39

Your approach to testing mediation appears to conform to the "causal steps approach" described in the classic methods paper by Baron & Kenny (1986). This approach to mediation entails the following steps: Test whether X and Y are significantly associated (the c path); if they are not, stop the analysis; if they are... Test whether X and M are ...


21

What does it mean that ACME (treated) is 0.0808? 0.0808 is the estimated average increase in the dependent variable among the treatment group that arrives as a result of the mediators rather than 'directly' from the treatment. The dependent variable in this example is the probability of sending a message to a congress member, the mediator is the emotional ...


13

I believe the quick one-sentence answer to your question, When is it appropriate to control for variable Y and when not? is the "back-door criterion". Judea Pearl's Structural Causal Model can tell you definitively which variables are sufficient (and when it's necessary) for conditioning, to infer the causal impact of one variable on another. Namely, ...


9

This does suggest a "full" mediation, in which all of the IV's influence is mediated. The ACME being significant shows that the mediating process appears to be present. On the other hand, you don't have evidence that there is an ADE (insignificant result). The reason the total effect is larger than the ACME alone is that the estimate still includes the ...


8

Baron and Kenny are indeed outdated, though that does not make them wrong in all cases. The concerns divide into broadly statistical limitations and assumptions which are discussed in the reference your reviewer suggests and in the literature alluded to by @PeterFlom, and broadly non-statistical concerns about the definition and causal identification of ...


8

This should probably be migrated to StackOverflow since it is about software, but: You could do this in the R package lavaan. In your model, you would first specify models for M1, M2, and Y. We will want to label all the paths, as well. I will label c' as cp, for "c-prime": M1 ~ a1 * X M2 ~ a2 * X + d21 * M1 Y ~ cp * X + b1 * M1 + b2 * M2 The indirect ...


7

This is more a discussion of concerns I have firstly with the approach of Baron and Kenny (which has some bearing on your question), and with a number of more recent papers (I haven't seen them all, so my comments may not apply to everything). It may also relate to the 2011 paper you mention, which I have only had the chance to skim through just now. From ...


7

Conditioning (i.e. adjusting) the probabilities of some outcome given some predictor on third variables is widely practiced, but as you rightly point out, may actually introduce bias into the resulting estimate as a representation of causal effects. This can even happen with "classical" definitions of a potential causal confounder, because both the ...


6

Moderating variables (aka effect modifier, interacting variable, etc.) alter the effect of another variable. In your example increased stress causes decreased well-being, but this effect is altered in the presence of locus of control. Of course, nothing prevents a moderating variable from having its own direct effect (aka main effect). An example here would ...


6

No, "mediation" and "indirect effect" are not synonymous. For example, when analyzing complex causal systems in which any variable in the system at time $t$ contributes causally to every variable in the system at times $>t$, the term "mediator" is largely meaningless, whereas indirect effects may or may not exist due both to the qualitative structure of ...


6

I posted this question in another location and was provided the answer by Terry Jorgenson. Question and answer: I would like to calculate the conditional indirect effects of X on Y given a set of values for the moderator W. Both X and the moderator are continuous. Could you provide some direction for structuring the model syntax? Assuming X and W are ...


5

The lavaan package is an R package for SEM. You can use it to test for multiple mediation hypothesis, and there is boostrap.


5

The question has been answered by Jake Westfall, but, as I find the question (and the answer) interesting, I decided to develop further. Let say you have the mediation model : $x \rightarrow m \rightarrow y $ and you know the correlations $a=\rho(x,m)$, $b=\rho(m,y|x) $ and $ c=\rho(x,y)$. You then know the indirect effect $ab=a*b$ and the direct effect $c' ...


5

From definitions, I feel that a variable can not simultaneously function as mediator and moderator. Let's try to investigate both effects: Mediaiton Mediation is a hypothesized causal chain in which one variable affects a second variable that, in turn, affects a third variable. The intervening variable, $M$, is the mediator. It mediates the relationship ...


5

ACME is an acronym for "Average Causal Mediation Effects" ADE is an acronym for "Average Direct Effects" Total Effect is the sum of ACME and ADE So the "Indirect Effect" that you are seeking is simply the row forACME


5

Interaction and mediation are different things. In mediation, we have a causal pathway where one variable causes the mediator and the mediator causes the outcome. In interaction, we have a joint action, where two variables are associated with an outcome, but the "effect" of one variable depends on the value of the other variable. Clearly these are ...


4

Baron and Kenny is distinctly old fashioned these days. They see mediation as a "yes-no" "present-absent" quality; more recent approaches (lots of work by MacKinnon and others) treats it as a continuum. This makes more sense to me.


4

From a mathematical standpoint, there's nothing wrong with doing a Sobel test with survey data (by the way, and slightly off-topic -- you should consider using a bootstrapping method to test your indirect effects instead of a Sobel test; bootstrapping methods are uniformly more powerful than Sobel tests). The real question is what conclusions you would be ...


4

This is quite straightforward. The reason you have no relationship between $x$ and $y$ using your approach is because of the code: y <- 2.5 + 0 * x + .4 * med + rnorm(100, sd = 1) If you want some relationship between $x$ and $y$ even when ${\rm med}$ is included (that is, you want partial mediation), you would simply use a non-zero value for $b_{32}$...


4

I guess your supervisor is correct and the model is implying mediations. In general, most of SEM and Path analysis involve some mediation or indirect effects. The whole point, of these models is to say, instead of everything being related to everything, i.e. like in a correlation matrix, is to say that some relations are good enough to explain the whole ...


4

A few thoughts come to mind. I hope they are helpful. Lets say you have exposure X, outcome Y, and mediator X. 1) Baron and Kenny is, in my opinion, not a very good way to address mediation, at least not without a lot of thougfulness. The main problem is potential "collider bias" REF. If there are confounders of the Z-Y relationship ( Z <-- C --> Y ), ...


4

Report standardized parameter estimates. These are analogous with some sort of effect size. That's what they do in the paper. (You might also ask your supervisor what they actually want - people sometimes ask for an effect size because they think that things should have effect sizes, but what is meant by an effect size can be complex - and it is not always ...


4

In mediation analysis, whatever the estimation procedure, it is totally fine to obtain direct and indirect effects with opposite directions. This situation is sometimes referred to as "inconsistent mediation", as it produces one between direct or indirect effect to be larger than the total effect (see here for further details: MacKinnon, David P., Amanda J. ...


4

If I'm understanding right, you may be wanting to do an analysis of covariance (ANCOVA). There is a helpful tutorial here (http://r-eco-evo.blogspot.com/2011/08/comparing-two-regression-slopes-by.html). Specifically, read down to the bottom where comparing slopes and intercepts is discussed. This was achieved through a slightly different version of your ...


4

If you are interested in interpreting coefficients and significance/p-values, don't use stepwise regression. See this post. In fact, stepwise regression is basically always a bad idea in this day and age when better model-selection techniques are easily computable. If you want to easily interpret the model and say which coefficients are truly "significant," ...


4

You have to be a bit more specific in your language. drug use may help explain this association Do you mean that there is a correlation between the 3 variables in the population, but the correlation between neuroticism and performance disappears or changes when you condition on drug use ? If that is what you mean then it is consistent with drug use being a ...


4

This seems entirely plausible. If the direct effect of $X$ on $Y$ is negative, while the mediated effect is positive, these will oppose each other and, if they are of similar magnitude, result in a weakly positive or negative total effect. This is easy to demonstrate with an example based on your causal diagram/DAG with R (hopefully this is sufficiently self-...


3

Trying different methods seems wise to me for the sake of understanding any differences in the results you receive from different analyses. Baron & Kenny's approach has received a fair amount of criticism (e.g., Pardo & Román, 2013; Hayes, 2009; Zhao, Lynch, & Chen, 2009; Krause et al., 2010), so alternatives to that would seem particularly worth ...


3

There seem to be two distinct aspects to this question: Are the terms 'indirect effect' and 'mediation' synonymous? I would say largely yes (edit: assuming--cf. @Alexis' answer--that you are referring to a terminal causal model). Bear in mind that you can have full or partial mediation. That is, the effect of A on C can flow only through B, or partly ...


3

Sounds like a job for structural equation modeling. Changes can be modeled as latent variables, used to predict one another, and lined up in a chain as you've described. For an introduction to latent change modeling, see McArdle (2009); it's quite readable and thorough. Given three latent change variables, you can use one to predict the next, and that next ...


Only top voted, non community-wiki answers of a minimum length are eligible