1. ## further causal analysis

Hi All,

So I did this analysis on a dataset by exploiting partial correlation matrix. I can see how the variables are related to each other and to the target.

I am wondering what other statistical method I can use to validate this output; because it considers the effects between two variables corrected for others variables...

I appreciate any guidance on this!
Thanks a lot!

2. ## Re: further causal analysis

There are many methods depending on the data and your specific question. To me the best is structural equation models because it allows testing of indirect effects and a much more complex set of relationships than say ANOVA or Regression.

A minor point. You never show causality with statistics. Ever

3. ## Re: further causal analysis

Thanks noetsi; what do you exactly refer to structural equation models? I do some symbolic regression on data which has numerical components, with few of them being categorical variables.
What I was basically trying to is validation in terms of causality/relationships etc.
What is the shortcoming of showing causality with statistics?

4. ## Re: further causal analysis

The second question first. By definition causality, which is a basic phenomenon in nature such as heat leading to more rapid movement of molecules, can never be shown by an analysis of stochastic processes. Even if you find a million times that X lead to Y in samples that does not prove causality. A variation of this is David Hume's comments that just because none of the swans you saw were black, does not mean that no swans exist. Most statistic text will have a much more in depth discussion of this, but the bottom line is that there is universal agreement that statistics never can prove a causal relation (causality). Only theory can.

Structural equation models are methods such as confirmatory factor analysis or path analysis that allows you to create a series of variables and then test the relationship between them (as well as whether your overall theory reproduces the covariance matrix). They are complex, but very useful. If you would like a recommendation for a text I can provide that.

5. ## The Following User Says Thank You to noetsi For This Useful Post:

viktus (09-06-2012)

6. ## Re: further causal analysis

Whenever I am trying to convey a relationship between variables, I usually try to incoporate as many components of the Bradford Hill Criteria (for Causality) as I can along with discussing chance, confounding, and errors.

7. ## Re: further causal analysis

I tend to go with the Shaddish, Cook, and Campbell approach, but even their best models (rarely usable in the real world - that is outside research) don't prove causality. They just make it more likely you discovered it.

8. ## The Following User Says Thank You to noetsi For This Useful Post:

viktus (09-06-2012)

9. ## Re: further causal analysis

That's intersting.. I need to arrive at more concrete conclusion if Im honest; so causalty is not that good then...
Hey noetsi, could you recommend a text for the structural models you mentioned?

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts