I am sure that you will find this rude and I apologize but a doctorally qualified Associate Professor should not be asking such a question. What does the literature say? What are your ideas? Find a useful question and plan how you can answer it. David Booth

I wasn't going to post in this thread because I'm not familiar with the literature about the use of QCA for panel data. But after David Eugene Booth's very unscientific reply, I felt obliged to express myself.

First of all, Yimin Mao was very clear about having surveyed the relevant literature about this subject — which, by the way, seems to be very incipient. QCA is a fairly new analytical tool, and there are many new and exiting advancements being made everyday. He's not asking about something that has been discussed over and over for centuries.

Second, he opened a discussion thread, with the clear intention of pushing this topic further for the sake of scientific advancement, so it's not like he's asking for an answer on a silver platter. And even if it was the case, there's no such thing as a dumb question. I'm sure there's a lot you don't know, Mr. David, and if every question could be answered simply by literature review, the very existence of universities and professorship would be futile.

Science is a social endeavor, being by the cumulative nature of the scientific work, or by the discussions it stimulates. A scientist must never act in a patronizing fashion, because science teaches us to humble ourselves in the face of an infinitely complex nature.

That said, hopefully this thread will incite some good discussions.

Best,

Breno

Breno A. H. Marisguia It was not meant to be scientific but rather vocational. I have never been associated with a university that would tenure/promote a doctorally qualified professor who could not advise a beginning PhD student on starting a research project. He should be aware of that fact. Presumably he has done his own dissertation. That exercise is intended to teach one to start, carry out and publish research. ALL of my doctoral graduates have done that. An Associate Professor that is unaware of that is endangered in higher education.

BTW all research is incipient or it would not be research. If you/he checks my recent postings I have been writing about starting research projects to beginning research students. BTW the reason nobody does research in Euclidean geometry today is because it has been discussed for centuries and centuries. Wishing you the best, David E. Booth, PhD, PSTAT, Emeritus Professor

Dear David Eugene Booth and Breno A. H. Marisguia , thank you for your discussion on this topic.

By opening this discussion, my primary goal is to attract the attention from those scholars who have the same interest in this topic. And then, we can have a more detailed discussion about the methodological issues, step by step. Therefore, the content I have posted in this discussion is simple and informal.

I do think this topic is worth further discussion. Most of QCA research relies on the cross-sectional data. Garcia-Castro and Ariño (2016) have done an attempt to extend QCA to the panel data context. Although only a few of studies (see below) have adopted their approach, Misangyi et al. (2017, p. 274) argued that it offer a promising approach to incorporating time into QCA to enable “longitudinal set-theoretic research.”

I have published some papers with the application of QCA method. I think that the advancement of panel QCA method needs more discussion. Now, I am doing the literature review. And after that, I will post more technical questions.

Beynon, Malcolm J., Paul Jones, and David Pickernell. 2019. ‘Country-Level Entrepreneurial Attitudes and Activity through the Years: A Panel Data Analysis Using FsQCA’. Journal of Business Research. Guedes, Maria João, Vítor da Conceição Gonçalves, Nuno Soares, and Marieta Valente. 2016. ‘UK Evidence for the Determinants of R&D Intensity from a Panel FsQCA’. Journal of Business Research 69(11):5431–36.

Yimin

Dear Yimin,

first of all, I think your question is a pertinent one, Mr. Booth's reply a very odd one that borders on personal insult, and Breno's intervention in your support very laudable.

Now to the topic: as you've rightly pointed out, panel data techniques for QCA are virtually non-existent. In fact, even techniques for simple time-series with QCA are still a fledgling field. Here're my thoughts on Garcia-Castro and Ariño (2016):

1. The paper stops short of addressing a pressing follow-up question to its procedural protocol: while claiming that it proposes "a general method for applying set-theoretic analysis to panel data" (p.63), it does not show how to perform the optimization of necessary and sufficient conditions once all consistency and coverage scores have been computed. If this paper claims that it extends QCA to panel data, it should demonstrate that, and if so how, the connection between BECONS, WICONS, and POCONS on the one hand, and Boolean optimization on the other can be established.

2. An important issue with panel data and STMs is not addressed: how should the effect of time be accounted for? Some particular year t may simply turn out inconsistent under WICONS because one data series shows an increasing trend. Would the calibration discount this trend? Or would within-consistency always be assessed relative to some baseline calibration? These question are important because high consistency may simply be the result of parallel trends in data series rather than some causal connection.

3. I am unsure what the distance measures are for. If it has to be known when time effects occur, then having an average distance measure is not of much use. For example, (0.1,0.5,0.9) yields the same distance as (0.9,0.5,0.1), but in order to detect the problem that affects the panel structure, the researcher would have to go back into the data to search for its occurence.

That said, you have to consider that QCA is barely 30 years old, and even it's mathematical and algorithmic basics are still contested these days (just witness the fierce debates about solution types, model ambiguities, etc.). Any advice on how to run panel analyses with QCA must necessarily be experimental at this stage. I wouldn't go as far as to say that anything goes at the moment, but I would certainly be extremely careful in setting up empirical QCA analyses with panel data.

Best wishes,

Alrik

Alrik Thiem I apologized in my first post and offered vocational advice in the second. My record is clear. Yimin Mao Now what about classical longitudinal models. Look at the AIDS treatment data, There is a ton of it. One of the leading researchers in this area developed a lot of this methodology and served as President of the American Statistical Association She has an additional book on Nonlinear Longitudinal Models. Here are the course notes for her doctoral class on the basic subject:

https://www4.stat.ncsu.edu/~davidian/st732/notes.html I would see what I could do there with QCA There are a lot of sick people that you might be able to assist.

Also

BTW, you might also find this link of use: https://www.google.com/search?rlz=1C1CHBF_enUS874US874&biw=1600&bih=827&ei=xfglXofAEs63tQbo1JHQCQ&q=statistical+measurement+of+effictiveness+of+Aids+treatment+over+time&oq=statistical+measurement+of+effictiveness+of+Aids+treatment+over+time&gs_l=psy-ab.3...4539.50625..51458...4.2..0.505.7370.22j19j2j3j3j1......0....1..gws-wiz.......0i71j0i67j0j0i22i30j33i22i29i30j33i160j33i299j33i22i10i29i30j33i10i299j33i10i160.A91cAY6lJ-s&ved=0ahUKEwjHtsK57pLnAhXOW80KHWhqBJoQ4dUDCAs&uact=5

and https://www.google.com/search?rlz=1C1CHBF_enUS874US874&q=statistical+measurement+of+effictiveness+of+Aids+treatment+over+time+Viral+load&nfpr=1&sa=X&ved=2ahUKEwivqZTH8JLnAhVGCc0KHTacDRYQvgUoAXoECAwQJw&biw=1600&bih=827

as well as:

https://www.google.com/search?rlz=1C1CHBF_enUS874US874&biw=1600&bih=827&ei=V_wlXsHxII3KtQaV3ZqYAQ&q=statistical+measurement+of+effectiveness+of+aids+treatment+over+time+measuring+viral+load&oq=statistical+measurement+of+effectiveness+of+Aids+treatment+over+time+measuringViral+load&gs_l=psy-ab.1.0.33i10.10386.14472..19257...0.2..0.309.1735.1j5j2j1......0....1..gws-wiz.......0i71.6zbUIN7Lmxg

Should you choose this topic you may find our latest attempt of some interest, see attached. Unfortunately this is not longitudinal data. One further reflection, survival studies might be considered as longitudinal with censoring. If you would care to look at our work in that area, it is attached as well. And perhaps one more. Sort of panel data. that data is available from the first author. She can be reached by contacting me.

Best wishes: David Booth

Dear Alrik Thiem , thank you for your nice reply. The panel QCA can be the long run goal. However, an intermediate goal is to check the robustness of QCA solutions over time.

Here, I offer a concrete example. For instance, I want to explain which combination of conditions can lead to D (outcome) at the country level. By reviewing the existing literature, we can find that there are three primary conditions, namely A, B, C. I can conduct three analysis with data at the year 1990, 2000, and 2010.

The results can be hypothetically as the follows:

for 1990 data A * B → D

for 2000 data A * ~B * C → D

for 2010 data A * ~B → D

It is reasonable that the configurational condition of D is changing over three decades. however, it also means that the QCA results with cross-sectional data may not be robust. I guess that scholars may encounter inconsistent solutions like this if they attempt to apply QCA into panel data. Indeed, it may cause problem for interpretation.

Yimin

I think the current trend for panel/longitudinal data is towards multivariate distribution models. Tractable models for such data may be Matrix variate distributions,. For example matrix-variate Bingham distribution has already been used for quantitative data.