European Journal of Epidemiology

, Volume 32, Issue 1, pp 21–29

Statistical inference in abstracts of major medical and epidemiology journals 1975–2014: a systematic review

  • Andreas Stang
  • Markus Deckert
  • Charles Poole
  • Kenneth J. Rothman
REVIEW

DOI: 10.1007/s10654-016-0211-1

Cite this article as:
Stang, A., Deckert, M., Poole, C. et al. Eur J Epidemiol (2017) 32: 21. doi:10.1007/s10654-016-0211-1

Abstract

Since its introduction in the twentieth century, null hypothesis significance testing (NHST), a hybrid of significance testing (ST) advocated by Fisher and null hypothesis testing (NHT) developed by Neyman and Pearson, has become widely adopted but has also been a source of debate. The principal alternative to such testing is estimation with point estimates and confidence intervals (CI). Our aim was to estimate time trends in NHST, ST, NHT and CI reporting in abstracts of major medical and epidemiological journals. We reviewed 89,533 abstracts in five major medical journals and seven major epidemiological journals, 1975–2014, and estimated time trends in the proportions of abstracts containing statistical inference. In those abstracts, we estimated time trends in the proportions relying on NHST and its major variants, ST and NHT, and in the proportions reporting CIs without explicit use of NHST (CI-only approach). The CI-only approach rose monotonically during the study period in the abstracts of all journals. In Epidemiology abstracts, as a result of the journal’s editorial policy, the CI-only approach has always been the most common approach. In the other 11 journals, the NHST approach started out more common, but by 2014, this disparity had narrowed, disappeared or reversed in 9 of them. The exceptions were JAMA, New England Journal of Medicine, and Lancet abstracts, where the predominance of the NHST approach prevailed over time. In 2014, the CI-only approach is as popular as the NHST approach in the abstracts of 4 of the epidemiology journals: the American Journal of Epidemiology (48%), the Annals of Epidemiology (55%), Epidemiology (79%) and the International Journal of Epidemiology (52%). The reporting of CIs without explicitly interpreting them as statistical tests is becoming more common in abstracts, particularly in epidemiology journals. Although NHST is becoming less popular in abstracts of most epidemiology journals studied and some widely read medical journals, it is still very common in the abstracts of other widely read medical journals, especially in the hybrid form of ST and NHT in which p values are reported numerically along with declarations of the presence or absence of statistical significance.

Keywords

Statistics Confidence intervals Statistics and numerical data 

Supplementary material

10654_2016_211_MOESM1_ESM.docx (51 kb)
Supplementary material 1 (DOCX 51 kb)

Funding information

Funder NameGrant NumberFunding Note
Bundesministerium für Bildung und Forschung
  • 01ER1305.

Copyright information

© Springer Science+Business Media Dordrecht 2016

Authors and Affiliations

  • Andreas Stang
    • 1
    • 2
  • Markus Deckert
    • 1
  • Charles Poole
    • 3
  • Kenneth J. Rothman
    • 4
  1. 1.Center of Clinical Epidemiology, Institute of Medical Informatics, Biometry and EpidemiologyUniversity Hospital of EssenEssenGermany
  2. 2.Department of EpidemiologyBoston University School of Public HealthBostonUSA
  3. 3.Department of Epidemiology, Gillings School of Global Public HealthUniversity of North CarolinaChapel HillUSA
  4. 4.RTI Health SolutionsDurhamUSA

Personalised recommendations