Hypothesis

Mortality cues interact with childhood socioeconomic status to shift risk acceptance and delay discounting, as found in US student samples.

What failed

Three replications, one with British university students and two with British online samples, using priming material modified to be credible to a British audience, found no interaction between childhood socioeconomic status and mortality-priming condition for either outcome; the only consistent trend was an interaction between sex and priming condition for delay discounting.

Broken assumption

The mortality-priming by childhood-SES interaction is robust enough to appear outside US student samples.

Why

The authors list demographic differences, their alterations to the prime and other study limitations as possible causes, while noting that priming effects are considered fragile and that the original interaction may not be robust or generalisable.

From the published paper: Pepper et al. PeerJ, 2017. ↗CC BY

Hypothesis

SYTO9 and propidium iodide staining, read in microtiter plates, gives an accurate high-throughput measure of live versus dead bacteria.

What failed

With SYTO9 alone, live and dead Staphylococcus aureus gave equal signal, and dead Pseudomonas aeruginosa gave an 18-fold stronger signal than live cells — still stronger after PI counterstaining. SYTO9 bleached strongly over time, and PI's maximum signal was weak relative to SYTO9 and background.

Broken assumption

SYTO9 stains live and dead cells equally, so the SYTO9/PI ratio reads out viability.

Why

The authors attribute the errors to SYTO9 bleaching, its different binding to live and dead cells, and background fluorescence, which make small inaccuracies in the weak PI signal change the outcome.

From the published paper: Stiefel et al. BMC Microbiology, 2015. ↗CC BY

Hypothesis

Chloroquine, proposed as a repurposed Ebola treatment, would improve survival in animals infected with Ebola virus.

What failed

Chloroquine treatment did not improve survival in either mouse or hamster models. In vitro it inhibited viral replication only at cytotoxic concentrations.

Broken assumption

In vitro antiviral activity of chloroquine against Ebola reflects a usable therapeutic window in vivo.

Why

The authors conclude that the in vitro activity appears only at concentrations toxic to cells, making chloroquine unpromising and directing effort to other drug classes.

From the published paper: Falzarano et al. Emerging infectious diseases, 2015. ↗CC BY

Hypothesis

Adding managed Apis mellifera or Bombus impatiens hives to pumpkin fields would raise bee visits to flowers and fruit yield beyond what wild bees provide.

What failed

In a two-year comparison of fields supplemented with honey bee hives, bumble bee hives, or nothing, neither kind of supplementation increased visitation by those bees or pumpkin yield.

Broken assumption

Placing managed hives next to a crop increases pollination of that crop above the wild-bee baseline.

Why

Yield did rise with visits by honey bees and bumble bees across all fields, so the bees matter as pollinators; adding hives simply did not raise their visit rate, and wild bees such as Peponapis pruinosa were also prominent visitors.

From the published paper: Petersen et al. PLoS ONE, 2013. ↗CC BY

Hypothesis

Coding how much effort respondents put into answering an open-ended question would proxy their real behaviour rather than merely stated intention, letting a vignette experiment on rewards and workplace knowledge sharing measure behaviour within subjects.

What failed

With participants answering three vignettes in a within-subject design, a strong order effect appeared: answers to the first vignette were longer on average than to later ones regardless of treatment, confounding the effort measure.

Broken assumption

Effort invested in an open-ended answer reflects the respondent's likely behaviour and not the question's position in the survey.

Why

The authors conclude the order effect makes this operationalisation unsuitable for within-subject designs, though it may still work between subjects with a single vignette each.

From the published paper: Fischer et al. Journal of Trial and Error, 2021. ↗CC BY

Hypothesis

Heavier lifetime use of glyphosate would be associated with higher cancer incidence, as IARC's 2015 'probably carcinogenic' classification and some earlier studies suggested.

What failed

In a prospective cohort of 54,251 licensed applicators (44,932 glyphosate users, 5,779 incident cancers among them) followed to 2012–13, glyphosate use was not significantly associated with cancer at any site. Only a non-significant raised risk of acute myeloid leukaemia appeared in the highest exposure quartile (RR 2.44, 95% CI 0.94–6.32).

Broken assumption

Glyphosate use raises cancer incidence in the people most heavily exposed to it.

Why

The paper reports no statistically significant association at any site after controlling for other pesticide use; the AML signal did not reach significance under any lag and is flagged as needing confirmation.

From the published paper: Andreotti et al. JNCI Journal of the National Cancer Institute, 2017. ↗Public domain

Hypothesis

Twenty-five-month-olds take an agent’s false belief into account in where they look in anticipation, as the original studies reported and as implicit theory-of-mind accounts require.

What failed

The effect failed to appear in 2- to 3-year-olds (N = 48) and again in an independent, tightly controlled, preregistered and adequately powered replication with 25-month-olds (N = 78); the same paradigm did work in adults (N = 115). In every sample a surprisingly large share of participants failed to anticipate the agent’s action during familiarisation, so applying the original exclusion criteria removed much of the data.

Broken assumption

Anticipatory looking is a dependable index of implicit false belief understanding in infants.

Why

The authors question the reliability of non-verbal anticipatory-looking paradigms themselves, noting that the exclusion rates strongly limit how the test phase can be read.

From the published paper: Schuwerk et al. Royal Society Open Science, 2018. ↗CC BY

Hypothesis

Standard isoprene emission algorithms, built from a few vascular plant species and widely applied to moorland, would reproduce isoprene fluxes from a moss-dominated peat bog.

What failed

Over six weeks of flux and enclosure measurements above a Scottish peat bog, the light response never saturated, and both big-leaf and canopy-level algorithms failed to match observed fluxes: they overestimated emissions below about 1000 μmol m⁻² s⁻¹ PAR and underestimated them on clear days.

Broken assumption

Light-response curves derived from vascular plants transfer to non-vascular, moss-dominated vegetation.

Why

The authors attribute the non-saturating response to the architecture of moss hummocks, which probably attenuate light differently from a grass canopy.

From the published paper: Langford et al. Environmental Research Ecology, 2022. ↗CC BY