glm {safeBinaryRegression} | R Documentation |
This function overloads the glm
function so that a check for the existence of the maximum likelihood estimate is computed before fitting a ‘glm’ with a binary response.
glm(formula, family = gaussian, data, weights, subset, na.action, start = NULL, etastart, mustart, offset, control = glm.control(...), model = TRUE, method = "glm.fit", x = FALSE, y = TRUE, contrasts = NULL, ..., separation = c("find", "test"))
The arguments are identical to the arguments of the glm
function provided in the ‘stats’ package with the exception of
separation |
either “find” or “test”. Both options prevent the model from being fit to binary data when the maximum likelihood estimate does not exist. Additionally, when separation = "find" , the terms separating the sample points are identified when the maximum likelihood estimate is found not to exist. |
formula |
see glm |
family |
see glm |
data |
see glm |
weights |
see glm |
subset |
see glm |
na.action |
see glm |
start |
see glm |
etastart |
see glm |
mustart |
see glm |
offset |
see glm |
control |
see glm |
model |
see glm |
method |
see glm |
x |
see glm |
y |
see glm |
contrasts |
see glm |
... |
see glm |
This function checks for the existence of the maximum likelihood estimate before the ‘glm’ function is used to fit binary regression models by solving the linear program proposed in Konis (2007).
See the return value for the glm
function.
Kjell Konis kjell.konis@epfl.ch
Kjell Konis (2007). Linear programming algorithms for detecting separated data in binary logistic regression models. DPhil, University of Oxford http://ora.ouls.ox.ac.uk/objects/uuid:8f9ee0d0-d78e-4101-9ab4-f9cbceed2a2a
glm
.
## A set of 4 completely separated sample points ## x <- c(-2, -1, 1, 2) y <- c(0, 0, 1, 1) ## Not run: glm(y ~ x, family = binomial) ## A set of 4 quasicompletely separated sample points ## x <- c(-2, 0, 0, 2) y <- c(0, 0, 1, 1) ## Not run: glm(y ~ x, family = binomial)