Congratulations to Annals of Statistics
Here is a good news story about an academic journal that is prepared to set the record straight.
The October issue of Annals of Statistics had a paper by Weishen Wang about smallest confidence limits. The main Theorem 4 gives a formula for a largest possible lower limit for a scalar parameter in an arbitrary discrete distribution. Unfortunately, this construction was first given by Robert Buehler (JASA,1957). The proof of its optimality was given by Jobe and David (JASA,1992) under a non-trivial restriction. The proof was generalised to remove this restriction by Lloyd & Kabaila (ANZJS,2003).
The paper by Wang includes reference to Bol’shev (1965). That paper refers to Buehler (1957). Indeed it is mainly about Buehler (1957).
The paper by Wang also contained an additional Proposition 2 which, in non-technical terms, says that a more informative statistic can never generate a worse Buehler lower limit than a less-informative statistic. This result was proven in Kabaila & Lloyd (ANZJS,2004) where we also gave conditions under which the lower limit will be strictly better rather than only no worse.
I sent the details of this sorry saga to the editor of Annals, Peter Buhlmann, who responded within two days that that the record could be set straight with a two page paper authored by Paul Kabaila and myself. The publication process was hurried it should appear in the December issue. Well done Peter Buhlmann and Annals of Statistics! My faith is renewed. I am not sure all editors would be prepared to do this.
In fact, I know from personal experience that they don’t.
About 7 years ago Biometrika published a paper, Using logistic regression procedures for estimating receiver operating characteristic curves, by Jing Qin and Biao Zhang. The paper was virtually identical to a paper I had published a year earlier. However, I had submitted this same paper to Biometrika two years earlier and had it rejected. You join the dots yourself.
The editor of the time, Mike Titterington, told me that Biometrika do not publish letters. Instead, he got the authors themselves to publish an erratum. This was totally inadequate in my view and does not correct the record. If you don’t believe me, try Googling “Using logistic regression procedures for estimating receiver operating characteristic curves” and see what you get. You sure don’t get the erratum or anything with my name in it.
It is hard to know what to do when you have your work plagiarised. If the journal just hunkers down then what avenues are left to you? Do you get up at a conference and denounce the offender and look like a looney? Or do you just blog about it…..
You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.
December 9th, 2010 at 1:07 pm
This is one of the most heartening pieces of news I have heard in statistics for a long time. I think it is wonderful that The Annals has decided to allow factual corrections. Well done Peter Buhlmann!!
In my career I have seen a number of examples of published papers which rediscover results, or reinvent ideas that are clearly published previously. Sometimes these subsequent papers come to over-shadow the original, especially if the authors of the rediscovering manuscript are very well known or connected so that they accumulate more citations. This is a problem in statistics, but an even bigger one in econometrics.
Sadly, many econometricians who work on inferential methodology believe that it is legitimate to republish ideas from statistics under the guise of “introducing the idea to the field”. This is just a cheap scam. All decent econometricians would read the prominent statistical journals for relevant ideas and methods. In fact, I cannot think of any established econometrician who has not published at least one paper in a statistical journal anyway.
The pressure to publish seems to have forced people to forget their obligation to cite and acknowledge the insights of others whose work came before theirs. Sadly, this seems to have affected many senior researchers, who can sometimes identify too much with their research, rather than be a truly objective scholar doing what they do not for the glory, but for the interest.
Over-taxed volunteer referees and editors are going to make mistakes. So, what is the solution? Clearly in our reports to editors as referees, we should request the authors to more accurately cite previous work. But the move by the The Annals in publishing a 2 page letter of this type is a more effective approach. It creates an incentive to keep authors honest. If JASA, JRSS B & Biometrika adopted this policy, and we as authors write to the journals, you will find that accurate citation and acknowledgement would really take hold.
December 9th, 2010 at 1:28 pm
Chris,
The author indeed does not properly cite the relevant literature (i.e. Buehler, 1957; Jobe and David, 1992; Lloyd & Kabaila, 2003) when building the ground for introducing the method, but the method itself is NEW! So…., almost new since it was in the center of my thesis two years ago (though never published other than in a book). By the way, I properly acknowledged both authors who contributed to the method calling this new limits B-square, i.e. for *B*arnard and *B*uhler. Section 3 of Wang (2010) repeats my explanations almost word for word!
Should I claim plagiarism as well? I don’t think so - too busy with the new things and the procedure is so intuitive that it was likely to be rediscovered! Plus the method never appeared in the mainstream literature, which of course is in contrast to Buehler (1957), Jobe and David (1992) and Lloyd & Kabaila (2003).