Using Scan Statistics for Cluster Detection: Recognizing Real Bandwagons
Bandwagons are ubiquitous in social life. No one doubts that people vote at least sometimes for political candidates simply because they are winning and or embrace many fashions simply because they want to “follow the crowd.” But estimating how much a bourgeoning trend owes to pure “bandwagon effects” can be very difficult. Often other factors motivate the people taking action to an unknown degree. In this paper we investigate the use of two variable window scan statistics, the minimum P value scan statistic and the generalized likelihood ratio test (GLRT) statistic, to analyze one important form of the bandwagon problem. We show how these scan statistics can be used to detect the clustering of bandwagon events in a time interval. Once the events are identified, the information can be used to set boundaries on the extent of bandwagoning. The method is illustrated by reference to data on political contributions in the 2016 U.S. Senate elections.
Chen, J., Ferguson, T. & Jorgensen, P. Using Scan Statistics for Cluster Detection: Recognizing Real Bandwagons. Methodol Comput Appl Probab 22, 1481–1491 (2020). https://doi.org/10.1007/s11009-019-09737-1
Methodol Comput Appl Probab