Regression with circular dependent variable and Euclidean or categorical independent variables.
spml.reg(y, x, rads = TRUE, xnew = NULL, seb = FALSE, tol = 1e-07)
circpurka.reg(y, x, rads = TRUE, xnew = NULL)
cipc.reg(y, x, rads = TRUE, xnew = NULL, tol = 1e-06)
gcpc.reg(y, x, rads = TRUE, reps = 20, xnew = NULL)A list including:
The runtime of the procedure.
The number of iterations required until convergence of the Newton-Raphson algorithm.
The regression coefficients.
The standard errors of the coefficients.
The value of the maximised log-likelihood.
The fitted values expressed in radians if the obsereved data are in radians and in degrees otherwise. If xnew is not NULL, i.e. if you have new x values, then the predicted values of y will be returned.
The dependent variable, a numerical vector, it can be in radians or degrees.
The independent variable(s). Can be Euclidean or categorical (factor variables).
If the dependent variable is expressed in rads, this should be TRUE and FALSE otherwise.
How many starting values shall the algortihm use? By default it uses 20 different starting values.
The new values of some independent variable(s) whose circular values you want to predict. Can be Euclidean or categorical. If they are categorical, the user must provide them as dummy variables. It does not accept factor variables. If you have no new x values, leave it NULL (default).
a boolean variable. If TRUE, the standard error of the coefficients will be be returned. Set to FALSE in case of simulation studies or in other cases such as a forward regression setting for example. In these cases, it can save some time.
The tolerance value to terminate the Newton-Raphson algorithm.
Michail Tsagris.
R implementation and documentation: Michail Tsagris mtsagris@uoc.gr.
For the spml.reg(), the Newton-Raphson algorithm is fitted in this regression as described in Presnell et al. (1998). For the cipc.reg(), the Newton-Raphson algorithm is fitted in this regression as described in Tsagris and Alenazy (2023). Note that the cipc.reg() is the same as the wrapped Cauchy regression. For the circpurka.reg() the optim() function is employed. For the gcpc.reg() the optim() and the optimise() functions are being used.
Presnell Brett, Morrison Scott P. and Littell Ramon C. (1998). Projected multivariate linear models for directional data. Journal of the American Statistical Association, 93(443): 1068-1077.
Purkayastha S. (1991). A Rotationally Symmetric Directional Distribution: Obtained through Maximum Likelihood Characterization. The Indian Journal of Statistics, Series A, 53(1): 70-83
Tsagris M. and Alzeley O. (2023). Circular and spherical projected Cauchy distributions: A Novel Framework for Circular and Directional Data Modeling. https://arxiv.org/pdf/2302.02468.pdf
circlin.cor, circ.cor1, circ.cor2, spher.cor, spher.reg
x <- rnorm(100)
z <- cbind(3 + 2 * x, 1 -3 * x)
y <- cbind( rnorm(100,z[ ,1], 1), rnorm(100, z[ ,2], 1) )
y <- y / sqrt( rowSums(y^2) )
y <- ( atan( y[, 2] / y[, 1] ) + pi * I(y[, 1] < 0) ) %% (2 * pi)
a <- spml.reg(y, x, rads = TRUE, xnew = x)
b <- cipc.reg(y, x, rads = TRUE, xnew = x)
Run the code above in your browser using DataLab