A. Murphy Univ. of Michigan Chicago: May, 2005.
Experiments and Dynamic Treatment Regimes S.A. Murphy Univ. of Michigan January, 2006.
© 2000 Prentice-Hall, Inc. Chap Forecasting Using the Simple Linear Regression Model and Correlation.
Research problem, Purpose, question
Hypothesis Testing and Adaptive Treatment Strategies S.A. Murphy SCT May 2007.
Chapter 1 Psychology as a Science
Clinical Trials. What is a clinical trial? Clinical trials are research studies involving people Used to find better ways to prevent, detect, and treat.
Overview of Adaptive Treatment Regimes Sachiko Miyahara Dr. Abdus Wahed.
Epidemiology The Basics Only… Adapted with permission from a class presentation developed by Dr. Charles Lynch – University of Iowa, Iowa City.
Understanding Statistics
Chapter 2 Research in Abnormal Psychology. Slide 2 Research in Abnormal Psychology  Clinical researchers face certain challenges that make their investigations.
Abstinence Incentives for Methadone Maintained Stimulant Users: Outcomes for Those Testing Stimulant Positive vs Negative at Study Intake Maxine L. Stitzer.
 Used to observe and describe behavior  Help to answer questions such when do certain behaviors occur  How often does the behavior occur  Is the behavior.
Thomson South-Western Wagner & Hollenbeck 5e 1 Chapter Sixteen Critical Thinking And Continuous Learning.
Research Methods It is actually way more exciting than it sounds!!!!
Reliability, Validity, and Bias. Reliability Reliability Reliability is the extent to which an experiment, test, or any measuring procedure yields the.
Critical Appraisal (CA) I Prepared by Dr. Hoda Abd El Azim.
An Experimental Paradigm for Developing Adaptive Treatment Strategies S.A. Murphy NIDA Meeting on Treatment and Recovery Processes January, 2004.
Research Design Evidence Based Medicine Concepts and Glossary.
Introduction to Research for Physical Therapy Students.
Designing An Adaptive Treatment Susan A. Murphy Univ. of Michigan Joint with Linda Collins & Karen Bierman Pennsylvania State Univ.
DESCRIPTIVE METHODS Methods that yield descriptions of behavior but not necessarily causal explanations.
Jump to first page Inferring Sample Findings to the Population and Testing for Differences.
Research in Psychology Chapter Two 8-10% of Exam AP Psychology.
Public Finance and Public Policy Jonathan Gruber Third Edition Copyright © 2010 Worth Publishers 1 of 24 Copyright © 2010 Worth Publishers.
The Practice of Statistics, 5th Edition Starnes, Tabor, Yates, Moore Bedford Freeman Worth Publishers CHAPTER 10 Comparing Two Populations or Groups 10.1.
Topic 2: Types of Statistical Studies
CHAPTER 10 Comparing Two Populations or Groups
Alcohol, Other Drugs, and Health: Current Evidence
CHAPTER 10 Comparing Two Populations or Groups
Chapter 14 Repeated Measures
Empirical Tools of Public Finance
Reliability, Validity, and Bias
CHAPTER 10 Comparing Two Populations or Groups
External Validity.
Critical Appraisal วิจารณญาณ
CHAPTER 10 Comparing Two Populations or Groups
Thinking critically with psychological science
CHAPTER 10 Comparing Two Populations or Groups
CHAPTER 10 Comparing Two Populations or Groups
CHAPTER 10 Comparing Two Populations or Groups
CHAPTER 10 Comparing Two Populations or Groups
CHAPTER 10 Comparing Two Populations or Groups
CHAPTER 10 Comparing Two Populations or Groups
Scientific Method Review.
CHAPTER 10 Comparing Two Populations or Groups
Presentation transcript:

Causal Inference and Alternative Explanations S.A. Murphy Univ. of Michigan May, 2004

Outline 1)Fundamental Problem of Causal Inference 2)Time Independent Treatments Example, Composition and Alternative Explanations, Ideal Trial 3)Time Dependent Treatments Example, Composition and Alternative Explanations, Ideal Trial

Fundamental Problem of Causal Inference

We have developed a new behavioral program for smokers. Is it better than standard care? Joe’s days abstinent if we provide the new behavioral program == Y 1 Joe’s days abstinent if we provide standard care==Y 0 If Y 1 > Y 0 then our answer is yes!

The fundamental problem of causal inference is that we can never observe both Y 1 and Y 0 and thus can not answer this question! We average Y 1 for people who appear like Joe and received new program. We average Y 0 for people who appear like Joe and received standard care. If average Y 1 > average Y 0 then our answer is yes!

Time Independent Treatments

Example: Does treatment improve abstinence one year later among smokers? Researchers compare smokers who receive standard care to smokers who receive the new behavioral program. Control for demographics, addiction severity, social network characteristics, stage of change.

Problem: Standard care smokers may differ from treated smokers in terms of unmeasured characteristics. There may be a compositional difference between smokers receiving standard care and smokers receiving the new behavioral program and this compositional difference may have led to observed differences in average days abstinent. Maybe difference in abstinence is due to difference in pretreatment motivational levels not difference in treatment?

Ideal Solution: Randomize subjects to new behavioral program or standard care

Example: Does the new behavioral program improve abstinence among smokers like Joe?

Example: Does the new behavioral program improve abstinence among smokers like Joe? But people who appear like Joe may differ from Joe in terms of unmeasured characteristics. Ideally we’d obtain the average effect of the new behavioral program for all smokers in our population who appear like Joe on measured characteristics: demographics, addiction severity, social network characteristics, stage of change

Randomize Treatment in our trial; control for measured demographics, pretreatment addiction severity, social network characteristics, stage of change Problem: People who appear like Joe in our trial may differ from people who appear like Joe in our population in terms of unmeasured characteristics. There may be a compositional difference between people like Joe in the population and people like Joe in our study.

Ideal Solution: Sample subjects from an explicitly defined population (Joe is a member of this population).

Time Dependent Treatments

Example: We want to evaluate a time varying treatment for smokers. Smokers are randomized to receive group therapy over 6 months or to standard care. In the treatment group, staff use clinical judgment that repeatedly assesses the smoker’s need for therapy and provides group therapy in response to this need. We would like to know if more group therapy translates into improved abstinence.

We compare smokers who are randomized to treatment and receive more group therapy to smokers who were randomized to treatment and who receive less group therapy. We control for demographics, addiction severity, social network characteristics, stage of change. We see a negative relationship between dose and days abstinent!

Problem: There may be unmeasured compositional differences between heavily treated and lightly treated smokers and these compositional differences may have led to observed differences in average abstinence rather than the dose of treatment. Perhaps smokers who show great need for treatment are getting the most treatment while smokers who show the least need for treatment are getting the least amount of group therapy.

X=measured characteristics U=unmeasured characteristics

Randomized Dose of Group Therapy X=measured characteristics U=unmeasured characteristics

For the Connoisseur!

Example: We want to inform clinical practice which would use measures of ongoing response in order to decide whether to provide more group therapy. Is it useful to provide more group therapy to those who show evidence of need? Regress days abstinent on measured characteristics X 1 and X 2 and on amounts of group therapy provided at times 1 and 2. Coefficient of group therapy at time 1 reflects more than the effect of group therapy at time 1 on days abstinent!

X=measured characteristics U=unmeasured characteristics

To assess the effect of and the usefulness of tailoring group therapy we need different kinds of regressions. This is what I do!