/
Optimizing Stratification through Blocking Backdoors Optimizing Stratification through Blocking Backdoors

Optimizing Stratification through Blocking Backdoors - PowerPoint Presentation

tatiana-dople
tatiana-dople . @tatiana-dople
Follow
342 views
Uploaded On 2020-01-24

Optimizing Stratification through Blocking Backdoors - PPT Presentation

Optimizing Stratification through Blocking Backdoors Farrokh Alemi PhD Causal Analysis of Observational Data Misleading Comparisons Causal Analysis of Observational Data Potential Outcomes Misleading Comparisons ID: 773669

Share:

Link:

Embed:

Download Presentation from below link

Download Presentation The PPT/PDF document "Optimizing Stratification through Blocki..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Optimizing Stratification through Blocking Backdoors Farrokh Alemi, Ph.D.

Causal Analysis of Observational Data Misleading Comparisons

Causal Analysis of Observational Data Potential Outcomes Misleading Comparisons

Causal Analysis of Observational Data Simple, Easy, Fast, Graphical Conditioning or Stratification Potential Outcomes Misleading Comparisons

Conditioning on Strata     T Cause Absent Cause Present Ceteris Paribus

Different Names for Same Thing Pearl’s “Do Operation”

Do Operation: Backdoors, Markov Blanket, & Directed Separation

Find variables that d-separate every pair of non-adjacent nodes R: Renal Complication D: Diabetes H: Dehydration N: Neuropathy T: Treatment S: Survival

Find variables that d-separate every pair of non-adjacent nodes R: Renal Complication D: Diabetes H: Dehydration N: Neuropathy T: Treatment S: Survival

When Z are Variables that Block Backdoor Then:   Causal Impact of X on Y

When Z are Variables that Block Backdoor Then:   Propensity Score

Impractical Many Few Explosion Ouch

Path Overlapping Pairs of Connected Variables

Backdoor Path Starts from Outcome & Points to Treatment

Stratify Only Variables that Block All Backdoor Paths

Surprise g All Backward Paths Mediator, Common Cause, but not Common Effect

Obesogenic Anti-Depressant Obesity Diabetes Infection Antibiotic Treatment ? Does Antibiotic Use Increase Diabetes

Obesogenic Anti-Depressant Obesity Diabetes Infection Antibiotic Treatment ? Does Antibiotic Use Increase Diabetes

Obesogenic Anti-Depressant Obesity Diabetes Infection Antibiotic Treatment ? Does Antibiotic Use Increase Diabetes

Obesogenic Anti-Depressant Obesity Diabetes Infection Antibiotic Treatment ? Does Antibiotic Use Increase Diabetes

Obesogenic Anti-Depressant Obesity Diabetes Infection Antibiotic Treatment ? Does Antibiotic Use Increase Diabetes

Obesogenic Anti-Depressant Obese Patients Diabetes Infection Antibiotic Treatment ?

Obesogenic Anti-Depressant Obese Patients Diabetes Infection Antibiotic Treatment ?

Obesogenic Anti-Depressant Obese Patients Diabetes Infection Antibiotic Treatment ?

A B C F D G Y U V ? Can You Find the Backdoor Paths? What is causal impact of D on Y

A B C F D G Y U V ? First Backdoor Path

A B C F D G Y U V ? Second Backdoor Path

A B C F D G Y U V ? Minimal Block Set

Blocking Back Door Paths No Need to Condition on All Alternative Explanations

Focus on Parents of Markov Blanket of Treatment All backdoor paths go through parents of treatment

Algorithm for Minimal Block Set Step 1: Remove events that occur after treatment

Algorithm for Minimal Block Set Step 1: Remove events that occur after treatment Step 2: Regress treatment on covariates

Algorithm for Minimal Block Set Step 1: Remove events that occur after treatment Step 3: Check for new backdoor paths Step 2: Regress treatment on covariates

Stratify a Subset of Covariates: Parents on Markov Blanket of Treatment