Dear Hanieh,

There is a method called Coincidence Analysis (CNA), which is closely related to QCA, but has native capabilities for analyzing multiple outcomes. In fact, CNA imitates a batched QCA run on multiple outcomes (see Baumgartner_Thiem_2015_RJ.pdf, which is attached, as well as article publication entry 13 on my website at http://www.alrik-thiem.net/about/; if you want a copy of that latter study, or any other, please simply send a private request via ResearchGate). The result of CNA is a so-called "complex solution formula", which combines the individual solution formulas for each outcome by conjunction.

Note in this connection, however, that you should not confuse QCA's "complex" / "conservative" solution with CNA's "complex solution formula". CNA only imitates QCA's so-called "parsimonious solution", for a very good reason. The complex / conservative and intermediate QCA solution are demonstrably incorrect procedures of causal inference (unfortunately, many applied users of QCA continue to use these two solution types; for a study on this problem, see article publication 22 on my website at http://www.alrik-thiem.net/about/).

In any case, all necessary details about CNA can be found in its package documentation.

If you have further questions, please don't hesitate to ask.

Best wishes,

Alrik

Dear Alrik,

Thank you very much for the very helpful information. It was great to see that the version 2.0.0 of CNA can also be used for MV and fs.

I will start using the package and if you don't mind I may get back to you with some new questions.

Thank you for all your academic contributions,

Hi, dear friend, have a good time. The qca results are different. In each design, your output circuit will be generated. good luck.

Dear Hanieh,

QCA can deal with multiple outcomes, at least in the QCA package in R (version 3 and above), which offers a function called causalChain().

CNA is indeed very similar to a batched QCA, and they give exactly the same solutions given the proper options. On comparing QCA with CNA, as well as to reveal the deficiencies of CNA (among which the most serious it is not guaranteed to be exhaustive), see here:

https://bookdown.org/dusadrian/QCAbook/QCAextensions.html#cna

The affirmation that "complex / conservative and intermediate QCA solution are demonstrably incorrect" is obviously false, and has been rejected by the community of QCA researchers, see here:

http://www.compasss.org/files/compasss-rejection-policy-statement-20170814.html

I hope this helps,

Adrian

PS: note that package CNA version 2.0.0 can indeed deal with mv data, but not with combinations of cs, mv and fs data. It can only deal with mv data when "all" data are mv, which is far from the usual research situations.

dear adrian

Please explain the reasons for publishing qca articles?

Dear Hanieh,

The reply by Adrian Dusa now leads us into a little bit of the sociology of science, but it seems necessary to delve a little bit into this lest you fall into a QCA trap in which you waste your precious data. In addition, it is important if you would like to understand the current situation in the QCA literature a little better, which can be confusing for outsiders and beginners. Maybe others will also read this post.

As you can easily see from my publications, I have been a co-author and the main promoter of the QCA package from 2012 to 2015. However, when I discovered that this package tricks users, just like the popular fs/QCA software, by standardly presenting them with results that are often much cleaner than the analyzed data warrant, and Adrian Dusa refused to correct this, I left the package to launch my own package, called QCApro, in 2016. This package is guaranteed to present users with correct solutions, and it warns them against using unsuitable procedures such as the conservative or the intermediate solution.

The COMPASSS statement that Adrian Dusa also points to, is, as you can see, not signed by the "QCA Community", but by a small circle of researchers, and their acdemic dependants, who have a lot to lose (both financially and reputationally) from the correctness of Michael Baumgartner and my publications. This statement indirectly calls on researchers to ignore Michael Baumgartner and my publications, but it does not present a scientific argument. As such, it simply is propaganda intended to influence applied users.

You are of course free to decide which people in the QCA community to believe. It can be a hard choice at the moment for applied researchers. But the very fact that Adrian Dusa and his peers from the inner circle of COMPASSS felt compelled to take such drastic measures like the COMPASSS statement only shows that more and more researchers, including the many reviewers of all my publications and all the applied researchers I have worked with and currently work together with, are, fortunately, not tricked any longer by their writings.

I assume Adrian Dusa will again reply to this post of mine, as he has done repeatedly in the past, but in the interest of not making this thread here another sociology-of-science battlefield, I will not respond again. To reiterate, I am happy to help out with any further questions you may have.

Best wishes,

Alrik

Indeed, this kind of "debate" is not new and need not be prolonged here. In short, it is about the opinion of a very restrictive minority of the QCA community against the opinion of the larger community (what Alrik Thiem names the "inner core of COMPASSS" is actually comprised of the most respected theoreticians in the field).

The QCA package in R was developed by myself long before AT joined as a second co-author (since 2007, to be exact), and continues to be developed in line with the community established standards, reaching a very mature state with cutting edge developments after version 3.

What AT is right about, is only one thing: that you should make your own opinion.

Best wishes,

Adrian

Dear Alrik and Adrian, I would like to thank both of you for this interesting and helpful debate. I think these debates are necessary for any scientific community and there will be no progress in case all scientists agree with each other.

Sorry Hanieh for using your thread again, which was initially about multiple outcomes, but since this topic has come up now, I thought I could as well invite more people to contribute so as to try and achieve some progress for the benefit of science.

Dear Adrian and "most respected theoreticians in the field",

Could any of you who have signed the argument-free COMPASSS statement urging people to ignore my publications please explain why applying the conservative solution to the attached test data yields the finding that all of the ten exogenous factors A to J are causally relevant to the endogenous factor Z (see screenshot)? I know of no other method that infers causal relevance for 10 factors on the basis of only 2 cases, and for factors that do not even vary across cases! To anyone only slightly trained in causal inference, with whatever method, this should sound alarm.

Here's my perspective:

Obviously, to only J can causal relevance with respect to Z be ascribed; it is the only factor that varies when Z varies, holding everything else constant. Try running the parsimonious solution, and there you go: J{0} is the only condition that is returned! Well done parsimonious solution! (see screenshot)

So why does the conservative solution seem to mess it up? Because it introduces a massive amount of artificial data through the back door, data that even often violate the very data-generating structure researchers are after.

How does the conservative solution introduce such data? By blocking access to remainders through assigning them a "0" on the output, the minimization algorithm (Quine-McCluskey in this case) is now forced to treat every remainder X as not sufficient for Z.

But under which circumstances is the sufficiency statement X -> Z false? Only when X exists, that is when X is treated as if it was a real case, and when it occurs in conjunction with Z{0} (build your own truth table to check this!). Only on the basis of these artificial, yet to the user unfortunately invisible data can the conservative solution ever conclude that all ten factors in a dataset of only two empirical cases, of which nine factors do not even vary, matter for explaining Z.

All this is consistent with the millions of simulations Michael Baumgartner and I have carried out here: Article Often Trusted but Never (Properly) Tested: Evaluating Qualit...

The screenshots are from the fs/QCA software, but you can also use Adrian's QCA package, or the Tosmana software...the results will be the same.

Best wishes,

Alrik

An RG thread is definitely not the place for such a response, but please rest assured that I am preparing a formal reply to the SMR paper.

As an advance conclusion for the other readers of this thread, the B&T paper from SMR can easily be dismissed as false. In order to arrive at their conclusions, B&T used a very custom definition of what is considered "correct".

In a parallel universe (parallel to the rest of the community), anything goes. But if we employ the community agreed standards, their "findings" fall flat.

Not to mention their perspective come in direct contradiction with the 60 years old, well established Quine-McCluskey (QMC) minimisation procedure (which is independent from QCA). Should a naive reader believe what they claim, this is equivalent to claiming the QMC algorithm is plain wrong, and this is obviously a very bold claim (good luck to AT in convincing the engineers about that).

Best,

Adrian

I think that as long as everyone remains polite, a place such as RG is perfect for debate because it is open to everyone and free, whereas journals have to be paid for and are walled by reviewers and editors. But I'm looking forward to your formal contribution in a journal if you don’t want to sketch out your argument on RG.

In any case, the Quine-McCluskey algorithm (QMC) in fact does not know "conservative" solutions at all, contrary to what you claim. It only knows "parsimonious" solutions. Switching functions that show so-called "don't cares" (remainders) are referred to as "incompletely specified functions" in electrical engineering. For deriving a minimal sum from such incompletely specified functions, QMC draws on the logical remainders in the same way the parsimonious solution in QCA does, for a very good reason: because it leads to the correct solution.

“Conservative” and “intermediate” solutions have only been invented by Charles Ragin (in 1987 for the conservative solution; and then in 2005 for the intermediate solution) because Ragin thought QMC needed to be tweaked in order to make the algorithm fit for social research. That was a huge mistake, and has led to the current state of affairs.

Have you ever asked an electrical engineer whether s/he has heard of “conservative solutions”, or whether s/he has made QMC work according to the logic of conservative solutions? I can guarantee you that you will not find anyone. Instead, QMC is perfectly imitated by QCA’s parsimonious solution, plenty of proof for which can be found in all major electrical engineering textbooks. Here're some of them:

• Edwards, Frederick H. 1973. The Principles of Switching Circuits. Cambridge, MA.: MIT Press. (page 95 onwards)
• Hohn, Franz E. 1966. Applied Boolean Algebra: An Elementary Introduction. 2nd ed. New York: Macmillan. (page 213 onwards)
• Lewin, Douglas, and David Protheroe. 1992. Design of Logic Systems. 2nd ed. London: Chapman & Hall. (page 83 onwards)
• Roth, Charles H., and Larry L. Kinney. 2014. Fundamentals of Logic Design. 7th ed. Stamford, CT: Cengage Learning. (page 103 onwards)

In particular, you may want to consult Edward McCluskey himself:

McCluskey, Edward J. 1965. Introduction to the Theory of Switching Circuits. Princeton: Princeton University Press. Page 126: "Any d terms [don't cares] which are present are treated as 1 terms in forming the prime implicants". That single sentence should sum it all up very neatly.

Hence, it is you and your statement co-signers from COMPASSS who not only have to argue against what you pejoratively call "a restrictive minority" living "in a parallel universe", but also all electrical engineers.

Best wishes,

Alrik

I am not certain if this reply is a joke or a serious one.

But just to clarify things: have you ever tried to (personally) perform a programming exercise of the QMC algorithm?

Because I have done that, and contributed not only one but three different implementations of this algorithm. And before a certain reply is given, my answer is "no!" you have not contributed to eQMC, that has been single authored by myself from 2007. You co-authored the paper describing it (in 2015), but apparently that wasn't enough.

Until you have done this programming exercise, I believe you are not qualified enough to have a serious opinion. Throwing out literature references does not make one an expert, those have to actually be read and properly understood.

Now back to the issue: indeed, QMC is unaware about "conservative" and "parsimonious", but on the other hand your opinion that QMC "only knows parsimonious solutions" is as false as it can be.

QMC is a function of input minterms. And engineers (just like in QCA), have an option whether to include "don't cares" (or not).

If that function contains only the observed (positive) gates, the engineering solution is exactly the conservative solution. If "don't cares" are included in the function, the solution is parsimonious. Just like in QCA, we have the option to include (or not) the remainders.

This is precisely the procedure described in all those references (should you bother to properly read them), but just in case your time is precious, here is a quick random description of QMC from Google:

http://pami.uwaterloo.ca/~basir/ECE124/QL.pdf

On slide 4, there are two sums (one over "m" and one over "d") and the author does mention in the round brackets "and don't cares". Therefore it is an option to include the don't cares into the minimization, not an automatic procedure as you seem to suggest.

It makes sense for the engineers to include the don't cares, because they are looking for the simplest possible (lowest cost) solution to a switching circuit. Just as it makes sense for QCA researchers to include (some) remainders, in the attempt to further refine solutions and eliminate causally irrelevant factors.

To conclude, QCA is perfectly compatible with QMC, therefore your opinion that QCA is "wrong" is the same thing as stating QMC is wrong.

Quod erat demonstrandum.

Yes, exactly what I said. When you make use of only the positive terms, and do not allow QMC to access logical remainders, you get the conservative solution. BUT: no engineer would ever, ever do this, because s/he would not be guaranteed any longer to obtain all prime implicants...which means that the circuit would become more expensive than it needed to.

Whereas in electrical engineering, using all remainders is an economic imperative when optimizing with QMC, in causal data analysis with QCA, it is an inferential imperative! If you bar QMC from accessing remainders, you are at high risk of introducing into the final solution conditions for which there is no empirical evidence whatsoever that they are causally relevant to the outcome (see the deliberately extreme example of my test data above and the corresponding explanation). In the worst case, you thereby commit a whole raft of causal fallacies (as I showed in the Sociological Methods & Research article). With the parsimonious solution, that never happens (given the usual proviso of non-violation of background assumptions).

Best wishes,

Alrik

You previously said:

Now you say:

These two sentences of yours are logically contradictory. One thing is to say engineers care more about the "parsimonious" solutions (that also happens in QCA, I have rarely seen publications restricted to the conservative solution), and a completely different thing is to claim that QMC knows the parsimonious ones.

But absolutely, and make no mistake about it: the parsimonious solutions commit causal fallacies just as well!

As I am going to demonstrate in the upcoming workshop in Cologne, the parsimonious solution is simply too greedy and tricks users into believing that insufficient conditions are in fact sufficient.

If the conservative solution is considered inefficient into removing causally irrelevant factors, the parsimonious solution is aggressive, over-zealous and eliminates too much. That leaves the intermediate solution as the closest possible one to the truth.

Oh and by the way, this can and will be demonstrated against your own data from the SMR article, using your own replication code.

Hi Adrian and Alrik,

I have following questions and would very much appreciate your response:

- Are there any publications (apart from the helpful book-chapter suggested by Adrian) on similarities and differences of the causalChain() and CNA()?

-What are limitations around maximum number of factors (conditions (independent) and outcomes (dependent)) for QCA (since it seems Quine-McCluskey is solving an NP-hard problem) and for causalChain() and CNA()? Does treating a condition or outcome as an mv help?

-I think there is a high possibility that I end-up using both causalChain() and CNA() and compare the results. Alrik has already suggested some helpful applied publications that use cna(). I would appreciate if Adrian can also suggest an applied publication that uses causalChain().

Regards,

Dear Hanieh,

Although I had already shown in 2015 that causal chains are easily implementable in QCA in my 2015 SMR article "Using Qualitative Comparative Analysis for Identifying Causal Chains in Configurational Data" (see http://smr.sagepub.com/content/44/4/723.abstract), there's no publication yet that has used the causalChain() function in the QCA package (some rudimentary capability for causal chains had already existed in the QCA package for quite some time).

Nor does there exist yet a formal comparison between causalChain() and cna() beyond the isolated example provided in Adrian's book. That would be something Michael Baumgartner and Adrian would have to slug out, I guess. After all, Adrian claims that his causalChain() function is superior to cna(). With empirical data, you have no means of finding out which function is superior. That requires a formal method evaluation with simulated data, or a mathematical proof.

Best wishes,

Alrik

Hi Hanieh,

It would indeed be difficult to compare causalChain() with cna(). In some respect they are very similar (both use a bottom up search strategy), but they are fundamentally different in the sense that causalChain() uses a truth table as an input (QCA style) while cna() uses something called a coincidence list (CNA style). My function is relatively new, and time will tell what it will become.

As already pointed in that chapter, the function cna() is not guaranteed to be exhaustive, therefore one never knows if it finds all possible solutions for a particular data (which is precisely what CNA is all about).

According to the CNA package, that function can only derive the parsimonious solution. But since it is already clear this solution type creates causal fallacies, in my honest opinion it is inferior. And CNA is definitely incapable of deriving an intermediate solution: this is visible and needs no further testing.

As for the maximum number of factors, I believe (by far) the superior package is QCA, given it employs a cutting edge new algorithm called CCubes:

Working Paper Consistency Cubes: A Fast, Efficient Method for Boolean Minimization

Solving these problems is indeed NP-hard, but for this purpose CCubes is even faster than Espresso (3.5 times faster, to be exact), the all time champion software from engineering. It also depends on the number of observed truth table configurations, but I have successfully obtained solutions even for 25 factors, and I believe it can work up to 30. This should be quite enough for any social science analysis.

Best regards,

Adrian

Dear Hanieh,

There are a couple of important points to consider in Adrian's answer:

• Unless causalChain() and cna() have the same search target, it's pointless to compare them. However, if they have the same search target, which I assume they have since both target configurational causal chains and since Adrian has argued that causalChain() is superior to cna(), they should be comparable. I would bet all my money on cna() in such a comparison, for at least the following reason: Adrian's causalChain() function is apparently only guaranteed to produce optimal results with intermediate solutions, which are highly inferior to parsimonious solutions (yet still better than conservative solutions).
• Regarding the number of factors: the number of (possible) prime implicants is a (partial) function of the number of exogenous factors k, and it can reach a number as high as 3k / k1/2 . If you have a research design that requires 15 variables or more, you're very likely to end up with no informative results whatsoever unless you have an enormously large amount of data. This problem is called "model ambiguity", on which you find papers here http://erx.sagepub.com/content/38/6/487.abstract and

  Article Model Ambiguities in Configurational Comparative Research

  . Also, if you require so many variables to resolve inconsistencies in your data, you're very likely to have a highly uncontrolled research situation. And if you seek to analyze causal chains, this problem in fact multiplies with each additional level of the chain.
• Adrian's speed comparisons are inadequate, and because of the previous point, pretty irrelevant to social research. First, you cannot simply claim that some algorithm is so many times faster than some other. You can only perform benchmarks tests for very specific minimization problems. This is why electrical engineers have developed standardized benchmark tests. Espresso, for example, is an algorithm that was devised for heuristic minimization with hundreds of variables, not for exact minimization with few variables (in electrical engineering, 30 variables is not many). Similarly, Quine-McCluskey performs pretty well when the ratio of 1-terms to d-terms is of a high magnitude. In contrast, as this ratio approaches 0, QMC begins to break down (recall the test data above; I could have told you with my bare eyes, in the blink of an eye, that the solution is J Z; QMC would have taken a couple of seconds; or to push it to the extreme: add 30 more constants to these test data and my eyes will easily outperform QMC on any super-fast computer).

Best wishes,

Alrik

To be honest, it is getting increasingly tiring to keep correcting AT's misunderstandings, and I am starting to wonder whether it is worth it. The only reason for which I am (still) responding is to protect possible other readers from believing nonsense (just because there is no "counter" response).

The only valuable thing in his reply is the so called model ambiguity, in fact a fancy term for equifinality. That outcomes are produced via (many) other causal paths was well known in the literature, it needed no new term.

The fact that "...intermediate solutions... are highly inferior to parsimonious solutions..." is (again and again) a marginal opinion coming from very few people, and already dismissed by the community.

As for the "...causalChain() function is apparently only guaranteed to produce optimal results...", I would suggest AT to refrain from having opinions on things he does not understand, or did not test.

The same advice about algorithms, which he continues to prove ignorance about. For instance, he does not seem to know that Espresso (despite its heuristic nature) has been programmed, from the very beginning, with an "exact" version as well. See, for instance, Rudell's (the author of Espresso) original thesis, section 2.4:

http://www.cs.columbia.edu/~cs6861/handouts/rudell-PhD-thesis.pdf

It is absolutely obvious that I don't compare CCubes with the heuristic Espresso, but with its exact version. This reinforces my point from an earlier reply that, in the absence of a programming exercise of the QMC algorithm, AT is definitely not qualified enough to have a meaningful opinion.

Dear Adrian, I find your continuing personal attacks heavily unprofessional. That style of communication is perhaps what you're used to, but I don't see any value to this. Either you have an argument and present it in a clear way or you don't have an argument. I think you well know that you can't dismiss my publications simply as "nonsense" or argue that "AT is definitely not qualified enough to have a meaningful opinion." I understand you find my publications discomforting because they shake your long-held convictions to their foundations, but you should have means other than derogating remarks to express your views.

Equifinality is not the same as model ambiguity, by the way. The former refers to the presence of multiple causal paths within one model, the latter to multiple models within one solution.

In any case, I suggest we leave it at that here. I think we have made some progress on this important front of inquiry in QCA, if not as much as I had wished for. I will certainly be looking forward to your presentation in Cologne in about a week. So as to be able to prepare some points for a more in-depth discussion of your criticism on my SMR paper, would it be possible that you pass around your conference paper among participants? I have already sent mine to Ingo, but it’s not on solution types.

Best wishes,

Alrik

The "nonsense" was rightfully attributed to the (false) opinion about CCubes vs. Espresso, to the (false) opinion that QMC only knows about the parsimonious solution and to the (again false) opinion that causalChain() is "apparently only guaranteed to produce optimal results with intermediate solutions". AT's subsequent silence about these issues is tale telling.

People with limited and incomplete information about a certain topic tend to suffer from confirmation bias, as I believe this is the case.

My replies are meant to clarify things (for the readers), but I am not surprised they are interpreted as "attacks": it is typical for someone who frequently uses this kind of language. I urge AT to (please!) stop intoxicating the public space with this kind of "fake news" opinions.

Dear Adrian, I don't know where your deep-seated frustration comes from, but it doesn't seem to be a coincidence that you only started criticizing my writings as "non-sense" right after I had ended collaborating with you in 2015.

Whatever the reason, I think you should now get your act together and focus on publishing scientific arguments, instead of keeping stalking me with your Trump-tweet-style rants about "intoxications of the public space" and "fake news opinions". However, should you decide to bring your style of communication to a more professional level, please let me know and I'd be happy to discuss topics of scientific interest with you again here on ResearchGate.

Best wishes,

Alrik

This is a little too close to an autistic discussion: I keep writing one thing, and it is repeatedly mis-read and misinterpreted. I am confident this will be evident for the other readers.

For the record, and (hopefully!) for the last time: I did not criticize your writings as nonsense. For those, I will have my response in a dedicated publication (by now you have already read the paper sent to Cologne, therefore you know I have my arguments).

What is considered "nonsense" was clearly explained in my previous reply, namely your opinions which lack a solid theoretical foundation and induce obvious confusion.

Let us not deviate a scientific discussion into personal matters: you are welcome to defend these specific opinions, should you bring scientific arguments to support them. Otherwise, my advice stands still: please refrain from polluting the public space, or expect this kind of reaction.

Yes, I have received your paper. However, I was surprised to find the first capital mistake already on page 2. Neither Mackie's INUS theory nor any related theory of causation, nor any version of the Quine McCluskey algorithm for that matter, would ever consider a solution such as A*B -> C if it still held in the data (or the function table for QMC) that A -> C.

For Mackie, B can never be an INUS condition in the presence of A and thus it cannot be a cause of C in the presence of A, and for Quine-McCluskey, an AND-gate that includes A and B is always more costly to build than an equivalent circuit consisting of only A, so QMC would never output A*B because C could be operated by A alone, or by B alone. Your argument thus falls apart right at the beginning.

I was told that you have already been alerted to this fundamental problem at the Zurich workshop, but you seem to be immune against any way of reasoning. Thus, in order not to having to repeat in Cologne what you seem to have already been told, what else shall I say about your paper?

Best wishes,

Alrik

The paper is not written for you, but for the community. Since it directly contradicts the conclusions from the SMR article, your reluctance to accept my arguments is unsurprising. It remains to be seen what the community's opinion is.

However, and I must stress again: your limited experience with the Quine-McCluskey algorithm is not exactly a landmark for a serious opinion (it is actually amusing to read how someone with zero experience in QMC programming preaches what QMC really is, to someone with a solid and most importantly proven expertise in the field).

This is not the place to explain why A cannot operate alone, it would suffice to say it is part of the prerequisites of the experiment. We already know (from the initial, known setup of the problem) that both A and B are needed to trigger the outcome, therefore what you claim is false by definition.

In a serial circuit, both electrical switches are needed to activate the outcome (turn on the light), while the parsimonious solution states the switch A alone can do that, which is obviously impossible. This is one example from many, where the parsimonious solution fails.

But please keep on reading, for the rest of the paper is even more interesting: using your own code, and your own data, the parsimonious solution (PS) is outperformed by the intermediate solution (IS), even for the initial setup involving a parsimonious starting solution. That is, the IS has a higher recovery rate than the PS itself.

Oh and by the way, you must have seriously misunderstood Mackie. Should you have read the entire paper, you would notice that he in fact contradicts the parsimonious "superiority". See for instance note number 7, where Mackie (1974, p.61) himself writes: "... All ABC are followed by P, but it is not the case that all AB are followed by P ..."

Which means it is not (always, as you imply) advisable to overly-minimize. Adapting to the example in my paper, both AB are followed by P, but it is not the case that A alone is followed by P.

Quod erat demonstrandum (again). I understand it must be very frustrating to have your "grand" theory dismissed with such ease, but that is life, you'll have to learn how to deal with it.

Reference:

Mackie, J.L. 1974. The Cement of the Universe: A Study of Causation. 2002 reprinted ed. Oxford, UK: Clarendon Press.

Dear Adrian, would you then mind uploading your paper to RG so that the community also has a chance to evaluate your argument and my counter-argument? That'd be great, thanks. You can, of course, also wait until you've managed to publish it somewhere if you don't want to share it now, but then I suggest you mince your words as long as you haven't succeeded in doing so.

Just because you've programmed something, that doesn't mean you've understood what you've programmed, let alone programmed correctly what you thought you had programmed. Anything can be programmed in R by anyone. It doesn't take much. Just see the long list of bugs that were corrected since you've published your first version of the QCA package! If someone had not noticed them at some point, they would still negatively affect QCA analyses with your package, and maybe some errors haven't even been detected yet. In other words, having programmed what one thought is QMC is neither necessary nor sufficient for having understood QMC.

Now to your argument regarding my SMR paper, which I have now read in detail. My verdict: your list of errors gets longer the further you develop your argument. First of all, let's go to your very first argument, on which you already base your first conclusions about the incorrectness of QCA-PS.

If two professors have to give a pass (A and B) for a student to pass (C), but professor B always does what A does, why should the correct causal structure be A*B -> C as you claim? You merely posit this, but it's far from clear that A*B -> C is the correct representation of this mechanism. To show you why, let me first pose a counter-question: what would the causal structure be if B were NOT deferential to A, but decided on her own? The answer: A*B -> C! Hence, either your mechanism is not correctly represented, or the fact that B is deferential to A is a superflous fact in the description of your mechanism.

You're not convinced? Ok, then let me try a second option. Because A and B are perfectly correlated since A is the one and only cause of B, you can also summarize A*B as Z. Whenever Z = 1, A = 1 and B = 1; whenever Z = 0, A = 0 and B = 0. Since neither A = 1, B = 0 nor A = 0, B = 1 is ever possible, that is a perfectly legitimate transformation. What you're left with is a simple table

Z C

1 1

0 0

Can you prove QCA-PS incorrect here? I can't. You're of course welcome to modify the set-up and try hard to make it more complicated.

Now to your reference to Mackie. Mackie here refers to evidence for declaring some condition to be an INUS condition. He simply means that sufficient conditions must be free of redundancies before being causally interpreted. If "... All ABC are followed by P, but it is not the case that all AB are followed by P ...", then there simply is evidence for the causal relevance of C. That is all there is in this sentence. Mackie is not an easy read for those who've done it the first time; I suggest you do it again.

By the way, what do your colleagues from the COMPASSS rejection statement say to all this? I mean, they've never programmed QMC in R, which, according to your words, one must have done before being able to understand QMC. Effectively, this then leaves only you as a serious signatory of this letter?

Best wishes,

Alrik

Finally, you read my paper in detail...! That you rutinely practise superficial reading I already suspected, but do you at least read what write?

Is this for real? How is it possible that I don't understand what I am programming? And this priceless one: “...let alone programmed correctly what you thought you had programmed...”

I've written three different implementations of the QMC, and all of them give exactly the same results (which is in itself a validation proof, small bugs don't count), and additionally they are cross-validated with Tosmana, fs/QCA and Espresso. Through which miraculous thinking process did you infer that my programming is incorrect? And if it is, how come you use a copy of my software into your package?

But funny enough, that made me thinking: what if you are right...? Since I do not understand my programming, I must have wrote that under some sort of a spell. Of course...! Now that you mention it, I do vaguely remember (like a mist in a dream) there was a force pushing my fingers on the keyboard. Yes, it's becoming so clear now! Thank you, oh, Great AT! Praised be thy name! Thank you for revealing the real reality to the unworthy myself, I will forever cherish this revelation moment and honor thy presence. This is better than Matrix, what pill would you now recommend, the blue or the red one?

I've given up explaining the results from QMC to you. It is by now clear that some sort of mental blockage prevents you from accepting arguments that are different from yours. You still seem unable to comprehend the correct causal structure (CNA style, multiple outcomes) is not just A*B -> C, but (A B)(A*B C), and that is not something to question why or if it should be like that, it is a fact. It is the initial, known setup of the experiment, and the PS fails to recover it. Incidentally, your ZC example leads to the conservative, not to the parsimonious solution. And yes, there is a second example in the paper where B is semi-deferential to A and the PS still fails to recover the true causal structure.

Before taking any suggestion from you, let me offer a small piece of advice: if you want respect, you should first respect others. Displaying such an impertinent disrespect, not only to my work but also to an entire community will not get you any further.

To reach the recognition you so desperately seek, you should first do your own homework (copying one's software not acceptable) and do proper reading to solve your ignorance (remember Espresso-exact, that you had absolutely no idea about, yet had strong opinions about my speed comparison with CCubes).

Given the time you have spent in the academia, you should have learned by now that ignorance is not an excuse. Rather, it is the explanation for such strong headed opinions.

Dear Adrian,

new arguments that deeply threaten conventional wisdom always meet fierce resistance from those who've established this wisdom. Kuhn's fascinating Structure of Scientific Revolutions or Mukherjee's magnificent The Emperor of All Maladies are highly recommendable readings in this respect. Unfortunately, the social sciences do not attract that much attention from historians of science as physics, chemistry, or oncology, but I didn't expect anything less when we published the SMR paper.

However, I'm not interested in sociology of science arguments here nor your pseudo-psychological assessments. I'm interested in your argument regarding the incorrectness of the parsimonious solution type, and the supposed superiority of conservative and intermediate solutions. So, I repeat my questions, and please, don't digress again in your next reply but try to respond to them.

Best wishes,

Alrik

This is not for you, but for the readers:

1. No, I have my own reasons.

2. I don't have to think, I know! It is the initial setup of the problem. If B would not be deferential to A, we would have observed dissimilar gradings. And even in the unlikely case we did not observed those, the CS is still more plausible than the PS, because at least it uses observed data, as opposed to counterfactuals like the PS. Using counterfactuals (remainders) is always a risk factor: one can be right or wrong, but never know for sure.

3. By saying that A (mid-day) is always related to B (high temperature, compared with night temperatures) one cannot possibly assume that A*B is the same as B*B, for A is not the same with B. They are simply always related, yet different.

For the millionth time, I don't have to assume. We know, because we can decide the initial causal structure in order to test the PS. This is exactly, but exactly the very same strategy adopted by you and Michael in the SMR paper. It was valid when it suited you, but not valid now as a counter-argument. Oh well, this is yet again a very sterile discussion.

And by the way, new arguments must be presented with respect and deference. If you think one can force an entire community to a different direction, good luck. Not going to happen in your lifetime.

Dear Adrian,

1. This is your decision, but you're expecting people to simply believe your public claim that Baumgartner and Thiem publish non-sense, while you deny people access to your arguments. That's a pity because it deprives interested readers of any possibility to evaluate your public claims, and it deprives me of the possibility to counter your public claims.

2. Your explanation misses my point. A*B -> C is the representation of a structure in which profs A and B grade independently of each other: unless prof A gives a pass, C (the student passing) will not obtain; unless prof B gives a pass, C (the student passing) will not obtain; A*B is a minimally sufficient condition in the very sense of Mackie. However, if B always defers to A, A*B can never be a minimally sufficient condition for C; in such a situation, an adequate representation is (A -> B)*(B -> C). Whenever A gives a pass, B takes her cues from A and also gives a pass; and whenever B gives that pass, the student passes.

3. You're not doing the same as Michael and I did in the SMR paper. You're linking a logic function (A*B -> C) to some causal story that doesn't fit this function. All we did was to stipulate logic functions that had a very specific form, namely those of minimally necessary disjunctions of minimally sufficient conditions, exactly the kind of structure Mackie says causal structures in our world have. However, we did not invent any story for these structures. We simply tested under which circumstances QCA could recover (at least a correct part of) these functions.

I don't want to enforce anything on anyone, and I find it unfortunate that you think this way. All I do is to try and improve methods, with no leanings whatsoever. I'm neither a proponent of QCA nor an opponent, I'm just a scientist who thinks that QCA and related methods are fascinating and deserve attention.

Thus, if I discover something new that withstands others' attempts of refutation, then this is good news for me and for science. But if I discover something new that proves incorrect, then this is bad news for me, but still good news for science. I would have no problem admitting that something I did was false. After all, we're all just humans, and science would not exist at all if we all knew all truths from the beginning.

However, before that happens you can expect me to defend my argument as long as it takes you to prove me wrong, and I expect you and your colleagues from the COMPASSS rejection letter to do the same. From this perspective, I hope we can continue to have a tough yet always fair scientific argument on this issue.

Best wishes,

Alrik

What you need to understand is that I have no intention to prove anything to you. Simply because your line of thinking seems to be caught in a feedback loop that distorts your vision in a single direction.

What I write is for the rest of the community, drawing the examples from the community (those examples are taken from Zhang, as you have certainly read in the paper).

I am wrong, Zhang is wrong, basically everyone else in the community is wrong, but yourself. And the real trouble is how you define correctness, in order to judge what is "incorrect". As previously mentioned, in your own private universe, you make the rules and then dream of electric sheep.

But in this universe, one cannot be both advocate and judge at the same time. Correctness is defined by common agreement, and the only possible way to redefine is to make it convincing, rather than forcing. Constant (and quite violent, I must say) negation of the work of others is far from convincing.

I am not the only one bothered by your relentless attempts to silence everyone through sheer volume of impertinent messages. I just don't shy away from responding in kind.

Dear Adrian,

Since you digress again into all-out personal attacks I take this reply as you having no counterargument to points 2 and 3 (point 1 probably needs no further discussion). Ok, then we leave it at that and see whether the meeting in Cologne will take us any further.

Best wishes,

Alrik

Of course I do have counter-arguments, but I believe it is pointless since you do not accept anything different from your own. Incidentally, what you write is against the very principle of Boolean minimization, specific to QMC (which you say you know).

In order to minimize A*B to B, one needs at least two cases that differ in A with respect to the same positive outcome:

A B C

1 1 1

0 1 1

Only (and only) in this situation could A*B be reduced to B. While the first row is observed, the second is a remainder. Even more, it is an impossible remainder since B can never take the value of 1 when A has the value of 0, because of the initial constraint that A B.

In order to reach a conclusion such as B -> C, you apply a very different logic than the Boolean minimization. Which means you have your own logic, in your own universe, completely unrelated to everything we (the rest of us, mortals) know from the past 60 years of Boolean logic.

So it all boils down to what is defined by "correct". And since you seem to be the only person in the world to determine what is "scientifically correct", you are also the only judge of everything, and this cannot be accepted.

Understandably, you believe this is an attack. On a closer inspection, you will realise it is a mirror of yourself: if you don't like it, then stop doing it! This approach of denying the work of others, against common sense logic and driven by ignorance, is unacceptable by any standards (think again about Espresso-exact, or that I don't understand my "incorrect" programming). I for one am fed up, reached a certain limit of tolerance and refuse to have my work trashed that way.

Despite being absolutely disgusted for having to do this, I am determined to maintain this treatment until you adopt a more respectful attitude. Certainly don't like it either, but someone has to serve you the same kind of poison you deliver to everyone else. Maybe, just maybe, you will understand exactly what you are doing.

http://www.alrik-thiem.net/blog/debate-over-qca-s-three-solution-types-intensifies-at-cologne-meeting/

There is a reason why I have not made my working paper public (yet), and that is precisely because it was still a draft version. Since the Cologne meeting, the paper got improved, having further clarified some aspects and inserted another section to respond to your "explanations". The title is now also changed.

In the meantime, as this is my own text and you have made it public without my express permission, I am now formally asking you to remove it from your website.

Please wait your turn until it gets published or until I make it public myself, and then you will have the opportunity to make comments.

As you chose to ignore my request and continue to make my paper public, I will now address a formal letter of complaint to your university.

You yourself have made your argument public because it is already used, with your confirmation, at methods schools, by other QCA scholars in their teaching and in reviews, and in fact by yourself here in this public forum. But I don't need to upload it directly on my website, of course. Everyone interested in your argument and my review of it at the Cologne meeting can send me a private e-mail. The main points are now summarized at http://www.alrik-thiem.net/blog/debate-over-qca-s-three-solution-types-intensifies-at-cologne-meeting/.

Maybe you should wait until your paper gets accepted somewhere before you bang the drum in public again as you did here.

Thank you for removing my working paper from your website. It will soon be made public of course (this was a work in progress), most likely in the form of a published paper. It has not been submitted yet on any journal, as I am still polishing the final arguments.

Soon enough, you will have your opportunity to comment the final version (rest assured it discusses and clarifies all of your arguments).

This sounds very good to me. I (honestly!) cross my fingers it will be accepted soon (perhaps at SMR?) as I think this debate is highly important for moving QCA forward.

I am glad we are on the same page here. As far as I can assess this debate, there is no reason whatsoever for any camp to claim the others are completely wrong (as per your conclusion in the SMR paper). It is just a different kind of logic, as it will become evident in the final paper.

Boolean minimization stands valid, just as before, and for the QCA community it is business as usual.

In fact fsQCA is not symmetric, so you have to do each outcome alone even for the same variable ( I mean the negation).

Please find attached my recent paper on Board characteristics and MENA banks' credit risk: A fuzzy-set analysis.

I need your feed back

thank you

Article Board characteristics and MENA banks' credit risk: A fuzzy-s...