Time Series Analysis of Census Internet Response

Abstract

Collection of census data over the internet promises to be more respondent friendly, accurate and cost effective than other methods. However, the characteristics of internet response over time are not well understood limiting the ability to accurately model, predict and manage take-up. This work theorizes that census internet response can be modeled as a time series of exponentially decreasing response probabilities associated with stimulating events. The response can be modeled as statistical distributions that characterize the events. This modeling is applied to census internet response data provided by Canada and the US to determine the common characteristics of response events and modeling parameters. The approach provides a means to model internet response patterns over time when calibrated to population and the survey methodology characteristics.