From f30c1f09497d1d5bb5c36440856c5ce787ac535d Mon Sep 17 00:00:00 2001 From: "Documenter.jl" Date: Mon, 16 Dec 2024 02:33:17 +0000 Subject: [PATCH] build based on 04d6a68 --- dev/.documenter-siteinfo.json | 2 +- dev/index.html | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/dev/.documenter-siteinfo.json b/dev/.documenter-siteinfo.json index 571dc74..c037474 100644 --- a/dev/.documenter-siteinfo.json +++ b/dev/.documenter-siteinfo.json @@ -1 +1 @@ -{"documenter":{"julia_version":"1.6.7","generation_timestamp":"2024-12-15T02:34:38","documenter_version":"1.8.0"}} \ No newline at end of file +{"documenter":{"julia_version":"1.6.7","generation_timestamp":"2024-12-16T02:33:11","documenter_version":"1.8.0"}} \ No newline at end of file diff --git a/dev/index.html b/dev/index.html index cae06e1..69e980f 100644 --- a/dev/index.html +++ b/dev/index.html @@ -570,4 +570,4 @@ end # submit job run(`qsub tmp.sh`) -end

Proportional Odds Assumption

The ordered multinomial model used in this package is a cumulative logit model where one effect size is estimated for each covariate. Using the logit link yields the commonly used proportional odds model which assumes the proportional odds assumption holds for your data/model. The proportional odds assumption assumes that the effect of a covariate $\boldsymbol{\beta}$ is constant across different groupings of the ordinal outcome i.e.

The SNP's effect (on the odds ratio) is the same for the odds ratio of mild vs {medium, severe} as it is for the odds ratio of {mild, medium} vs severe.

The consequences/neccesity of this assumption has been thoroughly discussed, as an example see this blog post. In simualtions, we did not find violation of this to increase type I error. As stated in the post, when the assumption is violated (meaning there are different effect sizes for different groupings), the estimated effect size is like a weighted average of the different effect sizes. Formal tests for proportional odds are likely to reject the assumption for large n (see p. 306, Categorical Data Analysis 3rd edition by Agresti). Violation of this assumption means the effect size $\boldsymbol{\beta}$ differs across outcome strata, but should not imply a significant finding is not significant.

+end

Proportional Odds Assumption

The ordered multinomial model used in this package is a cumulative logit model where one effect size is estimated for each covariate. Using the logit link yields the commonly used proportional odds model which assumes the proportional odds assumption holds for your data/model. The proportional odds assumption assumes that the effect of a covariate $\boldsymbol{\beta}$ is constant across different groupings of the ordinal outcome i.e.

The SNP's effect (on the odds ratio) is the same for the odds ratio of mild vs {medium, severe} as it is for the odds ratio of {mild, medium} vs severe.

The consequences/neccesity of this assumption has been thoroughly discussed, as an example see this blog post. In simualtions, we did not find violation of this to increase type I error. As stated in the post, when the assumption is violated (meaning there are different effect sizes for different groupings), the estimated effect size is like a weighted average of the different effect sizes. Formal tests for proportional odds are likely to reject the assumption for large n (see p. 306, Categorical Data Analysis 3rd edition by Agresti). Violation of this assumption means the effect size $\boldsymbol{\beta}$ differs across outcome strata, but should not imply a significant finding is not significant.