We’ve seen papers retracted for lots of reasons, but this is a new one.
A researcher at the University of Ruhuna in Sri Lanka has been forced to retract a paper in the International Journal of Public Administration after evidently failing to properly install the computer software used to process the data.
Here’s the retraction notice for the 2010 article, by Chamil Rathnayake:
The Editors and Publisher, Taylor & Francis, are retracting the article “Romance of Leadership in the Public Sector Higher Education in Sri Lanka” from publication in the International Journal of Public Administration. Upon further investigation, the author discovered that there were installation errors with the software used to run the factor analysis in the paper. We greatly regret the inconvenience the publication of erroneous results might have caused other researchers.
We’ve contacted the author to find out more, and will update with anything we learn.
.Instalation error? I had read about statistical incongruences, but this is new for me. I thought that the program one time instaled will not appear problems into the code.
I suppose it is not impossible. If you are using custom software (i.e. you write your own programs) and you move from one system to another, it is possible that different compilers and libraries might interpret code slightly differently. You might call this an ‘installation error’. It might also be very hard to detect if it was causing erroneous results, but no system errors.
I agree with Dan Zabetakis. There may, however, be another explanation: software from another country did not work properly on a computer bought in another country. I have heard of this problem occurring on laptops bought in China and then brought to the US. If the software were pirated, that could cause problems as well.
I just cannot reconcile the title of this paper with the need to use factor analysis. If the retraction was needed, why not do it on February 14?
As a grad student, I provided statistical support at the university level. I helped a faculty member transition his programs to the new version of an application. He reran his old programs to check things and found that a set of results generated for a published paper were very different. I spent a great of time verifying this with him. We confirmed that his old code had a programming error and that the published results were wrong. His response to all of this was interesting. He decided not to do the followup paper and I suspect that he never retracted the first one.
In spite of the reason for retracting the paper, I give the authors credit for owning up to the mistake.
The lesson I took from working with the faculty member is to verify results. Journals should ask and maintain the programs used to generate empirical results.
With respected to pirated programs, the journal articles for some disciplines (e.g., Public Health) include the name of the statistical application used to generate the results. Making this a universal rule and submitting code to the editor should help program developers locate pirated programs and keep pirated programs out of academic work.
“If the software were pirated, that could cause problems as well.”
Possible, but I suspect that the most common case of pirating involves removing whatever protection the original software has (not a rewritten version of the software). It is possible that the process of removing protection may break some of the data crunching code, but the little I’ve seen involves disassembling to find password input/checking routines and bypassing them.
At least one attempted pirated version of Stata will randomly delete a subsection of rows of data each time a statistical test is done. There is no notification to the user that this is happening. I suspect googling for this will give you more details if you are interested. In any case, it would be nice if more of us could transition to using free and open source solutions like R. This would be good for easing replication and allowing people in less economically advantaged situations to more easily participate.