gert storms
17 december 2017
philip abelson
editor of science
1962-1984
99,9999 % of
[scientific] papers
are accurate and
truthful
15%
10%
67 studies in oncology (47), gynaecology (12),
and cardiovascular studies (8)
only 20-25% replications
53 ‘landmark’ studies in
haematology & oncology
11 % reproduced
Yoshitaka Fujii (anesthesia, Japan): 172
Dipak K. Das (heart surgery, USA): 145
John Darsee (medicine, USA): 104
F. Hermann & M. Bruch (medicine, GER): 94
Diederik Stapel (psychology, Ndl): 55
Honorable mention:
Jan Hendrik Schön (Germany, Bell labs)
9 papers in Science
7 papers in Nature
28 papers in total
… all published in 2000 & 2001
the million $
question !!!
Allow a small chance for false positive results:
p < .05 but …
extreme flexibility in
data collection
data peeking
data-analysis
what is an outlier ?
when do you use covariates ?
pre-transform data ?
reportering
report all variables ?
report all tried-out analyses ?
all conditions? all experiments?
often many possibilities tested, but only
reported those that ‘worked’ (p < .05)
sometimes no bad intensions, but
a) there are no clear rules
b) we all want significance !
p < .05
”‘usual’ undisarable practices probably larger threat for science
than ‘real’ fraud”
- f0irst anonymous measurement of
‘questionable research practices’ (QRP)
- 2002 survey: NIH sponsored researchers
1768 mid-career (52% response rate)
1479 early-career (43% response rate)
non-response bias ?
- first measurement in psychology of
‘questionable research practices’ (QRP)
- electronic questionnaire
- 2155 respondents (response rate 36%)
non-response bias ?
American Economic Journal: Applied Economics,
Volume 8, Number 1, January 2016, pp. 1-32(32)
Button et al (2013), Nature Neuroscience
median power in neuro sciences = 21%
neuroimaging studies: median 8 %
water maze studies: median 18-31 %
Galilei Newton Mendel Bernoulli
Dalton Millikan HewishAlexis Carrel Elckerlyc ??
mean = 4
mean = 6
theoretical courses don’t have an effect,
but whoever sleeps with the dog gets his flees
30.000
participants
18.000
‘so-so’
150$
12
‘bad guys’
36.000$
“Our conclusion, in brief, is that science bears
little resemblance to its conventional portrait.
(…) In the acquisition of new knowledge,
scientists are not guided by logic and objectivity
alone, but also by such nonrational factors as
rhetoric, propaganda, and personal prejudice.
Scientists do not depend solely on rational
thought, and have no monopoly on it. Science
should not be considered the guardian of
rationality in society, but merely one major form
of its cultural expression.”
Broad & Wade, 1982, p. 8-9.

Replicability and questionable research practices

  • 1.
  • 2.
    philip abelson editor ofscience 1962-1984 99,9999 % of [scientific] papers are accurate and truthful
  • 4.
  • 5.
    67 studies inoncology (47), gynaecology (12), and cardiovascular studies (8) only 20-25% replications
  • 6.
    53 ‘landmark’ studiesin haematology & oncology 11 % reproduced
  • 7.
    Yoshitaka Fujii (anesthesia,Japan): 172 Dipak K. Das (heart surgery, USA): 145 John Darsee (medicine, USA): 104 F. Hermann & M. Bruch (medicine, GER): 94 Diederik Stapel (psychology, Ndl): 55
  • 8.
    Honorable mention: Jan HendrikSchön (Germany, Bell labs) 9 papers in Science 7 papers in Nature 28 papers in total … all published in 2000 & 2001
  • 9.
  • 11.
    Allow a smallchance for false positive results: p < .05 but … extreme flexibility in data collection data peeking data-analysis what is an outlier ? when do you use covariates ? pre-transform data ? reportering report all variables ? report all tried-out analyses ? all conditions? all experiments?
  • 12.
    often many possibilitiestested, but only reported those that ‘worked’ (p < .05) sometimes no bad intensions, but a) there are no clear rules b) we all want significance ! p < .05
  • 13.
    ”‘usual’ undisarable practicesprobably larger threat for science than ‘real’ fraud” - f0irst anonymous measurement of ‘questionable research practices’ (QRP) - 2002 survey: NIH sponsored researchers 1768 mid-career (52% response rate) 1479 early-career (43% response rate) non-response bias ?
  • 15.
    - first measurementin psychology of ‘questionable research practices’ (QRP) - electronic questionnaire - 2155 respondents (response rate 36%) non-response bias ?
  • 19.
    American Economic Journal:Applied Economics, Volume 8, Number 1, January 2016, pp. 1-32(32)
  • 21.
    Button et al(2013), Nature Neuroscience median power in neuro sciences = 21% neuroimaging studies: median 8 % water maze studies: median 18-31 %
  • 22.
    Galilei Newton MendelBernoulli Dalton Millikan HewishAlexis Carrel Elckerlyc ??
  • 24.
  • 26.
    theoretical courses don’thave an effect, but whoever sleeps with the dog gets his flees
  • 27.
  • 29.
    “Our conclusion, inbrief, is that science bears little resemblance to its conventional portrait. (…) In the acquisition of new knowledge, scientists are not guided by logic and objectivity alone, but also by such nonrational factors as rhetoric, propaganda, and personal prejudice. Scientists do not depend solely on rational thought, and have no monopoly on it. Science should not be considered the guardian of rationality in society, but merely one major form of its cultural expression.” Broad & Wade, 1982, p. 8-9.