Article has been wikified. KarenAnn 17:01, 27 June 2006 (UTC)Reply

Alternative Views

edit

Are there any plans to add criticisms of the RCM, such as the anti-experimentalist view of Heckman? The kerfuffle between Heckman and Sociological Methodology might make for fun reading! SkipSmith 05:53, 22 August 2006 (UTC)Reply

Great idea. Add away. David.Kane 02:30, 25 August 2006 (UTC)Reply

History of the model

edit
Is this article an example of Stigler's Law? I am thinking of David Freedman's criticism, e.g., in Statistical Models (page 199):

The model was proposed by Neyman (1923)...[it] has been rediscovered many times: see, for instance, Roy (1951) or Hodges and Lehmann (1964, Section 9.4). The setup is often called "Rubin's model": see for instance Holland (1986), who cites Rubin (1974). That simply mistakes the history.

I don't know enough to judge Freedman's claim. But if he's right, it seems that the title of this article should be changed, or at least that a new section about Neyman's work should be added.{John G Bullock 14:28, 25 December 2006 (UTC)}Reply

Feel free to add a new section about the history. Rubin and others have written on this topic, giving due credit to Neyman and others. (Whether or not Roy deserves much credit is a matter of dispute.) But Rubin is the single person most associated with this approach, at least for the last 3 decades, so the title should stay the same. David.Kane 19:36, 26 December 2006 (UTC)Reply

Note that Kane and Rubin both hold (or have held) positions at IQSS in Harvard. —Preceding unsigned comment added by 216.254.15.72 (talk) 05:48, 25 November 2010 (UTC)Reply
Some people do refer to it as Neyman-Rubin or Neyman-Rubin-Holland (or Neyman-Holland-Rubin). But "Rubin Causal Model" seems the most common by some way. Qwfp (talk) 18:48, 20 February 2008 (UTC)Reply
It seems mistaken to me to claim that Pearl (2000) was the first to point out the equivalence of the RCM and approaches in econometrics. Heckman in particular has been insisting since at least the early nineties that the RCM should be renamed because it's basically the same ideas as Neyman (1923), Thurstone (1927), Haavelmo (1943), Roy (1951), Quandt (1958) and others. I haven't read Shpitser and Pearl (2006), but it also seems strange to me to credit them with the analysis of identifiability in this context: Heckman and Honore 1990 covered that ground thoroughly. Sked123 (talk) 23:31, 12 August 2008 (UTC)Reply
I'd like to include some references to the other names, but haven't found a good citation for the use of the "Neyman-Rubin model" or "Neyman-Rubin-Holland model." Aaronshaw (talk) 06:30, 25 September 2009 (UTC)Reply

It is important to add Pearl (2009)[1] to the last line, because Morgan and Winship (2007)[2] do not deal with the crucial formal connection between the Rubin causal model and structural equation modeling,

  1. ^ Pearl, J. Causal inference in statistics: An overview. UCLA Computer Science Department, Technical Report R-350. To appear in Statistics Surveys, 2009.
  2. ^ Morgan, S. and Winship, C., Counterfactuals and causal inference: Methods and principles for social research, Cambridge University Press, 2007.

Needs work

edit

I have corrected several of the tables in minor but useful ways. The "means" are now correctly calculated and the unknowns "?" now match the text. Names in the tables and text do not yet match. The entire Extended example would benefit from some continuity which is now lacking.

The entire article is nominally limited to experimental assignment. It doesn't make clear that the assignment metaphor may be useful in observational studies.--P64 (talk) 19:22, 11 March 2011 (UTC)Reply

In this sentence, "That view, which has long been argued by Heckman (2005)[6] has received a formal confirmation in Pearl (2000)[5].," it doesn't make sense that something that was argued in 2005 then received confirmation 5 years earlier. — Preceding unsigned comment added by Jeff987 (talkcontribs) 16:28, 9 September 2011 (UTC)Reply


This contends that Pearl argued that RCM is subsumed by SEM (which are in fact ad hoc and incompletely defined collections of competing algorithms for making sense of observations, rather than a unified logical or mathematical model) in his 2000 book. But I can find no reference to this, and it simply does not make sense. If this section is to stay in the article, I think it needs to be significantly clarified. J.christopherwestland (talk) 04:58, 24 November 2015 (UTC)Reply

Definitely still needs some work (I made one edit; please see the subject names in sec 2.3- i think that susie should probably be sallie) but otherwise very clear and well written. Thank you! — Preceding unsigned comment added by 69.229.153.101 (talk) 18:53, 29 August 2012 (UTC)Reply

Paul Holland

edit

I noticed this article links to the wikipedia page for Paul Holland, the footballer. Is that REALLY the same Paul Holland who named RCM? I am not editing the page b/c I do not know. It simply seems unlikely. — Preceding unsigned comment added by 128.118.14.253 (talk) 15:09, 26 January 2012 (UTC)Reply

Well spotted. No, it's not the same person. I've unlinked it. Qwfp (talk) 15:41, 26 January 2012 (UTC)Reply

Cox?

edit

The section on SUTVA begins: 'We require that "the [potential outcome] observation on one unit should be unaffected by the particular assignment of treatments to the other units" (Cox 1958, §2.4).'

Should there be a reference identifying the source? Gammalcyklist (talk) 09:31, 10 June 2019 (UTC)Reply

Would it make sense to reframe the extended example in terms of raw blood pressure

edit

Seems more confusing than it ought to be, with a focus on changes. Such a framing means that the causal effect is a difference in changes. Rubincausalmodel (talk) —Preceding undated comment added 22:27, 22 October 2019 (UTC)Reply

Giving this a shot now. Rubincausalmodel (talk) 20:42, 23 October 2019 (UTC)Reply
Better now, I hope. Rubincausalmodel (talk) 22:24, 25 October 2019 (UTC)Reply