Premium
Assessing stationarity in web analytics: A study of bounce rates
Author(s) -
Poulos Marios,
Korfiatis Nikolaos,
Papavlassopoulos Sozon
Publication year - 2020
Publication title -
expert systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.365
H-Index - 38
eISSN - 1468-0394
pISSN - 0266-4720
DOI - 10.1111/exsy.12502
Subject(s) - computer science , dynamic time warping , metric (unit) , analytics , sampling (signal processing) , process (computing) , data mining , series (stratigraphy) , web analytics , image warping , data science , artificial intelligence , web page , world wide web , filter (signal processing) , web development , paleontology , operations management , web application security , biology , economics , computer vision , operating system
Evidence‐based methods for evaluating marketing interventions such as A / B testing have become standard practice. However, the pitfalls associated with the misuse of this decision‐making instrument are not well understood by managers and analytics professionals. In this study, we assess the impact of stationarity on the validity of samples from conditioned time series, which are abundant in web metrics. Such a prominent metric is the bounce rate, which is prevalent in assessing engagement with web content as well as the performance of marketing touchpoints. In this study, we show how to control for stationarity using an algorithmic transformation to calculate the optimum sampling period. This distance is based on a novel stationary ergodic process that considers that a stationary series presents reversible symmetric features and is calculated using a dynamic time warping algorithm in a self‐correlation procedure. This study contributes to the expert and intelligent systems literature by demonstrating a robust method for sub‐sampling time‐series data, which are critical in decision making.