Academic disciplines vary widely in their authorship attribution norms. Some fields employ contribution-insensitive conventions (alphabetical ordering or senior-author-last), while others use contribution-sensitive norms where author order signals relative input. Using game-theoretic modeling, we analyze both the evolutionary dynamics of these competing norms and their efficiency implications for scientific collaboration. Our results reveal surprising relationships between expected contribution distributions and norm emergence: fields where senior researchers typically contribute more are paradoxically more likely to evolve norms that obscure these contributions. Furthermore, we demonstrate that all attribution systems create collaboration inefficiencies by discouraging potentially valuable partnerships, but with different patterns of distortion. Contribution-sensitive norms generally support more collaboration overall, particularly when collaboration benefits are modest. These findings illuminate how seemingly arbitrary social conventions in science can significantly impact knowledge production through their effects on collaborative incentives and strategic behavior.