Tuesday, September 26, 2006

Contributions of Co-Authors

Guest Post from Jérémy Barbay.

In multi-authors publications from other fields (biology, physics), the order of the author's names roughly indicates the importance of the contribution from each author, while this is not the case in CS. I was told that it was to avoid meaningless fights about the order, but in my personal experience I found out that it created other conflicts, mainly the frustration of the main author toward co-author who did not "contribute enough."

The fact is, not recognizing the difference of importance in relative contributions pushes to two quite negative behaviors:

  1. Some authors will expect all co-authors to contribute an equal part to the paper, and be disappointed and frustrated when it is not the case. This is bad either way: having a frustrated co-author is not a nice experience, but on the other hand trying to balance the work so that each co-author contributes the same amount does not necessarily corresponds to the balance of skills for the particular publication, and may not be the most efficient way to work on a paper.
  2. Authors have an incentive to play the game of the "minimum contribution deserving authorship", eventually (for the highest in the academic hierarchy) pushing the other authors to contribute more in exchange of immaterial promises (such as future recommendations, or the "teaching" by experience). I was explained by a senior faculty that "students have to accept to eat a lot of shit (meaning, do a lot of stupid work), and young faculty to eat some shit but to pass down to their student most of it".
Both those behaviors are bad for the community in general. I saw it literally destroying a brilliant Ph.D. student, and I see its perverse effects on some promising younger students. Ranking the names of the authors is but a poor palliative: they are many measures of contributions on which to argue and I still see frustration among my fellow biologists and physicists.

One obvious solution would be to add a paragraph at the end of each paper, in a similar way to the "acknowledgement" paragraph commonly added, describing the contribution of each author. It would be a natural way to create an incentive for each author to contribute as much as this paper is worth to him, inversing the "minimum contribution deserving authorship" incentive, and remove most of the frustration. It would remove the ambiguity on "who did what", that we have anyway to explicitly remove when writing recommendation letters or applying for "best student paper". From a game theory point of view (for the little of game theory that I know), it would make the publication mechanism "truthful", and anybody opposing this mechanism would risk to look bad.

Now, I assume that this solution has some hidden drawback(s) (other than taking an additional five lines in each publication), as otherwise some field or other would already have adopted it. Or is it just that senior faculty members would not support such a measure?

37 comments:

  1. The obvious negative of having a paragraph describing individual contributions is that it takes all the negatives of name ordering and exacerbates them tenfold. If people have trouble arguing what order they should go in, what makes you think they'll be able to agree on a full paragraph describing their contributions?

    ReplyDelete
  2. I think the main drawback is that it may encourage fights among strong-willed people. Right now all they have to fight about is whether someone contributed enough to be an author, which is often clear-cut. But it's not always clear-cut who was responsible for a theorem; maybe someone had the main idea, but another figured out all the details, and there are disagreements about which part was harder or more central to the paper. If it started to be the case that salaries and tenures were decided based on stuff like this, I could see friends unwillingly turning into enemies very quickly.

    ReplyDelete
  3. If you think one of your coauthors is not contributing enough, just don't work with him/her. This is how the community weeds out people who do not contribute their fair share.

    If the "main" co-author complains about undue co-authors yet keeps coming back for further collaboration, then perhaps he really isn't getting as raw deal as he likes to claim in public.

    To complicate things further it has been shown that if you ask each author to estimate as a percentage their contribution to the paper it adds to well over 125%. So clearly one is not the best judge of one's own contribution to a paper.


    Authors have an incentive to play the game of the "minimum contribution deserving authorship".

    Not really. As I said above, if someone gets to be known for not pulling their weight then most people will avoid that person as a collaborator.

    I was explained by a senior faculty that "students have to accept to eat a lot of shit (meaning, do a lot of stupid work), and young faculty to eat some shit but to pass down to their student most of it".

    Sure, "eat s**t" but on in exchange of immaterial promises! They do menial work in the same way as apprentices of a trade do minor jobs while they watch and learn how to do the hard ones. A good student will spend very little time in that stage before being promoted to senior tradesman, while a junior faculty might spend a long time there if s/he doesn't learn quickly enough.

    ReplyDelete
  4. As well as the comments above, particularly in theoretical areas it is very hard to quantify or describe in any sensible way what the individual contributions are.

    For instance, I tend to work by generating a lot of wrong ideas/arguments leading eventually (one hopes!) to correct ones. A "shotgun" approach if you will. In some instances, a co-authors main contribution has been to filter the rubbish that I provide. Personally, I think this is hugely important, but I know from personal experience that others have very different feeling about this.

    ReplyDelete
  5. but on in exchange of immaterial promises!

    There's a nasty typo there. Corrected version:

    Sure, "eat s**t" but not in exchange of immaterial promises! They do menial work in the same way as apprentices of a trade do minor jobs while they watch and learn how to do the hard ones.

    ReplyDelete
  6. I am very much for the standard TCS convention of alphabetizing (or other agreed-upon schemes such as listing the student-authors first, as in many networking papers). To me, joint work is about the joy and learning that comes from collaboration, which would be harmed by distracting notions of who comes where in the author-list. Also, if a person tries to game the system by constantly making minimal contributions as suggested in the post, they will soon find themselves without collaborators.

    Aravind Srinivasan

    ReplyDelete
  7. I can't wait to read this paragraph for the Cao-Hamilton-Perelman-Yau-Zhu article on some recent results in low dimensional topology.

    ReplyDelete
  8. There are authors who mostly don't have any such problems with their (nonexistent) co-authors. I wonder if they have problems of their own.

    ReplyDelete
  9. I don't buy the drawbacks in this post. As others have pointed out, these issues tend to balance out in the majority of cases since the community is, well, a community. How you treat people matters.

    I'm not claiming that there are no difficulties, but on the whole I prefer the CS system to the alternative -- I think it tends to encourage much more collaboration rather than competition, and keeps people from worrying about whether they'll get "enough credit" rather than focusing on the paper itself.

    I can't see the "contribution section" leading to anything but miscommunication and competition. For one thing, sometimes it genuinely is ambiguous who "proved" a particular theorem, or even individual parts of it. For another, it would be far too easy for people working on the same project at the same time to both feel they had the essential insight that solved a particular difficulty. Right now, this doesn't matter -- both authors will get credit for important contributions. But under such a system, this could make a lot of new enemies.

    The primary situation where I can see the CS system becoming a legitimate problem is when a major result is proved, but one "coauthor" has contributed enough to be credited on the paper, but very little to the major result. This could also lead to precedence arguments, but is a drastically rarer situation than potential fights about authorship order or the "contributions section."

    ReplyDelete
  10. Anonymous 3,5: I am talking about ambitious students, who consider themselves at least as good as some faculty members. For them the promise to learn through doing menial tasks seems immaterial, as well as the promise of a "good recommandation", which they think they deserve anyway.

    Anonymous 6: I actually like the convention to put the students first: they are the ones who need their ego boosted the most, when more senior should know better.

    Anonymous 9: I don't buy the ambiguity argument. A paper where each author contributed the same to each part can state so. That's not stating anything which creates the ambiguity.

    Maybe I should make my point more clear. Because of the lack of a way to express the level of participation in a publication, some students assume that this level should be the same for everybody. Those students are wrong, but allowing them to claim that they did this and that would make them feel better. And maybe help them to realize how much they did not do, or what they did not know how to do.

    I don't truly believe that we will start to do that soon (too much of a cultural change), but I am curious to hear the drawbacks of it.

    ReplyDelete
  11. For instance, I tend to work by generating a lot of wrong ideas/arguments leading eventually (one hopes!) to correct ones.

    Shimura of Taniyama-Shimura said that Taniyama made a lot of "good mistakes". He seemed to revere Taniyama's ability to search the problem space.

    ReplyDelete
  12. A lot of times it is worth having coauthors just so they can appreciate all the cool stuff you did. They are probably the only ones who really get how difficult an obstacle was or how ingenious an idea.

    The main problem with liberal co-authorship barriers is that people become afraid to talk about their ideas, lest others make a menial contribution and receive equal credit.

    Finally, I'm sure that many of us have felt disheartened when we send our ideas day after day to a "coauthor" who is too distracted by some other project even to give feedback. But if you find the right people (who really care about the problems you're working on), they surprise you enough to make the collaboration worthwhile. And in the worst case, you can try to guilt them into writing the journal version (re: previous post)!

    ReplyDelete
  13. I am talking about ambitious students, who consider themselves at least as good as some faculty members. For them the promise to learn through doing menial tasks seems immaterial, as well as the promise of a "good recommendation", which they think they deserve anyway.

    I do not think there is ever a promise (implicit or explicit) of a good recommendation letter in exchange for menial labour, as a quid-pro-quo. You seem to suggest a system of X hours in exchange for one recommenedation letter. It is not about recommendation letters, it's about learning. They do it until they learn. Students are encouraged to give talks, for example, not with the recommendation letters in mind, but with the intent that they get practice on this. Some do so form the outset while others falter and sometimes even resist advice, having to come back and redo things over and over until they are ready for the next challenge.

    ReplyDelete
  14. The problem seems to be that the order of author names gives us Information.
    So why not try to reduce or eliminate this information.
    Here's one way to do it:
    Use some kind of meta-tags in the documents such that every online view of the paper will create a randomly ordered list of author names.
    We could even extend it to print versions too: Just print different copies of papers with different ordering of author names.

    ReplyDelete
  15. Maybe I should make my point more clear. Because of the lack of a way to express the level of participation in a publication, some students assume that this level should be the same for everybody. Those students are wrong, but allowing them to claim that they did this and that would make them feel better. And maybe help them to realize how much they did not do, or what they did not know how to do.

    This seems somewhat different from what you wrote first, thanks for the clarification. Still it wouldn't work. If a student has an overinflated perception of his/her contributions, trying to apportion credits will only exacerbate the problems you mention. Such a student will still believe that his contribution is dominant, and will even misremember who proposed what and where.

    ReplyDelete
  16. I can't wait to read this paragraph for the Cao-Hamilton-Perelman-Yau-Zhu article on some recent results in low dimensional topology.

    Well, what will happen here is quite clear. S***ting Yau will claim that he made 50% of the contribution, as he has done in the past.

    ReplyDelete
  17. The problem seems to be that the order of author names gives us Information.
    So why not try to reduce or eliminate this information.
    Here's one way to do it:
    Use some kind of meta-tags in the documents such that every online view of the paper will create a randomly ordered list of author names.
    We could even extend it to print versions too: Just print different copies of papers with different ordering of author names.


    I think the complaint is the current system doesn't offer enough information. A simple alphabetical ordering of authors' names already serves to give no information about contribution.

    ReplyDelete
  18. Which coauthor did more work on this recent ECCC submission?

    TR06-123
    Venkatesan Guruswami, Venkatesan Guruswami:
    Iterative Decoding of Low-Density Parity Check Codes (A Survey)

    I've heard that Venkat is a slacker, so Venkat ended up doing almost all of the writing by himself.

    ReplyDelete
  19. Perhaps the problem is not that big in theoretical cs, but in many empirical or "laboratory" sciences, research groups are usually quite large. In those groups, it is often the case that the work of every group member is an integral part of the results in every paper!

    Especially nowadays, when multiple-paper dissertations (vs. monographs) are common, putting the main author's name first is a way to indicate that "the results described in this paper come predominantly from my niche in this research". In your dissertation you can use only those papers in which your name appears first (they're about your niche in that project).

    However, the tendency to put every group member's name on every paper easily leads to "gift", "ghost", "guest", or "honorary authorships", which are against research ethics. There again, naming always one person as the main author places the responsibility for following research ethics to the main author!

    ReplyDelete
  20. " ... Some do so form the outset while others falter and sometimes even resist advice, having to come back and redo things over and over until they are ready for the next challenge."

    Advice is advice--it is not an order to do something. It should be used by the student to the extent that the student feels comfortable.

    ReplyDelete
  21. I always treated any information on how joint research is performed as private.

    To say truth about it, you frequently need a separate paper rather than just one paragraph.

    ReplyDelete
  22. In the sciences, it is not uncommon for multiple authors to share "first authorship". THat is, authors may be listed in any order, but a footnote indicates that two or more authors are joint "first authors" because they contributed equally to the crux of the paper. (Other people, like advisers, are simply there not because they did any work, but because the work was funded by their grants).

    ReplyDelete
  23. Our co-authorship conventions were addressed in Martin Tompa's 1989 SIGACT News piece Figures of Merit based on an earlier FOCS Follies presentation. In this he introduced the important "splotlight factor" and "coefficient of obliviousness". With the passage of time, these statistics probably need to be updated.

    ReplyDelete
  24. One reason for students to put their advisor's name on their papers is that it increases the chances of the paper being accepted at a conference. (At least if the advisor is famous enough).

    ReplyDelete
  25. It is difficult for a stundent to refuse to put the advisor's name on their papers anyway. And guess who would have the last word on the contribution paragraph if the advisor is not playing fair...

    ReplyDelete
  26. Anonymous 25: You can't really do anything against "unfair" advisors, and they are only a few anyway.
    Anonymous 15: Same argument, you can't do much for students with overinflated egos, and once again they are only a few anyway.
    Let's restrict the discussion to the general case: a fair advisor and a bunch of honest students.

    Consider the (imaginary) contexts where one advisor works on paper A with a student who get a good part of the results by themselves, and on a paper B with another student who need constant supervision and help; or where one advisor is working with a group of students of unequal strenghts on one common paper.
    Diferent levels of contribution are expected and normal, but it can (and does) bring resentment among students.
    We end up describing their participation in recomandation letters anyway, but this is an "immaterial promise" and more transparency would reduce some of the frustration that I witnessed. It is also more fair: a student did remind me of one contribution that I could have forgotten otherwise when writing his recomandation letter.

    The problem should not be discarded because "students have to learn and accept the system": being a scientist is more about improving the world than accepting it as it is: otherwise we would still be hanging on trees and showing our asses to each other when upset... ;)

    ReplyDelete
  27. Diferent levels of contribution are expected and normal, but it can (and does) bring resentment among students.

    What you are saying is that we should handle the students who tend to be resentful by asking them to participate in a difficult and delicate discussion of who did what.

    Can you see a problem in there?

    ReplyDelete
  28. I think people can categorize their papers into "selected publications" and "other publications" (or sth like that) in their papers list in the CV or on the web. "Selected publications" means that they think they made major contributions to these papers, and "other publications" means that they think other co-authors made major contributions to the papers. In this way, co-authors don't have to fight each other in order to decide who's done most of the work; rather, each co-author evaluates their contributions individually: if they themselves feel that they made enough contribution in the paper, they can put the paper in their "selected publications" category.

    Although I would expect most senior researchers won't care too much about categorizing their publications this way, this does help the students to claim their rightful credit.

    ReplyDelete
  29. My recent forays into computational biology have led me in contact with the world of the biologists (who order their papers in ways to indicate level of contribution and seniority) and let me tell you, any problems and resentments over co-authorship in this community are a friendly walk in the park in comparison!!! I don't know if it's our author ordering conventions or something about the type of science, but the theory community is SOOO friendly, honorable, cooperative and nice to each other in comparison to the stories I hear from biology postdocs, graduate students, and even junior faculty! Don't wish us in their world-- please!!!

    Cheers, Lenore Cowen

    ReplyDelete
  30. I personally know, even senior persons, who have misused the alphabetical ordering. The minimal contribution argument does not hold to such people because these people write their most papers with students, who do not even have a choice. Their contribution is often zero to negative. Negative because they do not even filter out the errors but anyway consume student times. Students do this possibly in exchange of transfering NSF funding, which they happen to get because of a virtuous cycle they happen to create for themselves. Their successful students are visible. But those who fail are not noticed by anybody. Such students suffer. Otherwise qualified students are in the situation of no help. I personally know one such student, whose research career is pretty much killed.

    Like we do not put any price on one's life, we should not put any price on one's career. I agree this is a small set of people, but the society must be open to weed them out. Because the cost otherwise a set of unknown but demoralized students. A bigger cost is perpetual unjustice such a person cause. This injustice is compounded when such a person has last name starting with a letter in the end of the alphabet. Because even if such person had last name starting with "A", his/her name would come last in the authorship list.

    Note that there have been award winning papers in Computer Science which do not have alphabetical listings. In such cases the person in the last does not get equal credit. But if such a person had the last name starting with "Z", then he/she does not lose even this credit!!!

    What's the solution?

    I see one solution to this. And the solution is to not let these set of people grow. But if they grow but their extra-technical skills, then filter out this small set of people. But how? Two words.

    Modern Technology!

    ReplyDelete
  31. A bad advisor certainly can hurt a student's research career, but won't kill it if the student is persistent enough. I am still recovering, careerwise, from a bad choice of phd-school. It takes time, patience and energy, but it can be done. You get a post-doc at a mediocre place, work hard, move to a post-doc in a better place, work hard some more, and so on.

    ReplyDelete
  32. @31.

    You are right, if a student is good and independent then a bad advisor won't kill the student career. But you should realize that not all students are like this to begin with.

    Many students who have intention to complete PhD leave after earning Master's. Yes, they realize that PhD is a tough nut to crack.

    Note that like in any large collection, the quality of students (whatever crude measure we take) could also be a bell-shaped curve. So a lot of students are just average. These majority of students need advisor support more than the name of their advisor on their paper.

    When I read these advisor's recommendation letters, they even fail in those career building letters to mention that it was student who did it.

    Also do not forget the loss of credit to student when the advisor just have a policy to be on the students paper (because of routing NSF funding). In CS we have several recognition awards for students, which gets ruined when an advisor push himself/herself on the student paper.

    I always support that the papers who have only advisor as a non-student author be eligible for these rewards. Because many professors have high ethical standard who do not share student credit even if they helped a lot but there are some advisor who won't hesitate pushing their name on students paper -- because otherwise they won't have many papers left.

    I repeat, the answer is still two words -- "modern technology".

    ReplyDelete
  33. Also do not forget the loss of credit to student when the advisor just have a policy to be on the students paper (because of routing NSF funding). In CS we have several recognition awards for students, which gets ruined when an advisor push himself/herself on the student paper.

    In other areas of CS the same student awards are available to papers with the supervisor as coauthor provided that (i) supervisors name appears at the end and (ii) supervisor agrees that the majority of the work was done by student. People in TCS have been talking about moving to this system for sometime, particularly since nowadays most conferences have one, sometimes two, rarely three papers solely by students. This means that the best student paper award often (but not always) ends up being "the only student paper" award.

    ReplyDelete
  34. @33 wrote, "i) supervisors name appears at the end and (ii) supervisor agrees that the majority of the work was done by student."

    I do not agree with these two criteria. Obtaining any agreement from advisor works with good advisors, who anyway do not put their name on students paper if the majority of the work is actually done by the student.

    Also in some cases advisor name appears in the last because the advisor name starts with Z. So these advisors have easier time putting an agreement in the note section of paper submission email. On the other hand if the advisor name starts with A, then such an agreement appears in public. So such an advisor is saying to the public that his/her student did most of the work but still he/she chooses to share the authorship. Not a dilemma for Z initialed advisors.

    Plain and simple, if the non-student author(s) is(are) only the student's advisor then the paper must be considered for best student paper awards.

    BTW, the last name for most people is inherited. My last name is "Jain" because my parents had the same last name. So ordering authorship based on last name is a form of discrimination.

    One could have used this arbitrary ordering if it had no implications (such as names in a list of class rolls). But for the authorship on a paper it has implications. If an "A" initial person is the last author in a multi-authored paper then the person does not get as much credit (suppose the paper wins a big award!). On the other hand if a "Z" initial person is the last author in a multi-author paper because he/she was given a courtesy authorship, the the person keeps equal credit.

    I think this implied "discrimination" even happens in 4% of the cases is a big enough reason to not order the authors by last name. If one just needs a default no-dispute ordering then why not choose a random ordering? Just take out a coin from the pocket when the concerned authors are there. Or simply write a cryptographic program which takes the title of the paper as the crypto key and output a random ordering on the authors. This way not even all the concerned authors have to be there in the same room.

    This is non-discriminatory. Theory community is one of the communities who could take randomness in a good spirit.

    Ordering by seniority is acceptable, but often two people could have the same ordering. Ordering by age (D.O.B) is acceptable too but leaks a bit of personal information. Writing authors name in a circular order is good too but causes problems in citation.

    A reproducible secure pseudo random number generator seems the best choise in my mind.

    I have no idea of crypto-programming but I am very sure somebody could write a short program and distribute it widely. I will use it!

    ReplyDelete
  35. Anonymous 28: I did not realize that indeed the "selected publications" was allowing the students to emphasize their participation to some papers. It is not perfect but it does the job well enough. Shame on me: I knew about those sections, had guessed what they meant, but did not link it to the problem of authorship.

    Kamal: The alphabetical order does a good job at hiding the contributions, there is no need to perform a random permutation. My question was about the need to hide the respective contributions, as opposed to transparency and clear statements, and I did not get your positions on this point?

    ReplyDelete
  36. Jeremy, it is human nature that whistle blowers are sometimes taken as problem creators. So, I did not use the explicit examples.

    I thought since this is a TCS blog, hypothetical counter-example could negate a proposition anyway. I guess it did not.

    So here are some explicit examples.

    In my upcoming paper in SODA, I have three co-authors whose last initials are Z (student), H (just graduated), J(myself), W(senior professor). The paper will list the authors in this order ZHJW. You may think this is seniority order or you may think we wanted to give more credit to Z. But if Z's initial were instead A, then you would definitely give equal credit to all 4. So having last initial A or Z have different implications, even though the persons involved and their work had remained the same. This is precisely the definition of discrimination. Agreed, it is not discrimination based on one's ethinic origin (i.e., parental attribute) but is a discrimination based on one's last name (again parental attribute, though my last name also represents my ethinic origin).

    You would find even more serious examples, where a paper won a big award but the credit sharing depends upon the last initials.

    Sometimes even A as a last initial is a boon too. If it is a 10 authors paper, in various talks it will be cited as Aaa et. al.

    I do not know why it is hard for anybody to see this discrimination. By definition, discrimination is an implication decided by an external factor.

    Most universities have a set date of graduation. Why not order the authors by the graduation date of their last degree, in the order of seniority (PhD is more senior than Master's). Juniors/students come first because they need more exposure. Also, date of graduation is not really an external factor either. This also goes well with various cultures, where seniors are more considerate for juniors. Also in your early career you are the receiving end and at some point in your career you are at the paying end. So overall things balances out.

    Now, your question, whether I like the credit been reflected on the authorship ordering. Well, I try to avoid subjective criteria. How do you say which is a higher credit? I proved theorem A and you proved theorem B? May be A is more important but B was much harder to prove? How do you compare apples to oranges? In such cases I prefer to choose conventional ordering (and as I said, I do not like aphabetical ordering. Choosing a more appropriate conventional ordering requires the change of culture, which could be initiated by more senior members of TCS.)

    On the other hand, you could compare a basket with full of apples with an empty or nearly empty basket. In such cases, ideally, nearly empty basket holders must be mentioned in the acknowledgement section only. In case of doubt, the decision whether the basket is empty lies with the basket holder only.

    Often I have generous co-authors. I asked them to be co-authors but they decline to share the authorship. I try to convince them to share the authorship, occasionally, I am able to convince them to share the authorship but often they insist to be the last authors.

    I myself have appeared in non-aphabetical ordering towards the end. Occasionally there are examples of such things, even by more junior persons.

    So my opinion in a brief: Yes the contribution must reflect in the author's ordering, as long as, it is objectively clear. In case it is not objectively clear then there must be a conventional ordering. There is no point creating disputes or bad feelings among the members of society. The loss is bigger than the gain. About the conventional ordering, I like to change it to ordering by seniority (total order by the highest degree and the date of the degree).

    ReplyDelete
  37. Many coauthors insist alphabetic ordering becuase it is often to their advantage (i.e., their names come first). So establishing a credit system (such as declaring rough percent cotributions) seems a great idea. This will also discourage authors adding too many coauthors whose contributions are not substantial.

    ReplyDelete