Residual sum of square

#1
As the whole number of the Y entry gets bigger, for instance if I use 1000 (g) rather than 1(Kg), the RSS gets bigger.
So even if the data is same ( one is written in g and the other one is written in kg), RSS is different.
So in that case what does the RSS mean (when it is ideal case if RSS is close to 0)
In other words, if I enter the y data in kg, RSS is smaller so it looks better than entering the data in g even though they are same data?
Can any one explain how to interpret RSS in that case?
 

hlsmith

Not a robit
#2
Well, you can mean center or standardize your values. I believe estimates would be comparable if using standardized version of the two different formatting. Otherwise you just need to accept, that if I have heights of people in cm and m, a standard deviation in both would be different just based on units. That doesn't mean one model is better than the other when looking at RSS, since they are based on not the same formatted data. If you were comparing two models with the same data formatting and one had better fit, you may select it. But its like you have two different datasets when you reformat, so you should be comparing RSS between them.

Overall, this is really a moot issue.
 

ondansetron

TS Contributor
#3
This is, as @hlsmith said, a moot point.

It's illogical to compare models with different dependent variables, generally speaking. Weight in grams is a different DV than weight in kg (even though it is a transformation), and only RSS should be compared from the same DV, but even this is missing the degrees of freedom/model complexity to achieve the RSS. This is why the model standard deviation and R-squared adjusted are useful for comparing between two models of different predictor sets for the same DV.