Including day of week in a logit model Let's say that I am putting together a logistic regression model where I am predicting 
something (y) based on the day of the week. However, the model needs to account for each single day.
Therefore, instead of:
y = B0 + B1*(day)

Where day is a categorical variable with 7 levels.
It would be:
y = B0 + B1*(monday) + B2*(tuesday) + B3*(wednesday) + ... + B7*(sunday)

I'm basically thinking that each day needs a separate coefficient because each 
has a different affect on y. However, I think each will need to be a dummy variable 
so that for monday, 1 is for monday, and 0 for not monday, and so forth. 
I'm just curious if there is a statistical logic to doing it the second way 
with separate days? What's the best way to do this?
 A: The model with day as a categorical variable with seven levels does account for each single day; you don't need to do it "by hand", so to speak.
For example:
library(MASS)

# Construct sample data: 700 observations, 100 on each of 7 days of week
Day <- factor(rep(c("Monday","Tuesday","Wednesday","Thursday","Friday","Saturday","Sunday"), 100),
              levels=c("Monday","Tuesday","Wednesday","Thursday","Friday","Saturday","Sunday"),
              ordered=TRUE)
Day.effect <- rep(rnorm(7), 100)
y <- rbinom(700, 1, 1/(1+exp(-Day.effect)))

# Estimate logit model without intercept (captures each day's effect)
foo <- summary(glm(y~Day-1, family=binomial))

# compare actuals to estimates
coefs <- foo$coefficients
coefs <- cbind(Day.effect, coefs)
colnames(coefs)[1] <- "Actual"
options(digits=3)

> coefs
             Actual Estimate Std. Error z value Pr(>|z|)
DayMonday     0.520    0.490      0.206   2.376 1.75e-02
DayTuesday   -0.230   -0.323      0.203  -1.593 1.11e-01
DayWednesday -0.247   -0.447      0.205  -2.182 2.91e-02
DayThursday  -1.156   -1.266      0.241  -5.243 1.58e-07
DayFriday     0.282    0.160      0.201   0.799 4.24e-01
DaySaturday  -0.383   -0.405      0.204  -1.986 4.70e-02
DaySunday    -0.357   -0.447      0.205  -2.182 2.91e-02

This would appear to do just what you want.
