Generalised Additive Mixed Model: Summary table differences between mgcv and brms

Interfaces brms
1

If you run an identical Generalised Additive Mixed Model with the mgcv vs. brms package in R, the summary tables for the factor smooth terms (e.g., te(..., bs='fs', ...) are different. For example, the generic model could be:

gam1 <-gam(DV ~
    ExpTrialOrder +
    FactorA +
    s(CovariateM) +
    s(ExpTrialOrder, Participant, bs='fs', m=1),
    ...)

brms_gam1 <- brm(DV ~
    ExpTrialOrder +
    FactorA +
    s(CovariateM) +
    s(ExpTrialOrder, Participant, bs='fs', m=1),
    ...)

For the traditional GAMM, the s(ExpTrialOrder, Participant, bs='fs', m=1) part of the table could look like this:

Approximate significance of smooth terms:
                                 edf  Ref.df     F p-value
s(ExpTrialOrder,Participant) 134.282 368.000 59.88  <2e-16

while the output from brms could look like this:

Smooth Terms:
                                 Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sds(sExpTrialOrderParticipant_1)     5.91      0.61     4.77     7.13 1.00     2333     4650
sds(sExpTrialOrderParticipant_2)   139.72     15.71   112.74   174.20 1.00      634     1720

How should we read the summary table from brms? Are the two components (i.e., _1 and _2) smooths for the two components of the s(); that is, Estimate == edf, for ExpTrialOrder and Participant, respectively?

  • Operating System: macOS
  • brms Version: 2.17.0

Thanks!

2 Likes