SSIS Postgraduate Conference, Exeter, 1 May 2015 - PowerPoint Presentation

Slide1

SSIS Postgraduate Conference, Exeter, 1 May 2015

Cause, Causatives, and Theories of Causation

Julian Reiss, DurhamSlide2

BackgroundMain concern: Provide a satisfactory account of causation in the sciences

My own work focuses on the biomedical and social sciences

Previously: detailed examination of the

methods of causal inference in these sciences

Here: taking a closer look at the causal language that is employed in science, taking cancer causation as a main case studySlide3

Background

Two things:

Take scientific language at face value

(Instead of starting with metaphysical intuitions and developing a theory that satisfies the given constraints)

Main target: Any attempt to explicate causation by means of ‘C causes E iff “…”’

‘

Straightjacket Theories of Causation

’

No matter what is on the RHS of the definition

In particular (for instance) whether or not ‘cause’ appearsSlide4

Feature 1: Scientific language is full of causatives such as push, bond, attract, crunch, prolong, dampen, deflateFamiliar from Anscombe and Cartwright

A problem: it’s not clear that you can always translate these into ‘cause +

x’ (buy)

This does constitute a problem for the straightjacket theory: they cannot account for the large majority of causal claims in science

(‘Cause’ itself is used quite rarely!)

Feature 1: Indispensability of Causatives

.

‘To buy’ =

To cause to acquire?

To cause to establish ownership?

To cause to have ownership?

To cause to have ownership by transferring money?

To cause to have ownership by causing someone to have money?Slide5

Straightjacket Theories often necessitate the assumption of further metaphysical principles (e.g., counterfactual theory); e.g.:Independent events

No absence causation

Causal order is identical to temporal order

Feature 2: Scientific language is extremely flexible with respect to the

C’s and E’s that are being causally related or indeed with respect to what is represented by a causal claim

Hypothesis: there is no general metaphysical principle to which one cannot find a counterexample

So ‘

C

φ

s

E

’, where

φ

is a causative, doesn’t work either

A theory of causation must be a theory of

causal claimsFeature 2: Metaphysical Anarchy

C

and

E

must be distinct events––and distinct not only in the sense of nonidentity but also in the sense of nonoverlap and nonimplication. It won’t do to say that my speaking this sentence causes my speaking this sentence, or that my speaking the whole of it causes my speaking the first half of it, or vice versa; or that my speaking it causes my speaking it loudly, or vice versa.Slide6

The third feature is that causatives are polysemousSome causatives can be used both causally and non-causallyWhat they mean depends on context

Therefore: a theory of causation must be a theory of causal claims

in a context

Feature 3: Polysemy

With these considerations in mind, it is essential to determine whether the use of PREPS [potential-reduced exposure products] actually lowers carcinogen dose.

(Hecht 2002, ‘Biomarkers For Investigating Tobacco and Cancer’)Slide7

The ‘Straightjacket Theory of Causation’:

‘C causes

E iff “…”’

Indispensable causativesMetaphysical anarchy

Polysemy

🍸 Interval 🍸

in a context

iff “…”’Slide8

What are the chances of providing truth conditions for a complex statement such as ‘[Claim using φ-causative] in context

K’ iff…?

I have no knock-down argument against it but doubt whether it can be done

Q&A if you want to know more

A pluralist truth-conditional theory?Slide9

The rationale behind this proposal is that so-called ‘monist’ theories of causation according to which, for instance, ‘C causes

E iff P(

E | C.

K) > P(E

| K)’ don’t seem to workProbability raising:

Probability lowering causes

Causes that are connected to their effects via two mechanisms that mutually cancel

Indeterministic causes that do not screen off their effects

Aside: ‘Monist’ Theories of CausationSlide10

Mechanistic connectedness:Causation by absence

Intervention:

Causes on which one cannot intervene ideally

Fragile causal relations

Note: none of the criticisms are absolute; but on ‘the balance of probabilities’ monist theories should be rejected

Aside: ‘Monist’ Theories of CausationSlide11

Idea: while no criterion is true of all causal relations,

some criterion will be true of each

of them

Two-fold problem:Still counterexamples?

The most successful of the alternative theories are realist theories; and realist theories involve either an infinite regress (Baumgartner and Drouet 2013) or further counterexamples

(To assume that the causal relations assumed to be known to hold on the RHS are of the same kind as the one characterised would be implausible under a disjunctive account)

A pluralist truth-conditional theory?

To determine whether X causes Y requires that there be an intervention I on X with respect to Y; to determine whether I is an intervention on X with respect to Y requires I to be a cause of X, which in turn requires that there be a further intervention I´ on I with respect to X; to determine whether I´ is an intervention on I with respect to X requires I´ to be a cause of I, which in turn requires that there be a further intervention I´´ on I´ with respect to I; and so on,

ad nauseam

. Slide12

Note also: The disjunctive account is a non-starter anyway as it makes at best sense of the causal content of these claims – but their causal content is entangled with additional descriptive content

My best shot: ‘[Claim using

φ-causative] in context

K’ iff there is a φ

-mechanism or activityProblems:

Dormative virtue theory

Doesn’t work for abstract causal claims such as ‘smoking causes lung cancer’

Highly contextual; no account of how context picks out appropriate mechanisms, activities and other kinds of causal relations

A pluralist truth-conditional theory?Slide13

An Alternative: Inferentialism

Main idea:

Causation and inference are clearly related

(Justified/accepted) Causal claims, in conjunction with other knowledge such as observations, license inferences to future and past states of affairs

Observations, in conjunction with background knowledge, license inferences to causal claimsProposal: The content of causal claims consists in the inferential network of which they are a part

Essentially: the content of a causal claim is given by the set of propositions from which it follows and those which follow from itSlide14

Inferential systems

More precisely: The content of a causal claim is given by its inferential system

; that is, by the propositions from which an epistemic community is entitled to infer the causal claim, and those the community is

entitled to infer from it

Divides into ‘inferential base’, ‘inferential target’ and causal claim CC itselfInferential base: essentially, the evidence – RCTs, controlled experiments, statements describing experimental design, observational studies, statements describing how confounders and biases are ruled out, background knowledge etc.Slide15

Evidence

I can only sketch the story here

Roughly, ask: under what conditions are we justified in inferring a causal claim from the evidence?

Answer, roughly: when there is a study that uses a reliable method and shows the claim to hold

Experimental methods often achieve reliability by design; but even they have to be checked for confounders, biasesObservational methods achieve reliability to the extent that alternative accounts can be ruled out

All this is highly context dependentSlide16

Inferential Targets

Causal claims are rarely established for their own sake

Rather, for the cash value:

explanations

attributions of blame and praisepredictions

propositions about effective strategies

To determine the content of a causal claim, ask: What is its inferential system?

CC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

EC

Inferential base

Inferential target

TC

TC

TC

TC

TC

TC

TC

Context

‘Inferential system-CC’Slide17

Here is an example of how the account worksWhat is the content of a sentence such as ‘Billy’s throw caused the iPad to shatter?’

Ask: what is the inferential system of the sentence?

Answer

Inferential base: critical observation

That is: observation plus critical background assumptions that are contextually justified

Causal Redundancy

Outcome

Cause

Backup

CauseSlide18

Inferential target (e.g.): Billy is responsible

;

Billy’s throw explains the shattering;

BUT NOT: Had Billy not thrown, the iPad wouldn’t have shattered

Why? We’re entitled to infer the counterfactual only in contexts where there are no backup causes (among other things)!

Causal Redundancy

Lung

Cancer

Smoking

Asbestos

Compen-

sationSlide19

By way of concluding, let me point out how the account deals with the three features about causal language in scienceFeature 1: Indispensability of causatives

It makes no difference between sentences in which ‘cause’ appears and those in which it doesn’t

For any sentence in a scientific publication we can ask, ‘What is the inferential system for this sentence?’

Neither does ‘cause’ vs causative make a principled difference, nor whether the sentence is a causal claim at all (though: the inferential networks of predictive sentences, for instance, may be a lot harder to understand)

ConclusionsSlide20

Feature 2: Metaphysical anarchyAs causation has to do with our reasoning practices and not with what the world is like, anything goes metaphysically speaking

The account, as presented here, can be underwritten by a realist metaphysics (which would mean that a representationalist account would eventually have to be found) or by anti-realism (the option I favour but don’t presuppose)

No general metaphysical principles are assumed

Feature 3: Polysemy

There is no difference-in-principle between causal and non-causal claims; to what extent a claim is causal depends on family resemblance of inferential systems

And family resemblance is much more than hand-waving in this case: we know the typical kinds of proposition in inferential base and target

Conclusions