Closed austinwpearce closed 1 year ago
After discussion about Cate Nelson functions, it seems the rcompanion
package uses a sort of hybrid approach that blends the graphical and statistical approach. Differences with soiltestcorr
will likely be minor in many cases, and users should just be aware.
Austin,
The differences between soiltestcorr::cate_nelson_1965()
and the rcompanion::cateNelsonFixedY()
are:
soiltestcorr::cate_nelson_1965()
is based on Cate & Nelson (1965), which finds the CSTV simply based on minimizing the number of observations into negative quadrants (I, III) and maximize them into the positive quadrants (II, IV). rcompanion::cateNelsonFixedY()
is NOT based on Cate & Nelson (1965). It is simply a hybrid form of Cate Nelson (1971), which tries to find the CSTV as the segregating point that maximizes the sum of squares (SS) of the anova of 2 classes (below- and above-CSTV). While the 1971 approach does do that first from a sequential division of STVs, and then finds the value of RY that maximizes the points into positive quadrants, the 1965 starts from drawing a line at a fixed RY, and then maximizes the points on positive quadrants looking for the CSTV. As a hybrid, the rcompanion::cateNelsonFixedY()
starts from drawing a line at a fixed RY value, and then looks for the STV that maximizes the SS. Thus, rcompanion::cateNelsonFixedY()
would tend to have slightly better R2 values compared to soiltestcorr::cate_nelson_1965()
, but it is not the original method described by Cate & Nelson (1965).Hope this explanation helps to explain the difference,
Best,
ADRIAN
The
cate_nelson_1965()
andcateNelsonFixedY()
functions from thesoiltestcorr
andrcompanion
packages, respectively, should ideally compute the same critical X value. However, in the example I included, they don't.Can we explore why this is?
R Code included in folder cate_nelson_problem.zip
R code pasted directly