2
$\begingroup$

I am following Wieling (2018) to fit a gam model. After checking for the autocorrelation of residuals, I added the autocorrelation corrections suggested by Wieling. Although the autocorrelation is corrected for, the QQplot obtained with the gam.check() function shows a marked difference between the red line and the residuals. Can anyone help me interpret this?

I copy below the two model fits, the autocorrelation plots and the QQ plots of the two models.

MODEL A

gam_dfw1_f0st_phon_alt_a <- gam(
  f0_st ~ 
    s(norm_index, by = PhonInt, k = 15) + 
    PhonInt +
    s(itemCode, bs = "re") + 
    s(speaker, bs = "re"),
  data = dfw1_alt_3syl,
  method = "ML"
)

gam.check(gam_dfw1_f0st_phon_alt_a)

acf_gam_dfw1_f0st_phon_alt_a <- 
    acf_resid(gam_dfw1_f0st_phon_alt_a)

MODEL B (with correction for autocorrelation)

gam_dfw1_f0st_phon_alt_b <- bam(
  f0_st ~ 
    s(norm_index, by = PhonInt, k = 15) + 
    PhonInt +
    s(itemCode, bs = "re") + 
    s(speaker, bs = "re"),
  rho = acf_gam_dfw1_f0st_phon_alt_a[2],
  AR.start = dfw1_alt_3syl$start.event,
  data = dfw1_alt_3syl,
  method = "ML"
)

gam.check(gam_dfw1_f0st_phon_alt_b)

acf_gam_dfw1_f0st_phon_alt_b <- 
    acf_resid(gam_dfw1_f0st_phon_alt_b)

Autocorrelation model A

Autocorrelation model B

model A

model B

Improve this question
$\endgroup$
0

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.