PSLE Science Reality Lab Vol No.492 | “The Funnel Plot Looks Asymmetrical” — Does That Prove Studies Were Hidden?

Reality Lab ID: PSLE-SCI-REALITY-0492

Wait, what? A scientific review shows a graph shaped roughly like an upside-down funnel. The large studies gather near the top. The smaller studies spread out below. But the dots are not balanced: there seem to be more small studies on the right than on the left. A caption on social media says, “This proves scientists hid the studies that disagreed.”

The picture is intriguing. The missing-looking side seems to invite a detective story. Yet the graph has not shown us a pile of secretly discarded papers. It has shown us a pattern. The next scientific job is to work out which explanations can produce that pattern and which evidence would distinguish them.

This PSLE Science Reality Lab teaches one precise evidence-transfer habit for Primary 5 and Primary 6 learners: how to evaluate an asymmetrical funnel plot without turning a visual pattern into proof of publication bias or hidden studies. It is not a statistics textbook, not a medical-advice page and not a replacement for the existing eduKateSengkang owners of graphs, sampling, alternative explanations or systematic reviews. It takes one real scientific communication object and asks the question that strong inquiry always asks: what does this evidence actually support, and what still needs to be checked?

The current 2026 PSLE Science assessment framework continues to assess both Knowledge with Understanding and Application of Knowledge and Scientific Inquiry. That second area includes interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning. The 2023 Primary Science syllabus also values healthy scepticism: learners should question observations, methods, processes and data, recognise assumptions and uncertainty, and remain open to more than one plausible explanation. A funnel plot is an unusually good training object for that habit because it is easy to see a shape and much harder to prove why the shape occurred.

Quick Answer

No. An asymmetrical funnel plot does not by itself prove that studies were hidden, suppressed or left unpublished. Funnel plots are used in research reviews to display how estimates from different studies relate to a measure of study size or precision. Smaller, less precise studies often spread more widely; larger, more precise studies tend to cluster more tightly. If the pattern is uneven, that can be a reason to investigate small-study effects or possible missing evidence.

But several different causes can produce asymmetry. Selective non-reporting is one possible cause. Others include genuine differences between smaller and larger studies, differences in study quality or methods, the way the effect measure behaves, chance, or other features of the evidence. Cochrane’s current handbook explicitly cautions that funnel-plot asymmetry should not be treated as diagnostic proof of non-reporting bias.

The correct scientific conclusion is therefore usually not, “The missing studies were definitely hidden.” It is closer to: “The plot is asymmetrical, so we should investigate why smaller studies appear to give different results and whether missing evidence is one plausible explanation.”

The Owned Learner Job — and the Boundary

This article owns one learner job: separating the observation “this funnel plot is asymmetrical” from the stronger causal claim “studies must have been hidden because of their results”.

It does not re-teach generic graph reading. It does not turn Primary learners into meta-analysis statisticians. It does not own publication ethics, research methods generally, or the mathematics of standard errors. It also does not replace the existing Reality Lab on forest plots, which asks a different question about whether a summary diamond means every study agrees. For that neighbouring job, use PSLE Science Reality Lab Vol No.147 | “The Diamond Is on the Right” — Did Every Study in the Forest Plot Find the Same Result?

For the broader discipline of separating what is seen from what is inferred, route to How to Tell Observation, Inference, Prediction and Explanation Apart in PSLE Science. For the meaning and limits of a systematic review label, use PSLE Science Reality Lab Vol No.436 | “Systematic Review” — Does That Mean It Found Every Study Ever Done?. For a different meta-analysis representation, use PSLE Science Reality Lab Vol No.465 | “I² = 80%” — Does That Mean 80% of the Studies Disagree?

The Original Composite Case: The Seedling Review

Imagine a fictional group of researchers reviewing experiments on a simple plant question: does lining a seedling tray with a reflective material increase average seedling height after fourteen days? The review is invented for this lesson. The studies, numbers and graph described here are original teaching material, not copied from an exam paper, research article or commercial worksheet.

The review contains twelve studies. Some are small classroom-style experiments with twenty to forty seedlings. Others are larger greenhouse studies with hundreds of seedlings. Each study compares reflective trays with ordinary trays and reports an estimated difference in average height.

Study	Approximate size	Estimated height difference	Precision
A	20 seedlings	+7.0 cm	Low
B	24 seedlings	+5.8 cm	Low
C	28 seedlings	+6.3 cm	Low
D	36 seedlings	+4.9 cm	Low
E	48 seedlings	+4.5 cm	Moderate
F	60 seedlings	+3.8 cm	Moderate
G	90 seedlings	+3.1 cm	Moderate
H	120 seedlings	+2.8 cm	Higher
I	180 seedlings	+2.6 cm	Higher
J	240 seedlings	+2.5 cm	High
K	320 seedlings	+2.4 cm	High
L	500 seedlings	+2.5 cm	Very high

When these studies are plotted, the larger studies cluster around an effect of roughly +2.5 cm. The smaller studies spread farther away, but in this invented set most of them lie on the high-positive side. The plot therefore looks asymmetrical.

A student sees the empty-looking lower-left region and says, “The negative studies must have been hidden.” That is one possible story. It is not yet the only story, and the graph alone has not established it.

First Identify the Scientific Object

A funnel plot is usually a scatter plot in which each dot represents a study. One axis represents the study’s estimated effect. The other represents a measure related to study size or precision, often the standard error. In a common design, larger or more precise studies appear toward the top and smaller or less precise studies toward the bottom.

The name “funnel” comes from a broad pattern that can arise because estimates from small studies vary more widely, while estimates from large studies tend to be more precise and therefore cluster more tightly. The graph is not a photograph of journals, researchers or filing cabinets. It is a representation of study estimates and their precision.

That object identification matters. If you mistake a funnel plot for a count of “published versus unpublished studies”, you will read information into the picture that it does not directly contain.

Observed, Claimed and Inferred

Observed from the plot: the dots are distributed unevenly; smaller studies tend to show larger positive effects; one lower region has few or no dots.
Supported interpretation: there may be a small-study pattern that deserves investigation.
Possible explanation: studies with certain results may be missing because they were not reported, published or found.
Too-strong inference: the plot itself proves that researchers deliberately hid studies.

The scientific discipline is in that third-to-fourth step. A pattern can raise a question without proving one cause.

Why Smaller Studies Spread More Widely

Suppose two classes estimate the average height of a school’s Primary 6 pupils. One class measures only four pupils. Another measures two hundred. The four-pupil average can move a lot depending on which four children happened to be selected. The two-hundred-pupil average is usually more stable because one unusually tall or short pupil has less influence on the total.

The same broad idea helps explain why estimates from small scientific studies can scatter more widely. Small studies are not automatically bad. Large studies are not automatically correct. But, all else being equal, estimates based on less information tend to be less precise.

This is why the bottom of a funnel plot can be wider than the top. The shape is connected to precision, not to a rule that every real evidence set must form a perfect triangle.

Asymmetry Is a Clue, Not a Confession

If one side of the funnel looks sparse, a careful reviewer may ask whether some results are missing. That is a sensible scientific question. But a sparse region is not the same thing as a confession from a researcher saying, “I hid my study because I disliked the result.”

The learner should therefore translate the visual cue into a question rather than a verdict:

Why do the smaller studies appear to give different results from the larger studies, and could missing evidence be one reason?

That wording keeps the inquiry alive. It makes room for evidence to narrow the explanations instead of deciding the cause before the investigation begins.

Alternative Explanation 1: The Small Studies Were Done Under Different Conditions

Return to the reflective-tray example. Imagine the small studies were mostly carried out on very sunny classroom windowsills, while the large studies were conducted in commercial greenhouses with more controlled lighting. Reflective lining might genuinely have a larger effect when light arrives from one strong side than when illumination is already well distributed.

The funnel plot could then be asymmetrical because study size is linked to experimental setting. Smaller studies are not merely noisier versions of the same experiment; they are partly different experiments. This is one form of genuine heterogeneity.

Notice what happened. We produced the same visual pattern without hiding a single study.

Alternative Explanation 2: Smaller Studies Used Different Methods

Suppose the small studies measured only the tallest five seedlings in each tray, while the larger studies measured every surviving seedling. The smaller studies might report larger effects because their measurement rule selects unusually strong plants.

Again, asymmetry can appear because study method is associated with study size. A funnel plot can warn us that smaller-study results differ, but we need the methods sections to understand why.

Alternative Explanation 3: Study Quality Differs

Imagine some small studies allowed the person measuring height to know which trays received the reflective lining. Perhaps that knowledge unintentionally influenced which bent stem was treated as the “top”. Larger studies used a fixed measuring procedure and blinded labels.

If weaker methods occur more often in small studies, the small-study estimates may be systematically different. The pattern would matter, but the explanation would be method quality, not necessarily missing publication.

Alternative Explanation 4: The Effect Measure Itself Can Matter

Funnel plots can be built using different effect measures and different vertical axes. Some choices behave better than others for particular kinds of data. In some circumstances, a mathematical relationship between the effect estimate and its standard error can create or exaggerate asymmetry even when there is no selective reporting.

A Primary learner does not need to calculate these statistics. The transferable lesson is simpler: before interpreting a graph shape, check how the graph was constructed. A representation can have properties caused partly by its design.

Alternative Explanation 5: Chance

With a small number of studies, random scatter can look dramatic. If you toss ten beans onto a sheet of paper, they may cluster more on one side even though nobody removed any beans. A sparse region can appear simply because the evidence set is limited.

This is one reason authoritative guidance warns against over-interpreting funnel plots with too few studies. A shape that feels meaningful to the eye can be unstable when only a handful of dots are available.

Alternative Explanation 6: Evidence Really Is Missing

Selective non-reporting is still a real possibility. A small study with an exciting positive result may be more likely to be written up, submitted, accepted, publicised or easily found than a small study with an unexciting result. Researchers might also report some measured outcomes but not others. Search systems may miss reports written in another language or stored outside major databases.

The important point is not to deny publication bias. It is to avoid claiming that one graph has uniquely identified it. Good science keeps the possibility open and then looks for additional evidence.

Representation Check: Read the Axes Before Reading the Story

A learner should never interpret “funnel shape” from appearance alone. Check the axes.

What effect measure is on the horizontal axis?
What quantity is on the vertical axis: standard error, sample size, precision or something else?
Does the vertical scale increase upward or downward?
Are studies from different groups shown with different symbols?
Is there a summary-effect line?
Are shaded regions confidence regions, significance contours or simply decorative backgrounds?

Two graphs can both look like funnels while encoding different quantities. The legend and axis labels are part of the evidence.

Comparison Check: Are Small and Large Studies Really Comparable?

Before saying “small studies disagree with large studies”, ask whether they studied the same question in comparable ways. Did they use the same species, age group, dose, environment, duration, measuring method and outcome definition? Did the small studies happen in laboratories while large studies happened outdoors? Did one group study a stronger treatment?

If study size is tangled together with another variable, the funnel plot may be showing that hidden difference. This is a classic scientific reasoning problem: an observed association between two features does not by itself tell us the causal mechanism.

Baseline Check: What Pattern Would We Expect Without Selective Reporting?

The phrase “missing side of the funnel” can make it sound as though every honest evidence set must be perfectly symmetrical. That is too strong. Real studies differ. Their populations, methods and conditions differ. Random variation remains. A perfectly mirrored shape is not a universal law.

The better baseline question is: given the way these studies were done and the amount of random variation expected, is the observed asymmetry surprising enough to investigate? That is a much more careful claim than “not symmetrical equals hidden studies”.

Method Check: How Were the Studies Found?

A funnel plot can only display the studies that entered the review. So the review’s search method matters. Did the reviewers search several databases? Did they look for conference abstracts, theses, trial registries or reports that were not published in ordinary journals? Did they include more than one language? Did they contact researchers or organisations for missing results?

If the search was narrow, missing studies become more plausible. If the search was broad and pre-specified, concern may be reduced, though never reduced to absolute zero. The graph and the search method should be read together.

Provenance Check: Where Did This Funnel Plot Come From?

A screenshot on social media may crop away the title, axes, study labels and caption. A news graphic may simplify the original. A scientific review may include a sensitivity analysis that is missing from the screenshot. Before evaluating the claim, find the original source if possible.

Is the image from the original review?
Is the review current or has it been updated?
What question was the plot designed to examine?
How many studies are included?
What does the caption say about interpretation?
Do the authors discuss other causes of asymmetry?

Provenance protects the learner from evaluating a graph that has been detached from the information needed to understand it.

Evidence That Would Strengthen Concern About Missing Studies

No single item automatically proves publication bias, but concern becomes stronger when several independent pieces of evidence point in the same direction. Examples include:

the funnel plot shows a small-study pattern consistent with missing unfavourable results;
registered or announced studies can be identified but their results cannot be found;
study protocols list outcomes that disappear from the final report;
small studies with striking results are easy to find while similarly designed null-result studies appear absent;
the pattern remains after plausible methodological and population differences are examined;
reviewers document extensive searches and still identify signs that outcomes or studies were selectively unavailable.

The key word is convergence. A stronger conclusion comes from different evidence sources agreeing, not from repeating the same visual clue in several sentences.

Evidence That Would Weaken the Simple “Hidden Studies” Story

small studies were conducted in systematically different environments;
small studies used a stronger treatment or different measurement rule;
the asymmetry disappears when unlike subgroups are examined separately;
the plot contains very few studies and the apparent gap is unstable;
the effect measure used is known to produce a relationship with its precision under the data conditions;
registered studies and unpublished reports are found on both sides of the effect estimate;
a second reasonable funnel-plot specification does not show the same pattern.

This does not prove there is no reporting bias. It shows why a scientific conclusion should be proportional to the total evidence.

How Far Can the Conclusion Travel?

From an asymmetrical funnel plot, a careful learner may say: “The smaller studies appear to give different results from the larger studies, so possible causes such as missing evidence, different study conditions, methodological differences and chance should be investigated.”

The learner should not jump straight to: “The scientists lied,” “The journal hid the truth,” “Every missing dot is a suppressed study,” or “The whole review is invalid.” Those claims name people, motives and events that the graph has not directly observed.

A strong scientific conclusion is often narrower than a viral caption. That narrowness is not weakness. It is accuracy.

Worked Case 1: The Sunny Windowsill Effect

Six small seedling studies are carried out on bright windowsills and all show a large benefit from reflective tray lining. Six large greenhouse studies show a smaller benefit. The funnel plot is asymmetrical.

Tempting conclusion: Small negative studies were hidden.

Better reasoning: Missing studies are possible, but the small and large studies also differ in lighting environment. The reviewer should investigate whether the treatment genuinely works differently under directional sunlight before assigning the asymmetry to non-reporting.

Worked Case 2: The Eight-Dot Funnel

A review contains only eight studies. Five happen to lie on one side of the summary estimate and three on the other. Someone draws a diagonal line around the dots and declares a “missing wedge”.

Better reasoning: With few studies, random patterns are easy to over-read. The visual imbalance may be unstable. Authoritative review guidance treats funnel-plot interpretation cautiously and statistical tests for asymmetry generally require a reasonably large set of studies. The learner should not treat an eight-dot picture as proof of suppressed evidence.

Worked Case 3: The Cropped Social-Media Screenshot

A screenshot shows dots but no axis labels. The post says, “The empty left side proves bad results were deleted.”

Better reasoning: Without axes, caption and source, the learner cannot even establish what the graph encodes. First recover the original review and identify the effect measure, precision measure, number of studies and authors’ interpretation. Provenance comes before causal judgement.

Worked Case 4: The Missing Outcome

A registered study says it will measure plant height, leaf count and root mass. The final paper reports plant height and leaf count but not root mass. Several similar studies also omit root mass when results are small.

Better reasoning: This is additional evidence relevant to selective outcome reporting. A funnel plot alone did not prove missing evidence, but protocol-to-report comparisons provide a separate line of evidence that can strengthen concern.

Worked Case 5: All Studies Are Present, but Methods Differ

Researchers locate laboratory notebooks for all twelve planned studies, including unpublished ones. No study appears missing. Yet the funnel plot remains asymmetrical. The small studies used manual ruler measurements; large studies used automated imaging.

Better reasoning: The persistence of asymmetry does not force a publication-bias conclusion. Method differences linked to study size provide another plausible cause that now deserves focused investigation.

Worked Case 6: A Symmetrical Plot Does Not Prove No Bias

A second review has a visually symmetrical funnel plot. A learner says, “Good. That proves no studies were hidden.”

Better reasoning: Symmetry does not guarantee the absence of reporting bias. Some forms of selective reporting can still produce a roughly symmetrical pattern, and a funnel plot cannot reveal studies that leave no identifiable trace. The graph is one piece of evidence, not a universal detector.

Worked Case 7: The Review Finds an Unpublished Study

A reviewer contacts a research group and obtains a small unpublished study with a near-zero effect. Adding it fills part of the sparse side of the funnel.

Better reasoning: This new evidence makes missing-results concerns more concrete because an actually missing-from-publication study has been identified. Even so, the reviewer should avoid assuming that every remaining empty region corresponds to another hidden study.

Worked Case 8: The Two Subgroups

When all studies are plotted together, the funnel is strongly asymmetrical. When studies are separated by plant species, each species forms a much more balanced pattern around a different average effect.

Better reasoning: The combined asymmetry was partly created by mixing groups with genuinely different effects. The plot was not “wrong”; the first grouping hid an important source of variation.

Tempting but Invalid Reasoning

“One side is empty, so studies were definitely deleted.” The shape has more than one possible cause.
“A symmetrical funnel proves the literature is unbiased.” Symmetry cannot certify the absence of all reporting problems.
“Small studies are bad studies.” Small size affects precision but does not automatically make a method poor.
“Large studies are always correct.” Large studies can have systematic bias, weak measurements or irrelevant populations.
“Every dot is equally informative.” Studies differ in precision, design and relevance.
“The empty space tells us exactly how many studies are missing.” A visual gap is not a literal count of invisible studies.
“Publication bias means fraud.” Selective evidence availability can arise through many stages and incentives; motive cannot be read directly from the plot.
“If the plot is asymmetrical, the scientific conclusion must be false.” The asymmetry affects confidence and interpretation; it does not mechanically reverse the result.

A Useful Three-Layer Reading Routine

When a learner meets a funnel plot in an article, infographic or news discussion, use three layers.

Layer 1: What is physically on the page?

Name the axes, dots, groups, reference line and any shaded regions. Count the studies. Describe the asymmetry without explaining it.

Layer 2: What scientific pattern is being suggested?

Ask whether smaller studies appear systematically different from larger ones. Identify which region is sparse and whether the effect changes with precision.

Layer 3: Which explanations remain possible?

List missing evidence, genuine study differences, method quality, effect-measure behaviour and chance. Then ask what extra evidence could separate them.

This routine keeps description, interpretation and causal explanation from collapsing into one sentence.

Why “Could Be Missing” Is Not Weak Language

Students are sometimes taught to sound certain because confident answers feel stronger. Science works differently. When several explanations fit the current evidence, saying “could be” or “is consistent with” can be more accurate than saying “proves”.

Uncertainty language is not an excuse to avoid conclusions. It is a way to grade conclusions by evidence strength. If later evidence identifies an unpublished study, the wording can become stronger. If method differences explain the pattern, the hypothesis can be revised. Good reasoning leaves room to update.

Model and Measurement Limits

A funnel plot compresses many complicated studies into a small number of coordinates. That is useful, but compression removes detail. The plot may not show whether a study used poor instruments, lost many samples, changed its outcome definition, included a different population or measured at a different time. Those details remain in the underlying reports.

The vertical axis is also a model of precision, not a direct “quality score”. A study can be very precise and still systematically wrong. Another can be imprecise yet carefully conducted. Precision describes the spread expected around an estimate; it is not the same as truth.

Finally, visual judgement itself is imperfect. Two people may disagree about whether a plot looks asymmetric. This is why scientific reviewers may use formal tests, sensitivity analyses and contextual evidence rather than relying only on eyeballing the dots.

PSLE-Style Transfer Case: What Can the Plot Support?

A fictional review combines fourteen experiments on whether a special transparent cover changes the rate at which water evaporates from identical dishes. The largest studies cluster around almost no change. Several small studies show a large decrease in evaporation. The lower-left part of the funnel plot contains few dots.

A student writes: “The funnel plot proves that small studies showing increased evaporation were hidden because researchers wanted the cover to look effective.”

Strong answer: The funnel plot shows that smaller studies tend to give different effect estimates and that the pattern is asymmetrical. Missing or selectively unreported studies are one possible explanation, but the plot alone does not prove that studies were hidden or reveal anyone’s motive. The reviewer should also check whether smaller studies used different temperatures, air movement, dish sizes or methods, and should search for registered or unpublished studies before deciding why the asymmetry occurred.

The answer earns its strength by doing four things: it describes the pattern, refuses an unsupported causal leap, offers plausible alternatives and names evidence that could resolve the uncertainty.

Practice 1: Pattern or Cause?

A funnel plot has more small positive studies than small negative studies. What is the safest immediate conclusion?

Answer: There is an asymmetrical small-study pattern worth investigating. The plot alone does not identify the cause.

Practice 2: Missing Studies?

The lower-left region has no dots. Does that tell you exactly how many studies are missing?

Answer: No. Empty graph space is not a literal count of unpublished studies. It can arise for several reasons.

Practice 3: Search Evidence

Which would strengthen concern about missing evidence more: a dramatic-looking funnel alone, or the funnel plus three registered studies whose final results cannot be located?

Answer: The second. The registrations provide an independent line of evidence that planned studies existed but their results may be unavailable.

Practice 4: Different Conditions

All small studies used 30°C while all large studies used 22°C. Why does that matter?

Answer: Study size is mixed up with temperature. A real temperature-dependent effect could make small and large studies differ even if none were hidden.

Practice 5: Symmetry

A funnel plot looks symmetrical. Can you conclude that no reporting bias exists?

Answer: No. Symmetry reduces one visual concern but cannot prove that all relevant evidence was reported or found.

Practice 6: Large Study

A very large study appears near the top of the plot. Does its size prove its measurement method was valid?

Answer: No. Large size can improve precision but does not automatically remove systematic measurement or design problems.

Practice 7: The Axis Trick

Two funnel plots use different vertical axes. Should you assume the shapes mean exactly the same thing?

Answer: No. Check what each axis measures and how the graph was constructed before comparing the shapes.

Practice 8: Motive

A graph is asymmetric. Can you identify which scientist deliberately hid a study?

Answer: No. The graph does not identify a person, motive or act of concealment. Those claims would require separate evidence.

Practice 9: What Would You Check Next?

Name three useful next checks after finding funnel-plot asymmetry.

Answer: Examples include checking whether small and large studies used different methods or conditions, examining study registrations and protocols for missing results, and reviewing the search strategy for unpublished or hard-to-find evidence.

Practice 10: Best Conclusion

Which is stronger scientific writing: “The empty side proves publication bias” or “The asymmetry is consistent with possible missing evidence, but other small-study explanations should also be examined”?

Answer: The second. It states what the evidence supports and preserves plausible alternatives.

Delayed Independent Return

Tomorrow, without rereading this article, draw an upside-down funnel shape and place twelve dots inside it. Make the lower-left side sparse. Under the drawing, write four different reasons the asymmetry might occur. One must involve missing evidence; the other three must not.

Then add one sentence beginning, “To distinguish these explanations, I would check…” If you can name useful evidence rather than merely repeat “more research”, you have transferred the habit.

For Parents and Tutors: The Pattern–Cause–Test Routine

Do not begin by teaching the phrase “publication bias”. Begin with the reasoning move.

Pattern: Ask the learner to describe the dots without explaining them.
Cause: Ask for at least three possible reasons for the pattern.
Test: Ask what extra evidence would make one explanation more likely than the others.

A learner who says “asymmetry means publication bias” has memorised a shortcut. A learner who says “asymmetry makes me investigate missing evidence, method differences, genuine study differences and chance” is doing inquiry.

For a simpler transfer exercise, replace funnel plots with everyday evidence. Put six red counters mostly on one side of a page and four blue counters elsewhere. Ask, “What happened?” The learner should realise that the pattern itself does not reveal the process that created it. Someone might have removed counters; the counters might have been dropped differently; the groups might have come from different containers; chance might matter. Then return to the scientific plot.

The purpose is not to make a child suspicious of all research. It is to build disciplined trust: understand what a representation can show, know what it cannot prove alone, and look for additional evidence before making a strong causal claim.

Why This Matters Beyond Funnel Plots

The same habit appears everywhere in science. A cluster on a map does not identify its cause. A before-and-after image does not prove the treatment caused the change. A sensor signal does not prove only one substance triggered it. A trend break does not prove the environment changed rather than the measuring system. A headline does not prove the underlying study supports every word in the headline.

Funnel plots are simply a demanding version of a core Primary Science habit: observe first, infer carefully, preserve alternatives, then ask what evidence would separate them.

Authoritative Sources and Further Reading

The Quiet Habit to Keep

When a scientific picture looks as though it is pointing to one obvious cause, do not argue with the picture and do not surrender to it. Name the pattern. Ask which explanations can make that pattern. Then look for evidence that separates them.

A funnel plot may tell you, “Something here deserves a closer look.” That is already valuable. The mistake is making it say, “I have proved exactly what happened,” when the evidence has not yet earned that sentence.

Small Group Tutorials