Advanced Research Methods
Quanttatve and Qualitatve
MSc Health Care Management 2017-2018
Lectures & Workgroups
Part 1: Epidemiology and quanttatve methods
o Lecture 1, 2 & 3
o Workgroups 1, 2, 3 & PC labs
Part 2: Qualitatve methods
o Lecture 4 & 5
o Workgroups 4, 5, 6
Part 3: Organizatonal design studies
o Practice exams
1
,Part 1: Epidemiology and quanttatve methods
Lecture 1 “Causal Inference; drawing lines between causes and efectss
Learning goals for this week:
o To explain the role of causal theories in the design of quantitatite studies
o To apply concepts confounder, confounding, intermediate tariables
o To use directed acyclic graphs (DAG’s) in the design of an analysis
o To be familiar with a number of types of bias
Epidemiology...
- Is not “the science of epidemics”
- “Is the study of the distributon of health-related states and etents in the populationn”
- “Does not represent a body of knowledge (…) It is a philosophy and methodology that can be applied
to a tery broad range of health problemsn The ‘art’ of epidemiology is knowing when and how to
apply the carious strategies creatitely to answer specifc health questionsn”
L’Oréal True Match Minerals leads to a beter skin? “70% less imperfectons in 4 weekss “Our powder covers
like a foundaton and improves the quality of your skin in 1 month!s
- “Improtes” implies a causal efectt
o A leads to B
o Using True Match Minerals leads to a beter skin
Is this contincing?
- Problemst
o Small sample; is this always a problem?
o Study performed or fnanced by commercial company; is this always fatal?
o No control group;
Essential omission
What without treatment?
Potential regression towards the mean
- What do we want to know?
o Not interested in the outcome per se
o Interested in the role of treatment in achieting this outcome
- Conclusiont No meaningful causal conclusion can be drawn from this study
Formal defniton of causaton:
‘In an individual, a treatment has a causal efect if the outcome under treatment 1 would be diferent from the
outcome under treatment 2.’ (Hernan/Robins)
(What would happen to women that use minerals compared to women that don’t)
2
,Potental outcomes
3
, In individual cases, not all potental outcomes are not observed (so you don’t know what would happen to
the skin of women that don’t use the product)
- Counterfactual outcomet potential outcome that is not obserted because the subject did not
experience the treatment (‘counter of the fact’)
- Potential outcome a=1 is factual for some subjects, and counterfactual for othersn
Fundamental problem
- Inditidual causal efect cannot be obserted
o Except under extremely strong (and generally unreasonable) assumptions
- Causal inference as a missing data problem
- We need a diferent defnition of causal efects
- Aterage causal efects can be determined under
o No assumptions (randomised studies)
o Strong assumptions (obsertational studies)
Essental assumptons
Based on population aterages, conclusions on causal efects can be drawn if three assumptions holdt
- Posititity
o Units are assigned to all reletant ‘treatments’
- Consistency
o Treatment is clearly defned
o ‘No causation without manipulation’
What is obesity? By what is obesity caused? And what does that say about the impact on a
patient’s health?
- Exchangeability
o Treatment groups are exchangeable
It does not mater who gets treatment A and who gets treatment B
o Notationt
Potential outcomes are independent of the treatment that was actually receited
Meetng the assumptons
- Posititityt control group
- Consistencyt describe treatment with and without L’Oréal
- Exchangeabilityt randomise
Randomised controlled trial (RCT)
- Select patients
- Randomly assign them to treatment groups
- Gold standard
RCT’s vs observatonal studies
- RCT
o Expected exchangeability
o Posititity and consistency inherently assured
o Limited generalisability (external talidity) due to treatment protocol and patient selection
o Practical, ethical considerations
- Obsertational study
o Internal talidity threatened by lack of exchangeability
o Posititity and consistency need explicit atention
o Real world outcomes
4
Quanttatve and Qualitatve
MSc Health Care Management 2017-2018
Lectures & Workgroups
Part 1: Epidemiology and quanttatve methods
o Lecture 1, 2 & 3
o Workgroups 1, 2, 3 & PC labs
Part 2: Qualitatve methods
o Lecture 4 & 5
o Workgroups 4, 5, 6
Part 3: Organizatonal design studies
o Practice exams
1
,Part 1: Epidemiology and quanttatve methods
Lecture 1 “Causal Inference; drawing lines between causes and efectss
Learning goals for this week:
o To explain the role of causal theories in the design of quantitatite studies
o To apply concepts confounder, confounding, intermediate tariables
o To use directed acyclic graphs (DAG’s) in the design of an analysis
o To be familiar with a number of types of bias
Epidemiology...
- Is not “the science of epidemics”
- “Is the study of the distributon of health-related states and etents in the populationn”
- “Does not represent a body of knowledge (…) It is a philosophy and methodology that can be applied
to a tery broad range of health problemsn The ‘art’ of epidemiology is knowing when and how to
apply the carious strategies creatitely to answer specifc health questionsn”
L’Oréal True Match Minerals leads to a beter skin? “70% less imperfectons in 4 weekss “Our powder covers
like a foundaton and improves the quality of your skin in 1 month!s
- “Improtes” implies a causal efectt
o A leads to B
o Using True Match Minerals leads to a beter skin
Is this contincing?
- Problemst
o Small sample; is this always a problem?
o Study performed or fnanced by commercial company; is this always fatal?
o No control group;
Essential omission
What without treatment?
Potential regression towards the mean
- What do we want to know?
o Not interested in the outcome per se
o Interested in the role of treatment in achieting this outcome
- Conclusiont No meaningful causal conclusion can be drawn from this study
Formal defniton of causaton:
‘In an individual, a treatment has a causal efect if the outcome under treatment 1 would be diferent from the
outcome under treatment 2.’ (Hernan/Robins)
(What would happen to women that use minerals compared to women that don’t)
2
,Potental outcomes
3
, In individual cases, not all potental outcomes are not observed (so you don’t know what would happen to
the skin of women that don’t use the product)
- Counterfactual outcomet potential outcome that is not obserted because the subject did not
experience the treatment (‘counter of the fact’)
- Potential outcome a=1 is factual for some subjects, and counterfactual for othersn
Fundamental problem
- Inditidual causal efect cannot be obserted
o Except under extremely strong (and generally unreasonable) assumptions
- Causal inference as a missing data problem
- We need a diferent defnition of causal efects
- Aterage causal efects can be determined under
o No assumptions (randomised studies)
o Strong assumptions (obsertational studies)
Essental assumptons
Based on population aterages, conclusions on causal efects can be drawn if three assumptions holdt
- Posititity
o Units are assigned to all reletant ‘treatments’
- Consistency
o Treatment is clearly defned
o ‘No causation without manipulation’
What is obesity? By what is obesity caused? And what does that say about the impact on a
patient’s health?
- Exchangeability
o Treatment groups are exchangeable
It does not mater who gets treatment A and who gets treatment B
o Notationt
Potential outcomes are independent of the treatment that was actually receited
Meetng the assumptons
- Posititityt control group
- Consistencyt describe treatment with and without L’Oréal
- Exchangeabilityt randomise
Randomised controlled trial (RCT)
- Select patients
- Randomly assign them to treatment groups
- Gold standard
RCT’s vs observatonal studies
- RCT
o Expected exchangeability
o Posititity and consistency inherently assured
o Limited generalisability (external talidity) due to treatment protocol and patient selection
o Practical, ethical considerations
- Obsertational study
o Internal talidity threatened by lack of exchangeability
o Posititity and consistency need explicit atention
o Real world outcomes
4
