1
$\begingroup$

just want to get my head around mediator and how to properly interpret its effect on the relationship between predictor and response.

From https://journals.lww.com/jnpt/fulltext/2019/04000/mediators_and_moderators,_confounders_and.1.aspx, the authors explained that the mediators "lie on causal pathway" whereas the covariates and confounders "do not lie on the causal pathway". As such, it kind of mean that there's no way to tease the mediators from the predictor if the mediators significantly contribute to the response, i.e. there's some "interaction" between the predictor and mediators which leads to a causal effect on the response? Is this correct?

To help the above discussion, let's say the mediator is overweight and the response is hypertension.

Now if I ask a different question such that the mediator interchange with the response (overweight becomes response and hypertension becomes mediator), and the association is significant, this then means that there's an interaction between the predictor with hypertension (formerly analysed as response) causing an effect on overweight (formerly analysed as mediator). Consequently, doesn't this mean there's an interaction between the mediator and response since significance will be observed irrespective to if they are treated as mediator or response? And would be incorrect to say overweight is mediating the effect of predictor on hypertension because we can also say that hypertension is mediating the effect of predictor on overweight?

I'm very confused, can someone help to shed light into this darkness? Thanking in advance.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

One says that there is an 'interaction' between the moderator and the predictor, not between the mediator and the predictor, as you are doing.

To make the example of a mediator let us assume that:

Obesity (the predictor) increases the risk for diabetes (what you call the response).

Now you might ask, what is the mechanism underlying this (supposed) causal relationship? One hypothesis is that insulin resistance mediates this relationship, that is, is the mediator. Obesity increases insulin resistance, which in turn increases the risk for diabetes. There needs to be no interaction between obesity and insulin resistance. Obesity remains a predictor both for insulin resistance and diabetes. Insulin resistance might provide a deeper/biological understanding of how/why obesity increases the risk for diabetes.

In this context, an example for a moderator could be genes.

There might be some specific genes, such that if one person has them, despite being obese the risk for diabetes is not increased. In this case, one speaks of an interaction between obesity (the predictor) and genes (the moderator).

Hope this helps.

Improve this answer
$\endgroup$
6
  • $\begingroup$ Thanks for ur explanation and the example given is good but it doesn't quite help me to understand a research i came across recently. It's a study where the predictor is chemical exposures, mediator is sex hormones and the response is sleep disorder. Why do they not have the sleep disorder as mediator and sex hormones as response? because if one is deprived of sleep for days, it will also disrupts hormonal balance, isn't it? As such, can i say that the chemical exposures can have influence on both health outcomes, in this context, sex hormones and sleep disorder, so cannot tease them out? $\endgroup$
    – Catalyst
    Commented Feb 9, 2022 at 14:32
  • $\begingroup$ Just to add clarity to my comment above... from start, an investigator may want to examine the effect of a predictor on responses and found this effect is significance on either sex hormones or sleep disorder. Next he starts to ponder if sex hormones or sleep disorder can act as a mediator, which he later found either model produces significance as well. As such, how to interpret this finding? or we will never see this happening? If the effect of predictor on either responses is significant, is it correct to say there's an effect of chemical exposures on both sex hormones and sleep disorder? $\endgroup$
    – Catalyst
    Commented Feb 9, 2022 at 23:25
  • $\begingroup$ @Catalyst I think they could also do it that way. Mediation analysis is an instrument, which you can use to test a hypothesis, but then it is you as a scientist who decide which hypothesis you want to test. So yes, nothing forbids you to test the hypothesis that sleep disorder is mediating the relationship between chemical exposures and sex hormones. What the data will give as an answer I do not know. $\endgroup$
    – fabiob
    Commented Feb 10, 2022 at 15:50
  • $\begingroup$ and to answer to your second comment: before you test whether sleep dx mediates the relationship between chem exposures and sex hormones (or vice-versa), you need as a requirement (at least in Baron&Kenny's formulation, I suggest you read this on wikipedia: en.wikipedia.org/wiki/Mediation_(statistics) ) that chem exp is a predictor for both sleep dx and sex hormones. $\endgroup$
    – fabiob
    Commented Feb 10, 2022 at 15:59
  • $\begingroup$ and it could happen that both sex hormones and sleep dx work mathematically as mediators. it is you as a scientist who should have a hypothesis well formulated and biologically meaningful before analyzing the data. otherwise your study is exploratory and can conclude that both hypotheses make sense and that further studies are necessary to compare them. $\endgroup$
    – fabiob
    Commented Feb 10, 2022 at 16:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.