single world intervention graphs (swigs)€¦ · returns as output a swig g(x). each instantiation...
TRANSCRIPT
![Page 1: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/1.jpg)
Single World Intervention Graphs (SWIGs):Unifying the Counterfactual and Graphical Approaches to Causality
Thomas Richardson
Department of Statistics
University of Washington
Joint work with James Robins (Harvard School of Public Health)
Data, Society and Inference Seminar, Stanford
19 May 2014
![Page 2: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/2.jpg)
Outline
Review of counterfactuals and graphsA new unification of graphs and counterfactuals vianode-splitting
I Simple examplesI General procedureI Factorization and Modularity PropertiesI Contrast with Twin Network approach
Further Applications:I Adjustment for ConfoundingI Sequentially Randomized Experiments / Time Dependent
ConfoundingI Dynamic Regimes
Concluding remarks
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 1
![Page 3: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/3.jpg)
The counterfactual framework: philosophy
Hume (1748) An Enquiry Concerning Human Understanding:
We may define a cause to be an object followed by another, andwhere all the objects, similar to the first, are followed by objectssimilar to the second, . . .
. . . where, if the first object had not been the second never hadexisted.
Note: this is not one of the 3(!) causal theories Hume is famous for.Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 2
![Page 4: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/4.jpg)
The potential outcomes framework: crop trialsJerzy Neyman (1923):
To compare v varieties [on m plots] we will consider numbers:
plots︷ ︸︸ ︷U11 . . . U1m
......
Uv1 . . . Uvm
varieties
Uij is crop yield that would be observed if variety i were planted in plot j.
Physical constraints only allow one variety to be planted in a given plot inany given growing season⇒ Observe only one number per col.Sometimes called the ‘Rubin causal model’, owing to Rubin (1974).Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 3
![Page 5: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/5.jpg)
Potential outcomes with binary treatment
For binary treatment X and response Y, we define two potentialoutcome variables:
Y(x = 0): the value of Y that would be observed for a givenunit if assigned X = 0;
Y(x = 1): the value of Y that would be observed for a givenunit if assigned X = 1;
WIll also write these as Y(x0) and Y(x1).Implicit here is the assumption that these outcomes arewell-defined. Specifically:
I Only one version of treatment X = xI No interference between units / Stable Unit Treatment Value
Assumption (SUTVA)
Will use ‘potential outcome’ and ‘counterfactual’synonymously.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 4
![Page 6: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/6.jpg)
Drug Response ‘Types’:
In the simplest case where Y is a binary outcome we have thefollowing 4 types:
Y(x0) Y(x1) Name0 0 Never Recover0 1 Helped1 0 Hurt1 1 Always Recover
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 5
![Page 7: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/7.jpg)
Assignment to Treatments
Unit Potential Outcomes ObservedY(x = 0) Y(x = 1) X Y
1 0 1 12 0 1 03 0 0 14 1 1 15 1 0 0
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 6
![Page 8: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/8.jpg)
Observed Outcomes from Potential Outcomes
Unit Potential Outcomes ObservedY(x = 0) Y(x = 1) X Y
1 0 1 1 12 0 1 0 03 0 0 1 04 1 1 1 15 1 0 0 1
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 7
![Page 9: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/9.jpg)
Potential Outcomes and Missing Data
Unit Potential Outcomes ObservedY(x = 0) Y(x = 1) X Y
1 ? 1 1 12 0 ? 0 03 ? 0 1 04 ? 1 1 15 1 ? 0 1
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 8
![Page 10: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/10.jpg)
Average Causal Effect (ACE) of X on Y
ACE(X→ Y) ≡ E[Y(x1) − Y(x0)]
= p(Helped) − p(Hurt) ∈ [−1, 1]
Thus ACE(X→ Y) is the difference in % recovery ifeveryone treated (X = 1) vs. if noone treated (X = 0).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 9
![Page 11: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/11.jpg)
Identification of the ACE under randomization
If X is assigned randomly then
X ⊥⊥ Y(x0) and X ⊥⊥ Y(x1) (1)
hence
E[Y(x1) − Y(x0)] = E[Y(x1)] − E[Y(x0)]
= E[Y(x1) | X = 1] − E[Y(x0) | X = 0]
= E[Y | X = 1] − E[Y | X = 0].
Thus if (1) holds then ACE(X→ Y) is identified from P(X, Y).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 10
![Page 12: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/12.jpg)
Inference for the ACE without randomizationSuppose that we do not know that X ⊥⊥ Y(x0) and X ⊥⊥ Y(x1).What can be inferred?
X = 0 X = 1Placebo Drug
Y = 0 200 600Y = 1 800 400
What is:
The largest number of people who could be Helped?400 + 200
The smallest number of people who could be Hurt? 0
⇒ Max value of ACE: (200 + 400)/2000 − 0 = 0.3
Similar logic:
⇒ Min value of ACE: 0 − (600 + 800)/2000 = −0.7
In general, bounds on ACE(X→ Y) will always cross zero.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 11
![Page 13: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/13.jpg)
Summary of Counterfactual Approach
In our observed data, for each unit one outcome will be‘actual’; the others will be ‘counterfactual’.
The potential outcome framework allowsCausation to be ‘reduced’ to Missing Data⇒ Conceptual progress!
The ACE is identified if X ⊥⊥ Y(x1) and X ⊥⊥ Y(x0)
Independences implied by Randomization of Treatment.
Ideas are central to Fisher’s Exact Test; also many parts ofexperimental design.
The framework is the basis of many practical causal dataanalyses published in Biostatistics, Econometrics andEpidemiology.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 12
![Page 14: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/14.jpg)
Relating Counterfactuals and ‘do’ notation
Expressions in terms of ‘do’ can be expressed in terms ofcounterfactuals:
P(Y(x) = y) ≡ P(Y = y | do(X = x))
but counterfactual notation is more general.Ex. Distribution of outcomes that would arise among those whotook treatment (X = 1) had counter-to-fact they not receivedtreatment:
P(Y(x = 0) = y | X = 1)
If treatment is randomized, so X ⊥⊥ Y(x = 0) then this equalsP(Y(x = 0) = y), but in an observational study these may bedifferent.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 13
![Page 15: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/15.jpg)
Graphs
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 14
![Page 16: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/16.jpg)
Factorization Associated with a DAGWe associate the following factorization of a joint distribution P(V)with a DAG:
P(V) =∏X∈V
P(X | pa(X))
Example:
A D
B
C
E
P(A,B,C,D,E)
= P(A)× P(B | A)× P(C | A)× P(D | B,C)× P(E | D)
Graphical rule (d-separation) allows independence relations holding in adistribution that factorizes wrt a graph to be ‘read’ from the graph.Ex: C ⊥⊥ B | A D ⊥⊥ A | B,C E ⊥⊥ A,B,C | D.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 15
![Page 17: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/17.jpg)
Graphical Approach to Causality
X Y
No Confounding
X
H
Y
Confounding
Unobserved
Graph intended to represent direct causal relations.
Convention that confounding variables (e.g. H) are always includedon the graph.
Approach originates in the path diagrams introduced by SewallWright in the 1920s.
If X→ Y then X is said to be a parent of Y; Y is child of X.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 16
![Page 18: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/18.jpg)
Graphical Approach to Causality
X Y
No Confounding
Associated factorization:
P(x,y) = P(x)P(y | x)
In the absence of confounding the causal model asserts:
P(Y(x) = y) = P(Y = y | do(X = x)) = P(Y = y | X = x).
here ‘P(y | do(x))’ is defined as the distribution resulting froman intervention (or experiment) where we fix X to x.Q: How does this relate to the counterfactual approach?
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 17
![Page 19: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/19.jpg)
Linking the two approaches
X Y
X ⊥⊥ Y(x0) & X ⊥⊥ Y(x1)
X
H
Y
X 6⊥⊥ Y(x0) or X 6⊥⊥ Y(x1)
Unobserved
Elephant in the room:The variables Y(x0) and Y(x1) do not appear on thesegraphs!!
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 18
![Page 20: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/20.jpg)
Node splitting: Setting X to 0
X Y
P(X= x, Y= y) = P(X= x)P(Y= y | X= x)
⇒ X x = 0 Y(x = 0)
Can now ‘read’ the independence: X ⊥⊥ Y(x=0).Also associate a new factorization:
P (X= x, Y(x=0)= y) = P(X= x)P (Y(x=0)= y)
where:P (Y(x=0)= y) = P(Y= y |X=0).
This last equation links a term in the original factorization to thenew factorization. We term this the ‘modularity assumption’.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 19
![Page 21: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/21.jpg)
Node splitting: Setting X to 1
X Y
P(X= x, Y= y) = P(X= x)P(Y= y | X= x)
⇒ X x = 1 Y(x = 1)
Can now ‘read’ the independence: X ⊥⊥ Y(x=1).Also associate a new factorization:
P (X= x, Y(x=1)= y) = P(X= x)P (Y(x=1)= y)
where:P (Y(x=1)=y) = P(Y=y |X=1).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 20
![Page 22: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/22.jpg)
Marginals represented by SWIGs are identifiedThe SWIG G(x0) represents P(X, Y(x0)).The SWIG G(x1) represents P(X, Y(x1)).Under no confounding these marginals are identified from P(X, Y).In contrast the distribution P(X, Y(x0), Y(x1)) is not identified.Y(x=0) and Y(x=1) are never on the same graph.Although we have:
X ⊥⊥ Y(x=0) and X ⊥⊥ Y(x=1)
we do not assume
X ⊥⊥ Y(x=0), Y(x=1)
Had we tried to construct a single graph containing both Y(x=0)and Y(x=1) this would have been impossible.
⇒ Single-World Intervention Graphs (SWIGs).Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 21
![Page 23: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/23.jpg)
Representing both graphs via a ‘template’
X Y
G
⇒ X x Y(x)
G(x)
Represent both graphs via a template:
Formally the template is a ‘graph valued function’ (not a graph!):
Takes as input a specific value x∗
Returns as output a SWIG G(x∗).
Each instantiation of the template represents a different margin:SWIG G(x0) represents P(X, Y(x0));SWIG G(x1) represents P(X, Y(x1)).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 22
![Page 24: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/24.jpg)
Intuition behind node splitting:(Robins, VanderWeele, Richardson 2007)
Q: How could we identify whether someone would choose to taketreatment, i.e. have X = 1, and at the same time find out whathappens to such a person if they don’t take treatment Y(x = 0)?
A: Consider an experiment in which, whenever a patient isobserved to swallow the drug have X = 1, we instantly interveneby administering a safe ‘emetic’ that causes the pill to beregurgitated before any drug can enter the bloodstream.Since we assume the emetic has no side effects, the patient’srecorded outcome is then Y(x = 0).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 23
![Page 25: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/25.jpg)
Harder Inferential problem
A Z
H B
Y
Query: does this causal graph imply:
Y(a,b) ⊥⊥ B(a) | Z(a),A ?
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 24
![Page 26: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/26.jpg)
Simple solution
A Z
H B
YA
aZ(a)
H
B(a)
b
Y(a,b)
Query does this graph imply:
Y(a,b) ⊥⊥ B(a) | Z(a),A ?
Answer: Yes – applying d-separation to the SWIG on the right wesee that there is no d-connecting path from Y(a,b) given Z(a).More on this shortly...
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 25
![Page 27: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/27.jpg)
Single World Intervention Template Construction (1)Given a graph G, a subset of vertices A = {A1, . . . ,Ak} to be intervenedon, we form G(a) in two steps:
(1) (Node splitting): For every A ∈ A split the node into a randomnode A and a fixed node a:
A
· · ·
· · ·
⇒ A
a
Splitting: Schematic Illustrating the Splitting of Node A
The random half inherits all edges directed into A in G;
The fixed half inherits all edges directed out of A in G.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 26
![Page 28: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/28.jpg)
Single World Intervention Template Construction (2)
(2) Relabel descendants of fixed nodes:
a ⇒A
B C
D
FE
X
T
Y
Z
· · ·
· · ·
· · ·
· · ·
a
A(. . .)
B(a, . . .) C(a, . . .)
D(a, . . .)
F(a, . . .)E(a, . . .)
X(. . .)
T(. . .)
Y(. . .)
Z(. . .)
· · ·
· · ·
· · ·
· · ·
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 27
![Page 29: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/29.jpg)
Single World Intervention Graph
A Single World Intervention Graph (SWIG) G(a∗) is obtained fromthe Template G(a) by simply substituting specific values a∗ for thevariables a in G(a);
For example, we replace G(x) with G(x=0).
Resulting SWIG G(x) contains variables V(x) and represents thejoint: P(V(x))
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 28
![Page 30: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/30.jpg)
Factorization and Modularity
Original graph G : observed distribution P(V)SWIG G(a) : counterfactual distribution P(V(a))
Factorization of counterfactual variables: Distribution P(V(a)) overthe variables in G(a) factorizes with respect to the SWIG G(a)(ignoring fixed nodes):
Modularity: P(V(a)) and P(V) are linked as follows:The conditional density associated with Y(aY) in G(a) is just theconditional density associated with Y in G after substituting ai for anyAi ∈ A that is a parent of Y.
Consequence: if P(V) is observed then P(V(a)) is identified.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 29
![Page 31: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/31.jpg)
Applying d-separation to the graph G(a)
In G(a) if subsets B(a) and C(a) of random nodes are d-separated byD(a) in conjunction with the fixed nodes a, then B(a) and C(a) areconditionally independent given D(a) in the associated distributionP(V(a)).
B(a) is d-separated from C(a) given D(a) ∪ a in G(a) (2)
⇒ B(a) ⊥⊥ C(a) | D(a) [P(V(a))].
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 30
![Page 32: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/32.jpg)
Inferential Problem Redux:
A Z
H B
YA
aZ(a)
H
B(a)
b
Y(a,b)
Pearl (2009), Ex. 11.3.3, claims the causal DAG above does not imply:
Y(a,b) ⊥⊥ B | Z,A = a. (3)
The SWIG shows that (3) does hold; Pearl is incorrect.Specifically, we see from the SWIG:
Y(a,b) ⊥⊥ B(a) | Z(a),A (4)
⇒ Y(a,b) ⊥⊥ B(a) | Z(a),A = a (5)
This last condition is then equivalent to (3) via consistency.(Pearl infers a claim of Robins is false since if true then (3) would hold).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 31
![Page 33: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/33.jpg)
Pearl’s twin network for the same problem
A
Z
H
B
Y
A
Z
H
B
Y
a
Z(a,b)
H(a,b)
b
Y(a,b)
UZ
UH
UY
The twin network fails to reveal that Y(a,b) ⊥⊥ B | Z,A = a.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 32
![Page 34: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/34.jpg)
Pearl’s twin network for the same problem
A
Z
H
B
Y
A
Z
H
B
Y
a
Z(a,b)
H(a,b)
b
Y(a,b)
UZ
UH
UY
The twin network fails to reveal that Y(a,b) ⊥⊥ B | Z,A = a.This ‘extra’ independence holds in spite of d-connection because (byconsistency) when A = a, then Z = Z(a) = Z(a,b).Note that Y(a,b) 6⊥⊥ B | Z,A 6= a.
Shpitser & Pearl (2008) introduce a pre-processing step to address this.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 33
![Page 35: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/35.jpg)
Adjustment for Confounding
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 34
![Page 36: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/36.jpg)
Adjusting for confounding
X Y
L
X x Y(x)
L
Here we can read directly from the template that
X ⊥⊥ Y(x) | L.
It follows that:
P(Y(x)=y) =∑l
P(Y=y | L= l,X= x)P(L= l). (6)
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 35
![Page 37: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/37.jpg)
Adjusting for confounding
X Y
L
X x Y(x)
L
X ⊥⊥ Y(x) | L.
Proof of identification:
P[Y(x) = y] =∑l
P[Y(x) = y | L = l]P(L = l)
=∑l
P[Y(x) = y | L = l,X = x]P(L = l) indep
=∑l
P[Y = y | L = l,X = x]P(L = l)
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 36
![Page 38: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/38.jpg)
Multiple Treatments
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 37
![Page 39: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/39.jpg)
Sequentially randomized experiment (I)
A B C D
H
A and C are treatments;
H is unobserved;
B is a time varying confounder;
D is the final response;
Treatment C is assigned randomly conditional on the observedhistory, A and B;
Want to know P(D(a, c)).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 38
![Page 40: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/40.jpg)
Sequentially randomized experiment (I)
A B C D
H
If the following holds:
A ⊥⊥ D(a, c)
C(a) ⊥⊥ D(a, c) | B(a),A
General result of Robins (1986) then implies:
P(D(a, c)=d) =∑b
P(B=b | A= a)P(D=d | A= a,B=b,C= c).
Does it??
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 39
![Page 41: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/41.jpg)
Sequentially randomized experiment (II)
A a B(a) C(a) c D(a, c)
H
d-separation:
A ⊥⊥ D(a, c)
C(a) ⊥⊥ D(a, c) | B(a),A
General result of Robins (1986) then implies:
P(D(a, c)=d) =∑b
P(B=b | A= a)P(D=d | A= a,B=b,C= c).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 40
![Page 42: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/42.jpg)
Multi-network approach
A B C D
H
UHUB
UC
UD
a B(a) C(a) D(a)
H(a)
a B(a, c) c D(a, c)
H(a, c)
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 41
![Page 43: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/43.jpg)
Connection to Pearl’s do-calculus
Factorization and modularity are sufficient to imply all of theidentification results that hold in the do-calculus of Pearl (1995);see also Spirtes et al. (1993):
P(Y = y | do(A = a)) is identified ⇔ P(Y(a) = y) is identified.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 42
![Page 44: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/44.jpg)
Dynamic regimes
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 43
![Page 45: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/45.jpg)
Dynamic regimes
A1
a1
L(a1) A2(a1)
a2
Y(a1,a2)
H2H1
A1
A+1 (g)
L(g) A2(g)
A+2 (g)
Y(g)
H2H1
P(Y(g)) is identified.Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 44
![Page 46: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/46.jpg)
Dynamic regimes
A1
a1
L(a1) A2(a1)
a2
Y(a1,a2)
H2H1
A1
A+1 (g)
L(g) A2(g)
A+2 (g)
Y(g)
H2H1
P(Y(g)) is not identified.Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 45
![Page 47: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/47.jpg)
Conclusion: Eliminating a false trichotomy
Previously the only approach to unifying counterfactuals and graphs wasPearl’s approach via Non-Parametric Structural Equation Models withIndependent Errors:
This gave causal modelers three options:
Use graphs, and not counterfactuals (Dawid).
Use counterfactuals, and not graphs (many Statisticians).
Use both graphs and counterfactuals, but be forced to make a lot ofadditional assumptions that are:
I not experimentally testable (even in principle);I not necessary for most identification results.
SWIGs show that one can use graphs and counterfactuals without beingforced to take on additional assumptions.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 46
![Page 48: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/48.jpg)
Summary and Extensions
SWIGs provide a simple way to unify graphs andcounterfactuals via node-splitting
The approach works via linking the factorizations associatedwith the two graphs.
The new graph represents a counterfactual distribution that isidentified from the distribution in the original DAG.
This provides a language that allows counterfactual andgraphical people to communicate.
(Not covered) The approach also provides a way to combineinformation on the absence of individual and population leveldirect effects.
(Not covered) Also allows to formulate models whereinterventions on only some variables are well-defined.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 47
![Page 49: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/49.jpg)
Thank You!
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 48
![Page 50: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/50.jpg)
ReferencesPearl, J. Causality (Second ed.). Cambridge, UK: CambridgeUniversity Press, 2009.
Richardson, TS, Robins, JM. Single World Intervention Graphs.CSSS Technical Report No. 128http://www.csss.washington.edu/Papers/wp128.pdf, 2013.
Robins, JM A new approach to causal inference in mortality studieswith sustained exposure periods applications to control of thehealthy worker survivor effect. Mathematical Modeling 7,1393–1512, 1986.
Robins, JM, VanderWeele, TJ, Richardson TS. Discussion of“Causal effects in the presence of non compliance a latent variableinterpretation by Forcina, A. Metron LXIV (3), 288–298, 2007.
Shpitser, I, Pearl, J. What counterfactuals can be tested. Journal ofMachine Learning Research 9, 1941–1979, 2008.
Spirtes, P, Glymour, C, Scheines R. Causation, Prediction andSearch. Lecture Notes in Statistics 81, Springer-Verlag.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 49
![Page 51: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/51.jpg)
Assuming Independent Errors and
Cross-World Independence
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 50
![Page 52: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/52.jpg)
Relating Counterfactuals and Structural EquationsPotential outcomes can be seen as a different notation forNon-Parametric Structural Equation Models (NPSEMs): Example:
X→ Y.
NPSEM formulation: Y = f(X, εY)
Potential outcome formulation: Y(x) = f(x, εY)
Two important caveats:
NPSEMs typically assume all variables are seen as beingsubject to well-defined interventions (not so with potentialoutcomes)Pearl’s approach to unifying graphs and counterfactualssimply associates with a DAG the counterfactual modelcorresponding to an NPSEMs with Independent Errors(NPSEM-IEs) with DAGs.Pearl: DAGs and Potential Outcomes are ‘equivalent theories’.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 51
![Page 53: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/53.jpg)
Mediation graphIntervention on X and M:
M YX M(x) mX x Y(x, m)⇒
d-separation in the SWIG gives:
X ⊥⊥ M(x) ⊥⊥ Y(x, m), for x, m ∈ {0, 1}
Pearl associates additional independence relations with this DAG
Y(x1,m) ⊥⊥ M(x0),X
Y(x0,m) ⊥⊥ M(x1),X
equivalent to assuming independent errors, εX ⊥⊥ εM ⊥⊥ εY .
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 52
![Page 54: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/54.jpg)
Pure Direct Effect
Pure (aka Natural) Direct Effect (PDE): Change in Y had X beendifferent, but M fixed at the value it would have taken had X notbeen changed:
PDE ≡ Y(x1,M(x0)) − Y(x0,M(x0)).
Legal motivation [from Pearl (2000)]:
“The central question in any employment-discrimination case is whetherthe employer would have taken the same action had the employee beenof a different race (age, sex, religion, national origin etc.) and everythingelse had been the same.” (Carson versus Bethlehem Steel Corp., 70FEP Cases 921, 7th Cir. (1996)).
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 53
![Page 55: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/55.jpg)
Decomposition
PDE also allows non-parametric decomposition of Total Effect(ACE) into direct (PDE) and indirect (TIE) pieces.
PDE ≡ E [Y(1,M(0))] − E [Y(0)]
TIE ≡ E [Y (1,M(1)) − Y (1,M(0))]
TIE+ PDE ≡ E [Y(1)] − E [Y(0)] ≡ ACE(X→ Y)
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 54
![Page 56: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/56.jpg)
Pearl’s identification claim
M YX
Pearl (2001) shows that under the NPSEM with independent errorsassociated with the above graph:
the PDE is identified (!) by the following mediation formula:
PDEmed =∑m
[E[Y|x1,m] − E[Y|x0,m]]P(m|x0)
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 55
![Page 57: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/57.jpg)
Critique of PDE: Hypothetical Case Study
Observational data on three variables:
X- treatment: cigarette cessation
M intermediate: blood pressure at 1 year, high or low
Y outcome: say CHD by 2 years
Observed data (X,M, Y) on each of n subjects.
All binary
X randomly assigned
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 56
![Page 58: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/58.jpg)
Hypothetical Study (I): X randomized
Y = 0 Y = 1 Total P(Y=1 |m, x)
M = 0 1500 500 2000 0.25X = 0
M = 1 1200 800 2000 0.40
M = 0 948 252 1200 0.21X = 1
M = 1 1568 1232 2800 0.44
A researcher, Prof H wishes to apply the mediation formula toestimate the PDE.Prof H believes that there is no confounding, so that Pearl’sNPSEM-IE holds, but his post-doc, Dr L is skeptical.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 57
![Page 59: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/59.jpg)
Hypothetical Study (II): X and M Randomized
To try to address Dr L’s concerns, Prof H carries out animalintervention studies.
Y = 0 Y = 1 Total P(Y(m, x)=1)M = 0 750 250 1000 0.25
X = 0M = 1 600 400 1000 0.40
M = 0 790 210 1000 0.21X = 1
M = 1 560 440 1000 0.44
As we see: P(Y(m, x)=1) = P(Y=1 |m, x);Prof H is now convinced: ‘What other experiment could I do ?’
He applies the mediation formula, yielding PDEmed
= 0.Conclusion: No direct effect of X on Y.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 58
![Page 60: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/60.jpg)
Failure of the mediation formula
Under the true generating process, the true value of the PDE is:
PDE = 0.153 6= PDEmed
= 0
Prof H’s conclusion was completely wrong!
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 59
![Page 61: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/61.jpg)
Why did the mediation formula go wrong?
Dr L was right – there was a confounder:
M YX
H
but. . . it had a special structure so that:
Y ⊥⊥ H |M,X = 0 and M ⊥⊥ H | X = 1
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 60
![Page 62: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/62.jpg)
Why did the mediation formula go wrong?Dr L was right – there was a confounder:
M YX
H
but. . . it had a special structure so that:
Y ⊥⊥ H |M,X = 0 and M ⊥⊥ H | X = 1
M Y
HX = 0
M Y
HX = 1
The confounding undetectable by any intervention on X and/or M.
Pearl: Onus is on the researcher to be sure there is no confounding (=independent errors).
Causation should precede intervention.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 61
![Page 63: Single World Intervention Graphs (SWIGs)€¦ · Returns as output a SWIG G(x). Each instantiation of the template represents a different margin: SWIG G (x 0) represents P(X,Y 0));](https://reader034.vdocuments.mx/reader034/viewer/2022042218/5ec3ce5cccd7ff46906c3774/html5/thumbnails/63.jpg)
Summary of critique of Independent Error Assumption
The independent error assumption cannot be checked by anyrandomized experiment on the variables in the graph.
⇒ Connection between experimental interventions and potentialoutcomes, established by Neyman has been severed;
⇒ Theories in Social and Medical sciences are not detailed enough tosupport the independent error assumption.
Thomas Richardson DSI Seminar, Stanford 19 May 2014 Slide 62