From epasipanodya at psych.udel.edu Thu May 7 05:15:51 2015 From: epasipanodya at psych.udel.edu (Elizabeth Pasipanodya) Date: Wed, 6 May 2015 23:15:51 -0400 Subject: [Mediation-information] Interpreting the ACME Message-ID: <8DC1EA0EB7E3FD419D9B3BE76BD403867B15439C63@razor.psych.udel.edu> Hello All, I am a doctoral student working on data from a project focusing on couples coping with breast cancer. I have a mediation model with two simultaneous predictors, X1 (a positive relationship event) and X2 (a relationship conflict), a mediator, M (a measure of intimacy), and an outcome Y (a measure of anxiety). X1 and X2 are both binary while M is continuous. Additionally, Y follows a count distribution. These variables are repeatedly measured for each individual across a number of days and, thus, I used the R packages lme4 and mediation to conduct my analyses. I would like to double-check the meaning of the ACME. My understanding is that it represents the expected difference in the potential outcome when the mediator takes the value that it would have under the treatment condition compared to the control condition, while the treatment condition is held constant. Since I have a count outcome, should I report and interpret my ACME in the same units as my count outcome, as one would a rate ratio? That is, the ACME shall represent, in the logs of expected counts, the estimated average change in Y among the treatment group (those with a relationship event) as a result of M (intimacy) rather than directly from X1 (positive relationship event) and X2 (negative relationship event)? For instance, based on the mediation output below, could I say something like the following --- on a day in which participants reported experiencing at least one negative relationship event, their estimated average change in anxiety due to changes in intimacy was 1.08 times (e^0.07485) that of those without a relationship conflict, controlling for the occurrence of positive relationship events? > med2 <- mediate(apath, bpath, treat = "X2", mediator = "M", sims=5000, control.value = -0.5, treat.value = 0.5, dropobs=TRUE, method = "boot", boot.type = "bca") > summary(med2) Causal Mediation Analysis Quasi-Bayesian Confidence Intervals Mediator Groups: ID Outcome Groups: ID Output Based on Overall Averages Across Groups Estimate 95% CI Lower 95% CI Upper p-value ACME (control) 0.07890 0.02639 0.15107 0.00 ACME (treated) 0.07080 0.02354 0.13856 0.00 ADE (control) -0.06645 -0.32283 0.17712 0.50 ADE (treated) -0.07455 -0.35353 0.19218 0.50 Total Effect 0.00435 -0.25810 0.26892 0.97 Prop. Mediated (control) 0.16404 -11.47515 12.32893 0.97 Prop. Mediated (treated) 0.20262 -9.82740 10.87888 0.97 ACME (average) 0.07485 0.02583 0.14364 0.00 ADE (average) -0.07050 -0.33728 0.18478 0.50 Prop. Mediated (average) 0.18333 -10.80037 11.56373 0.97 Sample Size Used: 602 Simulations: 5000 Best, Elizabeth Pasipanodya -------------- next part -------------- An HTML attachment was scrubbed... URL: From kimai at Princeton.Edu Mon May 11 16:53:44 2015 From: kimai at Princeton.Edu (Kosuke Imai) Date: Mon, 11 May 2015 10:53:44 -0400 Subject: [Mediation-information] Interpreting the ACME In-Reply-To: <8DC1EA0EB7E3FD419D9B3BE76BD403867B15439C63@razor.psych.udel.edu> References: <8DC1EA0EB7E3FD419D9B3BE76BD403867B15439C63@razor.psych.udel.edu> Message-ID: All the results are presented on the same scale as the outcome variable. So, if you are running the count model (e.g., poisson), you can interpret the ACME as the expected count. Kosuke Kosuke Imai Department of Politics Princeton University http://imai.princeton.edu On Wed, May 6, 2015 at 11:15 PM, Elizabeth Pasipanodya < epasipanodya at psych.udel.edu> wrote: > Hello All, > > > > I am a doctoral student working on data from a project focusing on couples > coping with breast cancer. I have a mediation model with two simultaneous > predictors, X1 (a positive relationship event) and X2 (a relationship > conflict), a mediator, M (a measure of intimacy), and an outcome Y (a > measure of anxiety). X1 and X2 are both binary while M is continuous. > Additionally, Y follows a count distribution. These variables are > repeatedly measured for each individual across a number of days and, thus, > I used the R packages lme4 and mediation to conduct my analyses. > > > > I would like to double-check the meaning of the ACME. My understanding is > that it represents the expected difference in the potential outcome when > the mediator takes the value that it would have under the treatment > condition compared to the control condition, while the treatment condition > is held constant. Since I have a count outcome, should I report and > interpret my ACME in the same units as my count outcome, as one would a > rate ratio? That is, the ACME shall represent, in the logs of expected > counts, the estimated average change in Y among the treatment group (those > with a relationship event) as a result of M (intimacy) rather than directly > from X1 (positive relationship event) and X2 (negative relationship event)? > > > > For instance, based on the mediation output below, could I say something > like the following --- on a day in which participants reported experiencing > at least one negative relationship event, their estimated average change in > anxiety due to changes in intimacy was 1.08 times (e^0.07485) that of those > without a relationship conflict, controlling for the occurrence of positive > relationship events? > > > > > med2 <- mediate(apath, bpath, treat = "X2", mediator = "M", sims=5000, > control.value = -0.5, treat.value = 0.5, dropobs=TRUE, method = "boot", > boot.type = "bca") > > > summary(med2) > > > > Causal Mediation Analysis > > > > Quasi-Bayesian Confidence Intervals > > > > Mediator Groups: ID > > > > Outcome Groups: ID > > > > Output Based on Overall Averages Across Groups > > > > Estimate 95% CI Lower 95% CI Upper p-value > > ACME (control) 0.07890 0.02639 0.15107 0.00 > > ACME (treated) 0.07080 0.02354 0.13856 0.00 > > ADE (control) -0.06645 -0.32283 0.17712 0.50 > > ADE (treated) -0.07455 -0.35353 0.19218 0.50 > > Total Effect 0.00435 -0.25810 0.26892 0.97 > > Prop. Mediated (control) 0.16404 -11.47515 12.32893 0.97 > > Prop. Mediated (treated) 0.20262 -9.82740 10.87888 0.97 > > ACME (average) 0.07485 0.02583 0.14364 0.00 > > ADE (average) -0.07050 -0.33728 0.18478 0.50 > > Prop. Mediated (average) 0.18333 -10.80037 11.56373 0.97 > > > > Sample Size Used: 602 > > > > > > Simulations: 5000 > > > > Best, > > > > Elizabeth Pasipanodya > > > > > > _______________________________________________ > Mediation-information mailing list > Mediation-information at lists.r-forge.r-project.org > > https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/mediation-information > -------------- next part -------------- An HTML attachment was scrubbed... URL: