From kimai at Princeton.Edu Sat Dec 13 13:36:28 2014 From: kimai at Princeton.Edu (Kosuke Imai) Date: Sat, 13 Dec 2014 07:36:28 -0500 Subject: [Mediation-information] mediation in R In-Reply-To: <0A68F554-9BB5-479D-857C-41ED859D5C34@kcl.ac.uk> References: <0A68F554-9BB5-479D-857C-41ED859D5C34@kcl.ac.uk> Message-ID: Dear Tabea, Your description seems to indicate that there is a strong relationship (perhaps by construction) between T and M. So, I wouldn't be surprised that even if the relationship between M and Y conditional on T is rather weak the ACME may come out to be statistically significant. Now, since your mediator is ordinal variable, you may want to treat it as such rather than using it in the linear model (assuming that's what you did). Good luck with your research. Best, Kosuke --------------------------------------------------------- Kosuke Imai Office: Corwin Hall 036 Professor Phone: 609-258-6601 Department of Politics Fax: 609-258-1110 Princeton University Email: kimai at Princeton.Edu Princeton, NJ 08544-1012 http://imai.princeton.edu --------------------------------------------------------- On Thu, Dec 11, 2014 at 10:00 AM, Schoeler, Tabea < tabea.t.schoeler at kcl.ac.uk> wrote: > Dear Prof Kosuke Imai, > > I am a PhD student at King?s College London using the mediation package > for R. I am approaching you because I have a little question regarding the > assumptions that need to be fulfilled to run the proposed mediation > analysis. I can fully understand if you are too busy to respond to this > mail, but in case you had 5 spare minutes, I would be immensely grateful > for any advice in this regard. > > I have the dichotomized independent variable IV = cannabis use (yes if > used at least once out of 3 time points / no if never used) as the > predictor variable for risk of violence. I created the continuous mediator > variable M = frequency of use, ranging from 0 (used at 0 time points) up to > 3 (used at 3 time points). When I included M in the logistic regression > model, the effect of IV cannabis use (Yes/No) was no longer a significant > predictor for violence. The mediation effect was significant when tested in > R (cf. Table below). > > Causal Mediation Analysis > > > > Nonparametric Bootstrap Confidence Intervals with the Percentile Method > > > > Estimate 95% CI Lower 95% CI Upper p-value > > ACME (control) 0.29410 0.13608 0.49965 0.00 > > ACME (treated) 0.18529 0.10937 0.25479 0.00 > > ADE (control) -0.07367 -0.13916 0.00882 0.07 > > ADE (treated) -0.18247 -0.42868 0.01485 0.07 > > Total Effect 0.11163 0.02287 0.20054 0.02 > > Prop. Mediated (control) 2.63464 0.84232 13.87867 0.02 > > Prop. Mediated (treated) 1.65993 0.91547 5.70383 0.02 > > ACME (average) 0.23970 0.12452 0.36121 0.00 > > ADE (average) -0.12807 -0.28627 0.01183 0.07 > > Prop. Mediated (average) 2.14728 0.87885 9.53672 0.02 > > > > My question is now whether I am allowed to use the frequency variable as a > mediator since only subjects in the cannabis YES group would be exposed to > the ranges 1-3, while all subjects in the cannabis NO group would have the > score 0 in the mediator variable. This is to test whether the effect of > cannabis use on violence is driven by its frequency of use. > > King regards, > > Tabea Schoeler > Tabea Schoeler (PhD student) > Department of Psychosis Studies, Institute of Psychiatry, Psychology & > Neuroscience, > King?s College London > > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: