Identifying and characterizing extrapolation in multivariate response data

Abstract

Faced with limitations in data availability, funding, and time constraints, ecologists are often tasked with making predictions beyond the range of their data. In ecological studies, it is not always obvious when and where extrapolation occurs because of the multivariate nature of the data. Previous work on identifying extrapolation has focused on univariate response data, but these methods are not directly applicable to multivariate response data, which are common in ecological investigations. In this paper, we extend previous work that identified extrapolation by applying the predictive variance from the univariate setting to the multivariate case. We propose using the trace or determinant of the predictive variance matrix to obtain a scalar value measure that, when paired with a selected cutoff value, allows for delineation between prediction and extrapolation. We illustrate our approach through an analysis of jointly modeled lake nutrients and indicators of algal biomass and water clarity in over 7000 inland lakes from across the Northeast and Mid-west US. In addition, we outline novel exploratory approaches for identifying regions of covariate space where extrapolation is more likely to occur using classification and regression trees. The use of our Multivariate Predictive Variance (MVPV) measures and multiple cutoff values when exploring the validity of predictions made from multivariate statistical models can help guide ecological inferences.

Ecological prediction at macroscales using big data: Does sampling design matter?

“…4). Recent work on identifying when predictions will be extrapolation or interpolation suggests that this can be done by either examining distributions of predictor variables or comparing predictive variance at out‐of‐sample locations to a threshold (e.g., maximum predictive variance) based on in‐sample locations (Conn et al 2015; Bartley et al 2019). As our example shows, not all targeted sampling designs will result in extrapolation and it may be acceptable to include data from such targeted efforts in larger, compiled data sets.…”

A Multiscale Land Use Regression Approach for Estimating Intraurban Spatial Variability of PM2.5 Concentration by Integrating Multisource Datasets

“…As the model estimates the variation in PM 2.5 instead of the absolute concentration value, the resultant map had a data distribution across positive and negative. It is important to note that the regression model is only valid within the spatial range of road and street area (the centrality and accessibility variables representing traffic were only calculated at the road network) and the numerical range of input predictor variables (no “extrapolation” [ 74 ] was done for the model to reduce the uncertainty). The resultant map reflects the variation at a finer spatial scale, but it does not have full spatial coverage of the city.…”

Top ten hazards to avoid when modeling species distributions: a didactic guide of assumptions, problems, and recommendations

“…Additionally, the model's robustness to extrapolation can be estimated with evaluation schemes that include transfer, such as spatial blocks (Roberts et al 2017, Soley‐Guardia et al 2019) (Hazard 8). Fortunately, tools for tackling aspects of extrapolation continue to be developed (Mesgaran et al 2014, Bartley et al 2019, Cobos et al 2019b, Andrade et al 2020). These now include the ability to make separate decisions (whether or not to constrain the response) for each tail of every environmental variable (for example, depending on whether the response curve is increasing or decreasing at the point of truncation; Anderson 2013, Guevara et al 2018b, Kass et al 2021).…”