Abstract
Scientific communities confer many forms of credit on their successful members. The motivation provided by these forms of credit helps shaping a community’s collective attention toward different lines of research. The allocation of scientific credit, however, has also been the focus of long-documented pathologies: certain research questions are said to command more credit then they deserve; and certain researchers seem to receive a disproportionate share of the credit. Here we show that each of these pathologies can actually increase the collective productivity of a community. We consider a model for the allocation of credit, in which individuals pick a project among projects of varying importance and difficulty levels, and compete to receive credit with others who choose the same project. Under the most natural allocation mechanism, in which credit is divided equally among those who succeed at a project in proportion to the project’s importance, the resulting selection of projects by self-interested, credit-maximizing individuals will in general be socially sub-optimal. However, we show that there exist ways of allocating credit both out of proportion to the true importance of the projects and out of proportion to the relative contributions of the individuals, that lead credit-maximizing individuals to achieve social optimality. These results therefore suggest how well-known forms of misallocation of scientific credit can in fact serve to channel self-interested behavior into socially optimal outcomes.
Similar content being viewed by others
Notes
This is a kind of rich-get-richer phenomenon, and Merton’s use of the term “Matthew Effect” is derived from Matthew 25:29 in the New Testament of the Bible, which says, “For unto every one that hath shall be given, and he shall have abundance: but from him that hath not shall be taken away even that which he hath.”
A similar claim was independently proven for a related class of congestion games in [22].
Note that we establish (i) and (ii) by showing that our game is \((\lambda ,\mu )\)-smooth. Thus, by [29] our price of anarchy results automatically carry over to more general solution concepts such as mixed Nash equilibria and coarse correlated equilibria.
Actually, as was cleverly proved in [29] the fact that the our game is \((\lambda , \mu )\)-smooth implies that consequence (ii) extends to more solution concepts: worst mixed Nash equilibrium, correlated equilibrium and no regret solution. For example, the ratio between the worst mixed Nash equilibrium and the optimal solution is also at most \(2 - \frac{1}{n}\).
References
Aaronson, S., Borodin, A., Chazelle, B., Goldreich, O., Goldwasser, S., Karp, R., Kearns, M., Papadimitriou, C., Sudan, M., Vadhan, S: Statement on conceptual contributions in theory. http://scottaaronson.com/blog/?p=315 (2008)
Arrow, K. J.: Economic welfare and the allocation of resources for invention. In: National Bureau of Economic Research, editor, The Rate and Direction of Inventive Activity: Economic and Social Factors, pp. 609–626. Princeton University Press (1962)
Bachrach, Y., Syrgkanis, V., Vojnović, M.: Incentives and efficiency in uncertain collaborative environments. In: International Conference on Web and Internet Economics, pp. 26–39. Springer (2013)
Parimal Kanti Bag: Optimal auction design and R&D. Eur. Econ. Rev. 41(9), 1655–1674 (1997)
Banerjee, S., Goel, A., Kollagunta K.A.: Re-incentivizing discovery: mechanisms for partial-progress sharing in research. In: Proceedings of the Fifteenth ACM Conference on Economics and Computation, pp. 149–166. ACM (2014)
Bergstrom, C.T., Foster, J.G., Song, Y.: Why scientists chase big problems: individual strategy and social optimality. arXiv:1605.05822 (2016)
Bonilla, J.Z.: The economics of scientific knowledge. In: Mäki, U. (ed.) Philosophy of Economics, pp. 823–862. Elsevier, London (2012)
Borodin, A., Nielsen, M.N., Rackoff, C.: (Incremental) priority algorithms. Algorithmica 37(4), 295–326 (2003)
Bourdieu, J.-P.: The specificity of the scientific field and the social conditions of the progress of reason. Soc. Sci. Inf. 14(6), 19–47 (1975)
Burke, M., Kraut, R.: Mopping up: Modeling wikipedia promotion decisions. In Proceedings of the CSCW’08: ACM Conference on Computer-Supported Cooperative Work, pp. 27–36 (2008)
Caragiannis, I., Kaklamanis, C., Kanellopoulos, P.: Taxes for linear atomic congestion games. ACM Trans. Algorithms (TALG) 7(1), 13 (2010)
Che, Y.-K., Gale, I.: Optimal design of research contests. Am. Econ. Rev. 93(3), 646–671 (2003)
Cole, R., Dodis, Y., Roughgarden, T.: How much can taxes help selfish routing? J. Comput. Syst. Sci. 72(3), 444–467 (2006)
Dasgupta, P., E, Maskin: The simple economics of research portfolios. Econ. J. 97(387), 581–95 (1987)
Fotakis, D., Kontogiannis, S., Koutsoupias, E., Mavronicolas, M., Spirakis, P.: The structure and complexity of nash equilibria for a selfish routing game. Theor. Comput. Sci. 410, 3305–3326 (2009)
Garey, M.R., Johnson, D.S.: Computers and Intractability: A Guide to the Theory of NP-Completeness. W.H Freeman, New York (1979)
Goemans, M.X., Li, L., Mirrokni, V.S., Thottan, M.: Market sharing games applied to content distribution in ad hoc networks. IEEE J. Sel. Areas Commun. 24(5), 1020–1033 (2006)
Goldreich, O.: Innovations in Computer Science (ICS). http://www.wisdom.weizmann.ac.il/oded/ics.html (2009)
Gowers, T., Kalai, G., Nielsen, T.: The polymath blog. http://polymathprojects.org/about/
Kitcher, P.: The division of cognitive labor. J. Philos. 86(1), 5–21 (1990)
Kitcher, P.: The Advancement of Science. Oxford University Press, Oxford (1993)
Kuniavsky, S., Smorodinsky, R.: Greediness and equilibrium in congestion games. Econ. Lett. 121(3), 499–503 (2013)
Leskovec, J., Huttenlocher, D., Kleinberg, J.: Governance in social media: a case study of the Wikipedia promotion process. In: Proceedings of the 4th International Conference on Weblogs and Social Media (2010)
Merton, R.K.: The Matthew effect in science. Science 159(3810), 56–63 (1968)
Merton, R.K.: The Sociology of Science: Theoretical and Empirical Investigations. University of Chicago Press, Chicago (1973)
Monderer, D., Shapley, L.S.: Potential games. Games Econom. Behav. 14, 124–143 (1996)
Peirce, C.S.: Note on the theory of economy of research. In: Arthur, W.B. (ed.) Collected Papers of Charles Sander Peirce, vol. 7, pp. 76–83. Harvard University Press, Cambridge (1958)
Roughgarden, T.: Selfish Routing and the Price of Anarchy. MIT Press, New York (2005)
Roughgarden, T.: Intrinsic robustness of the price of anarchy. In: Proceedings of the 41st ACM Symposium on Theory of Computing, pp. 513–522 (2009)
Sent, E.-M.: Econmics of science: survey and suggestions. J. Econ. Methodol. 6(1), 95–124 (1999)
Shen, H.-W., Barabási, A.-L.: Collective credit allocation in science. Proc. Natl. Acad. Sci 111(34), 12325–12330 (2014)
Stein, K., Hess, C.: Does it matter who contributes: a study on featured articles in the German Wikipedia. In: Proceedings of the 18th ACM Conference on Hypertext and Hypermedia, pp. 171–174 (2007)
Strevens, M.: The role of the priority rule in science. J. Philos. 100(2), 55–79 (2003)
Strevens, M.: The role of the Matthew effect in science. Stud. Hist. Philos. Sci. 37, 159–170 (2006)
Tardos, É., Wexler, T.: Network formation games and the potential function method. In: Nisan, N., Roughgarden, T., Tardos, É., Vazirani, V. (eds.) Algorithmic Game Theory, pp. 487–516. Cambridge University Press, Cambridge (2007)
Taylor, C.R.: Digging for golden carrots: an analysis of research tournaments. Am. Econ. Rev. 85(4), 872–890 (1995)
Vetta, A.: Nash equilibria in competitive societies, with applications to facility location, traffic routing and auctions. In: Proceedings of the 43rd IEEE Symposium on Foundations of Computer Science, pp. 416–425 (2002)
Acknowledgements
We thank Harry Collins, Michael Macy, Trevor Pinch, and Michael Strevens for valuable insights and suggestions on the relevant connections to work in the philosophy and sociology of science, and Larry Blume, David Easley, Bobby Kleinberg and Rann Smorodinsky for valuable advice on the game-theoretic aspects of the problem.
Author information
Authors and Affiliations
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A preliminary version of the paper appeared in ACM Symposium on Theory of Computing 2011, (STOC’11). This work was partially supported by BSF Grant 2018206.
Appendix
Appendix
In this Appendix we prove Theorem 2.12 we show that, with \({\varepsilon }> 0\) sufficiently small and the re-weighting of players defined by \(z_i = {\varepsilon }^i\), all Nash equilibria of the resulting game are socially optimal.
Even given the informal argument above, the proof is complicated by the fact that, with positive weights on all players, their strategic reasoning is more complex than it would be under an actual ordering. To prove Theorem 2.12, we consider the relationship between the actual utilities of the re-weighted players for a given strategy vector \(\mathbf {a}\), denoted \({u_i}(\mathbf {a}; \mathbf {z})\):
and their “ideal” utility under the order we are trying to simulate, denoted \(\widehat{u_i}(\mathbf {a}; \mathbf {z})\) which is formally defined next:
Definition 5.1
-
\(\pi _{\small< i}(S)=\{h \in S | z_i < z_h\}\) where \(S \subseteq N\) – the set of players before player i.
-
\(\pi _{\small>i}(S)=\{h \in S | z_i > z_h\}\) where \(S \subseteq N\) – the set of players after player i.
-
\({\widehat{u}}_i(\mathbf {a}; \mathbf {z})= r_{a_i}(|\pi _{\small < i}(K_{a_i}(\mathbf {a}))|)\) – the marginal contribution of player i to the social welfare.
Recalling that the projects’ weights and success probabilities are rational, let d be their common denominator. We first show that if these two different utilities are close enough with respect to d, then our approximate simulation of an order using weights will succeed:
Claim 5.2
If for every player i, project j and strategy vector \(\mathbf {a}\) we have
then any Nash equilibrium in the game with the weights \(\{z_i\}\) is also an optimal assignment.
Proof
Consider some Nash equilibrium \(\mathbf {a}\) in the game with the weights \(\mathbf {z}\). Let \(\mathbf {o}\) be an optimal assignment that is the result of running the greedy algorithm from Claim 2.2 with the same players’ order. In case several such optimal assignments exist, we pick \(\mathbf {o}\) to be the one that shares the longest prefix with \(\mathbf {a}\): \(\max _{i} \forall i'\le i,a_{i'}=o_{i'}\). Assume towards a contradiction that \(\mathbf {a}\) is not an optimal assignment. By definition, player i is the first player in the order that works on a different project in \(\mathbf {a}\) (denote it by j) and in \(\mathbf {o}\) (denote it by l). Since \(\mathbf {a}\) is a Nash equilibrium with the weights \(\mathbf {z}\) we have that \({u_i}(j,a_{-i}; \mathbf {z}) \ge {u_i}(l,a_{-i}; \mathbf {z})\). By applying the assumption of the claim we have that:
This implies that \(\widehat{u_i}(j,a_{-i}; \mathbf {z}) + \dfrac{1}{2d^{n+1}} \ge \widehat{u_i}(l,a_{-i}; \mathbf {z})\). Also, since \(\mathbf {o}\) is the result of a greedy algorithm with the same order, we have that \({\widehat{u}}_i(l,o_{-i}; \mathbf {z}) > {\widehat{u}}_i(j,o_{-i}; \mathbf {z})\). Since, \(\mathbf {a}\) and \(\mathbf {o}\) are identical for all players prior to player i in the order, we have that \({\widehat{u}}_i(l,a_{-i}; \mathbf {z}) > {\widehat{u}}_i(j,a_{-i}; \mathbf {z})\). Hence,
To complete the proof, recall that d is the common denominator of all success probabilities and weights. As both \(\widehat{u_i}(j,a_{-i}; \mathbf {z})\) and \(\widehat{u_i}(l,a_{-i}; \mathbf {z})\) are products of at most \(n+1\) terms of common denominator d, and they are not equal, they must differ by at least \(\dfrac{1}{d^{n+1}}\). Thus, it is impossible that in the Nash equilibrium \(\mathbf {a}\) player i works on project j and in the optimal solution \(\mathbf {o}\) on project l and a contradiction is reached. \(\square \)
Next, we show that it is possible to choose \({\varepsilon }\) sufficiently small such that for every player i and project j: \(\widehat{u}_i(j,a_{-i}; \mathbf {z}) - \dfrac{1}{4d^{n+1}} \le {u_i}(j,a_{-i}; \mathbf {z}) \le \widehat{u_i}(j,a_{-i}; \mathbf {z}) + \dfrac{1}{4d^{n+1}}\) as required by Claim 5.2. We begin by presenting the following definitions that will be useful in simplifying the utility function:
Definition 5.3
For a strategy vector \(\mathbf {a}\), a project j and \(S \subseteq K_j(a_{-i})\) we define
This is the probability that the players in \(S \cup \{i\}\) succeed at project j, the rest of the players working on j fail and player i gets the credit for succeeding at the project.
Using the definition we now have \({u_i}(j,a_{-i}; \mathbf {z}) = w_j \sum _{S \subseteq K_j(a_{-i})} \psi _i(j;S;\mathbf {a})\).
By using \(\pi _{\small < i}(S)\) and \(\pi _{\small >i}(S)\) we can break up the player’s utility in the following manner:
This is a convenient representation of a player’s utility since it partitions the successful player sets into two types:
-
1.
\(S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))\): for such a set S, player i’s weight is dominant in S, and hence she gets most of the utility.
-
2.
\(S \subseteq K_j( a_{-i})\) and \(S \cap \pi _{\small < i}(K_j(\mathbf {a}))\ne \emptyset \): for such a set S, player i’s weight is dominated, and hence she gets only a very small fraction of the utility.
In the next two lemmas we bound player i’s utility with respect to each of these types separately. The bounds are later combined to bound \({u_i}(j, a_{-i}; \mathbf {z})\).
Lemma 5.4
\(\dfrac{1}{1+2 {\varepsilon }} {\widehat{u}}_i(j, a_{-i}; \mathbf {z}) \le w_j \cdot \sum _{S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))} \psi _i(j;S;\mathbf {a}) \le {\widehat{u}}_i(j, a_{-i}; \mathbf {z})\).
Proof
We first show that \(w_j \cdot \sum _{S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))} \psi _i(j;S;\mathbf {a}) \le {\widehat{u}}_i(j, a_{-i}; \mathbf {z})\). To do this, we write an alternative expression for \({\widehat{u}}_i(j, a_{-i}; \mathbf {z})\):
The resulting expression is an upper bound on \(w_j\sum _{S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))} \psi _i(j;S;\mathbf {a})\) since it assumes that \(z_i>0\) and the other weights are 0.
We now turn to show that \(\dfrac{1}{1+2 {\varepsilon }} {\widehat{u}}_i(j, a_{-i}; \mathbf {z}) \le w_j \cdot \sum _{S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))} \psi _i(j;S;\mathbf {a})\). By the definition of the weights, we have that for all \(h\in \pi _{\small >i}(K_j(\mathbf {a}))\), \(z_h \le {\varepsilon }z_i\). This allows us to bound the weight coefficients \(\dfrac{z_i}{z_i + \sum _{h \in S} z_h}\) in \({u_i}(j, a_{-i}; \mathbf {z})\):
where the last inequality holds for \({\varepsilon }< 0.5\). Hence we have that \(\dfrac{1}{1+2 {\varepsilon }} {\widehat{u}}_i(j, a_{-i}; \mathbf {z}) \le w_j \cdot \sum _{S \subseteq \pi _{\small >i}(K_j(\mathbf {a}))} \psi _i(j;S;\mathbf {a})\). \(\square \)
Lemma 5.5
\(\sum _{\begin{array}{c} S \subseteq K_j(a_{-i}) \\ S \cap \pi _{\small < i}(K_j(\mathbf {a}))\ne \emptyset \end{array} } \psi _i(j;S;\mathbf {a}) \le \dfrac{{\varepsilon }}{1+ {\varepsilon }} \).
Proof
Observe that since for each of the sets S included in the sum, \(S \cap \pi _{\small < i}(K_j(\mathbf {a}))\ne \emptyset \), then in each set S there exists at least a single player \(h\in S\) such that \(z_h>z_i\). This implies the following bound on the weight coefficients in \( \psi _i(j;S;\mathbf {a})\):
The proof is completed by observing that \(\sum _{\begin{array}{c} S \subseteq K_j(a_{-i}) \\ S \cap \pi _{\small< i}(K_j(\mathbf {a}))\ne \emptyset \end{array} } q_j^{|S|+1} \cdot (1-q_j)^{k_j(a_{-i}) - |S|} <1\). \(\square \)
It is not hard to see that given Eq. 3 and Lemmas 5.4 and 5.5 the following bounds hold:
Corollary 5.6
For every strategy vector \(\mathbf {a}\), player i and project j:
We can now use the previous corollary to compute a value of \({\varepsilon }\) for which the assumptions of Lemma 5.2 hold:
Lemma 5.7
Let l be a project with maximal weight (\(l \in \arg \max _{j \in M} w_j\)). For \({\varepsilon }\le \dfrac{1}{8d^{n+1}w_l}\) we have that: \( {\widehat{u}}_i(j,a_{-i}; \mathbf {z}) -\dfrac{1}{4d^{n+1}} \le {u_i}(j,a_{-i}; \mathbf {z}) \le \widehat{u}_i(j,a_{-i}; \mathbf {z}) + \dfrac{1}{4d^{n+1}} \).
Proof
Note that by the lower bound on \({u_i}(j, a_{-i}; \mathbf {z})\) from Corollary 5.6 and the choice of l and the choice of \({\varepsilon }\) we have that
Similarly, by the upper bound on \({u_i}(j, a_{-i}; \mathbf {z})\) from Corollary we have that \( {u_i}(j, a_{-i}; \mathbf {z}) \le {\widehat{u}}_i(j, a_{-i}; \mathbf {z}) +w_j \dfrac{{\varepsilon }}{1+ {\varepsilon }} \le {\widehat{u}}_i(j, a_{-i}; \mathbf {z}) + \dfrac{1}{4d^{n+1}}\). \(\square \)
This completes our proof of Theorem 2.12 showing that there exists \({\varepsilon }\) for which any equilibrium of the game with player-weights \(\{z_i={\varepsilon }^i\}\) is an optimal solution in the unweighted game.
Rights and permissions
About this article
Cite this article
Kleinberg, J., Oren, S. Mechanisms for (Mis)allocating Scientific Credit. Algorithmica 84, 344–378 (2022). https://doi.org/10.1007/s00453-021-00902-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00453-021-00902-y