Investigating coordinated account creation using burst detection and network analysis

Journal of Big Data

Table 1 The average marginal effects (AME) of the binary burst variable on each indicator of coordination

PFA	0.25		0.30		0.35
Dependent Var.	AME	\(\frac{\text {AME}}{\sigma }\)	AME	\(\frac{\text {AME}}{\sigma }\)	AME	\(\frac{\text {AME}}{\sigma }\)
Mean neighbor- hood similarity	0.050\(^{***}\)	1.38	0.040\(^{***}\)	1.11	0.017\(^{***}\)	0.46
Mean clustering coefficient	0.011\(^{*}\)	0.76	0.007\(^{**}\)	0.51	\(9\textrm{e}{-04}\)	0.06
Density amongst contemporaries	\(1\textrm{e}{-06}\)\(^{**}\)	0.62	\(9\textrm{e}{-07}\)\(^{***}\)	0.42	\(3\textrm{e}{-07}\)\(^{***}\)	0.15
Mean hashtag similarity	0.010	2.11	0.005\(^{*}\)	0.97	\(9\textrm{e}{-04}\)\(^{*}\)	0.19
Mail-in voting stance std. dev.	-0.016\(^{***}\)	-2.55	-0.009\(^{***}\)	-1.42	-0.002\(^{***}\)	-0.34
Mask wearing stance std. dev.	-0.011\(^{**}\)	-1.84	-0.005\(^{*}\)	-0.83	-0.001\(^{**}\)	-0.18
Prop. bots \(\ge \) 70%	0.022\(^{***}\)	0.41	0.016\(^{***}\)	0.31	\(3\textrm{e}{-04}\)	\(6\textrm{e}{-03}\)
Prop. sharing low- credibility sites	0.003	0.53	0.004\(^{**}\)	0.73	0.002\(^{***}\)	0.41

Each AME was calculated as the average difference in model predictions between all data points being assigned as bursts and all data points being assigned as non-bursts. They were then standardized by dividing by the standard deviation \((\upsigma )\) of the dependent variable. Three asterisks indicate p < 0.001, two indicate p < 0.01, and one indicates p < 0.05. All p-values were corrected to control for a false discovery rate of 5%. Complete model specifications and AMEs for all variables are available in Additional file 1: Results