HW1_Prob10_Discussion

Anna Starodubtseva

October 8, 2021

  • Linear regression (simple, GLM) is able to capture some spatial variation with a relatively simple model
  • Linear regression could be applied to data where general/regional trends are more important to identify than local behavior
  • The local polynomial/GLM method is more complex and less interpretable but offers greater potential to capture local behavior
  • A local polynomial/GLM model could be applied to data with a larger number of covariates where local variation is desired
  • GAM offers the greatest amount of model flexibility at the cost of having to manually identify different combinations of models
  • GAM could be applied for datasets with a small number of covariates and/or where the approximate desired relationship of the data to covariates is already known
  • Spatial models (Kriging, hierarchical) best capture spatial effects and interpretability of the spatial data but are more computationally expensive
  • Spatial models could be applied for spatial data where detailed local behavior is desired