lecture 20 (reasoning)

Question 1

Criteria John Stuart Mill for causation? (3)

Answer 1

• Priority: Change X precedes change Y
• Consistency: Change X varies systematically with change Y
• Exclusivity: There is no alternative explanation for the relationship (attention: exclusivity does not mean X is always sufficient and necessary for Y to occur)

Question 2

INUS condition? (1->4)

Answer 2

• an element of causation which is insufficient and non-redundant but part of an unnecessary but sufficient condition
• insufficient: gun alone does not kill someone
• non redundant: gun is a relevant part, adds explanatory power to a set of factors
• unnecessary: people can still be killed with other methods
• sufficient: all factors together (loaded gun, someone shooting it) are sufficient to kill someone

Question 3

how to check for non redundacy? (2)

Answer 3

• causal graphs (DAG)
• counterfactual

Question 4

counterfactual? (4)

Answer 4

• something that would have happened if something else was done during an experiment
• In research we compare observations with a counterfactual to know what the effect of the manipulation is
• e.g. what would the homicide rate be if there were no guns
• That “perfect” counterfactual is physically impossible

Question 5

experimental research and counterfactuals? (2)

Answer 5

• Experimental designs should aim to create a high-quality source of counterfactual inference and understand how this differs from the treatment condition
• can be made through random assignement (ethics concern), within group design, etc

Question 6

quasi experiments? (1+2)

Answer 6

• experiment not using random assignment to different conditions
• can be made through observing the same people over time or by matching participants from nonrandom control groups (however, there is always unknown systematic differences in this case)

Question 7

quasi experiments improvements over correlational studies? (2)

Answer 7

• quasi-experiments force the cause to occur before the effect, thereby establishing temporality, which is not possible in correlational studies
• quasi-experiments allow for controlling at least some of the possible confounds present in correlations

Question 8

threats to causality: outside factors? (2)

Answer 8

• history: certain events occurring at the same time as treatment and affecting performance negatively
• maturation: Natural changes that may be confused with effect treatment

Question 9

threats to causality: Effects of selection? (2)

Answer 9

• selection: systematic differences between conditions that obscure the treatment effect
• attrition: loss of participants could produce contrary or artificial effects if attrition is correlated with the conditions (war plane example)

Question 10

threats to causality: Unintended effects of study itself? (2)

Answer 10

• instrumentation: Change in measuring instrument resulting in a difference between pre- and
  post-measurement
• testing: being exposed to tests multiple times can affect the scores on the test

Question 11

threats to causality: statistics?

Answer 11

regression to the mean: if the subjects of an experiment initially had extreme scores because of certain characteristics, they may have scores closer to the mean later which could be confused for an effect of the manipulation

Question 12

falsification? (2+1->1)

Answer 12

• Scientists should try to falsify their conclusions because if a conclusion withstands this process it gains more credibility
• depends on the causal claim being clear and agreed upon in all of its details
• requires perfect observational procedures that reflect the theory to be tested, however this is rarely the case because of subjective elements being present

Question 13

randomization?

Answer 13

to diminish the affect of confounding factors in the manipulation variable

Question 14

3 main pillars of the swap of ambiguity in causality according to Foster?

Answer 14

• ignoring causality and only wrting down correlations found
• taking implausible assumptions for granted (e.g. linearity, a statistical assumption or no unobserved confounding, a conceptual assumption)
• No direct statements about causality, but clearly implied in the conclusion

Question 15

why is causal inference necessary?

Answer 15

• allows researchers to identify risk factors and come up with treatments to improve the well being of others
• causal thinking is unavoidable
• many people cannot distinguish between associations and causal relationships

Question 16

DAGs?

Answer 16

• directed acyclic graph (causal graphs)
• identifies what implications follow from certain associations for causality and the assumptions for these associations to imply causality
• assume a preference for simplicity (as few linkages as possible) and probabilistic stability (relationships are robust across different magnitudes)

Question 17

3 possible relationships between three variables?

Answer 17

• Effect of X on Y is mediated by Z: there is an indirect effect between X and Y
• X and Y have common cause Z: Z is the confounder of the relationship between X and Y -> no causal effect of X and Y
• X and Y have common effect Z: Z is the collider caused independently by X and Y
• check notebook for graphs

Question 18

backdoor path?

Answer 18

• any path from X to Y with an arrow pointing towards X
• we need to block all the back-door paths

Question 19

conditioning Z?

Answer 19

• Whether or not to condition on a third variable depends on how
  you think these variables are actually related to each other
• only control fr Z if the arrow is pointing towards x (common cause) -> e.g. if you check correlation for every level of socialeconomic status you won’t find a correlation between electornic equipments and birth control
• you could control for the mediator if you want to observe a direct relationship between X and Y but we usually dont
• conditioning for common effect Z will create spurious relationship between the otherwise independent X and Y variables (collider bias) -> e.g. you are only looking at a subset of the population (NBA players/people you date from tinder) which introduces a correlation that does not exist in the population

Question 20

confounding bias?

Answer 20

confounding variables impacting the found relationship / effect without the experimenter knowing

Question 21

overcontrolling?

Answer 21

• controlling for too many measured variables in one’s study
• one may control for the variable one wishes to measure by accident or downplay the actual effect size
• Purification principle: The mistaken idea that the more control variables are included in a model, the more accurate the estimation of the causal effect is