How to extract fitted values of GAM {mgcv} for each variable in R? -

i'm searching method add predicted (real, not standardized) values of every single variable in model

> model<-gam(ln_brutto~s(agecont,by=sex)+factor(sex)+te(month,age)+s(month,by=sex), data=bears)

this summary of model:

> summary(m13)  family: gaussian  link function: identity   formula: ln_brutto ~ s(agecont, = sex) + factor(sex) + te(month, age) +      s(month, = sex)  parametric coefficients:              estimate std. error t value pr(>|t|)     (intercept)   4.32057    0.01071  403.34   <2e-16 *** factor(sex)m  0.27708    0.01376   20.14   <2e-16 *** --- signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1  approximate significance of smooth terms:                     edf  ref.df      f  p-value     s(agecont):sexf  8.1611  8.7526 20.170  < 2e-16 *** s(agecont):sexm  6.6695  7.5523 32.689  < 2e-16 *** te(month,age)   10.3651 12.7201  6.784 2.19e-12 *** s(month):sexf    0.9701  0.9701  0.641    0.430     s(month):sexm    1.3750  1.6855  0.193    0.787     --- signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1  rank: 60/62 r-sq.(adj) =  0.781   deviance explained = 78.7% gcv = 0.048221  scale est. = 0.046918  n = 1093 

predicted values provided code:

> predict<-predict(m13, type = "terms") 

and result looks this:

    factor(sex) s(agecont):sexf s(agecont):sexm te(month,age)   s(month):sexf   s(month):sexm 1   0.2770806   0.000000000     0.111763696     -0.077845764    0.000000000     0.0007840912 2   0.2770806   0.000000000     0.240016156     -0.049143798    0.000000000     0.0007840912 3   0.2770806   0.000000000     0.034328752     0.046524454     0.000000000     -0.0058871897 4   0.0000000   -0.786533918    0.000000000     -0.067942427    0.021990192     0.0000000000 5   0.0000000   0.074434715     0.000000000     0.046524454     0.021990192     0.0000000000 6   0.0000000   0.161121563     0.000000000     0.089599601     0.021990192     0.0000000000 7   0.0000000   0.074434715     0.000000000     0.046524454     0.021990192     0.0000000000 8   0.2770806   0.000000000     -0.298597370    -0.007877328    0.000000000     -0.0058871897 ... 

but guess these standardized predicted values , not real values (the real ones should have no negative values!?).

so know have modify in code, real values? idea? thank you!

not quite sure if follow correctly, predict(model, type = "terms") might solution you're looking for.

update

i don't think these standardised. possibly of coefficients negative.

consider example file ?mgcv:::predict.gam:

library(mgcv) n<-200 sig <- 2 dat <- gamsim(1,n=n,scale=sig)  b<-gam(y~s(x0)+s(i(x1^2))+s(x2)+offset(x3),data=dat) 

the results below illustrate these in fact contributions being used each predictor calculate fitted values (by calculating sum of each of these contributions , adding intercept , offset).

> head(predict(b))         1         2         3         4         5         6   9.263322  2.822200  7.137201  4.902631 14.558401 11.889092  > head(rowsums(predict(b, type = "terms")) + attr(predict(b, type = "terms"), "constant") + dat$x3)         1         2         3         4         5         6   9.263322  2.822200  7.137201  4.902631 14.558401 11.889092