SUMMARY RESEARCH PROJECT 2020-2021
LECTURE 1
Just an introduction to the course.
LECTURE 2
Causal inference
Correlation does not imply causation. Several reasons why X and Y can correlate:
● X causes Y
● Y causes X
● Z causes both X and Y
● Spurious correlation
Reverse Causality: X appears to cause Y, but it is actually Y that causes Y
● → Sales of a brand of soda are higher during weeks of heavy advertising.
○ However, advertising is allocated when the stakes are greatest (e.g.,
during holidays and summer months). Thus, anticipated sales may
actually cause advertising.
Third Variable: X appears to cause Y, but both X and Y are actually caused by Z.
● On the average, the more toys a child has, the higher his or her IQ.
○ Both the number of toys and IQ may be caused by family resources such
as income (i.e., better nutrition and education).
When can we infer that X causes Y?
Three conditions for causality:
● Relationship between X and Y
○ X and Y vary together
● Time order
○ X cannot happen after Y
● Elimination of other possible causal factors
○ All other possible causes held constant or controlled
Experimental design
Independent variable is “manipulated” across groups or “between-subject”.
● Between-subject: one participant is assigned to one experimental condition of
the IV (e.g., ad A or ad B)
1
, ● Within-subject: one participant is assigned to several experimental conditions
(e.g., ad A and ad B)
Moderator
Experiments can also account for moderating variables, which affect the relationship
between the IV and the DV. The moderator can be either:
● Measured (e.g., with a Likert scale)
● Manipulated (e.g., with specific experimental conditions)
Experimental validity
Internal Validity
● Conclusions about the effects of IVs on DVs are valid
● How to ensure it? correct implementation of principles randomization, control of
extraneous factors, etc
● Lab studies are high in internal validity
External Validity
● Conclusions can be generalized outside the experiment e.g. lab participants →
consumers?
● Field studies are higher in external validity (although they tend to suffer from
internal validity)
Randomized Control Trial (RCT)
● In parallel groups: “between-subject experiment”
● In crossover: random allocation into different groups + DV measured several
times (e.g., before and after taking the drug).
● In cluster: pre-existing groups of participants (e.g., villages, schools) are randomly
selected to receive (or not receive) an intervention.
Designs that face threats to internal validity/causal claims
● Quasi-Experiment: Assignment to the experimental conditions is not random
(e.g., nutri-score in supermarket A vs. no nutri-score in supermarket B).
● Pre-post: The DV (e.g., stock price) is measured before and after an IV category
(e.g., before/after the new policy). Can be labeled a non-experimental
observational study.
WORKSHOP 1
Ensuring that all uncontrolled is random: A way to control for confounding factors
(“confounds”). Then you know that it’s actually X that causes your Y and not other
differences.
● Internal validity: A special case of “Holding everything the same”.
● External validity: Ensures that your effect holds under different conditions
2
LECTURE 1
Just an introduction to the course.
LECTURE 2
Causal inference
Correlation does not imply causation. Several reasons why X and Y can correlate:
● X causes Y
● Y causes X
● Z causes both X and Y
● Spurious correlation
Reverse Causality: X appears to cause Y, but it is actually Y that causes Y
● → Sales of a brand of soda are higher during weeks of heavy advertising.
○ However, advertising is allocated when the stakes are greatest (e.g.,
during holidays and summer months). Thus, anticipated sales may
actually cause advertising.
Third Variable: X appears to cause Y, but both X and Y are actually caused by Z.
● On the average, the more toys a child has, the higher his or her IQ.
○ Both the number of toys and IQ may be caused by family resources such
as income (i.e., better nutrition and education).
When can we infer that X causes Y?
Three conditions for causality:
● Relationship between X and Y
○ X and Y vary together
● Time order
○ X cannot happen after Y
● Elimination of other possible causal factors
○ All other possible causes held constant or controlled
Experimental design
Independent variable is “manipulated” across groups or “between-subject”.
● Between-subject: one participant is assigned to one experimental condition of
the IV (e.g., ad A or ad B)
1
, ● Within-subject: one participant is assigned to several experimental conditions
(e.g., ad A and ad B)
Moderator
Experiments can also account for moderating variables, which affect the relationship
between the IV and the DV. The moderator can be either:
● Measured (e.g., with a Likert scale)
● Manipulated (e.g., with specific experimental conditions)
Experimental validity
Internal Validity
● Conclusions about the effects of IVs on DVs are valid
● How to ensure it? correct implementation of principles randomization, control of
extraneous factors, etc
● Lab studies are high in internal validity
External Validity
● Conclusions can be generalized outside the experiment e.g. lab participants →
consumers?
● Field studies are higher in external validity (although they tend to suffer from
internal validity)
Randomized Control Trial (RCT)
● In parallel groups: “between-subject experiment”
● In crossover: random allocation into different groups + DV measured several
times (e.g., before and after taking the drug).
● In cluster: pre-existing groups of participants (e.g., villages, schools) are randomly
selected to receive (or not receive) an intervention.
Designs that face threats to internal validity/causal claims
● Quasi-Experiment: Assignment to the experimental conditions is not random
(e.g., nutri-score in supermarket A vs. no nutri-score in supermarket B).
● Pre-post: The DV (e.g., stock price) is measured before and after an IV category
(e.g., before/after the new policy). Can be labeled a non-experimental
observational study.
WORKSHOP 1
Ensuring that all uncontrolled is random: A way to control for confounding factors
(“confounds”). Then you know that it’s actually X that causes your Y and not other
differences.
● Internal validity: A special case of “Holding everything the same”.
● External validity: Ensures that your effect holds under different conditions
2
