对数几率与逻辑回归模型的解释。
Log Odds and the Interpretation of Logit Models.
机构信息
Department of Health Management and Policy, Department of Economics, University of Michigan, Ann Arbor, MI.
National Bureau of Economic Research, Cambridge, MA.
出版信息
Health Serv Res. 2018 Apr;53(2):859-878. doi: 10.1111/1475-6773.12712. Epub 2017 May 30.
Abstract
OBJECTIVE
We discuss how to interpret coefficients from logit models, focusing on the importance of the standard deviation (σ) of the error term to that interpretation.
STUDY DESIGN
We show how odds ratios are computed, how they depend on the standard deviation (σ) of the error term, and their sensitivity to different model specifications. We also discuss alternatives to odds ratios.
PRINCIPAL FINDINGS
There is no single odds ratio; instead, any estimated odds ratio is conditional on the data and the model specification. Odds ratios should not be compared across different studies using different samples from different populations. Nor should they be compared across models with different sets of explanatory variables.
CONCLUSIONS
To communicate information regarding the effect of explanatory variables on binary {0,1} dependent variables, average marginal effects are generally preferable to odds ratios, unless the data are from a case-control study.
摘要
目的
我们讨论如何解释逻辑回归模型中的系数，重点关注误差项的标准差（σ）对解释的重要性。
研究设计
我们展示了如何计算优势比，它们如何依赖于误差项的标准差（σ），以及它们对不同模型规格的敏感性。我们还讨论了优势比的替代方法。
主要发现
不存在单一的优势比；相反，任何估计的优势比都是有条件的，取决于数据和模型规格。不应该在不同的研究中使用来自不同人群的不同样本来比较优势比。也不应该在具有不同解释变量集的模型之间进行比较。
结论
要传达有关解释变量对二元{0,1}因变量的影响的信息，平均边际效应通常优于优势比，除非数据来自病例对照研究。
相似文献
1
Log Odds and the Interpretation of Logit Models.对数几率与逻辑回归模型的解释。
Health Serv Res. 2018 Apr;53(2):859-878. doi: 10.1111/1475-6773.12712. Epub 2017 May 30.
2
Using marginal standardisation to estimate relative risk without dichotomising continuous outcomes.使用边缘标准化来估计连续结果无需二分类的相对风险。
BMC Med Res Methodol. 2019 Jul 29;19(1):165. doi: 10.1186/s12874-019-0778-9.
3
Bridging conditional and marginal inference for spatially referenced binary data.用于空间参考二元数据的桥接条件推断和边际推断
Biometrics. 2013 Jun;69(2):545-54. doi: 10.1111/biom.12027. Epub 2013 May 31.
4
[Down with odds ratios: risk ratios in cohort studies and randomised clinical trials].[摒弃比值比：队列研究和随机临床试验中的风险比]
Ned Tijdschr Geneeskd. 2012;156(28):A4775.
5
Creation of Interpretable Summary Measures in Displaying Results from Mixed-effects Logit Models.混合效应逻辑模型结果展示中可解释性汇总指标的创建。
J Biom Biostat. 2016 Jun;7(3). doi: 10.4172/2155-6180.1000304. Epub 2016 May 31.
6
Use of conditional and marginal odds-ratios for analysing familial aggregation of binary data.使用条件和边际优势比分析二元数据的家族聚集性。
Genet Epidemiol. 2000 Mar;18(3):193-202. doi: 10.1002/(SICI)1098-2272(200003)18:3<193::AID-GEPI1>3.0.CO;2-W.
7
Interaction terms in nonlinear models.非线性模型中的交互项。
Health Serv Res. 2012 Feb;47(1 Pt 1):255-74. doi: 10.1111/j.1475-6773.2011.01314.x. Epub 2011 Aug 30.
8
Distance-based logistic model for cross-classified categorical data.用于交叉分类分类数据的基于距离的逻辑模型。
Br J Math Stat Psychol. 2022 Nov;75(3):466-492. doi: 10.1111/bmsp.12264. Epub 2022 Jan 25.
9
Use of odds ratio or relative risk to measure a treatment effect in clinical trials with multiple correlated binary outcomes: data from the NINDS t-PA stroke trial.在具有多个相关二元结局的临床试验中使用比值比或相对危险度来衡量治疗效果：来自美国国立神经疾病与中风研究所组织型纤溶酶原激活剂中风试验的数据
Stat Med. 2001 Jul 15;20(13):1891-901. doi: 10.1002/sim.841.
10
Marginal modeling of binary cross-over data.二元交叉数据的边际建模
Biometrics. 1993 Dec;49(4):997-1009.
引用本文的文献
1
miRNA-494 in Lymphocytes: A Promising Biomarker for Acute Ischemic Stroke.淋巴细胞中的miRNA - 494：急性缺血性中风的一个有前景的生物标志物。
Rev Neurol. 2025 Aug 29;80(7):37809. doi: 10.31083/RN37809.
2
Fine scale structural information substantially improves multivariate regression model for mRNA in-vial degradation prediction.精细尺度结构信息显著改进了用于mRNA瓶内降解预测的多元回归模型。
bioRxiv. 2025 Aug 23:2025.08.15.670605. doi: 10.1101/2025.08.15.670605.
3
Geographic Disparities by Rural-Urban Status and Drive Time to Care in Tobacco Treatment for COPD.慢性阻塞性肺疾病（COPD）烟草治疗中城乡状况及就医驾车时间的地域差异
JAMA Netw Open. 2025 Aug 1;8(8):e2528898. doi: 10.1001/jamanetworkopen.2025.28898.
4
Association of hearing and vision impairment with cognitive impairment in nursing home residents in Switzerland.瑞士养老院居民听力和视力障碍与认知障碍的关联
Eur J Ageing. 2025 Aug 7;22(1):39. doi: 10.1007/s10433-025-00880-y.
5
Evaluating the effectiveness of sociotherapy following psychiatric hospitalization: a target trial emulation protocol using German statutory health insurance data.评估精神科住院后社会心理治疗的有效性：一项使用德国法定医疗保险数据的目标试验模拟方案。
BMC Health Serv Res. 2025 Jul 31;25(1):1012. doi: 10.1186/s12913-025-13137-2.
6
Consumer Acceptance of Alternative Proteins: Exploring Determinants of the Consumer Willingness to Buy in Germany.消费者对替代蛋白的接受度：探索德国消费者购买意愿的决定因素。
Foods. 2025 Jul 9;14(14):2427. doi: 10.3390/foods14142427.
7
The contribution of non-primary caregivers in early stimulation in Kenya and Uganda: Implications for ECD parenting programs in low- and middle-income countries.肯尼亚和乌干达非主要照料者在早期刺激中的贡献：对低收入和中等收入国家早期儿童发展育儿项目的启示。
PLoS One. 2025 May 23;20(5):e0323830. doi: 10.1371/journal.pone.0323830. eCollection 2025.
8
Legalizing Youth-Friendly Cannabis Edibles and Extracts and Adolescent Cannabis Use.将对青少年友好的大麻食品和提取物合法化与青少年使用大麻的情况
JAMA Netw Open. 2025 Apr 1;8(4):e255819. doi: 10.1001/jamanetworkopen.2025.5819.
9
Absence of nonfatal suicidal behavior preceding suicide death reveals differences in clinical risks.自杀死亡前无非致命自杀行为揭示了临床风险的差异。
Psychiatry Res. 2025 May;347:116391. doi: 10.1016/j.psychres.2025.116391. Epub 2025 Feb 24.
10
Continuity of Prescription Medication Use Among Adults Leaving State Prison.刑满释放成年人员的处方药持续使用情况
JAMA Netw Open. 2025 Feb 3;8(2):e2461982. doi: 10.1001/jamanetworkopen.2024.61982.
本文引用的文献
1
Computation of standard errors.计算标准误差。
Health Serv Res. 2014 Apr;49(2):731-50. doi: 10.1111/1475-6773.12122.
2
Misuse of odds ratios in obesity literature: an empirical analysis of published studies.肥胖文献中比值比的误用：已发表研究的实证分析。
Obesity (Silver Spring). 2012 Aug;20(8):1726-31. doi: 10.1038/oby.2012.71. Epub 2012 Mar 22.
3
Interaction terms in nonlinear models.非线性模型中的交互项。
Health Serv Res. 2012 Feb;47(1 Pt 1):255-74. doi: 10.1111/j.1475-6773.2011.01314.x. Epub 2011 Aug 30.
4
What's the Risk? A simple approach for estimating adjusted risk measures from nonlinear models including logistic regression.风险是什么？一种从包括逻辑回归在内的非线性模型估计调整后风险度量的简单方法。
Health Serv Res. 2009 Feb;44(1):288-302. doi: 10.1111/j.1475-6773.2008.00900.x. Epub 2008 Sep 11.
5
Choice of effect measure for epidemiological data.流行病学数据效应测量指标的选择
J Clin Epidemiol. 2000 Sep;53(9):931-9. doi: 10.1016/s0895-4356(00)00210-9.
6
Misunderstandings about the effects of race and sex on physicians' referrals for cardiac catheterization.关于种族和性别对医生进行心脏导管插入术转诊影响的误解。
N Engl J Med. 1999 Jul 22;341(4):279-83; discussion 286-7. doi: 10.1056/NEJM199907223410411.
7
Odds ratios should be avoided when events are common.当事件很常见时，应避免使用比值比。
BMJ. 1998 Nov 7;317(7168):1318. doi: 10.1136/bmj.317.7168.1318.
8
Interpretation and choice of effect measures in epidemiologic analyses.流行病学分析中效应测量指标的解读与选择
Am J Epidemiol. 1987 May;125(5):761-8. doi: 10.1093/oxfordjournals.aje.a114593.
Zotero 插件