Friday, August 26, 2016

Certified Random: How To Co-Author If You Must

by Debraj Ray ® Arthur Robson

(For the full Monty, click here)

Many years ago, when Debraj worked at Boston University and his good friend Arthur visited there, we spent one of our many enjoyable lunches together railing against the indignities of alphabetical order, which is the dominant name-ordering convention for publications in economics. A quick perusal of our last names will explain why we railed. To add insult to injury, Debraj had just been enthusiastically recommended a “wonderful paper” by Banerjee et al, on which he was a co-author.

Alphabetical order is, in many ways, a good arrangement. Our colleagues from other disciplines express wonderment that such a self-centered subspecies — the academic economist — actually uses this civilized convention. Around 85% of two-author papers are written in alphabetical order. Compare this to the cutthroat nature of much of the sciences, in which there is often a tussle for first authorship, while other not-so-subtle signals such as lab leadership are sent through ancillary ordering conventions. It can be argued that the civility of alphabetical order lends itself to more joint work, as the possible rancor in settling on a name order at publication time is thereby avoided.

And yet, there are features of alphabetical order that create significant and unwarranted advantages for names earlier that appear earlier in the alphabet:

1. Psychologically, names that appear first are more likely to be given extra credit given that society as a whole appears to be attuned to merit-based rankings. This is certainly in line with research on marketing: products presented earlier exhibit higher probabilities of selection, as this aptly ordered article by Carney and Banaji (2012) observes. Even stocks with earlier names in the alphabet are more likely to be traded; see another aptly ordered paper by Jacobs and Hillert (2016), or the more staidly ordered contribution by Itzkowitz, Itzkowitz and Rothbort (2016).

2. Earlier names appear bunched together on a bibliographical or reference list, lending additional perceptual weight to how often they are cited. They also appear earlier on the reference list. Haque and Ginsparg (2009) --- yet another aptly ordered paper! --- note that article positioning in the ArXiv repository is correlated with citations of that article. Feenberg, Ganguli, Gaule, and Gruber (2015) demonstrate that the same bias exists in the downloading and citation of NBER “New This Week” Working Papers, which led to a change in NBER Policy.

3. There is at least one major journal in economics (the Review of Economic Studies) which publishes articles in alphabetical order (using the last name of the first author). Because many other journals use the convention that the lead article is to be regarded as special, and because many do not know that the Review of Economic Studies follows this policy — did you? — this confers an advantage on earlier names.

4. There is, of course, the et al convention, which, while strictly speaking is not a corollary of alphabetical order, is widely used in citations and especially on slides in seminars, completely swamping the identity of later authors. Even if et al were to be banned in journal publications (which it currently is not), it cannot be banned from slides. In addition, it is widespread practice in verbal presentations to mention the name of the first author and then add “and coauthors”: an understandable but nevertheless damaging shortcut.

Is there any evidence that these considerations matter at all, or is this the resentful fantasy of two disgruntled economists with surnames far down the alphabet? There is, actually, quite a bit of evidence. In a paper published in the Journal of Economic Perspectives, Einav and Yariv (2006) write:

“We present evidence that a variety of proxies for success in the U.S. economics labor market (tenure at highly ranked schools, fellowship in the Econometric Society, and to a lesser extent, Nobel Prize and Clark Medal winnings) are correlated with surname initials, favoring economists with surname initials earlier in the alphabet. These patterns persist even when controlling for country of origin, ethnicity, and religion.”

There are several other papers that are in line with the Einav-Yariv empirical findings; see, for instance, the appropriately hedged Chambers, Boath and Chambers (2001), or the impeccably ordered van Praag and van Praag (2008). This last article finds “significant effects of the alphabetic rank of an economist’s last name on scientific production, given that an author has already a certain visibility in academia ...Being an A author and thereby often the first author is beneficial for someone’s reputation and academic performance.” A recent survey by Weber (2016) summarizes the literature thus: “there is convincing evidence that alphabetical discrimination exists.”

It is possible to argue that alphabetical order is an efficient arrangement: earlier names are better off. So why care? Economists who eschew interpersonal comparisons of utility (and especially those with earlier surnames), are likely to oppose any change in alphabetical order, arguing that existing alternatives would lead to nasty fights, grumpy co-authors, and ultimately a disintegration of the happy system of joint research that we all admire.

There are at least four responses to this point of view. First, one might argue that the game in question isn’t zero-sum. After all, people put in effort into doing research. Equal division of the credits from that research might be better for each author than unequal division, as efforts adjust to the more equitable distribution of credits. 

Second, the fear of being relegated to second or third author, or even to the dreaded et al dungeon, might discourage authors from writing papers with those more fortunately placed in the alphabet.  Again, this is an efficiency loss.

Third, when ordering is alphabetical but relative contributions are not consistent with that ordering, there will be feelings of unfairness, guilt, disappointment, or outrage. Indeed, the fact that alphabetical order is occasionally reversed is circumstantial evidence that such feelings do exist. 

Finally, the recognition accorded to earlier authors appears to cumulate over time, which of course generates persistent inefficiencies, not to mention snowballing unfairness. Van Praag and Van Praag (2008) focus on this aspect of mistreatment, and conclude that “Professor A, who has been a first author more often than Professor Z, will have published more articles and experienced a faster productivity rate over the course of her career as a result of reputation and visibility.” Quite frankly, and apart from any efficiency consideration, we see no reason to continue to support such a system. 

Yet, as Vladimir Ilyich so sagaciously observed, what is to be done?

Of course, the randomization of name order comes to mind. But it is important to note that --- with the exception of a few fair-minded souls who have used it --- private randomization will not break the grip of alphabetical order.

Suppose, for instance, that Archimedes and Boethius, working together, (never mind that they were born some 750 years apart) gallantly agree to break the convention by randomizing their joint authorship. Will they agree to the randomization? There are clear difficulties. Given an “alphabetical society,” a change in name order is a clear signal that the newly christened first author has contributed the bulk of the work. Thus, for instance, “Boethius and Archimedes” would be a statement that Boethius has done most of the research for that paper, whereas “Archimedes and Boethius” would indicate very little, any such signal being swallowed in most part by the naming convention. Therefore Archimedes gains nothing over alphabetical order when his name comes first, while Boethius gains a lot when his does. Boethius will agree to the ex-ante randomization, but Archimedes will not. That is one reason it is hard to “invade” an alphabetical society with a mutant scheme.

In a new working paper, we address this question. In brief: institutions can help. Here is a simple variant of the randomization scheme — a mutant — which will set it apart from pure randomization. Our new scheme involves flipping a coin to determine first authorship. Subsequently any such randomized name order be presented with the symbol ® between the randomized names. We ask, moreover, that any citation of such a paper respect the use of this symbol; e.g., Ray ® Robson (2016) is the appropriate reference for our paper. 

We propose, in addition, that an august body such as the American Economic Association make an announcement to this effect. We only ask that they acknowledge that this alternative is available. There is, of course, no question of imposing it. 

In fact, our paper provides a very good reason why there is no question of imposing it. Briefly, we show that random order (with the protection of ®) will successfully "invade" alphabetical order. If this option is available, we show that both authors will have an incentive to use it --- initially in a relatively small but subsequently ever-widening set of circumstances. Not only does random order (protected by ®)) invade alphabetical order,  there is no such possibility of reverting to the old status quo once in the new equilibrium. Random order maintains both the civility of alphabetical order and offers equal treatment to all concerned. 

The beauty of the mechanism ® is that it does not demand any more of the agents than does the present economics convention, despite being fairer and more efficient. In summary, random order maintains all the ethical niceties of alphabetical order, but in addition: (a) it distributes the psychological and perceptual weight given to first authorship evenly over the alphabet, (b) it allows either author to signal credit when contributions are extremely unequal, (c) it will be willingly adopted even in an environment where alphabetical order is dominant, (d) it is robust to deviations, (e) it dominates alphabetical order on the grounds of ex-ante efficiency, and (f) barring the addition of a simple symbol, it is no more complex than the old system, and brings perfect symmetry to joint authorship. 

Endnotes, and a Request

Anecdotes to advance our cause were provided by Noah Williams and Stanley Zin; unsurprisingly, perhaps, given the ill-fated location of their surnames. Zin describes his unalloyed delight at the prospect of breaking the curse of the perennial last-listed author by writing with Irina Zviadadze of Stockholm. Zviadadze is the fourth to last listed Member of the Econometric Society, though our commiserations must go out to Yanos Zylberberg of Bristol who is listed dead last. It is no doubt a coincidence that he is the sole author on 3 of the 4 working papers listed for him on RePEc. Zin also relates how Noah Williams recently lost a first-name tie-breaker to remain in his customary last position. While on the subject of ties, we salute the contribution of Goodman, Goodman, Goodman, and Goodman (2014), a listing that is distinctly robust to all naming conventions. They note, for example, that a reference to Goodman et al (2014) disparages the contribution of no author. Noah Williams, in turn, refers us to the lexicographically blessed Georges Aad, of Marseille, who has appeared as the lead author on 458 scientific papers, on which alphabetic order was adhered to, including one with 5,153 co-authors. “Basically, this guy has won the academic lottery,” said Vincent Larivire, a professor of information science at the University of Montreal (quoted in the Wall Street Journal, August 10, 2015). The not-quite-so-blessed Leaat Yariv, together with Lirat Einav, wrote the paper that most motivated us (Einav and Yariv 2006). Leeat suggested to us that an alternative or supplementary remedy would be to randomize citations, a suggestion that we discuss towards the end of the paper.

And the request: send comments to debraj.ray@nyu.edu and robson@sfu.ca, and if nothing else, send us a one-word email: yes, if you like the idea; no, if you don’t.