1

Is is possible to decompose fitted linear model? What I mean by that:

I have parameters of fitted linear model as following:

y=2.3a-1.23b+1.65c+1.76d

Now I have (known) all variables a,b,c,d

How to obtain the net effect (contribution) of each individual variable?

Let's say that y (after pre-multiplying with variables a,b,c,d I get y=34,324. I would like to get the contribution (effect) of variable i.e. d so that I can plot y~d.

Maximilian
  • 245
  • 1
  • 3
  • 14

1 Answers1

4

Sounds like you are looking for either an added-variable plot or a component-plus-residual plot. If you're using R, the car and effects packages offer several ways of displaying these.

Russ Lenth
  • 15,161
  • 20
  • 53
  • Never used these component plus residual plots before. These are great! – shadowtalker Oct 26 '14 at 06:32
  • Thanks, yes I'm using also R!I would need the same for survival (cox) model, which is non-linear, how to approach it in this case? Thanks. – Maximilian Oct 26 '14 at 08:31
  • The answer is here: http://grokbase.com/t/r/r-help/084bsmm9m8/r-added-variable-plot-for-cox-regression – Maximilian Oct 26 '14 at 08:52
  • 7
    I'm really surprised this answer got so many upvotes. You should at least mention the relevant functions and show some examples. `car` package contains almost 100 different functions... – David Arenburg Oct 26 '14 at 09:08
  • See [What does an Added Variable Plot (Partial Regression Plot) explain in a multiple regression?](http://stats.stackexchange.com/q/125561/17230). – Scortchi - Reinstate Monica Oct 19 '15 at 11:33