This document reports convergence diagnostics and reusable posterior-check code for the Bayesian LR/RL arrangement models reported in Appendix A.
| model_id | study | outcome | family | formula |
|---|---|---|---|---|
| sentence_picture_match_accuracy | Garcia et al. (2023, 2025) | Sentence-picture match accuracy | bernoulli | accuracy ~ arrangement_bias * (language_group) + (1 | experiment) + (1 + arrangement_bias || participant_id) + (1 + arrangement_bias || item) |
| picture_match_response_time | Garcia et al. (2023, 2025) | Picture-match response time | shifted_lognormal | match_rt_ms ~ arrangement_bias * (language_group) + (1 | experiment) + (1 + arrangement_bias || participant_id) + (1 + arrangement_bias || item); ndt ~ 1 + (1 || participant_id) |
| word_order_priming_patient_initial | Garcia et al. (2023, 2025) | Patient-initial production | bernoulli | patient_initial ~ arrangement_bias * (language_group) + (1 | experiment) + (1 + arrangement_bias || participant_id) + (1 + arrangement_bias || item) |
| garcia_2020_sentence_picture_choice_accuracy | Garcia et al. (2020) | Sentence-picture choice accuracy | bernoulli | accuracy ~ arrangement_bias * (age_group) + (1 + arrangement_bias || participant) + (1 | item) |
| garcia_2021_who_question_accuracy | Garcia et al. (2021) | Who-question accuracy | bernoulli | accuracy ~ arrangement_bias * (age_group) + (1 + arrangement_bias || participant) + (1 + arrangement_bias || item) |
| garcia_kidd_2020_patient_initial | Garcia and Kidd (2020) | Patient-initial production | bernoulli | produced_pi ~ target_bias * (age_group + prime_action_direction) + (1 + target_bias * prime_action_direction || participant) + (1 + target_bias * prime_action_direction || item) |
| garcia_2018_patient_initial | Garcia et al. (2018) | Patient-initial production | bernoulli | patient_initial ~ arrangement_bias * (age_group) + (1 + arrangement_bias || participant) + (1 + arrangement_bias || item) |
| garcia_2019_picture_verification_accuracy | Garcia et al. (2019) | Picture-verification accuracy | bernoulli | accuracy ~ arrangement_bias * (age_group) + (1 + arrangement_bias || participant) + (1 + arrangement_bias || item) |
| garcia_2020_first_target_fixation | Garcia et al. (2020) | First image fixation to target | bernoulli | first_target_fixation ~ arrangement_bias * age_group + (1 + arrangement_bias || participant_id) + (1 | item) |
| garcia_2020_target_looking_duration_lognormal | Garcia et al. (2020) | Target looking duration | lognormal | target_look_duration_ms ~ arrangement_bias * age_group + (1 + arrangement_bias || participant_id) + (1 | item) |
| garcia_2021_first_fixation_agent | Garcia et al. (2021) | First entity fixation to agent | bernoulli | first_look_agent ~ arrangement_bias * age_group + (1 + arrangement_bias || participant_id) + (1 + arrangement_bias || item) |
| garcia_2021_agent_looking_duration_lognormal | Garcia et al. (2021) | Agent looking duration | lognormal | look_duration_ms ~ arrangement_bias * age_group + (1 + arrangement_bias || participant_id) + (1 + arrangement_bias || item) |
The primary convergence checks are the number of divergent transitions, maximum \(\hat{R}\), and the minimum bulk effective-sample-size ratio across parameters. All but one fitted model had no divergent transitions; the picture-match response-time model had four divergent transitions. The largest \(\hat{R}\) was 1.010.
| model_id | study | outcome | n_divergent | max_rhat | min_ess_bulk | n_draws |
|---|---|---|---|---|---|---|
| sentence_picture_match_accuracy | Garcia et al. (2023, 2025) | Sentence-picture match accuracy | 0 | 1.004 | 0.190 | 8000 |
| picture_match_response_time | Garcia et al. (2023, 2025) | Picture-match response time | 4 | 1.006 | 0.156 | 8000 |
| word_order_priming_patient_initial | Garcia et al. (2023, 2025) | Patient-initial production | 0 | 1.003 | 0.236 | 8000 |
| garcia_2020_sentence_picture_choice_accuracy | Garcia et al. (2020) | Sentence-picture choice accuracy | 0 | 1.002 | 0.265 | 8000 |
| garcia_2021_who_question_accuracy | Garcia et al. (2021) | Who-question accuracy | 0 | 1.003 | 0.259 | 8000 |
| garcia_kidd_2020_patient_initial | Garcia and Kidd (2020) | Patient-initial production | 0 | 1.003 | 0.296 | 8000 |
| garcia_2018_patient_initial | Garcia et al. (2018) | Patient-initial production | 0 | 1.003 | 0.284 | 8000 |
| garcia_2019_picture_verification_accuracy | Garcia et al. (2019) | Picture-verification accuracy | 0 | 1.004 | 0.159 | 8000 |
| garcia_2020_first_target_fixation | Garcia et al. (2020) | First image fixation to target | 0 | 1.002 | 0.272 | 8000 |
| garcia_2020_target_looking_duration_lognormal | Garcia et al. (2020) | Target looking duration | 0 | 1.003 | 0.155 | 8000 |
| garcia_2021_first_fixation_agent | Garcia et al. (2021) | First entity fixation to agent | 0 | 1.005 | 0.026 | 16000 |
| garcia_2021_agent_looking_duration_lognormal | Garcia et al. (2021) | Agent looking duration | 0 | 1.010 | 0.034 | 8000 |
The manuscript Bayes factors depend on the directional prior used to represent the LR-bias hypothesis. As a prior-sensitivity check, the table recalculates Bayes factors for effects with reported Bayes factors of at least 3 under half-width, reported-width, and double-width versions of the directional prior. This check shows how strongly the Bayes-factor interpretation depends on the chosen prior scale; the manuscript does not interpret smaller Bayes factors.
| Study | Outcome | Group | Direction | Half BF | Reported BF | Double BF |
|---|---|---|---|---|---|---|
| Garcia et al. (2021) | Agent looking duration | 5 years | against LR bias | 22.3 | 4.7 | 1.9 |
| Garcia et al. (2021) | Agent looking duration | 7 and 9 years | against LR bias | 45.9 | 9.8 | 4.0 |
| Garcia et al. (2021) | Agent looking duration | adults | for LR bias | 276.3 | 1299.1 | 3156.1 |
| Garcia et al. (2021) | First entity fixation to agent | 5 years | for LR bias | 1.5 | 7.1 | 17.2 |
| Garcia et al. (2021) | First entity fixation to agent | 7 and 9 years | for LR bias | 14.9 | 70.1 | 170.2 |
| Garcia et al. (2021) | First entity fixation to agent | adults | for LR bias | 21.2 | 99.8 | 242.4 |
| Garcia and Kidd (2020) | Patient-initial production | 3 years; prime LR | against LR bias | 6.3 | 3.2 | 1.6 |
| Garcia and Kidd (2020) | Patient-initial production | 3 years; prime RL | against LR bias | 18.9 | 9.4 | 4.7 |
| Garcia and Kidd (2020) | Patient-initial production | 5 years; prime LR | against LR bias | 7.9 | 3.9 | 2.0 |
| Garcia and Kidd (2020) | Patient-initial production | 5 years; prime RL | against LR bias | 24.7 | 12.3 | 6.2 |
| Garcia and Kidd (2020) | Patient-initial production | 7 and 9 years; prime RL | against LR bias | 7.0 | 3.5 | 1.7 |
| Garcia and Kidd (2020) | Patient-initial production | adults; prime RL | against LR bias | 9.7 | 4.8 | 2.4 |
| Garcia et al. (2023) | Patient-initial production | Tagalog | against LR bias | 62.9 | 31.5 | 15.7 |
| Garcia et al. (2025) | Patient-initial production | English L2 | against LR bias | 67.3 | 33.7 | 16.8 |
| Garcia et al. (2023, 2025) | Picture-match response time | English L2 | against LR bias | 12.2 | 6.1 | 3.1 |
| Garcia et al. (2019) | Sentence-picture verification accuracy | 7 and 9 years | against LR bias | 6.9 | 3.5 | 1.7 |
| Garcia et al. (2023, 2025) | Sentence-picture verification accuracy | English L2 | against LR bias | 6.2 | 3.1 | 1.5 |
| Garcia et al. (2023, 2025) | Sentence-picture verification accuracy | Indonesian | against LR bias | 20.8 | 10.4 | 5.2 |
| Garcia et al. (2023, 2025) | Sentence-picture verification accuracy | Tagalog | against LR bias | 38.6 | 19.3 | 9.6 |
| Garcia et al. (2020) | Target looking duration | 7 and 9 years | against LR bias | 114.3 | 24.3 | 10.0 |
| Garcia et al. (2020) | Target looking duration | adults | against LR bias | 35.1 | 7.5 | 3.1 |
The sensitivity check was generally stable in direction. The largest Bayes factors remained large across prior widths, whereas some borderline Bayes factors near 3 were more sensitive to the prior scale.
The fixation-duration outcomes were also fitted with Gamma likelihoods and log links. We retained the lognormal likelihood because approximate leave-one-out cross-validation favoured it for both duration outcomes.
| Outcome | Family | ELPD difference | SE difference | ELPD LOO | LOOIC |
|---|---|---|---|---|---|
| Target looking duration | lognormal | 0.0 | 0.0 | -4537.1 | 9074.2 |
| Target looking duration | gamma | -41.8 | 5.7 | -4578.9 | 9157.9 |
| Agent looking duration | lognormal | 0.0 | 0.0 | -27394.1 | 54788.1 |
| Agent looking duration | gamma | -74.4 | 11.6 | -27468.4 | 54936.8 |
The following plots overlay the observed data with data simulated from the posterior predictive distribution for every fitted model. Binary models are shown as observed-versus-posterior-predicted bar overlays; response-time and fixation-duration models are shown as observed-versus-posterior-predicted density overlays with log-scaled x-axes.
The following plots show Markov-chain traces for the population-level
parameters. For models with many population-level parameters, the plot
is limited to the first eight b_ parameters to keep the
supplementary document readable.