I've just finished reading Methodological Pitfalls in Social Network Analysis by Nicholas Marschall. The central theme is that current methods produce questionable results, which is precisely why I read the book.

Running to 86 pages the book is an easy one-sitting read. For what it is it's also expensive. The book is a translation from German so in some places the English is - well unusual. Looking beyond this small problem, it appears to be a student or scientist research justification, or perhaps a short synopsis of a PhD, which means the style is very academic, but it is interesting!

Marschall quite rightly says data collection approaches colour results, and are full of implicit assumptions. He comes to the conclusion that size reduction and transformation processes, which are quite common in published studies, can significantly change the results of an analysis. By size reduction he means the loss of data resulting from sampling methods or even forgetting. Transformation is when the researcher manipulates data such as block modelling, dichotomisation, and symmetrisation.

The boundary specification or missing data problem is well known, and extensively discussed in other literature. Marschall highlights the problems this can cause. In particular he highlights mathematically and graphically the problems this may cause with small networks. Figure 4-1 on page 39 is most enlightening. Using five actors he shows how betweenness centrality can give very different and almost polar results simply by removing one actor. In his example an actor who was the most peripheral becomes the most central. This highlights why a deep understanding of the network and data matters. My own work confirms this, which is why I am reluctant to deal with anything other than whole very tightly defined populations.

I highly recommend this book to serious network analysis students, researchers and practitioners. Understanding the limitations and methodological pitfalls of network analysis matters!

Regards, Graham