Longitudinal causal modelling

#1
Hi community!

I am asking for help primarily due to how poor my dissertation supervisor has been. I am a Neuropsychology student and am due to start my dissertation, which is specifically looking at the impact of gut health on our mood, and even behaviours/preferences. I designed the experiment, but did so on the back of guidance from my supervisor who claimed we could establish a causal relationship through a complex statistical procedure that he was willing to help me with. This all sounds great, but multiple emails and two months later and I have not had a reply since.

I was just hoping to pick your brains as to what the best statistical analysis may be for this following experiment (the analysis I am unsure on would be for experiment 2):

Experiment 1:
Individuals with temperamental guts complete the following:
-7 day food checklist to identify how 'inflammatory' their diet is
-Questionnaire on gut health
-Questionnaire on mood
-Questionnaire on sleep quality


The variables will all be correlated against factors of gut health (established via factor analysis) to attempt to establish significant relationships between the variables and gut health. The variables that are significant are then included in the second experiment.

Experiment 2:
Lasting for two weeks, each day the individuals have to-
-Complete a forced-choice paradigm choosing between inflammatory and anti-inflammatory food items (measuring food preference)
-Gut health measure
-Other measures which were found to significantly correlate in experiment 1

The issue lies in the second experiment. What I am wanting to establish is whether it is just that behaviours impact on our gut health (poor sleep, poor diet, psychopathology), which is already established in the literature, or whether it is actually the case that our gut is impacting on our mood, our behaviours and our preferences (this is a relatively new position in the literature, but with a strong theoretical backing due to things like the majority of serotonin being synthesised in the gut). So, in theory, what I am needing to show is that gut health today is predicting/causing behaviours/moods/preferences tomorrow.

Do any of you know of a way to longitudinally establish this causal relationship? Would something like longitudinal hierarchical linear modelling work?

Thank you so so much for any help!
 

noetsi

Fortran must die
#2
It is not possible to establish causality unless you do random assignment. You might look at structural equation models one form of which deals with time.
 
#3
Wouldn't you need to directly manipulate gut health today (and see how that affects the psychological factors tomorrow) in order to establish causality? I'm curious - I have heard people talking about causality in similar cases and I never quite understand how it is possible without a direct and randomly assigned manipulation. Otherwise we can just circle back around and claim that it was mood that affected gut health.

Also curious - since serotonin doesn't cross the blood brain barrier (I think?), what might be the proposed link between gut-based serotonin and brain-based?
 
#4
Wouldn't you need to directly manipulate gut health today (and see how that affects the psychological factors tomorrow) in order to establish causality? I'm curious - I have heard people talking about causality in similar cases and I never quite understand how it is possible without a direct and randomly assigned manipulation. Otherwise we can just circle back around and claim that it was mood that affected gut health.

Also curious - since serotonin doesn't cross the blood brain barrier (I think?), what might be the proposed link between gut-based serotonin and brain-based?
Thank you for your response! I agree that I was surprised when my supervisor suggested a causal modelling approach when the relationship between the factors are so complex. However, having read up on panel data models I think it is possible. If you read the response on this forum I am pretty sure it is possible? https://stats.stackexchange.com/questions/218620/what-is-cross-lagged-panel-design. Do either of you know much about a cross-lagged panel design?

In terms of serotonin and the blood-brain barrier the research is pretty limited. I only raised the point as an interesting insight into the role of our gut in mood, but things like probiotics having found significant improvements in mood (over-and-above SSRIs in some experiments) would suggest there is definitely a link between gut health and mood. My experiment will not be considering serotonin anyway, so the nuance of the BBB won't be needed.

Thanks,
David
 

hlsmith

Not a robit
#5
Without randomized treatment assignment, causality cannot be conclusively established. However, using quasi-experimental study designs and meeting/establishing causal assumptions (e.g., temporality, exchangeablity, etc.) can help to acquire evidence that supports a causal relationship. Though in complex causal relationships (e.g., exposure mixtures) this is obviously more difficult. But when there are strong relationships supporting evidence can be discerned (e.g., smoking and lung cancer, vaccination and prevention of diseases).
 

noetsi

Fortran must die
#6
Given that quasi-experimental treatments are hard to do most just assume correlation equals causality....which of course is a guess.