Solved – Hosmer-Lemeshow goodness of fit test, P=1

hosmer-lemeshow-testlogisticrregression

I'm trying to perform a Hosmer-Lemeshow goodness of fit test on a bunch of logistic regression models but for some reason the p-value that is returned is 1 on all of the models. Does anybody have a clue on why that is happening?

Here's a piece of my R-code:

model1 <- glm(c(variable1,variable2)~x+y+z, family="binomial")
model2 <- glm(c(variable1,variable2)~x+y,   family="binomial")
model3 <- glm(c(variable1,variable2)~x,     family="binomial")

hoslem.test(model1$y, model1$fitted)
hoslem.test(model2$y, model1$fitted)
hoslem.test(model2$y, model1$fitted)

and the output of the tests are:

X-squared = 0.35263, df = 8, p-value = 1
X-squared = 0.035221, df = 8, p-value = 1
X-squared = 0.29097, df = 8, p-value = 1

Best Answer

Does anybody have a clue on why that is happening?

I think the test is working as it should. The p values are just being rounded up to 1. E.g.:

> pchisq(0.35263, 8, lower.tail = F)
[1] 0.999965

Whether this test meets your needs is another matter.

Related Question