2.3 Problem 7
Wagenmakers Et Al.
Fairy-tale Factor- researchers don’t commit themselves to plan of
analysis, prior to seeing data, meaning they can fine tune analyses
to data, making the data more compelling than it really is
o Increased likelihood that finding is fictional and non-replicable
Confirmation bias- humans tend to seek confirmation rather than
disconfirmation of their beliefs
o 3 ways it works:
1. Ambiguous info interpreted to be consistent with ones prior
beliefs
2. Search for information that confirms, not disconfirms
preferred beliefs
3. Remember info that supports our prior beliefs, better
- Hindsight bias- tendency to judge an event as more predictable
after it has occurred
o Researchers typically seek confirmation, not falsification
Furthered by want for publication, chose methods most
likely to publish
Catastrophic: results in false publications
Replication rates lower than 50% in biomedical and cancer research
Bad Science
Virtually on psychological research conducted in confirmatory way
o Rarely specify specific analyses prior to data collection
Can cherry-pick only variables that obtain desired
results
Include in papers only experiments with desired
outcome
Use different statistical tests to tailor data to fit
Researchers often believe that they aren’t doing wrong, instead
allowing for deeper analyse of data
Issue partly rooted in law that data can only be used once
o Only for one hypothesis, so want this hypothesis to be right
Amount of exploration, data analysis, etc. varies greatly between
each psychological test
o So does reliability of statistical results
Good Science
Key that the researcher is honest
Researcher may think they’re honest, even when not, due to
confirmation and hindsight bias
o Should instead separate exploratory articles and confirmation
articles
If exploratory articles disguise themselves as
confirmatory, increases amount of bad science
Proposed solution: publish protocol and means of methods before even
starting the study
Eliminates fairy tale factor
Proposed Research:
, 1. Conduct exploratory studies, but don’t present them as strong
evidence for a claim: should instead determine interest data aspects
2. Confirmatory approach: use online repositories, and submit
document with the variables, data collection, methods, etc.
Should remove hindsight and confirmation bias
All findings should be mentioned in separate exploratory results
section
Estimating the Reproducibility of Psychological Science
Reproducibility= core principle of scientific progress
Scientific claims shouldn’t gain credit because of the status or
authority of originator
o Should get status from replicability of supporting evidence
Direct replication- attempt to recreate conditions believed
sufficient to obtain previously observed finding, means to establish
reproducibility of a finding
o Gives change to assess/improve reproducibility
o May not obtain original results as:
Differences between replication and original study may
change observed effect
Original result could be false positive*
Replication could create false negatives*
* create misleading info on the effects, fail to identify
necessary conditions to reproduce a finding
Reproducibility poorly understood, as greater incentives for sciences
in novelty, not reproducibility
Problematic practices (eg. selective reporting, selective analysis,
insufficient specification of conditions needed)
o May lead to greater chances of false-positives and
irreproducible results
Study Results
Effect size is significantly lower in replications than original studies
Replication success consistently related to original strength of
evidence: more than to team characteristics and implementation of
replication
Direct replication provides evidence for result reliability
Publication, selection, cultural differences, and reporting biases can
explain difference between original and replication effects
HARKing
Hypothesising After Results are Known
Hypothetico-deductive approach- deducing or deriving one or more
explicit and testable hypotheses from some plausible theory(ies) about
the phenomena of interest prior to designing one’s research (HD
approach)
Wagenmakers Et Al.
Fairy-tale Factor- researchers don’t commit themselves to plan of
analysis, prior to seeing data, meaning they can fine tune analyses
to data, making the data more compelling than it really is
o Increased likelihood that finding is fictional and non-replicable
Confirmation bias- humans tend to seek confirmation rather than
disconfirmation of their beliefs
o 3 ways it works:
1. Ambiguous info interpreted to be consistent with ones prior
beliefs
2. Search for information that confirms, not disconfirms
preferred beliefs
3. Remember info that supports our prior beliefs, better
- Hindsight bias- tendency to judge an event as more predictable
after it has occurred
o Researchers typically seek confirmation, not falsification
Furthered by want for publication, chose methods most
likely to publish
Catastrophic: results in false publications
Replication rates lower than 50% in biomedical and cancer research
Bad Science
Virtually on psychological research conducted in confirmatory way
o Rarely specify specific analyses prior to data collection
Can cherry-pick only variables that obtain desired
results
Include in papers only experiments with desired
outcome
Use different statistical tests to tailor data to fit
Researchers often believe that they aren’t doing wrong, instead
allowing for deeper analyse of data
Issue partly rooted in law that data can only be used once
o Only for one hypothesis, so want this hypothesis to be right
Amount of exploration, data analysis, etc. varies greatly between
each psychological test
o So does reliability of statistical results
Good Science
Key that the researcher is honest
Researcher may think they’re honest, even when not, due to
confirmation and hindsight bias
o Should instead separate exploratory articles and confirmation
articles
If exploratory articles disguise themselves as
confirmatory, increases amount of bad science
Proposed solution: publish protocol and means of methods before even
starting the study
Eliminates fairy tale factor
Proposed Research:
, 1. Conduct exploratory studies, but don’t present them as strong
evidence for a claim: should instead determine interest data aspects
2. Confirmatory approach: use online repositories, and submit
document with the variables, data collection, methods, etc.
Should remove hindsight and confirmation bias
All findings should be mentioned in separate exploratory results
section
Estimating the Reproducibility of Psychological Science
Reproducibility= core principle of scientific progress
Scientific claims shouldn’t gain credit because of the status or
authority of originator
o Should get status from replicability of supporting evidence
Direct replication- attempt to recreate conditions believed
sufficient to obtain previously observed finding, means to establish
reproducibility of a finding
o Gives change to assess/improve reproducibility
o May not obtain original results as:
Differences between replication and original study may
change observed effect
Original result could be false positive*
Replication could create false negatives*
* create misleading info on the effects, fail to identify
necessary conditions to reproduce a finding
Reproducibility poorly understood, as greater incentives for sciences
in novelty, not reproducibility
Problematic practices (eg. selective reporting, selective analysis,
insufficient specification of conditions needed)
o May lead to greater chances of false-positives and
irreproducible results
Study Results
Effect size is significantly lower in replications than original studies
Replication success consistently related to original strength of
evidence: more than to team characteristics and implementation of
replication
Direct replication provides evidence for result reliability
Publication, selection, cultural differences, and reporting biases can
explain difference between original and replication effects
HARKing
Hypothesising After Results are Known
Hypothetico-deductive approach- deducing or deriving one or more
explicit and testable hypotheses from some plausible theory(ies) about
the phenomena of interest prior to designing one’s research (HD
approach)
