Our analyses of three different online datasets confirm the layered structure found in offline face-to-face social networks. For all the online datasets, the scaling ratio for the various layers identified by the analyses, and the respective sizes of these layers, are extremely close to those observed in offline networks (Hill and Dunbar, 2003, Zhou et al., 2005 and Hamilton et al., 2007). These layers have previously been identified only from samples of quite modest size (grouping levels in small scale societies, Christmas card distribution lists: all N < 130,000, ∼6000 and ∼61,000 egos, respectively) from (b) three very different sources.
The sizes of the entire ego networks for the three datasets are smaller than the total size of conventional offline egocentric networks (typically about 150 alters: Hill and Dunbar, 2003, Roberts et al., 2009). This is true especially for the Facebook datasets, where the most external ego network layer is completely missing. This is, perhaps, not too surprising, since the outermost (150) layer in offline networks corresponds to people who are contacted only about once a year. At least as far as the Facebook datasets are concerned, early users (remember that datasets were collected in 2009, when Facebook was still new and yet to start its exponentially increasing diffusion) were not forced into ‘friending’ complete strangers and, instead, typically only sought out people they knew well. Moreover, as discussed earlier, the information we have about weak relationships in Facebook datasets could be not enough to identify all the relationships with very low contact frequency.
In Table 1, we match all the layers obtained from online communications datasets and those identified offline, according to their minimum frequency of contact. The contact frequency of the ego network layers in the two Facebook datasets is significantly lower than those in the corresponding layers in the Twitter dataset. This is not surprising since at the time of the download (2009) users were less active on Facebook than now. Moreover, tweets are very short messages (i.e., up to 140 characters), and the emotional investment in a tweet is likely to be lower than an interaction in Facebook or face-to-face. Despite this, the data on contact frequencies yielded by the online data are surprisingly similar to the face-to-face contact frequencies observed in offline networks. Although the Twitter contact rates are, generally, somewhat higher than both the Facebook and face-to-face contact rates (perhaps reflecting a lower time and emotional investment), nonetheless they are broadly similar, being only about double the frequency of the latter, and in similar ratios across the layers.
Aside from confirming the size and scaling ratios for the conventional first three layers of social networks (those associated, cumulatively, with 5, 15 and 50 individuals) for the Facebook datasets, and also the outermost layer (associated with 150 individuals) for the Twitter dataset, the three online datasets also identify an entirely new layer that was not visible from face-to-face communication data (layer 0 in Table 1). This is an innermost layer at ∼1.5 individuals, scaling perfectly with the layers outside it. The layer is visible in all the datasets. It is clear that this innermost layer has special relevance to egos since they contact these individuals at very high frequencies (on average at least once every five days in Facebook and every other day in Twitter), indicating a very high emotional investment.
This tendency for individuals to have one or two intensely intimate friends is evident in other smaller datasets where contact frequencies have been plotted in rank order (e.g., Saramäki et al., 2014). However, it is clear from even the data shown in Saramäki et al. (2014) that not everyone has such an innermost layer, perhaps explaining the rather odd fact that the scaling ratio itself predicts a decimal value for this layer. It is also possible, however, that this reflects a gender difference in attachment to intimates, such that men have 0–1 and women 1–2. Unfortunately, we do not know the gender identities of the egos in either dataset, so we cannot test whether or not this is so. However, the fact that the three large-scale datasets we have analysed identify this innermost layer suggests that it really is a robust phenomenon, and would merit closer attention.
Quite remarkably, the mean rates of contact in each layer are extremely close, especially for the Facebook datasets, to those found in (and, indeed, used to define: Dunbar and Spoors, 1995) the different layers in egocentric offline personal social networks (Sutcliffe et al., 2012). This suggests that the online environments may be mapping quite closely onto everyday offline networks, or that individuals who inhabit online environments on a regular basis begin to include individuals that they have met online into their general personal social network, treating the different modes of communication as essentially the same. This, of course, has important implications for both the design and promotion of online social environments. However, our present concern is with the sociological similarities between online and offline environments, as implied by these data.
Our analyses of three different online datasets confirm the layered structure found in offline face-to-face social networks. For all the online datasets, the scaling ratio for the various layers identified by the analyses, and the respective sizes of these layers, are extremely close to those observed in offline networks (Hill and Dunbar, 2003, Zhou et al., 2005 and Hamilton et al., 2007). These layers have previously been identified only from samples of quite modest size (grouping levels in small scale societies, Christmas card distribution lists: all N < <1000). The present analyses provide us with strong evidence both for the existence of these layers and for their relative sizes and scaling ratios based on (a) very large sample sizes (>130,000, ∼6000 and ∼61,000 egos, respectively) from (b) three very different sources.The sizes of the entire ego networks for the three datasets are smaller than the total size of conventional offline egocentric networks (typically about 150 alters: Hill and Dunbar, 2003, Roberts et al., 2009). This is true especially for the Facebook datasets, where the most external ego network layer is completely missing. This is, perhaps, not too surprising, since the outermost (150) layer in offline networks corresponds to people who are contacted only about once a year. At least as far as the Facebook datasets are concerned, early users (remember that datasets were collected in 2009, when Facebook was still new and yet to start its exponentially increasing diffusion) were not forced into ‘friending’ complete strangers and, instead, typically only sought out people they knew well. Moreover, as discussed earlier, the information we have about weak relationships in Facebook datasets could be not enough to identify all the relationships with very low contact frequency.In Table 1, we match all the layers obtained from online communications datasets and those identified offline, according to their minimum frequency of contact. The contact frequency of the ego network layers in the two Facebook datasets is significantly lower than those in the corresponding layers in the Twitter dataset. This is not surprising since at the time of the download (2009) users were less active on Facebook than now. Moreover, tweets are very short messages (i.e., up to 140 characters), and the emotional investment in a tweet is likely to be lower than an interaction in Facebook or face-to-face. Despite this, the data on contact frequencies yielded by the online data are surprisingly similar to the face-to-face contact frequencies observed in offline networks. Although the Twitter contact rates are, generally, somewhat higher than both the Facebook and face-to-face contact rates (perhaps reflecting a lower time and emotional investment), nonetheless they are broadly similar, being only about double the frequency of the latter, and in similar ratios across the layers.Aside from confirming the size and scaling ratios for the conventional first three layers of social networks (those associated, cumulatively, with 5, 15 and 50 individuals) for the Facebook datasets, and also the outermost layer (associated with 150 individuals) for the Twitter dataset, the three online datasets also identify an entirely new layer that was not visible from face-to-face communication data (layer 0 in Table 1). This is an innermost layer at ∼1.5 individuals, scaling perfectly with the layers outside it. The layer is visible in all the datasets. It is clear that this innermost layer has special relevance to egos since they contact these individuals at very high frequencies (on average at least once every five days in Facebook and every other day in Twitter), indicating a very high emotional investment.This tendency for individuals to have one or two intensely intimate friends is evident in other smaller datasets where contact frequencies have been plotted in rank order (e.g., Saramäki et al., 2014). However, it is clear from even the data shown in Saramäki et al. (2014) that not everyone has such an innermost layer, perhaps explaining the rather odd fact that the scaling ratio itself predicts a decimal value for this layer. It is also possible, however, that this reflects a gender difference in attachment to intimates, such that men have 0–1 and women 1–2. Unfortunately, we do not know the gender identities of the egos in either dataset, so we cannot test whether or not this is so. However, the fact that the three large-scale datasets we have analysed identify this innermost layer suggests that it really is a robust phenomenon, and would merit closer attention.Quite remarkably, the mean rates of contact in each layer are extremely close, especially for the Facebook datasets, to those found in (and, indeed, used to define: Dunbar and Spoors, 1995) the different layers in egocentric offline personal social networks (Sutcliffe et al., 2012). This suggests that the online environments may be mapping quite closely onto everyday offline networks, or that individuals who inhabit online environments on a regular basis begin to include individuals that they have met online into their general personal social network, treating the different modes of communication as essentially the same. This, of course, has important implications for both the design and promotion of online social environments. However, our present concern is with the sociological similarities between online and offline environments, as implied by these data.
การแปล กรุณารอสักครู่..
