I am having trouble interpreting the results of my Causal Mediation in SAS

[marksanter]

Hello SAS Community,

I am having some difficulty interpreting my results using Casual Mediation.

My mediator variable (wealthy) is binary (0 or 1). 0 being not wealthy and 1 being wealthy

My treatment variable (race) is binary as (0 or 1). 0 being white and 1 being minority

My outcome variable is test scores. This variable is continuous.

This is code I used 

proc causalmed data=scores;
class race wealthy;
model scores = race wealthy race*wealthy;
mediator wealthy = race;
bootstrap;
run;

These are my results and there is significance, but I am having trouble with the interpretation as I do not work with binary variables often. 

Summary of Effects
	Estimate	Standard Error	Bootstrap Standard Error	Wald 95% Confidence Limits	Bootstrap Bias Corrected 95% Confidence Limits	Z	Pr > |Z|
Total Effect	-7.2053	1.8438	1.7369	-10.8192	-3.5915	-10.6610	-3.7261	-3.91	<.0001
Controlled Direct Effect (CDE)	-15.2432	3.6224	3.9705	-22.3430	-8.1435	-22.8759	-7.1777	-4.21	<.0001
Natural Direct Effect (NDE)	-4.7966	1.9995	1.8347	-8.7156	-0.8777	-8.4364	-1.2785	-2.40	<.0001
Natural Indirect Effect (NIE)	-2.4087	0.9230	1.0044	-4.2178	-0.5995	-4.9271	-0.8741	-2.61	<.0001
Percentage Mediated	33.4290	14.8643	16.0116	4.2955	62.5624	11.1562	76.7381	2.25	<.0001
Percentage Due to Interaction	-93.0785	41.8649	48.0093	-175.13	-11.0247	-218.73	-31.5048	-2.22	<.0001
Percentage Eliminated	-111.56	53.7929	61.2548	-216.99	-6.1238	-267.02	-23.3474	-2.07	<.0001

Thank you very much for your help! 

Mark

[pink_poodle]

You are asking what percent of the effect of race on test scores is mediated by differences in income. The answer from the table is 33% (Percentage Mediated) and this is statistically significant (p<0.0001). In any model, the factor is always on the right and the outcome is on the left. The mediator statement shows the mediator model (wealthy = race). Hence, the direct pathway is race -> test scores and the indirect (mediated) pathway is race-> wealth -> test scores. This can be drawn as a triangle with 33% over “->wealth->” and the remaining 66% over the direct “race -> test scores” arrow.

[marksanter]

Hello, thank you for your guidance.

I do have a followup questions. I conducted a t test and found that individuals of minority status scores lower on the tests, and I wanted to know if income mediates that relationship. And it looks like it does according to my analysis, is thay correct? So essentially, wealth buffers the relationship?

[pink_poodle]

Yes, although we do not have directionality with causal mediation (but maybe there is - look at all the minus signs in the table!). It is tempting to say that part of the reason why individuals of minority status have lower test scores is because of lower income, but that would need further exploration. Right now, it would probably be more accurate to conclude that race has an effect on test scores partly due to differences in income.