Google Maps E-mail RSS

To Improve Network Measurement, Minimize Opportunities for Error

Published on May 28, 2018 by

Measuring social, biological, and other networks is often a difficult and expensive process, involving surveys, experiments, or other time consuming and costly procedures.  Given that no measurement is perfect, how do we get the greatest “bang for the buck” when trying to assess network ties?  A new paper by NCASD lab members Francis Lee and Carter Butts, published in the journal Social Networks, addresses this question.  Lee and Butts consider the common situation in which measurements on a potential edge are obtained from both parties to that edge, and these potentially discrepant reports are to be integrated.  When the reports agree, the problem is easy: go with the consensus.  But what happens when the reports disagree?  Is it better to require both parties to agree that the edge is present to count it (“mutual assent”), or is it better to count an edge if either party claims it (“unilateral nomination”)?  Applying a hierarchical Bayesian model to extensive data on networks from a variety of settings, Lee and Butts are able to assess the performance of these simple heuristics, and render a verdict: so long as the true network is fairly sparse, requiring mutual assent gives better results than unilateral nomination.  As they show, the reason for this is surprisingly simple.  In a sparse network (one with relatively few edges per individual), there are many more opportunities to invent spurious ties that are not actually present than to miss ties that are actually there – so, even if an informant is less likely to invent ties than to omit them, using the method that guards against that error turns out to yield better results.  These results provide direct and easily followed guidance for researchers working with social network data, and are also applicable to settings such as protein-protein interaction networks in which similar types of error also arise.  By improving the quality of our measurements, we can ensure that researchers get the most out of their hard-won data.

 Share on Facebook Share on Twitter Share on Reddit Share on LinkedIn
No Comments  comments 
© Copyright NCASD 2012