Causal Mechanism & Prediction v. Explanation

Question 1

what is the a causal logic (or mechanism)?

Answer 1

a set of statements about how or why a cause produces its effect.

it involves a causal chain that connects the cause to the effect:
C = e1 = e2 = e3 = E

it involves assumptions:
-what do we need to believe is true about the world to believe the steps in the logic?
-Ex; in health = democracy, we assume:
1. politicians actually want to stay in power.
2. voters actually vote based on health-related issues as opposed to any number of other things.

Question 2

what is the motivated reasoning in common causal mechanism?

Answer 2

Ex: why do citizens of different partisan stripes see objective facts about the world so differently?

• When we enter “motivated reasoning”:
  1. people seek to satisfy goals in gathering and interpreting information about the world.
  2. Accuracy is just one goals.
  3. seeing the world as we prefer/believe it to be = psychological pleasure!
  4. people are emotionally attached to prior beliefs.

Question 3

why causal mechanisms matter?

Answer 3

1. more complete understanding: tell us how.
2. suggests other causes of the effect:
   C = e1 = e2 = e3 = E
3. helps us develop better predictive models.
4. helps with prescription: sometimes we cant manipulate C. but we can manipulate e2, e2 etc.

Question 3

why causal mechanisms matter?

Answer 3

1. more complete understanding: tell us how.
2. suggests other causes of the effect:
   C = e1 = e2 = e3 = E
3. helps us develop better predictive models.
4. helps with prescription: sometimes we cant manipulate C. but we can manipulate e2, e2 etc.

Question 4

Example: Healthcare

suppose:
(C) allowing private hospitals =
(e1) competition among providers for patient’s business =
(e2) incentives to improve care =
(E) improvements in care!

Answer 4

we could:

manipulate C - allow private hospitals, OR

manipulate e2 - create other incentives to improve care. (e.g. pay doctors based on patients’ health outcomes).

Question 5

what is explanation?

Answer 5

it means causal explanation. it is generally focused on understanding what has already happened.

Question 6

what is explanatory models?

Answer 6

models are applied to data in order to test hypothesis inspired by a causal theory linking cause (x) to effect (y).

Question 7

what is prediction?

Answer 7

it is focused on predicting an outcome (y) in new or future observations, given a set of input values (x).

Question 8

what is prediction models?

Answer 8

it includes any method that produces predictions, regardless of underlying approach.

Question 9

what is the common between explaining and predicting?

Answer 9

1. both represent attempts to understand the world.
2. both require a model.
3. both can stand or fall based on how good that model is, including whether or not it makes reasonable assumptions.
4. both determining “good” means, to a degree, relying on “out-of-sample” verification.

Question 10

what is the difference between explaining and predicting?

Answer 10

1. causation v. association:
   - explanatory modelling intrinsically concerns x causing y.
   -predictive models depends only on x’s association (or correlation) with y.
2. relationship to theory:
   - in explanatory modelling, causes-effect relationship is established beforehand (based on theory) and tested against the data.
   - in predicting modelling, x-y association often emerges out of data analysis.
3. time horizon:
   -predictive modelling is prospective (forward looking), constructed for predicting y on new observations.
   -explanatory modelling is retrospective, and understands values of y already recorded.

Question 11

what is the difference between predictive accuracy vs. sound explanation?

Answer 11

the standard of judgment is different:
1. put differently:
- how good is the model at predicting the outcome we care about.
VS.
-how much evidence is there that there is a causal relationship between supposed “cause” and “effect”.

2. this can matter, because:
   - you can have a good predictive model that has no explanatory power whatsoever.
   -you can have a good explanatory model that dose a terrible job predicting the outcome.

Question 12

causal explanations and predictions?

Answer 12

1. causal explanations necessarily make predictions:
   - the counterfactual is essentially “predicts” the outcome in an alternate reality where cause is absent.
2. but predictive accuracy is not sufficient to ensure the validity of a causal argument.
   -the predictive validity the model has might be entirely coincidental
   -specific cause-and-effect relationship might be “backwards”.
   - Predictor variable may only be connected to outcome via spurious association.

Question 13

WFT rule?

Question 14

what the political scientists think about explanation and predicting?

Answer 14

1. they care more about explanation than they do about prediction.
   - because science is about understanding, which only comes about through building and testing causal theories about the world.
   -because predictive models generally valued for their practical utility.
2. different in terms of emphasis:
   - explanation and prediction are neither inconsistent nor incompatible.