Wishlist for Stata Journal

[River Huang]

Dear All, I wish that some Stata experts can offer a VERY USEFUL command to implement PSM+DID (propensity score matching with difference-in-differences) models for publication in Stata Journal. The method has been implemented in many finance/economics articles but with flaws. A possible (more complete) approach can be found here https://imai.fas.harvard.edu/research/tscs.html (Matching Methods for Causal Inference with Time-Series Cross-Sectional Data). An R code/package is available, but not Stata.

[Nick Cox]

As an Editor of the Stata Journal I can simply confirm that we would welcome such a submission. The Journal like most academic or research-related journals is mostly reactive, and downstream of what people submit, subject to what is accepted after review,. As a very broad comment I would summarize my experience as Editor in terms of ironic rules:

1. Researchers are cats, not dogs, and woefully independent. People write what they want, not what you want them to write.

2. If you really want something written, you may need to write it yourself. (A serious point, but familiar to every teacher, is that trying to explain something in simple terms is a very good way to understand it better.)

We do reach out to people sometimes and encourage a submission, say after a great talk at a Stata conference or a great posting here, and we have had some welcome successes there.

I've often felt that we should carry more papers on data management and on teaching (teaching Stata, and teaching with Stata). On data management, I have the same conversations with many people to the effect that (a) papers like that won't enhance my reputation with people back at my workplace; they want cutting-edge methodology to be written up (b) papers like that just underline how messy the data are in my field, which no one wants to hear. For overlapping reasons, good papers on teaching are hard to get.

[Joro Kolev]

I can add my bitter experience as a contributor to Stata Journal and to the American Statistician (Teachers corner), somewhat showing the other side of the coin of what Nick said above.

The beauty of Stata Journal is that it occupies a unique niche. The curse of Stata Journal from an author's perspective is that... it occupies a unique niche.

You write a paper for Stata Journal, you spend on it whatever months you spend, if the editors of Stata Journal reject the paper, your time goes down the drain. Nobody else in the whole wide world is interested in Stata related matters, because, well, Stata Journal occupies a unique niche. And you are left with a product that you put valuable time in, and you cannot "sell" to nobody because there is no market.

Same thing with the American Statistician (teachers' corner). You spend whatever months you spend to write a paper on teaching statistics, if the editors of the American Statistician reject it, you are done. You are left with a good on your hands that has no market.

Contrast this with the case where you write a standard paper in economics, statistics, finance, sociology, management, psychology, etc. There are thousands, of journals, if one journal rejects your paper, you just go neeeeext. After a couple of submissions you publish it... it is only a matter of how well ranked is the journal where you ultimately publish it.

So back to OPs request -- I am afraid it does not work like this. The only people that can put a couple of months of work to be considered by the editors of Stata Journal for publication, and in a very likely scenario to be left with unsellable good on their hands should the editors of Stata Journal decide to reject, are people with vested interests on the topic, or people who have the job almost done already. Typically these would be the authors of the method, with the vested interest their method to be popularised and used by more researchers.

[River Huang]

Dear @Nick Cox and @Joro Kolev, Thanks for the explanation. I see the points. I just wonder if it is a good idea for Editor of the Stata Journal (if you agree with the importance/popularity of the PDM+DID approach) to invite the original authors to submit a Stata command to the Stata Journal?

[Leonardo Guizzetti]

On the other hand, if it's very important for your work, you can spend the time to implement the model in Stata. You could reach out to the original authors to collaborate on writing an article for the Stata Journal, accepting that you will probably do most of the writing and handle the submission and review process. Or, don't write the code and release to the public but forgo a formal publication.

[Richard Williams]

On the positive side, if you do get into SJ, the rewards can be great. My two most-cited articles are in SJ. It can be especially good if you write a substantive piece for one outlet and an SJ piece that describes how to use the programs you wrote.

Whenever I review a piece where the author has come up with their own unique Stata code, I always encourage them to turn it into a program that can be downloaded from SSC. The easier you make it for people to use your method, the more likely people are to try it. And if they do, that can mean lots and lots of citations for you. Get an SJ article too, and that is even better,

In short, Stata programs -- especially when accompanied by SJ articles -- can be an excellent way to get your work known and to get others to adopt your methods. And, if you can make the SJ article be a companion piece for other articles you write, the SJ piece may not be that much extra work. Even if it isn't published, you can make it part of your program's documentation.

[River Huang]

Dear @Leonardo Guizzetti, and @Richard Williams, I agree with what you said. I have no experience in writing a Stata command. As such, the opportunity cost is especially large for me. However, I am really interested in writing a code, and I might begin to learn more in this respect. Thanks again for your suggestions.

[Felix Bittmann]

In general this is a very interesting topic and I would like to see an implementation of matching in longitudinal settings more often. I have used this in the past but with different data (person data with large N and small T, so it is not directly what the authors look at in the cited paper). I would like to add that I think it is not necessarily the case that each contribution to the SJ must be a complete package / ado. I would be happy with a "how to", that is, a general demonstration of a technique, broken down to the relevant steps. Writing and maintaining an ado is an intensive and time costly process, and I guess this is an important factor that prevents many people from writing or publishing ados. However, giving a transparent tutorial ("look what I did in my project and feel free to apply it to yours") is easier. Sure, more work is delegated to the reader who has to understand the code and adapt it. However, this also gives a greater flexibility that is often not possible with an ado that only serves a very narrow application. To sum up, I would like to see more "tutorials" with applied examples, which is also better for learning and understanding a technique.

[Maarten Buis]

The most important reasons for me to write a program is either

• because I need it (I have a project that requires a particular technique), or
• because I come across a technique that looks cool, and writing it up in a program is a (somewhat weird) way of learning more about it, or
• because the programming problem is interesting (some people solve crossword puzzles or sudokus, I write Stata programs...).

The impact factor does not play a big role. From talking to other "serial offenders" at places like Stata Conferences, I get the (incomplete and selective) impression that this is mostly true for them as well.

[Nick Cox]

I've identified myself as an Editor of the Stata Journal and made some informal comments. Beyond that, it's prudent not to make too many comments that might be misinterpreted as indicating policy or intentions, or quoted out of context. So, by all means, suggest to authors what they might do or (perhaps better) suggest collaborations. There will be a predictable diversity of responses.

Felix Bittmann made an excellent point. In my own work I've publicised various Stata commands I wrote but also in some cases rowed back from them whenever I see belatedly that there is a fairly simple way to do it that just involves a few basic steps, not a new command at all -- and learning the technique is a really good idea, because the trickery involved is useful for other problems. There is a phenomenon of YAUC -- look, yet another user-written command! -- which is double-edged. In Stata there must be a few thousand packages and some other software boasts many more than that, but this prolixity bites too. It's a jungle to traverse, for experienced users too.

Similarly in the Stata Journal we're very open to genuinely expository articles explaining step by step how to do something, so long as the task is generic.

[River Huang]

Dear Nick, Got it and thanks gain.

[Richard Williams]

In his articles Paul Allison often includes examples of Stata, SAS, and R code that implement the techniques he is showing. This is another excellent way to make your methods accessible to other people without writing a full-fledged ado file.

We did, however, team up to write xtdpdml, because hand-coding for the models it estimates is very difficult and error-prone. And, we got three articles out of this, one in SJ and two in other places. https://www.stata-journal.com/articl...article=st0523