Table 2.
Fit of different GLMM models in labeling experiment
| Model | df | AIC | Log-likelihood | χ2 tests | P (>χ2) | P (bootstrap) |
| Model 1 Null (see Materials and Methods) | 5 | 316.5 | −153.3 | — | — | — |
| Model 2: Model 1 +Type | 6 | 318.1 | −153.1 | 0.407 | 0.524 | 0.532 |
| Model 3: Model 2 + Position | 7 | 319.8 | −152.9 | 0.265 | 0.607 | 0.612 |
| Model 4: Model 3 + Type x Position | 8 | 312.5 | −148.2 | 9.378 | 0.0022 | 0.0031 |
| Model 5: Model 4 + Norm + Norm × Position | 12 | 317.1 | −146.5 | 3.397 | 0.494 | 0.535 |
The χ2 tests relate to comparisons with the null model 1. The best-supported model is in boldface. The χ2 tests were between models and the next-simplest model in the list. The sample size was 96 trials for 48 individuals, using 12 syllable types, with three note types (short, intermediate, long), two positions (initial, final), and three syllable-type norms (short, intermediate, long). x represents an interaction term.