Monitoring autoregressive binary social networks based on likelihood statistics. (November 2020)

• Record Type: Journal Article
• Authors: Taheri, Zahra; Esmaeeli, Hamid; Doroudyan, Mohammad Hadi
• Is Part Of: Computers & industrial engineering. Volume 149(2020)
• Publication Date: 2020-11
• Subjects: Social network -- Binary response -- Logit link function -- Autoregressive model -- Likelihood estimation; ACP average communication probability -- AD double values for all parameters -- AM negative values for all parameters -- AP positive values for all parameters -- AR autoregressive -- ARL average run length -- ARMA autoregressive moving average -- CEO chief executive officers -- CUSUM cumulative sum -- DBLP digital bibliography and library project -- DM directors and managers -- EWMA exponential weighted moving average -- FIFO first in first out -- GLM generalized linear model -- LRT likelihood ratio test -- MCS maximum common sub-graph -- MRL median run length -- PR presidents -- SDRL standard deviation of run length; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2020.106721
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: Considering an autocorrelation structure for a labeled network. Proposing three new statistics for dealing with the autocorrelation structure. Evaluating the performance of the proposed methods using simulation studies. Comparing the performance of the proposed methods with three counterpart methods. Applying the proposed methods to a real-life example. Abstract: Network monitoring is a new area in statistical process control applications. It aims at detecting assignable changes in the communication structure of a network. The probability of communications in social networks is usually based on the attributes of vertices. Moreover, due to the nature of human relationships, social networks are almost time-dependent. Neglecting this feature in control chart design reduces the chart performance. In this paper, communications are defined as autoregressive binary variables with the probability modeled by the logit link function. The explanatory variables of the model are the vertices' attributes and previous information of the network. Accordingly, we propose three likelihood ratio test-based methods, one static and two dynamic reference methods. The performance of the proposed methods is evaluated using simulation studies and real numerical examples from the email communications of Enron Corporation. Then, the effect of the autocorrelation structure on the link function is investigated. Also, the effect of parameter estimation on the ARL measure of the proposed methods is studied. Furthermore, the performance of the proposed methods is compared with three traditional methods. Finally, some practical suggestions are given for different out-of-control situations and statistical designs.