Social networks created with different techniques are not comparable

Research questions should drive edge definitions in social network studies

“…We feel the first assertion is misplaced, and we caution against the second. Our results were in agreement with those of the thornbills in some years for some social network metrics, where we also found correlations between some proximity and interaction methods (see Figure 3 and supplementary material in Castles et al, 2014). However, the correlation between the two methods was not found in other years.…”

“…To return to Farine's first assertion, we are not seeking to generalize patterns from our study but rather the principle that consistency between groups/years should not be assumed until it has been demonstrated. Thus, with respect to Farine's second assertion, we would reiterate our conclusion from Castles et al (2014): because of the dynamic nature of social networks, we recommend that researchers take care when assuming that proximity can be a proxy for interactions. This is distinct from the suggestions that (1) proximity can never be a proxy for interactions and (2) proximity cannot be used to create social networks, generalizations that we do not advocate.…”

“…We focused on one decision a researcher could make at the data collection stage, specifically, the behaviours that could be used to create edges in a social network. Yet, as we mentioned in our study (Castles et al, 2014), there are many considerations after the data collection stage, as highlighted by Farine) and outlined in detail elsewhere (Whitehead, 2008). We appreciate that Farine is using our study to illustrate some general points, and agree that had we analysed our data differently (e.g.…”

“…As we mention above, we are in agreement with Farine that the gold standard in network studies requires a multinetwork framework. In our original article (Castles et al, 2014), we were largely concerned with issues of comparability between studies that use different methods to define an association, and raised the issue of using proximity as a proxy for interactions in the discussion of our findings. Where we disagree with Farine is in his assertion that proximity edges can sometimes be used to infer interaction edges or vice versa without prior testing of this assumption.…”

An introduction to the special column on animal social networks

“…Understanding how interaction networks that represent different situations relate to one another may shed light on emergent properties of the group such as its social stability. However, there is only little theory on how to compare social networks, and methods of data collection can have great implications on the reliability of such comparisons (Castles et al, 2014).…”