Wait, what? A biology graphic shows a branching tree of five organisms. Beside one branch is the number 95. A student says, “That means there is a 95% chance this branch is the true evolutionary history.” The statement sounds mathematically neat. It is also more certain than the evidence allows.
Bootstrap support is a useful way of asking how consistently a branch reappears when the data are repeatedly resampled and the analysis is rerun. But a support value is not automatically the same thing as a direct probability that the branch is true. To interpret it, a learner must know what was resampled, what branch the number belongs to, what method and model were used, and what other evidence agrees or disagrees.
Quick Answer
No. “Bootstrap support = 95” usually means that the branch or grouping appeared in about 95% of the bootstrap analyses under the stated procedure. It is evidence that the grouping is stable to that resampling method. It is not, by itself, a guarantee or simple 95% probability that the branch is the one true evolutionary history.
The Owned Learner Job
This article owns one evidence-transfer job: how to evaluate a support number printed on a phylogenetic tree without turning it into a stronger probability claim than the method provides.
It does not replace lessons on DNA, inheritance, evolution, classification or the general idea of a phylogenetic tree. It also does not teach statistics as a standalone topic. The object here is a real scientific communication object: a branching diagram carrying branch-support numbers.
Start With the Diagram, Not the Number
Suppose a simplified tree contains organisms A, B, C, D and E. One internal branch groups C and D together. The number 95 is printed beside that branch.
Before asking what 95 means, ask what object is labelled. The value usually belongs to a branch or grouping, not to an organism, not to the whole tree, and not to “95% DNA similarity.” A number placed near a line can be misread if the learner does not identify the line it annotates.
This is the first representation check: locate the evidence label before interpreting the evidence.
What the Bootstrap Procedure Is Testing
A phylogenetic analysis may begin with many aligned positions from DNA, RNA or protein sequences. The original dataset produces one estimated tree. Scientists then create many bootstrap datasets by resampling the original positions with replacement. Each resampled dataset is analysed again. The question becomes: how often does the same branch or grouping reappear?
If a branch appears in 950 out of 1,000 bootstrap reconstructions, the bootstrap support may be reported as 95%. That repeated reappearance is useful evidence about the stability of the branch under resampling. But it remains conditional on the original data, analytical method, evolutionary model and other choices.
Observed, Constructed and Inferred
- Observed: sequence data were collected from organisms or samples.
- Constructed: a tree was estimated using an analytical method and model.
- Resampled: many bootstrap datasets were generated from the original alignment.
- Calculated: the branch reappeared in a stated fraction of bootstrap analyses.
- Inferred: the branch has stronger or weaker support under this analysis.
- Overclaim: the number directly gives the probability that the evolutionary event really happened exactly as drawn.
That last step is the one to resist.
For the more general PSLE Science distinction between what was observed and what is inferred, route to How to Tell Observation, Inference, Prediction and Explanation Apart in PSLE Science.
Why 95 Is Not Automatically “95% True”
The bootstrap asks how stable a result is when the available data are resampled. It does not create new independent evolutionary history. All the resampled datasets are generated from the same original evidence. If the original data have a systematic limitation, use an unsuitable model or contain too little information for a difficult branch, high resampling stability does not magically remove those issues.
Experts therefore treat bootstrap support as a branch-support measure whose exact statistical interpretation requires care. The European Bioinformatics Institute’s phylogenetics training materials describe bootstrap values as confidence estimates on branches and emphasise that interpretation depends on the analysis and appropriate evolutionary model. Research literature likewise warns against casually turning bootstrap support into a direct probability that a branch is true.
The Independence Check
A common error is to say, “We made 1,000 bootstrap trees, so we have 1,000 independent studies.” We do not. The procedure repeatedly resamples the same underlying alignment. It explores sensitivity to sampling of the characters in that dataset; it does not replace independent data collection.
This is a powerful transfer idea. Repetition can increase information about stability without becoming independent replication. Ten calculations from one dataset are not automatically ten separate biological experiments.
Worked Case 1: 95 and 60 on the Same Tree
Maren sees one branch with support 95 and another with support 60. She writes, “The 95 branch is definitely true and the 60 branch is definitely false.”
That is too absolute. The 95 branch is much more consistently recovered under the bootstrap analysis than the 60 branch. This supports greater confidence in that grouping under the stated method. But the correct language is about relative evidence and support, not certainty versus impossibility.
A stronger answer is: “The branch with support 95 is more strongly supported by the resampling analysis than the branch with support 60, but the support value does not by itself prove the branch is true.”
Worked Case 2: Same Data, Different Model
Iona analyses the same sequence alignment with two reasonable models. Model X gives one branch support of 93. Model Y gives 72. She asks which number is “the real support.”
The disagreement is itself evidence. It suggests the branch support is sensitive to analytical assumptions. The learner should not hide that sensitivity by choosing the larger number. Instead, investigate why the models differ, whether one model better fits the data, and whether other forms of evidence support the grouping.
This is a model-limits habit: a scientific output inherits assumptions from the model that produced it.
Worked Case 3: A Long Branch With a High Number
Leonie sees a visually dramatic long branch with support 99. She says, “The long line plus 99 proves this organism changed the most and the branch is certainly correct.”
Two different visual variables have been mixed. Branch length may encode an amount of change or another defined quantity depending on the tree. Bootstrap support encodes resampling support for a grouping. The two should not be multiplied into one intuitive story. First read what branch length means in the figure, then separately read the support value.
Tempting Reasoning That Fails
| Tempting claim | Why it fails | Better interpretation |
|---|---|---|
| “95 means 95% chance the branch is true.” | Bootstrap support is a resampling support measure, not automatically a direct truth probability. | “The branch appeared in 95% of the bootstrap analyses under the stated method.” |
| “1,000 bootstrap trees are 1,000 independent studies.” | They are generated by resampling the same original dataset. | “They test stability to resampling of the available data.” |
| “Low support proves the branch is false.” | Low support may also reflect insufficient or conflicting information. | “The data and analysis do not strongly stabilise that branch.” |
| “Every number on the tree applies to the whole tree.” | Support values usually label particular branches or groupings. | Identify the exact branch first. |
What Evidence Would Strengthen a Branch Claim?
- Strong bootstrap support under an appropriate method.
- Similar topology from an independently collected dataset.
- Support from different reasonable analytical methods.
- More informative sequence data.
- Agreement with relevant anatomical, fossil, biogeographic or other biological evidence when appropriate.
Notice how the evidence becomes stronger when it is not all the same calculation repeated in slightly different form.
What Would Weaken or Limit the Claim?
- Low or unstable support under small changes in method.
- Conflicting signals in different genes or datasets.
- Very little informative sequence variation.
- A poor-fitting model.
- Data-quality problems, contamination or misidentified samples.
How Far Can the Conclusion Travel?
A high bootstrap value supports a statement about the stability of a particular branch under a particular resampling analysis. It does not automatically support claims about exact divergence time, exact ancestor appearance, percentage DNA similarity, certainty of the whole tree, or the probability that every evolutionary event shown really occurred.
Strong scientific reasoning keeps the scope of the conclusion attached to the scope of the evidence.
PSLE-Style Transfer Case
A simplified phylogenetic tree was built from a DNA sequence alignment. A branch grouping Species P and Q has bootstrap support 92. A student writes: “There is a 92% probability that P and Q are definitely each other’s closest relatives in nature.”
Evaluate the statement.
Reasoned answer: The statement is too strong. The value shows that the P–Q grouping was recovered in about 92% of bootstrap resampling analyses under the stated method. This is strong support within that analysis, but it is not automatically a direct 92% probability that the branch is true in nature. Other data and analytical assumptions should also be considered.
Delayed Independent Return
Now imagine a computer model classifies images and reports “95% cross-validation accuracy.” Is that automatically a 95% chance that every future classification is correct? No. The technical method is different, but the evidence habit is similar: identify what was repeated, what data were reused, what quantity the number measures and how far the conclusion can travel.
The transfer is not “all percentages are suspicious.” The transfer is “name the quantity before interpreting the percentage.”
Explained Practice
- A branch has support 100 from 100 bootstrap replicates. Does that prove the branch cannot be wrong? Explain.
- Two branches have supports 88 and 42. Which has stronger resampling support, and what should you avoid claiming?
- A paper reports high bootstrap support, but a second independent dataset gives a different grouping. Why should the disagreement not simply be ignored?
- A number 90 is printed near a branch. What should you identify before interpreting it?
Check your thinking: Strong responses mention the exact branch, repeated resampling of the same underlying data, analytical assumptions and the difference between support and certainty.
For Parents and Tutors: Teach Number Identity
Students often see a percentage and immediately translate it into “probability,” “accuracy” or “amount.” The repair is to make them name the quantity first. Ask, “Ninety-five percent of what?” In this case: of bootstrap analyses in which that branch was recovered.
Then ask a second question: “What does that not tell you?” This forces a boundary around the evidence. After guided practice, give a new graph or report containing an unfamiliar percentage and require the learner to identify its denominator and meaning independently.
Canonical eduKate Routes
- Observation, inference, prediction and explanation
- Turn diagrams, tables and graphs into evidence
- Update explanations when new evidence arrives
- Use intuition without overriding evidence
Authoritative Sources
- EMBL-EBI Training — Confidence estimates in phylogenetic trees
- Statistics and Truth in Phylogenomics — discussion of bootstrap interpretation
- SEAB — 2026 PSLE Science syllabus
- MOE — 2023 Primary Science syllabus
The Quiet Habit
When a scientific figure prints a confident-looking number beside a branch, do not let the number name itself. Ask what was counted, repeated or compared. Bootstrap support is useful because it tells us something precise about resampling stability. It becomes misleading only when we make it say more than that.
