From 5e7bdd63b2a1e7e1be9ade542d6381080ef21c7e Mon Sep 17 00:00:00 2001 From: RobFryer Date: Sun, 5 Jul 2026 18:36:11 +0100 Subject: [PATCH 1/2] update change log --- CHANGES.md | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/CHANGES.md b/CHANGES.md index a4b9a98..19e8b27 100644 --- a/CHANGES.md +++ b/CHANGES.md @@ -1,5 +1,22 @@ ## Change log +### Version 1.0.6 + +#### Major bug fix - p_overall_trend + +For smooth models, the degrees of freedom used in the likelihood ratio test of +an overall temporal trend are now correct. Previously, they were too small (by +one) resulting in values of `p_overall_trend` that were too significant. +Fortunately, this had no effect on model selection as the chosen model is +based on AICc or AIC (depending on the distribution of the response) and the +p-values are calculated after the model is chosen. + +The statistical interpretation of smooth models in `report_assessment` is now +more nuanced. Smooth models chosen by AIC or AICc are not necessarily significant +at the conventional 5% level. The degree of significance (`p_overall_trend`) is +now characterised as weak, moderate or strong. + + ### Version 1.0.5 This release is used to run the OSPAR 2026 assessment. From 07f3e2db2eeb96fdce5c2b604814f026cd7b6c58 Mon Sep 17 00:00:00 2001 From: RobFryer Date: Mon, 6 Jul 2026 10:23:00 +0100 Subject: [PATCH 2/2] Further explanation about p-values for smooth models --- CHANGES.md | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-) diff --git a/CHANGES.md b/CHANGES.md index 19e8b27..2bdb274 100644 --- a/CHANGES.md +++ b/CHANGES.md @@ -13,8 +13,22 @@ p-values are calculated after the model is chosen. The statistical interpretation of smooth models in `report_assessment` is now more nuanced. Smooth models chosen by AIC or AICc are not necessarily significant -at the conventional 5% level. The degree of significance (`p_overall_trend`) is -now characterised as weak, moderate or strong. +at the conventional 5% level. The significance of the final model is given by +`p_overall_trend` which is based on a likelihood ratio test that compares the +smooth model and the mean model (and essentially tests for any evidence of a +change in concentrations over time). This degree of significance is now +characterised as weak (p >= 0.05), moderate (0.05 < p <= 0.01) and strong (p < +0.01). In theory, a smooth model chosen by AICc could have a +`p_overall_trend` as high as 0.135, but this can only happen when there are many +years of data and the improvement in AICc between the smooth model and the mean +model is marginal. + +Note that, for smooth models, the significance of the overall trend can be split +into the significance of the nonlinear component (`p_nonlinear_trend` based on a +likelihood ratio test that compares the smooth model with the linear model) and +the linear component (`p_linear_trend` based on a likelihood ratio test that +compares the linear model with the mean model). Both these p-values were +calculated correctly in previous releases. ### Version 1.0.5