repeated_measures_test: Perform test on two repeated measures

In fruciano/GeometricMorphometricsMix: Miscellaneous functions useful for geometric morphometrics

Perform test on two repeated measures

Description

Test based on Hotelling's T squared for the null hypothesis of no effect between two repeated measures (e.g., treatment/control)

Usage

repeated_measures_test(T1, T2, rnames = TRUE, shrink = FALSE)

Arguments

T1, T2	matrices (n x p of n observation for p variables) or arrays (t x p x n of n observations, t landmarks in p dimensions),
rnames	if TRUE (default) the rownames of the matrix or the names along the 3rd dimension (for arrays) will be used to match the order
shrink	if TRUE, a shrinkage estimator of covariance is used internally

Details

The function assumes that each individual observation (e.g., specimen) has been measured two times (e.g., at two time points, or between two treatments).

If rnames is TRUE (default), the rownames of the matrix or the names along the 3rd dimension (for arrays) will be used to match the order of observations (e.g., specimens) between the two datasets. Otherwise, the function will assume that the observations in T1 and T2 are in the same order.

This function is useful in various contexts, such as:

• testing the effect of preservation (Fruciano et al. 2020)

• testing for variation through time

For instance, in the context of the effect of preservation on geometric morphometrics, it has been argued (Fruciano, 2016) that various studies have improperly used on repeated measures data methods developed for independent observations, and this can lead to incorrect inference.

Value

The function outputs a matrix n x p of the original data projected to the subspace orthogonal to the vector

Notice

The function requires internally non-singular matrices (for instance, number of cases should be larger than the number of variables). One solution can be to perform a principal component analysis and use the scores for all the axes with non-zero and non-near-zero eigenvalues. To overcome some situations where a singular matrix can occurr, the function can use internally a shrinkage estimator of the covariance matrix (Ledoit & Wolf 2004). This is called setting shrink = TRUE. However, in this case, the package nlshrink should have been installed. Also, notice that if the matrices T1 and T2 are provided as arrays, this requires the package Morpho to be installed.

Citation

If you use this function please cite Fruciano et al. 2020

References

Fruciano C. 2016. Measurement error in geometric morphometrics. Development Genes and Evolution 226:139-158.

Fruciano C., Schmidt, I., Ramirez Sanchez, M.M., Morek, W., Avila Valle, Z.A., Talijancic, I., Pecoraro, C., Schermann Legionnet, A. 2020. Tissue preservation can affect geometric morphometric analyses: a case study using fish body shape. Zoological Journal of the Linnean Society 188:148-162.

Ledoit O, Wolf M. 2004. A well-conditioned estimator for large-dimensional covariance matrices. Journal of Multivariate Analysis 88:365-411.

fruciano/GeometricMorphometricsMix documentation built on Jan. 31, 2024, 6:24 a.m.