Hi Lawrence,

Being able to determine cause-effect relationships purely based on statistical methods rather than using experimental methods is definitely something that would be greatly appreciated by many of us. In line with this, statisticians have been working on the development of statistical analyses that can determine directional relationships based on correlational data. One method that has been showing success in this is called additive noise modeling. I haven't used this statistical analysis myself, but the following blog post is a good starting point to give you a general idea of the technique: https://medium.com/the-physics-arxiv-blog/cause-and-effect-the-revolutionary-new-statistical-test-that-can-tease-them-apart-ed84a988e. Unfortunately, the technique is only suitable for specific types of scenarios (and I'm doubtful whether it would work with emotions and performance), but it might still be worth reading.

Best,

Ilona

https://medium.com/the-physics-arxiv-blog/cause-and-effect-the-revolutionary-new-statistical-test-that-can-tease-them-apart-ed84a988e

Mediation :)

 Strictly speaking, if you have done a true experiment, you have some of the simplest/strong evidence we have for causal relationships. I see causality as one of those things that we simply approach asymptotically (we can't really know that we're there... at least with our current methods). Tests like mediation analyses test specific theoretical causal models and how they fit the data; however, the results are still cross-sectional (often times) and without any true manipulation. That being said, a true experiment (an experimental group in which the suspected cause is manipulated and a control group in which there is no manipulation) can provide strong evidence for a causal relationship (assuming you have not committed conceptual Type II errors). 

At a certain point, it comes down to what you consider a sufficient amount of evidence to establish a causal relationship. For some, a simple mediation test will do it. For others, they want to see a meta-analysis of randomized clinical trials. 

I know that makes it sound kind of wishy washy, but to some extent... it is. 

(This doesn't even begin to touch the fact that most of the effects we psychologists are interested in are bi/multi-directional. I'm sure that emotional states do effect performance, but I would also bet that the reverse is also true.)

Dear All, 

Thank you so much for the answers provided. I will be looking into the feedback provided and will get back to you if i find something interesting. 

Thanks once again 

Lawrence, causation is the hardest to predict. Other than using the statistical techniques as noted above, you need to be sound on your theoretical background. Your theory should suggest that there might be causation and you should be able to explain why A causes B and not just A causes B. Moreover, we need to strictly take care of time ordering- if A affects B, then A should be measured/ manipulated before B happens. I encourage reading up on books for research methods and not just focus on statistical techniques.

Path analysis tests the direct and indirect effects of a group of variables (mediating variables) to explain the relationship between a IV and a DV. Some would argue that implying causation in these cases is flawed. Judea Pearl writes about this, heavily cited in Econometrics.

Hi Lawrence, are you designing the experiment or considering the analysis of already collected data? If designing a new experiment, you could use a mood induction paradigm contrasted to behavioural performance prior to mood induction

At Gargi Bhaduri - Yes and to address that problem I am going to conduct a counterbalanced experiment with the order of the control and the experiment being swapped. 

At Justin Watts - I have been reading about mediation (and moderation). Sounds promising however you still have to assume the mediation model. I.e. the anlysis is based on the assumption that the mediation model is in fact true. 

At Tamsyn Elizabeth Van Rheenen - what i am after is the change in the emotional sate. The experiment has been practically designed and what I am consider is the analysis of already collected data

Aside from experiments, it is tricky (and they're tricky too, but not as tricky). Attributing causality from non-experimental data requires further assumptions. See

http://bayes.cs.ucla.edu/BOOK-2K/,

http://imai.princeton.edu/research/mediation.html,

and

http://www.amazon.com/Causal-Inference-Statistics-Biomedical-Sciences/dp/0521885884

ps. Also, on the causal versus associative relationships between emotions and cognition, see the attached. The example is that it is fairly clear that emotional events tend to be associated with clearer memories than non-emotional memories (for most populations of events), but it is less clear what the causal processes are.

Article CAUSAL AND ASSOCIATIVE HYPOTHESES IN PSYCHOLOGY: Examples Fr...

Dear Laurence, 

I have two concerns about your question. First, I believe that causation can only be tested by means of well-designed experiments. To do this, you need good theory. I think that statistics can never "prove" causation. 

My second concern is that, as far as I understand, you imply that there is a unidirectional causation. I doubt whether this can be the case in any psychological phenomenon. So, "proving" such a direction statistically could be misleading. 

Ilona, I will check the site you suggested. I am very excited about these kinds of analyses (although I doubt on their usefulness!). I hope I will understand something/anything!

Georgios. 

Dear Laurence,

If you really mean causation in the literal sense: causation is something we principally cannot access on an empirical basis. It is simply impossible, even in "hard" natural sciences. As observers of phenomena, we can only assess correlations. Refer to Wittgenstein, "On Certainty". This comment is not meant to be polemic; from a logical point of view, it just the way it is.

Ricarda

Hi there,

you might want to check out "Direction of Effects in Multiple Linear Regression Models" by Wiedermann & Von Eye (DOI:10.1080/00273171.2014.958429). A discussion of causality (e.g. granger causality) is also included in the introduction of "Direction of Effects in Mediation Analysis" (same authors, same year). Let me know how your are doing!

Best, Simon

Dear Simon - thanks for your answer i will have a good look at it. 

So far the mediation model suggested by others looks promising. 

Further reading about mediation : 

Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach (Methodology in the Social Sciences) Hardcover – 14 Jun 2013

by Andrew F. Hayes (Author)

I found this book to be quite useful :)

Best Regards and Thanks 

Lawrence