Formalism or pluralism? A reply to commentaries on ‘Causality and causal inference in epidemiology’
Year: 2016
Type: letter
Abstract: We thank the commentators on our paper.1–5 We have learned a great deal from this exchange. These papers do much to clarify previous statements, and show that there is much that we agree on. They also show that there remain important points of disagreement. In this response we do not expect that we can resolve all these points of disagreement; rather, our intention is to clarify them and our position in relation to them. We approach these commentaries by identifying what we take to be an authoritative version of the Potential Outcomes Approach (POA), which we believe that most of our interlocutors would accept, given what they have written. In addition, we define the Restricted Potential Outcomes Approach (RPOA), which (as we have indicated from the outset) is characterized by a further commitment concerning the role of interventions (whether or not humanly feasible). We have concerns about certain aspects of the POA, but we regard the RPOA as more seriously mistaken. Our approach in what follows is to identify, as clearly as we can, our points of agreement and disagreement on the substantive elements of the POA and RPOA. Having dealt with these ‘hard’ theoretical issues, we seek to relate these to the various ‘soft’ downstream issues relating to pedagogy, emphasis, prioritization of study designs and neglect of important exposures. These ‘soft’ aspects of the debate are by no means less important—in fact they may even be more important for the characteristics that epidemiology takes on as it develops. But if clarity on the substantive points of agreement and disagreement can be achieved before discussing the ‘soft’ downstream issues, then those discussions will be more fair and fruitful. The version of the POA that we believe emerges from the exchange has the following elements. Counterfactual dependence of E on C is not necessary for C to cause E, but it is sufficient (POA’s Basic Metaphysical Stance). Sufficient evidential conditions currently exist for attributing the counterfactual dependence of E on C, but necessary conditions currently do not; the POA identifies some (but not all) of these sufficient conditions (POA’s Basic Epistemological Stance). Causal inference includes two distinct aspects: causal identification, in which the truth value of a claim of the form ‘C causes E’ is determined; and quantitative causal estimation, in which a numerical value n is estimated for a claim of the form ‘C has n effect on E’ (the Identification/Estimation Distinction). Adequately well-defined counterfactual contrasts are necessary for giving meaning to quantitative estimates of causal effect (POA’s Semantic Stance on Estimation). These four elements characterize what we regard as a ‘standard’ version of the POA. Different authors emphasize different elements of these in their responses to our original paper and in their writings preceding this discussion. However, we believe that most of our interlocutors accept these elements. In his considered and especially clear response,4 Tyler VanderWeele touches on almost all of these, and is particularly at pains to emphasize that counterfactual dependence may be sufficient for causation even if it is not necessary, part of (i), and that there is an important difference between causal identification and quantitative estimation, as stated in point (iii). Hernán has emphasized (iv) in particular, as well as point (v) below.6 Perhaps because of the POA’s resonance with the theoretical basis of randomized controlled trials, some have taken it implicitly to involve interventions. This is not a logically essential element of PO theory (as we pointed out in our original paper), and it is certainly not inherent in the counterfactual approach (as Pearl and others make clear7). This gives rise to an additional fifth element which, taken with (i)-(iv), defines the Restricted Potential Outcomes Approach (RPOA). (v) Counterfactual contrasts are adequately well-defined if and only if we can specify a corresponding adequately well-defined intervention on the putative cause, by which the counterfactual contrast would be (or would have been) brought about. (RPOA’s Restriction to Interventions). This additional clause imposes both a necessary and a sufficient condition on adequately well-defined counterfactual contrasts. Hernán is particularly closely associated with (v), having made statements to the effect that causal questions are well-defined when interventions are well-defined,6 and having critiqued the work of epidemiologists who do not define interventions.8 However, because the difference between (i)-(iv) and (i)-(v) is not generally appreciated, all of our interlocutors have at some stage assumed elements (i)-(v): that is, they have assumed or committed themselves to the RPOA approach, as we show in Section B1 below. It is the additional element involving interventions that has given rise to the most debate, since ‘intervention’ has not been defined. In particular, there has been discussion about interventions be humanly or However, there are with (v), even the of is, even the that ‘intervention’ not We clear on this in our original paper and we this below. In of this we of our are as We accept We because we do not believe that there are logically sufficient for Causal inference is not means in which we accept but regard it as because we of causal inference before the POA, or in We accept for the of the to be to We accept (iv) that are in a and that it is not with the claim that well-defined contrasts are sufficient for causal We (v) as a is a when this is not and a in the discussion. there is no to on interventions when to specify counterfactual contrasts. indicated in our original interventions are not sufficient for well-defined counterfactual we may not in a intervention is and interventions are not necessary for well-defined counterfactual contrasts. In of the paper we to more In B1 we the which we have of we in which VanderWeele and Hernán to the RPOA and ‘intervention’ as We to do because what and is not to the substantive at However, of our concerns with the RPOA concerns of there has been a to define and a to between different POA, humanly what has been in and commentaries is not to what before we our original We believe that these is important for the of the In Section we that is a and we our that the POA not the of the of that we and from which we believe epidemiology as a has and will to We regard the POA’s Basic Metaphysical Stance as (i), epidemiologists are not with as they are to take a in to on with their We take this to be on all in the The most for epidemiologists to given that is not their is to the the is the of if that for epidemiologists with the The with in that do not on their as when the not on the because there a who would have the if the not a may have for the with Tyler VanderWeele in that if and and on then causes of in this from the of the role of in the truth conditions of We accept (i), for these VanderWeele as a that are well sufficient conditions for attributing causation and the of sufficient We and the claim as We can have certain of which are sufficient for to and the POA or (or of the POA sufficient conditions to counterfactual and by for a counterfactual is for We that the be from counterfactual dependence to and not from to because it is of the POA that it not from to causation that do not specify about counterfactual dependence it not the of as some of the commentaries have We take this to be on all The by is clearly by VanderWeele when that there are sufficient conditions for attributing It that this is those by the POA. We that there may be at our to accept which may even be in some as or as the of an of we do not accept the in and in The meaning of point on the of in be that is that is the we that logically sufficient conditions exist for the of causation or for the of counterfactual because all are the of an or has been at great pains to of of the of that a is only as well as the which the of it The may be given the and in this the causal is from the the the are themselves These have and the of all from the of that is necessary to different the that the are in the about and of and more of this point is the much of a effect (or a in randomized controlled the have a of because the is by or for This may be the of the of the of this is not because the the that have to make that more of these is with the These of as and make on a basis about what to what to when to and when to and these, can the the a causal a given a of but no and the the of is no the of the that are to can be even at some basis may a in that in is from an point of and but do not the of a These points show that randomized controlled do not sufficient conditions for causal the sufficient condition includes a of which be by the of the no These points to of sufficient conditions for the of from In particular, they to the of the approach by the POA and the associated with We take the point of that to is but we are not that the of the POA do in fact epidemiologists to all the The that they to the be since it the fact that remain we that even a of our clear can be Our is not with that but with the to that by some of our more we have the for a theory and We the as to They that inference to the the of these by which they to that which questions to However, they believe that certain and is no in our that a theory is to may seek and for this a for causal inference is But this not that it is we have because we we Causal do not from a of and (as these are themselves not of the may be that is not and dependence on not We are also about who may what it is that causal inference has been and it is not that the as it may is only as as the that then causal will they We that VanderWeele and others would agree that the POA not to a logically sufficient of conditions for from The in which be is the the more in the to a causal or for an inference to a causal the or we can this The claim that there are sufficient conditions for causal inference is However, it is not It to inference is for or we can make causal This is and has been since the It a inference is given our to or even to However, it certainly that we have made causal on We that the that we can make causal is much what POA would and is also what would We that the intention is to that the POA has a of logically sufficient before we not have a We do not accept We agree that the POA has to our causal inference by to the and of sufficient conditions for causal but we that it has not logically sufficient and that the sufficient conditions it has a of conditions of this we a that of this may a of The fact that there are sufficient conditions for causal inference certainly has not been made by the of the POA or what POA would to in response is that POA a or an in our to make causal We to agree and we made a claim in our original that the POA a of and that may causal inference in However, this is not to the of and we that the is in the of discussions of the POA and some of the associated and as well as a of of this debate has been the or at the of a between two of causal Causal identification when we some as the causes of the Causal when we seek to the of some to an of with the a of for to the from a a theory of and who emphasized the of contrasts in causal and in causal of his with a make a Our of the between identification and is that it a for the POA to that counterfactual dependence of certain is necessary for certain of causal even it is not necessary for causal inference in We are it is in out a in which the POA may seek to necessary conditions on a certain of causal those relating to the quantitative of causal we our that even causal is the of an involving the of of from of the POA is an that it is not or at not to make quantitative of causal the counterfactual contrast which the of the putative cause to the effect is This is a the of certain but it has for what we can be (or and be can they be the of inference an inference a is to the truth of that (iv) that putative causal are if they to causal adequately counterfactual contrasts. We regard (iv) as a and important and a to Perhaps it to be but if is to specify much of a difference to be clear about the which the difference is However, we do two points of about we that is in a and not as the of the of closely associated with the and not be We remain of the that there are not logically sufficient conditions for Adequately contrasts are necessary to make of estimates of of However, the fact that we have adequately our contrasts not by that our causal is in or in the causal it to and from is a point we to in below. is the that be in the of well-defined This is the element of the in contrast with the POA, and the that we believe is most We the of the as we emphasize in our to a by VanderWeele the has not been by those who on In ‘intervention’ a of some if we have this at the with which the ‘intervention’ has been The commentaries that ‘intervention’ not to be as to humanly interventions. we the claim that interventions are necessary or sufficient for the of counterfactual contrasts. In our original we with the RPOA concerns on the of a humanly or we made clear that our a commitment to of the has our of the of intervention in as it in the on our of humanly interventions. We take this to seek to further clarify we that interventions or not have an essential or in the POA. In our original paper we with the role to interventions by the an intervention is not sufficient for adequately a causal may not be to in an intervention is and an intervention is not necessary for adequately a causal The point is by the interventions in Hernán and of includes different of which may have different on that an intervention not all the to a The intervention to be adequately by the of the RPOA has the of a and not The point is a if interventions necessary, we would be in because we would be to the study we about to on this or not the In of a we may not different of we may not (or may more which of the of The point is that interventions is, in not necessary for counterfactual contrasts. an intervention to a causal is not the as a different value for that an intervention to about a counterfactual value of the causal is not necessary for a counterfactual value of the causal This is clear even the POA. and Pearl essential of causation is the to to in of those We by our original to the which have not been in of the We that this may be because of a to distinct the causes of an of the effect of the on the of and the on the of of interventions on the may have as may have on and may to different interventions which may have different on both their on and their different on of this means that is a as VanderWeele it is a cause of which can be by and can be by interventions. it be that different of different and that we study more of but this to most exposures. We are for the effect on of a of or the effect on of This not on interventions. even if we are to on to the estimates we can of the of remain and to what can be achieved by the causal effect of is important in this is the only effect that can be estimated in on interventions is not we can (or to the causal of and we may also to (or to the of interventions to The is in the is Hernán and these issues in their They point out that it is to what to means a counterfactual contrast in which the is However, their contrast is in which an intervention is on an to In the counterfactual contrast of in the effect of is not between a with much and of has been by a The is the counterfactual contrast for the effect of an intervention on The counterfactual contrast for the effect of is by the if there (or more if this of a of in the that the effect of is by if not if they the effect of giving not the effect of In for all about that we believe to be it is a that it be a intervention interventions for have to and The original interventions on epidemiologists to believe that a cause, to for that the for means to which to the of what of causal is will on the of the in the it is the effect of a cause interventions be all of the is some counterfactual contrasts as in the can be as or when and of counterfactual contrasts may intervention (or a for because the of the counterfactual contrast is to be as by in a or counterfactual contrasts will be in between these there is a between and the of an is which at the part of the causal and has to do with the original of a at causes in that the of the effect of an with the effect of an intervention on the is particularly when important and The effect of a on is not well by the effect of a from the study of the of an intervention an for further Having out our on the in the of the debate, we to various ‘soft’ issues which, as stated may be more important in the of the from the various commentators is that we have a We have an of the RPOA approach, and then this We this point because it is to do but we we We are with the that we to some of the more of the we are that have been taken out of or of the commentaries is our as if the an In the that we have taken from VanderWeele and others are not or taken out of but have in papers on these We some of them to that our interlocutors have committed themselves to the RPOA and have assumed ‘intervention’ as their different point of in the a we are with a VanderWeele has the giving a role to for has been debate as to if is by the of The causal inference has generally of causal as a between or in the causal inference the position is taken that it is to of a contrast of counterfactual only to the that we can specify an VanderWeele has also an of ‘intervention’ that is to humanly in this follows from the this position is associated with the causation has to this of or in more However, is not we can and the associated counterfactual generally as The of what would a have been they been a to It is that not the of in when do to a as in the of we a causal of the by much a be by on a different which may be more This with are these authors for is the this a there with attributing to do we to to a counterfactual contrast can be in of This of to can be considered as a cause it is randomized at and may or may not be considered a cause, be (or at causal effect be the of the of this can be These are to more is about the for on if there is and from the theoretical point of of causal the It is much more to that causes (but are not well-defined counterfactual and that we can to the causal effect of contrast to for It is then to the of causes for which interventions can be if VanderWeele this point in his with different from approach is with a of causal questions that can be as a contrast of This is the to our that this point has been in the various discussions of this We have also the that there are no between our and those of are that it is all a in a We believe that it is even what the In particular, we the claim that it is necessary to specify an intervention or when to a causal on the the of interventions as for causal be much less in the of well the to out all causal not the (or of causal But even if this is as to quantitative of causal effect to our only in the commentaries on our original paper), we do not accept this We may accept an of counterfactual contrasts as necessary for quantitative but we do not regard the of interventions as necessary for counterfactual contrasts. In the of an intervention to be the as the of a counterfactual of causal effect well-defined counterfactual that is, a about the the two are and RPOA. We that this of the two is of the we have that we do not We that the difference is there is an to The is then is the point of a causal effect that is not well The will not be to who will be to relate it to a or who will be to it interventions.8 we that we have a and we also that we agree with our interlocutors on all as these are in previous our the by and the by VanderWeele are the on which it has been that the approach not all causes of and not a and theory of causal We agree with these and are for the may be of on causal inference and on the We believe that is not a but a theoretical that a certain is causal to a even if it is a for we believe that the of is not a and that is a a of of are We that all the to this debate have that they accept that a of of from is But the points about of for causal inference in with this the of the is on inference a or of for different of are if The of a cause of on but there are some in which epidemiologists have been to has been their are particular, and not theoretical this means that have a particularly and even It is the between and at the and which an of and this may in the of that are for in a is a cause of or for the of a But this at the of a with and we would that also be considered as a of and that estimates from a may not to because of different of effect there will almost be effect of the of difference or or the or of the in do not even estimates of the of causation in the in the most can is that if an causes an then expect it to also cause in but not even the of causal is not an quantitative we that it is to from even if it is a randomized It is the of the that to the of is a a theoretical and as of counterfactual contrasts or of interventions are and but are in a and even these are not but are They are not or from but from and the inference a of and in the an from a is not a or can be by of a of It is a theoretical even when in quantitative estimates for with theoretical we causal of causal as the of an all the a study can be in some but not because of to the a role in the of we have emphasized a of be taken on be the in a for our interlocutors have some of these we that there is a difference in between this approach to causal inference and that of the this this a in a In our it is because it two different of what epidemiology is, it be and it We believe that epidemiology is a with a to which we can a of In the we a of epidemiology as a of which are then to those questions which can be with these identifies causes of in It is part of a with and from to that to and the them. means as to the This can and to they may not (or may have different in We quantitative causal as an important part of but it is We the of the and of the but we do not believe that this is what epidemiology has been what it is, what it to Our has been to all of the different causal questions that epidemiologists may to and to the issues in them. and with a of the and of in and questions about this are the causes of or is In to these then a that by a of that do not are and those that are and the questions that can and the to all causal would the However, in of of We that an approach to in which we is not only but We agree with and that the essential of causation is to in of this is In all of of causes of are to causal and causal by not causes and to certain the issues in for a are and Perhaps the only that can be is that at causal of a of from a of no study can for of these of as there is no study even an which can in we that with and of the of and that epidemiology with this epidemiology a of study designs and of to to some may be more in as in the of by there are no as to which of are and which are In our original we the of the role of in the about and are not for in the of the debate they because they out to the (the all of the an in This is not for given in may be by as those of and the but they be because there are no logically sufficient conditions for causal as in is about it to all since all on In this the and by the various commentators on our paper are as and as which in the and of the of that is at the of a causal But these and the that are with do not define they are a of which can be We believe that the of from to at an is a much the of the of causal from a study or of of our commentators have But we our that the effect of interventions has been almost emphasized by the at the of the of causal inference a of is if only a part of causal inference is Our is that the but important of epidemiology be or given the of the and the of about We that this exchange of has to clarify the of these as to causal that they not the of this The for is by the of this by the of for The authors are to the authors of the commentaries for their of
DOI: 10.1093/ije/dyw298
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: bronze
Estimated APC paid: $3,172
OpenAlex ID: W2582919482