Data are available at greater volumes, varieties, and velocities than ever before, but there are still many useful datasets or data points missing. Challenges to understanding the world with data arise from dataset missingness. Missing data points reduce statistical power, bias estimation parameters, reduce the representativeness of samples, and generally complicate statistical analysis (Kang 2013; Rubin 1976). There are various methodologies for imputing or interpolating missing values - the appropriate method depends on the context and the missingness mechanism (Bertsimas, Pawlowski, and Zhuo 2018; Comber and Zeng 2019; Lam 1983; Li and Heap 2014; Mandel J 2015).
I distinguish between missing data points and sets. Missing data points occur when a variable is included in a dataset and partially complete. When I speak of missing datasets, I’m referring more to what Onuoha speaks of - “blank spots that exist in spaces that are otherwise data-saturated…something does not exist, but it should…an established system is disrupted by distinct absence. That which we ignore reveals more than what we give our attention to. It’s in these things that we find cultural and colloquial hints of what is deemed important. Spots that we've left blank reveal our hidden social biases and indifferences.”
Both spatial patterns within the data and GIS frameworks constrain interpolation. In the decision tree below methodologies are color coded according to similarities. For example, SK and OK are different only in their choice of mean (local, global). The major categories of difference for these spatial interpolation methods are: focus on local trends or global ones, exactness, incorporation of randomness or error (deterministic vs. stochastic), the smoothness of the output surface (gradual vs. abrupt), range of estimates within data range (convex) or not, number of explanatory variables incorporated (univariate vs. multivariate), and assumptions of linearity or normal distribution.
