0
$\begingroup$

I'm having trouble interpreting the diagnostic plots obtained from a gam modeled with family="scat".

The data seem to adjust reasonably well to the 45 degree line, but the red reference line is perpendicular to the x axis. Does this mean the results from the model are not reliable? I paste the rest of the diagnostic plots, which look ok to me.

Diagnostic plots

The code I used to fit the model is

model<- gam(pHT ~ s(T) + s(O2) + s(S) + s(pratio)+ s(T, by=station) + s(O2, by=station) + s(S, by=station) + s(pratio, by=station)+ month, data=datosresu14, method="REML", family="scat")

All variables are continuous except for station and month, which are factors.

ALso k.check() looks ok

`> k.check(m15GIg)
                          k'          edf   k-index p-value
    s(T)                   9 1.000001e+00 1.0394585  0.4950
    s(O2)                  9 1.398550e+00 1.0388868  0.4625
    s(S)                   9 1.000000e+00 0.9649325  0.3500
    s(pratio)              9 1.000005e+00 1.3053333  0.8975
    s(T):station1 km       9 1.000000e+00 1.0394585  0.4825
    s(T):station4 km       9 3.338502e-07 1.0394585  0.5075
    s(O2):station1 km      9 1.000000e+00 1.0388868  0.4900
    s(O2):station4 km      9 1.000003e+00 1.0388868  0.5375
    s(S):station1 km       9 4.601437e+00 0.9649325  0.3400
    s(S):station4 km       9 2.627084e-06 0.9649325  0.3550
    s(pratio):station1 km  9 2.971070e-07 1.3053333  0.8775
    s(pratio):station4 km  9 1.000001e+00 1.3053333  0.8800`

Thank you,

Mar.

Improve this question
$\endgroup$
13
  • 1
    $\begingroup$ Welcome to CV. It would be better to paste the actual diagram. Some people may not want to click on links like this one. $\endgroup$ Commented Feb 5 at 13:21
  • 2
    $\begingroup$ This looks both artificial and overfit, suggesting more details about your data would be useful. $\endgroup$ Commented Feb 5 at 13:41
  • $\begingroup$ Thank you Peter, I'm new to the site and didn't know how to do it! $\endgroup$ Commented Feb 5 at 13:56
  • $\begingroup$ Thanks @PeterFlom and whuber, I just edited the post so it has more information. $\endgroup$ Commented Feb 5 at 14:05
  • 5
    $\begingroup$ The model is doomed from the outset: it's hopeless to get reliable information from an analysis that fits less than 25 observations using nine explanatory variables, most of which are splined! $\endgroup$ Commented Feb 5 at 15:51

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.