non-causal interpretations of QCA results

[Yves Boulmer]

dear all, I am relying on QCA in a study performed in the context of innovation funding. Intellectually, being able to discover/learn/apply QCA has really been a game-changer. I hope to be able to contribute to the further development of the method. In my current paper though, I struggle a little bit on finding a good way to explain/justify a non-causal interpretations of my QCA results instead of claiming a causal relationship between the conditions and the outcome. As the whole approach is grounded in "causal complexity" it is challenging to both refer to causal complexity on the one hand and have a non-causal interpretations of the results on the other hand. 

Would you have any thoughts on this?

I thank you in advance

Best regards

Yves     

[Adrian Dușa]

Dear Yves,

Many start to avoid the term causal in QCA, if Iam not mistaken the term explanatory conditions is coined instead.

A true “cause” is not something a computer can detect, this is (at least to me) related to human understanding.

Computer algorithms detect regularities, that is how consistent the presence of a condition is to the presence of the outcome.

So I wouldn’t worry that much about causality, but instead focus on comprehensive explanatory models.

Otherwise, some more information is needed: what exactly do you mean by non-causal interpretation?

Best wishes,

Adrian

--
You received this message because you are subscribed to the Google Groups "QCA with R" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qcawithr+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/qcawithr/aedaa32b-1d4f-4e9a-b077-3ea0fcc0c3a6n%40googlegroups.com.

[Yves Boulmer]

Thank you Adrian for your swift answer! also because I realise that my inquiry is not specifically on QCA using R (but you should see my pc-screen, definitely a lot of QCA using R going on!).

I am trying to address a comment (that I guess many have received) about "being very careful not to claim causal relationships with QCA".

Very interesting to read that some have started refer to "explanatory conditions", I will look for examples in recently published papers. Thank you for sharing this! I guess one needs then to frame references to "causal complexity" appropriately and probably getting back to Ragin's work.

When referring to a "non-causal interpretation"  I am referring to "descriptive/exploratory purpose", concluding on different typologies based on the different configurations of conditions "associated with" an outcome.

Best

Yves

[Ingo Rohlfing]

The quote in your email seems to echo concerns that QCA is not useful for causal research. It would be interesting to know what the basis for the concerns are, but I guess this has not been explained. Michael Baumgartner and co-authors have written a series of papers on set-relational causal analysis (with a focus on his preferred algorithm coincidence analysis, but still interesting from a QCA perspective). 

I am skeptical that much would be gained by calling the conditions explanatory. The question is how one defines "explanatory". I think that in the social sciences explanations are meant to be causal, in particular in the case-based, small-n, process tracing line of work. There is philosophical work on explanations, including distinctions between causal and non-causal explanations, which can be mathematical. Another possibility would be to use QCA for predictive purposes. I do not know any work on predictive QCA, but I don't see a fundamental problem with that.

I guess it is easier to use QCA for typology building. The analysis would then stop with the truth table, where each row represents a different type. (Some may say that it is not a proper QCA study without minimization; I have no strong feelings in that regard.) In this book, https://mitpress.mit.edu/9780262572224/case-studies-and-theory-development-in-the-social-sciences/ George and Bennett have one chapter on typological theorizing based on what I would call a truth table. They do not explicitly link the approach to QCA, but this is not a problem and may be useful.

Kind regards

Ingo

[Adrian Dușa]

I think the discussion regarding causality is important, and the shift towards "explanatory" instead of "causal" is actually caused (sic!) by none other than the latest drama around Baumgartner and the CNA vs QCA dispute.

It all boils down that QCA is actually an analysis of sufficiency, and theory itself purports that sufficiency is not the same thing with causality. Not everything sufficient is causal.

Or, QCA and the algorithms behind can indeed detect sufficiency but in the light of the above it cannot be said with 100% certainty that QCA detects causality. I think this is an inferred, human conclusion.

Adding to the fact there are multiple ways to infer sufficiency (CNA uses the inferior, in my opinion, propositional logic), it makes one think twice equating the "sufficient" reported models with actual causality (as Baumgartner himself writes: "in the real world").

This whole back and forth methodological discussion makes me think of QCA in terms of "explanatory" models, and I need to mention this term is not my own. I picked it up because it seems to have face value, it just makes sense. Whether QCA models are causal, that is something the researcher can decide, but I would be cautious pointing to QCA discovering (true) causality.

With all my best,

Adrian

[Ian Greener]

Thank you for this thread. It's made me think hard about some really important things. Here's where I think I've got to.

When we are putting our explanatory/causal factors together we aren't doing so randomly - at least I hope not. We are looking at existing research, being very picky about which claims we think stand up, and only then incorporating the factors for inclusion. We might pick factors based on their place within a more general framework that captures different dimensions of what we are trying to explain, so we are coming at our outcome from alternative angles.

Often the claims others have made were not produced using QCA, but more linear approaches. That's fine - and it hopefully makes clear the value we might add. But we are starting with factors which others have generally made claims about - often causal ones.

When we put these factors into our QCA, we are typically looking for complex patterns of them in relation to necessity and sufficiency. Where we have factors that are based on existing research and which form strong sufficient solutions, and that corresponds with what we know about the cases which are in the sufficient solution, then I think we can say we have a claim to be making a causal claim. As Adrian says, we can't see causality , but instead only trace it through our imperfect factors and by worrying about our calibrations and other assumptions. But when it all adds up, and we get the 'aha!', then that's for me a causal claim.

Best,

Ian

--

You received this message because you are subscribed to the Google Groups "QCA with R" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qcawithr+u...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/qcawithr/CAJ%3D0CtCDGnz3CPAs58hT7dfdxmN9Otva-O5C21OdgZ_Mt2F1ug%40mail.gmail.com.

[Yves Boulmer]

thank you all, Adrian, Ingo, Ian for a very interesting conversation indeed and for sharing your thoughts on this topic. Extremely valuable!

thank you very much indeed.

Best

Yves