Talk:Mediation (statistics)

This should be moved to Mediation (statistics). If no objections are raised within 24h or so, I'll move it. Jobjörn (Talk ° contribs) 21:55, 24 December 2006 (UTC)

Estimates of the Mediated Effect
Article needs to include work by preacher, mackinnon, and dwyer regarding methods of estimating the effect. Asymptotic confidence limits for the indirect effect using PRODCLIN is the gold standard as of now. Bootstrapping is briefly described, however this could be expanded. As Baron and Kenny's method of testing the effect has been considered underpowered and inaccurate in describing the magnitude of the effect. When I get a chance I will add citations/contributions. However if others are familiar with this work go ahead.

does not make sense
This doesn't make any grammatical sense: "The inclusion of the mediating variable into the equation increases the relation between the treatment and outcome rather accounts for (decreases in terms of the size of the statistical relation)." I'm not sure what the author intended.
 * Would this be clearer? (Although, somewhat longer!) "'The inclusion of the mediating variable into the equation increases the significance of the relationship between the IV and DV. That is, the mediating variable can not explain or account for the variance in the DV itself, yet its inclusion in the equation increases the significance of the IV-DV relationship.'" I'm still grappling with the idea of suppression, so if someone could confirm (and/or streamline!) the above, that'd be great! Skittled 00:16, 12 April 2007 (UTC)

Mediation (quantitative psychology)
This entry is written in the tradition of mediation as it exists within quantitative psychology. It is a good entry, but I don't think that the entry reflected that it was only concerned with a particular account of the concept of mediation. I feel that a proper place for this entry should be Mediation (quantative psychology) as the present entry does not say a whole lot about the statistical views on mediation (nothing on counterfactuals, graphical models, no mentioning of problems inherent in the extrapolation of the standard psych methods to non-linear models etc), but primarily deals with the statistics used my people with quantative psychology to quantify mediation. To my (limited) knowlegde there are seperate traditions of mediation in statistics, epidemiology and sociology, which are not covered by this entry.

--Udansk 09:06, 1 April 2007 (UTC)

I think the defintion of full vs. partial mediation should be reconsidered per Baron and Kenny (1986). Full or complete mediation occurs under the condition where the IV/DV relationship becomes zero (in contrast to when it becomes statisically no different than zero). Partial mediation occurs when the IV/DV relationship falls below significance when the mediator is included but the value of the coefficient between the IV and DV is not actually different. A potential mediator can be said to have no effect when the IV/DV relatioship is not reduced to non-significance. The way in which full and partial mediation is currently being defined creates a situation where considering a variable as a mediatior, this variable will almost always be labeled as either having a mediation effect or having a suppressive effect depending on it effect on the IV/DV relationship. However, these effects are likely to either be very small and/or, more importantly, unreliable. By the current standard it is almost never the case that a potential mediation has no effect on the IV/DV relationship whatsoever, it will almost always be labeled as a partial mediator or suppressor. I submit that this be changed and a more conservative view be taken whereby the occurance of full mediation (the actual reduction of the IV/DV relationship is reduced to zero or near zero) be the more rare occurance rather than the rare occurance be that potential mediatiors have no effect at all.

Yes it should be merged.
I think that this page should be merged because it is only a short page and could be easily merged.

Also, on the other page (Mediation (statistics)), there is a diagram showing a triangle with one of the points being "Mediator Variable" but there is no section about it in the page.

It would make much more sense to merge it into this page under a subsection (subheading) Mediator Variable.

Obscure: "...and formally derived by Pearl"
Please edit this obscure, dangling part of a sentence if you understand what it means. I didn't. Sslevine (talk) 02:47, 16 August 2008 (UTC)

A synonym for mediator variable - "intervening variable"
In some works(e.g., see M.Katz. Multivariable analysis: a practical guide for clinicians. 1999) mediator variables are called intervening variables. This seems to be worth mentioning.

The problem is that there is already an article Intervening variable. Some additional editing needed.


 * I'm proposing that Intervening variable be merged here; every source I've seen that mentions both says that they're synonyms, and the other article is a ragged stub. Unless someone can draw a clear distinction between the two, I'm going to redirect intervening variable here. -Bryanrutherford0 (talk) 18:40, 15 January 2016 (UTC)

moderated mediation box figures
I believe "mediator" and "moderator" should be inverted in the box figures pertaining to the moderated mediation section. — Preceding unsigned comment added by 129.240.215.211 (talk) 11:32, 11 December 2013 (UTC)
 * The previous commenter is right. The figure captions are correct, but the labels inside the boxes are inverted. When I get a chance (and if I don't forget), I can make new figures to replace the current ones. But I'll have to learn how to embed figures; I assume this is tedious but doable and probably explained somewhere on the site. Someone else is welcome to beat me to it. 71.215.84.21 (talk) 09:04, 29 January 2014 (UTC)
 * There, I've edited the images and reuploaded them, so they should now be correct. If the images still show up wrong for you, just refresh the image page. Jeff H. (talk) 22:56, 7 July 2014 (UTC)

Correct usages
This edit and four edits that immediately followed it were devoted mostly to fixing large numbers of conspicuous usage errors. Let's recall some really basic stuff:
 * right: X = 1
 * wrong: X=1

The "1" should not be italicized and proper spacing should be used.
 * correct section heading: Direct versus indirect effects
 * incorrect section heading: Direct Versus Indirect Effects
 * correct TeX code: = NDE = C, \text{ independent of } m
 * incorrect TeX code: = NDE = C,~{\rm independent~of}~m
 * correct TeX code: E(Y\mid X=1) - E(Y\mid X=0)
 * incorrect TeX code: E(Y|X=1) - E(Y|X=0)
 * correct range of pages or years or other numbers: 241–275
 * incorrect range of pages or years or other numbers 241-275

These things are codified in WP:MOS and WP:MOSMATH and standard LaTeX usage conventions.

I also change this:
 * $$ 1- NDE / TE = b_1 (c_2 +c_3) / (c_1 + b_0c_3 + b_1 (c_2 + c_3)). $$

to this:
 * $$ 1- \frac{NDE}{TE} = \frac{b_1 (c_2 +c_3)}{c_1 + b_0c_3 + b_1 (c_2 + c_3)}. $$

and a few similar things. Michael Hardy (talk) 18:28, 26 January 2016 (UTC)