![]() ![]() If you have n observations, a saturated model has n parameters, it will always fit the data perfectly and leave 0 residuals. Other times you get non-integer degrees of freedom due to some generalisation people have made. The caveat is that if your two parameters are the intercept and 2*intercept then you actually still have n-1 degrees of freedom in the residual because the parameter statistics aren't linearly independent. When you then use that mean to calculate the residual, the random vector that is used in this calculation has 2 constraints imposed on it so you have n-2 degrees of freedom in the residual. ![]() In linear regression you might estimate 2 parameters, which you then use to estimate the mean. This is reduced for each parameter you estimate because each linearly independent function you fix for your set of observations reduces the number of free-varying RV by one.įor example you have a vector of 4 normally distributed RV, they are all free to vary and you have 4 degrees of freedom, but once you estimate and fix the mean, then you will have 3 values left that can freely vary and a 4th that can be set as a function of the first 3 and the mean. My understanding of degrees of freedom (and it may not be entirely correct) is the number of random variables in your random vector that are free to vary. R-bloggers - blog aggregator with statistics articles generally done with R software.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |