For more information on the data or methods

This data is described well in Viswanath et al. (2009). (see Viswanath et al. 2009). All code is available from Alec McGail [am2873 at cornell dot edu], don't hesitate to reach out.

Helpers and data organization

Random subnetworks

The subnetworks are defined by randomly selecting an ego, and their friends, and the friends of their friends (3rd-order egonet). If this network is sizable enough, but not too sizable (100 < N < 2000) we calculate a whole bunch of network statistics from it and save these.

Friendship network

g.friends <- graph_from_data_frame(
  friendships[,c("f","t")],
  directed = F
)
g.friends <- simplify(g.friends)

There are 63731 individuals in the network, with a total of 817090 edges among them. Thus the average degree is 2*E/V = 25.6418384.

##               feld_predicts
## feld_predicts     -493.2161

There are 508 subnetworks total. For each subnetwork we have the following attributes:

assortativity, crit_deg_overall_theory, crit_deg_overall_true, crit_deg_para_theory, crit_deg_para_true, crit_p_overall_theory, crit_p_overall_true, crit_p_para_theory, crit_p_para_true, deg_meanfdeg_b, deg_meanfdeg_corr, mean_degree, mean_friends_degree, mean_maxfdeg, mean_meanfdeg, mean_medfdeg, median_degree, n_degAtLeastTwo, n_edges, n_nodes, p_deg_eq_median, p_deg_gt_mean, p_deg_gt_median, p_deg_lt_mean, p_deg_lt_median, p_fdeg_eq_median, p_fdeg_gt_mean, p_fdeg_gt_median, p_fdeg_lt_mean, p_fdeg_lt_median, p_meanfdeg_eq_mean, p_meanfdeg_eq_mydeg, p_meanfdeg_gt_mean, p_meanfdeg_gt_mydeg, p_meanfdeg_lt_mean, p_meanfdeg_lt_mydeg, p_medfdeg_eq_median, p_medfdeg_gt_mean, p_medfdeg_gt_median, p_medfdeg_lt_mean, p_medfdeg_lt_median, sd_degree, skew_degree, variance_degree, feld_predicts, var_mean, var_mean2, mean_med_Omean, mPvOm

ASV

Assortativity

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Skewness

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Variance

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Combinations

var_mean2 (rows) X var_mean2 (cols)

## Warning in data.frame(val = as.numeric(l), x = (0:(length(l) - 1))%%10, : NAs
## introduced by coercion

var_mean2 (rows) X mean_med_Omean (cols)

## Warning in data.frame(val = as.numeric(l), x = (0:(length(l) - 1))%%10, : NAs
## introduced by coercion

var_mean2 (rows) X deg_meanfdeg_corr (cols)

mean_med_Omean (rows) X var_mean2 (cols)

## Warning in data.frame(val = as.numeric(l), x = (0:(length(l) - 1))%%10, : NAs
## introduced by coercion

mean_med_Omean (rows) X mean_med_Omean (cols)

## Warning in data.frame(val = as.numeric(l), x = (0:(length(l) - 1))%%10, : NAs
## introduced by coercion

mean_med_Omean (rows) X deg_meanfdeg_corr (cols)

deg_meanfdeg_corr (rows) X var_mean2 (cols)

deg_meanfdeg_corr (rows) X mean_med_Omean (cols)

deg_meanfdeg_corr (rows) X deg_meanfdeg_corr (cols)

## Warning in data.frame(val = as.numeric(l), x = (0:(length(l) - 1))%%10, : NAs
## introduced by coercion

Does assortativity matter at all, given these other attributes?

m1 <- lm(p_meanfdeg_gt_mydeg ~ skew_degree + assortativity, data=nets.friends)
m2 <- lm(p_meanfdeg_gt_mydeg ~ skew_degree + var_mean2, data=nets.friends)
m3 <- lm(p_meanfdeg_gt_mydeg ~ assortativity + var_mean2, data=nets.friends)
m4 <- lm(p_meanfdeg_gt_mydeg ~ assortativity + var_mean2 + skew_degree, data=nets.friends)

top <- rbind(
  data.frame(y=m1$residuals, m=1+rnorm(length(m1$residuals))/10),
  data.frame(y=m2$residuals, m=2+rnorm(length(m1$residuals))/10),
  data.frame(y=m3$residuals, m=3+rnorm(length(m1$residuals))/10),
  data.frame(y=m4$residuals, m=4+rnorm(length(m1$residuals))/10)
)

ggplot(top) +
  geom_point(aes(x=m, y=y), size=1) +
  xlab("Model") +
  ylab("Residual")

Viswanath, Bimal, Alan Mislove, Meeyoung Cha, and Krishna P. Gummadi. 2009. β€œOn the Evolution of User Interaction in Facebook.” In Proceedings of the 2Nd Acm Workshop on Online Social Networks, 37–42. WOSN ’09. New York, NY, USA: ACM. doi:10.1145/1592665.1592675.