The function calculates the predicted values and the difference of a range of cases with the confidence interval. It can be used for any glm, polr or multinom model.
predicts(model, values, position = NULL, sim.count = 1000, conf.int = 0.95,
sigma = NULL, set.seed = NULL, doPar = FALSE,
type = c("any", "simulation", "bootstrap"))The output is a data.frame with the predicted values and discrete changes.
the model-Object generated with glm(), glm.nb(), polr(), multinom(), mlogit() or tobit()
The values of cases as character in the order how they appear in the summary(model) Estimate. The values must be in the following way: "value1;value2;value3;...". Each one of the values can be one of the following:
"all": takes all unique values of that variable
"mean": takes the mean of that variable (can only be used when the variable is numeric)
"median": takes the median of that variable (assumes for factors that they are correctly ordered)
"mode": takes the mode of that variable
"Q4": takes the quartiles (0,0.25,0.5,0.75,1) of that variable (other number for other quantiles)
"min": takes the minimum of that variable
"max": takes the maximum of that variable
from-to,by: takes all values from "from" to "to" with the distance "by" (for example: "160-180,5" --> 160,165,170,175,180)
from-to: same as from-to,by with by=1 (for example: "2-8" --> 2,3,4,5,6,7,8); also works for factors and takes the given levels form their position
value1,value2,value3,...: takes the given values (for example: "160,180" --> 160,180); also works for factors and takes the given levels form their position
constant,|value1, value2, value3|, constant,...: for conditional logit models, all levels should be held constant besides one level which must be surounded by || and containing all values to simulate for. (for example: "5,|1,2,3,4,5|,5,5,5,5" for a choice with 7 levels (6 without the base category))
value1: takes the given value (for example: "5.34" --> 5.34); also works for factors and takes the given level form its position
log(from-to,by): takes the log of all values from "from" to "to" with the distance "by" (for example: "160-180,5" --> 160,165,170,175,180)
log(from-to): same as log(from-to,by) with by=1 (for example: "2-8" --> 2,3,4,5,6,7,8)
log(value1,value2,value3,...): takes the log of the given values (for example: "160,180" --> 160,180)
log(value1): takes the log of the given value (for example: "5.34" --> 5.34)
"F": takes all values of a factor/character
"F(1,4,7)": takes the first, fourth and seventh level of a factor/character
"F(2)": takes the second level of a factor/character
OPTIONAL which variable should be taken for the discrete change, the variable must have at least two values. default: only predicted probabilities
OPTIONAL numbers of simulations to be done by the function. default: 1000
OPTIONAL the confidence interval used by the function. default: 0.95
OPTIONAL the variance-covariance matrix, can be changed when having for example robust or clustered vcov. default: vcov(model)
OPTIONAL set a seed for the random number generator
OPTIONAL if the code should run parallel if more than 2 cores are detected
OPTIONAL choose between simulation and bootstrap, "any" chooses between those two according to the number of cases (bootstrap if n < 500)
Benjamin Schlegel, kontakt@benjaminschlegel.ch
The function makes a simulation for the all combination of cases and compares them to each other.
if (FALSE) {
model1 = glm(Sex ~ Height + Smoke + Pulse, data=MASS::survey, family=binomial(link=logit))
summary(model1)
# comparing person with hight 150 to 160, 160 to 170, 170 to 180, 180 to 190
# with all combination of(non-)smokers and a median of pulse
predicts(model1, "150-190,10;F;median", position = 1, doPar = FALSE)
}
Run the code above in your browser using DataLab