A seed lot can have a shorter average germination time among the seeds that germinated, yet produce fewer germinated seeds by a useful deadline. The two statements can both be correct. They describe different denominators and different questions.
To compare germination fairly, keep three pieces together: how many seeds began, when the declared event was observed, and how many remained unresolved when observation ended. A timing summary calculated only from successful events should not quietly become a claim about every seed in the lot.
This guide develops a focused comparison within The Gold Standard of Botany. It complements the existing guide to germination percentages, sampling and confidence intervals by working through an explicit ranking reversal and the meaning of unresolved times. All data are fictional teaching records. The wider educational purpose is learning that transfers into life, connected through How X Works.
eduKateSG · The Gold Standard · Botany
Choose the question you need to resolve
Start with the biological event, then decide which seeds a timing summary actually describes.
Full chapter contents · Transfer checks · Botany subject guide
Chapter contents
Define the observation · Chapters 1–3
Compare the same seeds · Chapters 4–6
Handle unresolved outcomes · Chapters 7–9
Practise and continue · Chapters 10–12
CHAPTER 1 OF 12 · Define the observation
1. Define the event before starting the clock
For a classroom observation, the event might be the first appearance of a visible radicle meeting a declared size criterion. A formal seed test may classify normal seedlings using a specified method. Germination, normal seedling development and emergence through soil are related but different observations.
Write the event definition before recording data. Include the starting point of elapsed time, observation schedule, environmental conditions and final planned observation. A clock beginning at wetting is not automatically comparable with one beginning after a pretreatment or transfer.
If seeds are checked once each day, an event first seen on day 3 usually occurred sometime after the day-2 inspection and before or at the day-3 inspection. The recorded day is an observation boundary, not a directly observed exact instant. That distinction becomes important when comparing schedules or fitting a detailed time model.
For the calculations below, use the inspection day as a consistent descriptive convention. This makes the arithmetic easy to check. We will later identify the interval uncertainty it leaves. The numbers do not prescribe test temperatures, durations or treatments for an actual species; those choices require suitable botanical methods and, where applicable, current official testing rules.
Let N be the number of seeds placed in the test and nᵢ the number newly meeting the event criterion at observation time tᵢ. The cumulative count by time t is the sum of the new-event counts up to that time. Dividing it by N gives the observed cumulative fraction of the original test group.
The final observed germination percentage is G/N × 100, where G is the total number observed germinating by the planned end. This is an observation-window statement. It does not establish that every seed outside G is dead or incapable of later germination.
A common descriptive mean germination time is MGT = Σ(nᵢtᵢ)/Σnᵢ. Its denominator is G, not N. It describes the recorded event times among the seeds that germinated within the window. A seed without an observed event contributes neither a recorded time nor a weight to that mean.
Neither denominator is intrinsically wrong. Each belongs to a different question. The problem begins when an average among observed germinants is presented as the average waiting time of all starting seeds. Naming the selected population is a small writing change with a large effect on the honesty of the conclusion.
CHAPTER 3 OF 12 · Define the observation
3. Read newly germinated counts and cumulative counts separately
A timing table can store either new events at each inspection or the cumulative total. Mean-time arithmetic requires the new-event counts. Using cumulative counts as weights counts early seeds repeatedly and produces a different, generally misleading summary.
Suppose new counts on days 1, 2 and 3 are 2, 3 and 5. The corresponding cumulative counts are 2, 5 and 10. The descriptive mean recorded time is (2 × 1 + 3 × 2 + 5 × 3)/10 = 2.3 days. Using 2, 5 and 10 as weights gives 42/17 ≈ 2.47 days and silently treats the record as though 17 events occurred.
Check the conservation statement: the sum of new events should equal the final cumulative count. If a seed can satisfy the event only once, a cumulative series cannot decrease without a documented correction, reclassification or change in the outcome definition.
For a useful record, retain the raw inspection counts and explain any corrections. The graph can be reconstructed later, and another reader can discover whether a calculation accidentally used the wrong type of count. This is often more valuable than reporting a timing coefficient to several decimal places.
Two fictional lots each begin with 100 seeds and are observed through day 7 under the same declared conditions. Lot A has 20 events on day 2 and no further observed events. Lot B has 40 on day 3 and another 40 on day 5.
| Lot | Starting seeds | New events | Observed events by day 7 | Mean recorded time among those events |
|---|---|---|---|---|
| A | 100 | 20 on day 2 | 20/100 = 20% | 2.0 days |
| B | 100 | 40 on day 3; 40 on day 5 | 80/100 = 80% | 4.0 days |
By MGT, the observed germinants in A were faster: their mean is two days rather than four. By the proportion of starting seeds producing an event by day 5, B is ahead: 80% rather than 20%. Nothing contradictory has happened. One comparison selects the seeds that produced events; the other retains the entire starting population.
An honest report says both. “Lot A’s 20 observed germinants had a shorter mean recorded time, while lot B produced substantially more observed germination by day 7.” Calling A simply “the better lot” suppresses the outcome most readers may care about. A practical deadline should be defined before choosing the summary used to judge it.
CHAPTER 5 OF 12 · Compare the same seeds
5. Distinguish two meanings of median time
The median event time among the seeds observed germinating asks when half of G had produced events. In lot A, half of G is 10 seeds, reached by the day-2 inspection. In lot B, half of G is 40 seeds, reached by the day-3 inspection.
The time to observed germination of half the original seeds asks when the cumulative count reaches N/2. Lot A never reaches 50 events within this record, so that time is not reached by day 7. Lot B reaches 50 between its day-3 and day-5 inspections.
The available observations therefore support different statements: B’s original-population halfway point falls within the interval after day 3 through day 5, while A’s does not occur within the observation window. A linear interpolation could supply a descriptive estimate within B’s interval, but would add an assumption about the distribution of events between inspections.
Labels such as T50 can refer to different definitions in different contexts. Read the equation or method instead of assuming the name identifies the denominator. A graph that rescales each curve to its own final observed percentage may make both curves reach 100%; it then shows progress relative to each lot’s observed germinants, not the fraction of all starting seeds that germinated.
CHAPTER 6 OF 12 · Compare the same seeds
6. Show what changes when observation continues
Continue the fictional observation of lot A. On day 10, another 60 seeds meet the event criterion. The mean recorded time among A’s now 80 observed events becomes (20 × 2 + 60 × 10)/80 = 8.0 days. Its observed germination percentage becomes 80%.
The early day-2 observations have not changed. The selected group included in the mean has expanded. A timing summary based on observed successes can therefore shift substantially when slow outcomes are added later.
This does not justify extending a test only when the result is disappointing. A predeclared observation rule preserves comparability. If an exploratory extension is scientifically useful, report both the original endpoint and the extension, with their reasons and methods. Do not present the later result as though it came from the original plan.
The example also shows why “eventual germination” can be an overstatement. At day 7 we know the observed total by day 7. We do not know an infinite-time total. A fitted asymptote is a model-based quantity, distinct from the final observed count. Give both their proper names so that an estimate does not quietly become a directly witnessed outcome.
CHAPTER 7 OF 12 · Handle unresolved outcomes
7. Treat unresolved seeds as unresolved
At the planned end, a seed without an observed event has a timing record that extends at least to that boundary. For time-to-event analysis it may be represented as right-censored there, rather than assigned a made-up germination day. Observation between inspections also introduces interval censoring for seeds that do germinate.
Censoring describes the observation, not the physiological cause. A nongerminated seed may be dormant, nonviable, exposed to unsuitable conditions or simply slower than the window. Follow-up viability assessment or a suitable dormancy investigation addresses a different question and needs its own evidence.
Research on time-to-event analysis in germination and emergence explains why these observation structures deserve explicit treatment. Some seeds may never experience the event, so a model assuming that all eventually germinate can be inappropriate.
Do not solve the problem by giving every unresolved seed the last observation time. That asserts an event the observer did not see. Nor should unresolved seeds silently disappear from a population-level conclusion. Preserve their count and status, and choose an analysis that matches the scientific question and plausible biological processes.
CHAPTER 8 OF 12 · Handle unresolved outcomes
8. Match the comparison to the decision
If the question is “How many seedlings can be expected by a particular deadline under these test conditions?”, cumulative percentage at that declared deadline is directly relevant. If the question concerns speed among seeds that germinated, a conditional timing summary may help. If the entire event-time process matters, analyse the curve and unresolved outcomes together.
The experimental design matters as much as the summary. Seeds within a tray share moisture, position and other conditions. One tray per treatment can confound the treatment with the tray. A hundred seed observations do not automatically create a hundred independent treatment replicates.
Use appropriate replication and allocation for the intended comparison. State the experimental unit, lot sampling, blocking where relevant, observation schedule and handling of missing records. A damaged or lost seed may require a different status from an intact seed observed without germination.
The existing guide to fair plant-growing experiments develops the control logic. This article’s denominator example is deterministic arithmetic, not a statistical proof that one real treatment is superior. A real comparison also needs variation and uncertainty to be represented at the level supported by its design.
CHAPTER 9 OF 12 · Handle unresolved outcomes
9. Report a compact account another reader can use
A useful result can fit into a few sentences. State the event, starting count, conditions and endpoint. Then report observed germination and any timing summary with its denominator. Preserve the number unresolved rather than compressing the whole outcome into a speed ranking.
For our example: “By the day-7 inspection, 20 of 100 seeds in A and 80 of 100 in B met the declared event criterion. The mean recorded event times among these observed germinants were 2.0 and 4.0 days respectively. Eighty A seeds and twenty B seeds remained without observed events at day 7. Physiological status was not established by this record.”
If observations occur daily or less often, describe the timing convention. Calling a mean “recorded inspection time” avoids implying that every event instant was measured continuously. If interpolation or a fitted event-time model is used, name it and retain the underlying counts.
The practical/evidence check is simple: can the reader recover both the speed account and the yield-by-deadline account from the report? If one is missing, the statement may invite a conclusion the data do not establish. Clear reporting keeps a useful result usable when it moves into a classroom, seed-bank discussion or future experiment.
CHAPTER 10 OF 12 · Practise and continue
10. Transfer checks with explained answers
Change the denominator. Lot C has 30 observed events, all on day 1, out of 100 starting seeds. Lot D has 90 events, all on day 2. Which is faster? C’s observed germinants have the shorter recorded time. Which has more observed events by day 2? D. A complete answer keeps those questions separate.
Change the target. In a 40-seed test, 12 events are observed by day 3 and 16 by day 6, with no more by day 10. Is the original-population median reached? No: half the original group is 20. A median among the 16 observed germinants answers a different question.
Change the schedule. One group is inspected every day and another only on days 3 and 6. Can their recorded means be compared as exact event-time means? No. The differing inspection intervals contribute different rounding and interval uncertainty. Preserve those intervals or use a consistent justified approach.
Change the missing status. Ten seeds are lost after a tray spill. Are they ten confirmed failures? No. The loss must be recorded and handled under a stated analysis, with sensitivity checks if needed. The observer did not witness ten physiological failures to germinate.
Is mean germination time a bad measure? No. It is useful when its selected population, timing convention and endpoint are explicit. The mistake is asking it to describe seeds with no observed event time.
Is an ungerminated seed dead? Not from that observation alone. Dormancy, environment, slow germination and nonviability require different evidence. The result is “not observed germinating under these conditions by this time”.
Can a curve be normalised to its final count? Yes, if that serves a declared question. Label it as the fraction of the final observed germinants and retain the original-population curve or counts. Otherwise normalisation can conceal a low observed total.
Does a censored seed receive an estimated event time? Not automatically. Its record supplies a boundary on what was observed. Model-based estimates must be identified as estimates, with appropriate assumptions.
Is laboratory germination the same as field emergence? No. A field seedling must encounter and survive additional conditions. A controlled test supports a claim at its own method and observation scale.
CHAPTER 12 OF 12 · Practise and continue
12. Connected reading and source boundaries
Understand the plant system Return to Botany and the botany mechanism guide. Continue into seed banks, dormancy and germination evidence for the biological context.
Build a fair quantitative comparison Read germination percentages, sampling and confidence intervals alongside fair plant-growing experiments. The first supplies wider statistical foundations; the second helps make treatment comparisons interpretable.
Research references and the next useful check The unified time-to-event framework and Germination Data Analysis by Time-to-Event Approaches provide specialist methodology. Links and source records were checked on 21 October 2026. The synthetic ranking examples and arithmetic here are educational; they do not certify a seed lot or replace an official testing procedure.
For the next germination claim you read, write down the denominator beside every timing measure. Then ask what remained unobserved at the endpoint. That is a small, transferable act of scientific reading. Continue through The Gold Standard of Education or How X Works.