Need Help on Analysis: How to know potential causality model from historical data

reynald

Quite Involved in Discussions
Hello All,
I have a problem. I only have limited resources to run a DoE to nail the causality.
The problem is that I have a 1 Response and 3 Potential Causes and they exhibit the following conditions
- For all 3 there is a strong 1-to-1 correlation against the Response.
- All three factors are strongly correlated as well.
- Practical knowledge says the relationship could either be any of this
-- Y = f(X1, X2, X3)
-- Y = f(X3), and X3 = f(X1, X2)
-- Y = f(X3), while X3 = f(X2), and X2 = f(X1).
-Running a General Linear model on these models says all factors are significant

My question is is there a way to figure out which model is the most probable? I want to know the mechanism on which drives which. Is it the complex one where X1, X2, X3 --> Y, or the simple one where X1 -->X2-->X3-->Y? And if the relationship X1 -->X2-->X3-->Y is true, how to know which factor i s the root trigger of the chain responses?

It matters because I want to know which one to resolve, is it all factors or just X1.

Any idea/help on how to proceed would be appreciated.

Thanks
 

Bev D

Heretical Statistician
Leader
Super Moderator
I assume you mean statistical significance, i.e. p<.05?

if so,remember that statistical significance doesn't mean practical importance; it does not correlate to the size of the effect. this is a common error many practitioners make. my recommendation is to plot the data - all of it not just the averages - and look for the factor tht produces the largest effect.
 

reynald

Quite Involved in Discussions
I assume you mean statistical significance, i.e. p<.05?

if so,remember that statistical significance doesn't mean practical importance; it does not correlate to the size of the effect. this is a common error many practitioners make. my recommendation is to plot the data - all of it not just the averages - and look for the factor tht produces the largest effect.

Thanks Bev!
The size of the effect sure made it easy for me to decide which one to address first. I can't believe I missed to look at that. Will now do a pilot test on factor#2 to confirm.
 

reynald

Quite Involved in Discussions
Thanks Steve,
I had a glance reading on Wikipedia and I find the concept appealing.
This is my first time to hear about it and will read more on the topic.

Care to share though how the conditional probabilities are estimated in practice?
Thanks.
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
Thanks Steve,
I had a glance reading on Wikipedia and I find the concept appealing.
This is my first time to hear about it and will read more on the topic.

Care to share though how the conditional probabilities are estimated in practice?
Thanks.

If you have direct observation data - for example in oil exploration - given that the geology structure is of a certain type, what is the probability there is oil there is a calculation of how many times there was oil found in a geology structure of that type divided by the number of times you looked at that geology type.

If you don't have enough direct data, then you've got to go for theory and or expert opinion - a published treatise on geology theorizes that 33% of the time you ought to find oil in geology of this type due to . . .
 
Top Bottom