[R-sig-Geo] Ordinary Kriging vs Kriging with external drift

[Justice Moses K. Aheto]

Hi Jose,Thanks for the question.First of all, the R-square value of 0.10 does suggests that some amount of variation is been explained in your external drift model by the inclusion of latitude. Also the R-square value alone is not sufficient to use in deciding whether or not to overlook the drift as some covariates could serve as a potential control variables in your model depending on your problem even if they did not affect your predictive power remarkably. If you plot the variogram for both the ordinary kriging (kriging with no covariates) and the external drift/universal kriging (including latitude as covariates), you will observe that the variance from the ordinary kriging will be higher than the variance in the universal kriging, suggesting that at least, inclusion of latitude has helped explained some spatial variations in your model, I think you need to do some consideration regarding the objective of your study, the outcome you are using in your modelling, and what are some potential/significant factors (covariates) associated with the outcome. The later information could be obtained from the literature on your study outcome. You may or may not be making any error by overlooking the drift depending on the problem you wish to solve. As a result, I will suggest that first of all, fit linear regression model (not spatial model) to your data and include the latitude as a covariate and find out if this covariate is statistically significant in your model. If it is significant, go ahead and include it in your spatial model. In case it is not significant, you have further considerations to make as suggested above (objective of your study, the outcome variable and its associated significant covariates from literature). Unfortunately, there is no exact answer to your problem in the absence of the data you are working with, your outcome variable under consideration, your study objectives and what the potential covariates associated with your outcome are from the literature.   However, my suggestions above could help you get this sorted.Hope this helps.Cheers 
  Kind regards

*****************************************
Justice Moses K. Aheto
PhD Candidate in Medicine (Statistics, Biostatistics, Public Health and Epidemiology). 
MSc Medical Statistics. 
BSc Statistics. 
HND Statistics 

Chief Executive Officer
Statistics and Analytics Consultancy Services Ltd.

Skype: jascall12
Mobile: +447417589148.
	[[alternative HTML version deleted]]