P094 - The effects of and treatments for measurement error in network analysis - a multidisciplinary systematic review

• In Event: Poster Session & Ice Cream Social

Abstract

Network analysis is an increasingly popular tool used to conceptualise and analyse relational data. Often applied to social networks, it nonetheless has applications ranging from analysing features of terrorist networks, to modelling interactions between protein complexes. A growing field within network analysis concerns the effects of and treatments for measurement error, which can present a challenge to researchers due to the interdependent nature of network data. This not only necessitates the use of different analytical methods than those seen in more traditional frequentist statistics, but also makes dealing with measurement error significantly more complex. This is problematic as measurement error can have significant consequences for network analysis, especially in cases such as highly structurally important nodes being missing. In cases such as this, measurement error concerning a single, highly connected node may be sufficient to alter the statistics of the entire network. However, whilst multiple studies have investigated various effects measurement error in a network analysis context, there currently is no systematic analysis collating these results, as well as no systematic framework detailing best practice for reporting. Furthermore, the use of graph theoretical methods is not limited to the field of social network analysis. Therefore, this review aims to: 1, Summarise findings regarding the effects of measurement error across network types. 2, discuss findings regarding the efficacy of treatment strategies across error and network types. 3, assess whether methodologies used by different disciplines may be applicable to network analysis, as used in the study of criminal networks.

Authors

• Ben Palfreeman-Watt, University of Manchester

• Ben Palfreeman-Watt, University of Manchester