Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Stepwise selection in small data sets: a simulation study of bias in logistic regression analysis.

E W Steyerberg1, M J Eijkemans, J D Habbema

  • 1Center for Clinical Decision Sciences, Department of Public Health, Erasmus University, Rotterdam, The Netherlands.

Journal of Clinical Epidemiology
|October 8, 1999
PubMed
Summary

Stepwise selection in regression models can lead to biased regression coefficients. This study shows significant overestimation, especially with low events per variable (EPV), highlighting the need for careful variable selection in statistical analysis.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Navigating Fairness in AI-based Prediction Models: Theoretical Constructs and Practical Applications.

medRxiv : the preprint server for health sciences·2025
Same author

Prediction of transforaminal epidural injection success in sciatica (POTEISS): a protocol for the development of a multivariable prediction model for outcome after transforaminal epidural steroid injection in patients with lumbar radicular pain due to disc herniation or stenosis.

BMC neurology·2024
Same author

Methodological guidance for the evaluation and updating of clinical prediction models: a systematic review.

BMC medical research methodology·2022
Same author

Recall bias in pain scores evaluating abdominal wall and groin pain surgery.

Hernia : the journal of hernias and abdominal wall surgery·2022
Same author

Instrumental variable analysis to estimate treatment effects: a simulation study showing potential benefits of conditioning on hospital.

BMC medical research methodology·2022
Same author

A tailored intervention does not reduce low value MRI's and arthroscopies in degenerative knee disease when the secular time trend is taken into account: a difference-in-difference analysis.

Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA·2022

Area of Science:

  • Statistics
  • Biostatistics
  • Clinical Research Methodology

Background:

  • Stepwise selection methods are common for identifying covariates in regression models.
  • A known issue with stepwise selection is the potential for biased estimation of regression coefficients.
  • This bias, termed 'selection bias,' can impact the reliability of model findings.

Purpose of the Study:

  • To quantify the selection bias in regression coefficients resulting from stepwise selection.
  • To investigate the impact of events per variable (EPV) on this selection bias.
  • To illustrate selection bias using logistic regression in a large clinical trial.

Main Methods:

  • Logistic regression analysis was performed on data from the GUSTO-I trial (40,830 patients).

Related Experiment Videos

  • Random samples with varying events per variable (EPV: 3, 5, 10, 20, 40) were generated.
  • Backward stepwise selection was applied to models with 8 or 16 pre-specified predictors of 30-day mortality.

    • Main Results:

    • Stepwise selection resulted in considerable overestimation of regression coefficients for selected covariates.
    • The magnitude of selection bias decreased as the EPV increased.
    • With EPV=3 and alpha=0.05, bias exceeded 25% for 7 of 8 predictors; this reduced to 1 predictor with EPV=40.

    Conclusions:

    • Stepwise selection methods can introduce substantial bias into estimated regression coefficients.
    • Higher events per variable (EPV) mitigate selection bias.
    • Researchers should be cautious when using stepwise selection, especially in low EPV settings.