Who do we acknowledge, and why?

Being acknowledged on a research paper tells something about you and about the author: The author perceives you as important.

Facebooktwittergoogle_plusredditpinterestlinkedinmail

In the last post I reported evidence that many more researchers contribute to a research paper than just authors. The most visible difference between authors and commenters — and our only way to judge on the effort — is that authors receive all the credit while commenters are “just” acknowledged.

Working with others is that common, that it naturally rises the question: Where do we meet other researchers to get their opinion? Most likely, when it’s not faculty colleagues and friends, it is conference guests or seminar visitors to your faculty.

It’s an open question. The fact however that an author  – call him A – receives advice from another researcher – call him C – reveals information about both: For one, it means that A perceives C as important in the very field, at least important enough to get insights. It may also be because A want to signal quality by showing that he knows C. Vice versa, the fact that C gives a comment for which he is acknowledged shows that he is obviously willing to spend time on someone else’s research.

Oettel (2012) has termed this behaviour as “helpful”, saying that a helpful colleague (one that is often acknowledged) matters more to you than a productive colleague (one that publishes many papers). In a similar fashion, we created a ranking of how often financial economists have been acknowledged in six financial economics journals. Our dataset consists of more than 5,600 full research articles.

The distribution of how often someone is acknowledged is very skewed. The vast majority of commenters is being acknowledged twice or less, with very few individuals being acknowledged very often (zoom in the upper tail by dragging a rectangle):

However, we can uncover more. When A and C agree to exchange ideas, we can think think of their relationship as a tie in a network. In network lingo, a tie is being formed when both nodes (here A and C) mutually. Now take many more researchers and eventually more form a tie with C. This likely puts C in a very central position, which both displays and transfers influence. For this reason we computed the rank according to eigenvector centrality.

The idea behind this measure is: When you know important people, you are probably important as well. Importance here is measured iteratively as the degree to which you know important people. Google uses the same idea for its PageRank.

Interestingly, being acknowledged often does not automatically put researchers in central position. In our paper we explore this relationship and find that Spearman correlations for different periods never exceed 0.65.

In a related blogpost on VOX from last year we propose this ranking as a new measure of influence. This makes sense because under the assumption of mutually beneficial tie formation the market automatically puts the “right” people in the center – the network as the aggregated market wisdom on which researchers matter more for financial economics.

Please refer to our website for an ranking for the period 2009-2011 and others.

References