Logistic regression

Go back

If $Y \in [0,1]$, then you may use the Logistic regression. The left of the formula is changing a bit, while the right-side will not.

@ \log(\frac{P(X=n)}{1-P(X=n)}) = a + b X + c Z @

You may notice that there are no residuals anymore.

Simple logistic regression

You will use the function glm.

# regression Y = a + b * X
# Y = quantitative variable
# X = quantitative/qualitative variable
model <- glm(Y ~ X, data=ech, family="binomial")
# or
model <- lm(ech$Y ~ ech$X, family="binomial")
# check the result table

Multiple logistic regression

# regression Y = a + b * X + c Z
model <- glm(Y ~ X + Z, data=ech, family="binomial")


You will need to evaluate exp(b) if you want to get b' value. Then b' is the same as the b you know in linear regression.

If you want all of them, then use


If you are using drop1, then the syntax changed a bit to

drop1(model, .~., test="Chisq")


If we are considering $|X|$ equals to

  • 1: if the variable is a quantitative or binary variable
  • else the number of levels minus 1 length(levels(qual))-1

Then the condition for $Z = a + bX + cT$

  • $(|X|+|Z|+|T|) * 5 \le n$
  • or better $(|X|+|Z|+|T|) * 10 \le n$w