I'm trying to determine the relative distance of certain points of my plant community to the general regression of the GLOPNET database (Wright) and I'm in doubt if the Hotelling's T-squared is an adequate way of doing so. If not, is there a way of determining that distance, by comparing with the data from the regression, asserts the difference between the groups?
I don´t tink so, because Hotteling requires independent samples, covariance matrices and homogeneous distribution of data with strong adherence to the multivariate normal distribution, additionally requiring a large number of observations. I suggest you to use nonparametric tests.
Well, I would suggest the RMSE (root mean square error) since it can be expressed in the unit of your data or percentage, as you like it. The accuracy of prediction/fitting is reflected with practical value in RMSE. You can use it in non-linear cases as well.
I can recommend the following code in R:
lr.test
I would like to have more information. What are the "groups" of which you speak? What is meant by 'the general regression of the GLOPNET" data base? What variables are measured, and are they measured on the "plant community" or on the individual plants within the community?
I wish to know whether that regression curve is fitted pooling the different data sets or based on only a single data set. If it is the former the root mean square calculated for each set from the general curve will give a measure of the closeness of each data set to the common regression and this will give you an idea of which all sets are more similar
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics