Association & Causation

Question 1

What are variables?

Answer 1

Variables are usually what we measure or manipulate

Question 2

What are the 4 types of variables?

Answer 2

-Dependent
-Independent
-Confounding/extraneous
-Control

Question 3

What is association?

Answer 3

Typically = statistical relationship between two variables

Question 4

What is correlation?

Answer 4

Refers to extent pair of variables are linearly related

Question 5

What can association & causation be useful for?

Answer 5

-Can allow us to infer potential causal relationships (maybe..)
-Predictive relationships more so
–> BUT need v. solid empirical design & justification

Question 6

What are spurious correlations?

Answer 6

Relationship between 2 otherwise completely unrelated variables
–> Variables can be correlated or associated without actually being related

Question 7

What are 3rd variable problems?

Answer 7

Variables might be linked as a third problem is causing both (confounds the data)

Question 8

What is Anscombe’s quartet?

Answer 8

-Statistical relationship is not indicative of distribution
-Careful about summary statistics

Question 9

What is the most important thing to remember about correlation?

Answer 9

Correlation does NOT imply causation!!!

Question 10

Fill in the table about correlations.

Answer 10

Question 11

What is cause & effect for association?

Answer 11

Exposure associated with outcome

Question 12

What is cause & effect for causation?

Answer 12

Exposure leads to outcome

Question 13

Fill in this common single comparison test chart - for causation.

Answer 13

Question 14

What are some of the issues surrounding causation?

Answer 14

-Finding a ‘result’ does not always reflect reality
-Comes from good design

Question 15

What are confounding variables?

Answer 15

-Influences exposure & outcome
-Should control for
—> to ensure exposure causes outcome & not confounder

Question 16

What are colliders?

Answer 16

-Exposure & outcome independently influence a 3rd variable
-Can obscure real or reveal false associations

Question 17

What is causation in terms of epidemiology?

Answer 17

Considers causes of disease

Question 18

What is a web of causation?

Answer 18

Shows multiple different factors contributing to disease

Question 19

What is the Bradford Hill Criteria for causation?

Answer 19

9 principles used to establish causal inference

Question 20

What is the sufficient-component cause model?

Answer 20

Causal pies

Question 21

Describe features of webs of causation.

Answer 21

-Multi-causal
-Non-linear
-Accommodates chronic disease
-Includes social factors

Question 22

Give the 9 guidelines of the Bradford Hill Criteria.

Answer 22

1. Strength
2. Consistency
3. Specificity
4. Temporality
5. Biological gradient
6. Plausibility
7. Coherence
8. Experiment
9. Analogy

—> are guidelines to determine if association is causal

Question 23

What is meant by 1- Strength - effect size of the Bradford Hill Criteria?

Answer 23

-Larger association = likely causal
-i.e. relative risk
-Larger associations unlikely to be due to
confounding and bias
-Weak associations are difficult to justify
-i.e. smoking strongly linked to lung cancer

Question 24

What is meant by 2- Consistency of the Bradford Hill Criteria?

Answer 24

-Reproducible
-Observed repeatedly across different samples
& designs
-i.e. smoking repeatedly linked to lung cancer

Question 25

What is meant by 3- Specificity of the Bradford Hill Criteria?

Answer 25

-Single exposure should cause single disease
-Specific population at specific site with disease
and no alt explanation likely
More specific an association = likely causal
-However, lung cancer results from more than
just smoking

Question 26

What is meant by 4- Temporality of the Bradford Hill Criteria?

Answer 26

-Causal factor must precede disease in time
-Rather obvious & always agreed upon
-i.e. cohort of smokers are recruited when
healthy and follows until subsequent lung
cancer

Question 27

What is meant by 5- Biological Gradient of the Bradford Hill Criteria?

Answer 27

-Dose response relationship between
exposure & disease
-More exposure = more risk of disease
-i.e. lung cancer rates rise with number of
cigarettes smoked

Question 28

What is meant by 7- Coherence of the Bradford Hill Criteria?

Answer 28

-Model exists to explain association
-How plausible & coherent is it
-i.e. cigarettes contain many carcinogenic
substances

Question 29

What is meant by 8- Experiment of the Bradford Hill Criteria?

Answer 29

-Experiment led intervention that modifies
exposure should result in less disease
-Treatments should work in other words
-i.e. smoking cessation programs result in
lower lung cancer rates

Question 30

What is meant by 9- Analogy of the Bradford Hill Criteria?

Answer 30

-Similar relationship between another exposure
and disease?

Question 31

What is the sufficient-component cause model?

Answer 31

-Similar to web of causation
-Shown by causal pie
-The sufficient cause is that - if a minimum no. of factors are present within a causal mechanism then this will inevitably result in disease
-Many factors play a part in having a sufficient cause on disease (no single factor)

Question 32

Explain HIV using the sufficient-component cause model?

Answer 32

-“Complete causal mechanism that inevitably produces disease” - i.e., is no way around it such as in HIV
-Is a set of factors (rather than a single one)

Question 33

What is collider bias?

Answer 33

Occurs when an exposure & outcome (or factors causing these) each influence a common 3rd variable & that variable or collider is controlled for by design or analysis

Question 34

What is Simpson’s paradox?

Answer 34

Phenomenon in probability and statistics in which a trend appears in several groups of data but disappears or reverses when the groups are combined