A pooled result can look precise while still leaving an important question unanswered: which research findings were available to be pooled? In cannabinoid research, as in other fields, the visible literature is not automatically a complete record. Publication bias is a reason to examine that record, not a reason to assume that any particular conclusion is false.
What is publication bias?
Publication bias occurs when whether a study is published depends on its findings. The broader problem of missing evidence also includes unavailable outcomes within reported studies, selected according to the direction, size or statistical significance of results. These gaps can distort a synthesis even when the included studies were assessed carefully. The Cochrane Handbook’s chapter on missing evidence explains why reviewers should consider evidence beyond journal publications.
For readers, the practical distinction is between “we found these studies” and “these are all the relevant studies.” The first statement describes a search result. The second requires a stronger account of how investigators looked for work that might not have reached a journal.
What can a funnel plot show?
A funnel plot places each study’s effect estimate against a measure of its precision, often its standard error. Less precise estimates tend to spread more widely, while more precise estimates cluster more narrowly. An uneven pattern can prompt questions about missing evidence. However, asymmetry can also reflect differences between studies, methodological problems or chance. A funnel plot is therefore an investigative tool, not a direct count of unpublished studies. See Cochrane’s discussion of funnel plots.
When reading a plot, first identify the labels. Which outcome is being analysed? What does the horizontal axis measure? Does the vertical axis show precision or uncertainty? A familiar funnel shape is not a substitute for understanding what each point represents. Our explanation of confidence intervals and statistical significance provides useful background for reading estimates and their uncertainty.
Keep a short note beside the graph: the outcome, the comparison and the question it is intended to investigate. This makes it easier to connect the visual display with the review’s written explanation.
A hypothetical cannabinoid review
Imagine a review of cannabinoid trials measuring the same prespecified outcome. Its smaller studies appear to show larger effects than its larger studies. The reviewers draw a funnel plot and find an uneven distribution.
That observation creates questions rather than answers. Were the studies conducted in comparable settings? Were outcomes measured in the same way? Did the search identify completed trials whose results could not be located? If a trial record has no accessible results, its findings remain unknown. Calling them unfavourable would turn an evidence gap into an invented result.
This example is entirely hypothetical. It illustrates why a review should explain its reasoning rather than attach a bias label to a picture.
Does “no detected bias” prove completeness?
No. Statistical tests for small-study effects have variable reliability, and a pattern associated with missing unfavourable studies is not uniquely explained by publication bias. A review needs to assess the concern in the context of the evidence for each outcome, as described in NHMRC’s guidance on assessing certainty. “Not detected” should not be rewritten as “ruled out.” Tests may miss a real pattern, particularly when few studies contribute.
What should a transparent review report?
Look for a description of registry searches and efforts to locate results outside conventional publications. WHO treats reporting findings as an important part of the clinical research record; its reporting guidance includes guidance on registry summary results. Our article on clinical trial registration explains how a recorded study plan helps readers ask follow-up questions.
Also look for a clear explanation of how missing evidence affected confidence in the conclusion. Publication bias is one domain considered in certainty assessments, and summary-of-findings tables can make those judgments visible. See Cochrane’s guidance on presenting findings and certainty and our overview of systematic reviews and evidence certainty.
The useful takeaway is a reading habit: ask what was searched, what remains unavailable and how that uncertainty changed the conclusion. A reassuring plot alone cannot answer all three questions.