Authors: Ikuho Yamada*, University of Tokyo, Hiroyuki Usui, University of Tokyo
Topics: Spatial Analysis & Modeling, Geographic Information Science and Systems
Keywords: sample data, sample size, spatial statistical analysis, local Moran’s I statistic
Session Type: Virtual Paper
Start / End Time: 4:40 PM / 5:55 PM
Room: Virtual 18
Presentation File: No File Uploaded
Regional data used in spatial analysis are often based on samples rather than the population. For example, data related to people’s opinions and/or behaviors may be collected through a questionnaire survey for individuals sampled from a study region. Even when we have “big data,” they do not include everyone so that we still deal with some kind of samples from the population.
Because characteristics of the population estimated from the sample inevitably have some discrepancies from its true characteristics and those discrepancies potentially influence results of statistical analysis, traditional, non-spatial statistics provide accumulated theories and discussions about samples and sampling. However, such discussions have been quite limited in the context of spatial statistical analysis, whereas those discrepancies in sample data could be more problematic there as multiple data values collected for individual areal units are often compared and integrated.
We conducted several simulation-based studies and revealed that small sample size would decrease the power of spatial statistical analysis more than expected from theories of non-spatial statistics. In the current study, we thus attempt to develop a methodology to determine appropriate sample size for spatial statistical analysis using local Moran’s I statistic as an example. We first discuss its theoretical development based on a probability distribution of local Moran’s I value and then present results of simulation-based experiments.