Two microbial count records can produce the same pooled concentration while disagreeing about how the observed colonies are distributed between assessed volumes. To examine that disagreement, we need a model of variation, not just another average. Under a declared independent Poisson model with one common concentration, conditioning on the total count turns the allocation into a binomial question.
This guide works from original fictional laboratory records supplied on paper. It gives no instructions for collecting, growing or handling microorganisms. The calculations test compatibility with a specific count model; they do not establish water safety, regulatory compliance, a cell count or the cause of a disagreement. The central lesson is how to preserve information that a pooled result leaves out.
eduKateSG · Water Microbiology
Find your next learning step
Keep the total count and its allocation visible before interpreting one pooled estimate.
ROUTE 1 · CHAPTERS 1–4
Define the reported records
Separate represented water, counted units and the assumed variation model.
ROUTE 2 · CHAPTERS 5–9
Derive the allocation
Preserve the total, remove the common rate and define a tail question.
ROUTE 3 · CHAPTERS 10–14
Calculate and challenge
Check exact probabilities, sample size, estimation, recovery and dependence.
ROUTE 4 · CHAPTERS 15–18
Change the evidence
Interpret zeros, extend to several records and solve a new exposure case.
ROUTE 5 · CHAPTERS 19–23
Practise and continue
Report bounded claims, repair the reasoning and follow the biological context.
Chapter contents
Define the reported records · Chapters 1–4
1. Ask what pooling has left unanswered
2. Reconstruct represented original-water volume
Derive the allocation · Chapters 5–9
5. Preserve the total and the allocation separately
6. Derive the conditional allocation in words
7. See the cancellation algebraically
Calculate and challenge · Chapters 10–14
10. Calculate the small-count tail exactly
11. Explain why the same proportions can carry different evidence
12. Keep estimation and model checking in their proper roles
Change the evidence · Chapters 15–18
15. Handle zero counts without inventing certainty
16. Extend to three or more records
Practise and continue · Chapters 19–23
19. Report the result in a bounded sentence
20. Repair and stabilise the learner's reasoning
21. Check which rescalings preserve the allocation question
CHAPTER 1 OF 23 · Define the reported records
1. Ask what pooling has left unanswered
Imagine two reported count records from one fictional water sample. The records represent different amounts of original water, even though the reported assessed liquid volumes might look similar after dilution. A pooled concentration combines the total count with the total represented original-water volume.
That pooled number answers a scale question: under a common-rate model, what concentration best matches the overall count? It does not by itself answer an allocation question: are the separate counts distributed in a way that is compatible with the same concentration and the declared exposure amounts?
Consider the distinction before reaching for a statistical calculation. If one record represents three times as much original water as the other, equal expected counts would not follow from a common concentration. Random realised counts can differ from their expectations, so a mismatch also cannot be judged simply by demanding an exact three-to-one observed ratio.
Our task sits between those two mistakes. We will neither discard the allocation by pooling immediately nor treat any departure from an exact ratio as proof of a problem. Instead, we declare a variation model, derive the distribution it predicts and ask a carefully bounded probability question. A finished answer explains what that question can reveal and what remains outside it.
CHAPTER 2 OF 23 · Define the reported records
2. Reconstruct represented original-water volume
Call the amount of original water represented by record i its exposure eᵢ. In a supplied dilution account, it equals the assessed volume multiplied by the fraction of original sample in that assessed liquid. This is bookkeeping on already reported records, not a laboratory preparation procedure.
For an invented example, an assessed amount of 0.10 mL with original-sample fraction 1/100 represents 0.001 mL of original water. An assessed amount of 0.30 mL at the same reported fraction represents 0.003 mL. The exposures have ratio one to three.
Write those amounts explicitly before interpreting counts. A dilution factor of one hundred is not itself an exposure and is not measured in millilitres. Confusing the factor with its reciprocal can alter a concentration by several orders of magnitude while leaving the final unit superficially familiar.
The USGS indicator-bacteria reference provides the professional method context for enumeration and reporting. Our fractional-volume arithmetic is an original teaching account based on stated record fields. A real result must retain its method, validity conditions and laboratory interpretation. Exposure is one part of that record; it cannot turn an unsuitable count into a valid one merely by supplying a denominator.
CHAPTER 3 OF 23 · Define the reported records
3. Distinguish counted units from individual cells
The supplied count is a count of reported colony-forming units under a stated method. It is not automatically the number of individual living cells in the original water. A cluster, incomplete recovery or another method-dependent feature can affect the relationship between biological entities and observed units.
This distinction matters even in a probability model. We can define a concentration λ of count-generating units per millilitre for the exercise without claiming that λ is a direct census of cells. The model’s observable is the reported count. Its connection to a biological quantity needs separate evidence.
The existing Yeast article examines cell multiplication, cluster breakup and colony counts in detail. In particular, it already explains why pooling counts by represented source volume differs from taking an unweighted average of concentrations. Our new question begins after that bookkeeping: does the allocation across records fit a single supplied count-rate model?
For a learning check, ask students to complete two sentences. The first says what the instrument or method counted. The second says what biological quantity someone wants to infer. If they use the same noun for both without justification, pause there. More sophisticated statistics cannot repair an unexamined change in the meaning of the thing being counted.
CHAPTER 4 OF 23 · Define the reported records
4. State the common-concentration model
Let C₁ and C₂ be the two reported counts. Our null model assumes they are independent Poisson random variables with means λe₁ and λe₂. The same unknown λ applies to both records, and the exposures are treated as known positive quantities.
Each part of that sentence matters. Common concentration gives the shared λ. Exposure determines how much expected count each record receives. Independence specifies the relationship between the random records. The Poisson assumption supplies a particular distribution of variation around each mean.
NIST’s probability references define the Poisson and binomial distributions used in our derivation. They do not establish that every microbial dataset follows this model. Here the distributional assumptions are declared for the classroom question and must be examined before any real application.
A useful comparison is a pair of fictional records with mean counts three and nine. The model does not require every realised pair to be exactly three and nine. It assigns probabilities to possible nonnegative integer counts. Observing six and six is therefore possible, even though it differs from the expected allocation. To judge how unusual an allocation is, we must state which outcomes are being compared and how the total count is treated.
CHAPTER 5 OF 23 · Derive the allocation
5. Preserve the total and the allocation separately
Define the total count N = C₁ + C₂. A separate variable, say J = C₁, tells us how that total is allocated to the first exposure. Knowing both N and J recovers the full pair: the second count is N − J.
Knowing only N does not. With N = 12, the pair could be three and nine, six and six, zero and twelve, or another nonnegative pair summing to twelve. All have the same total represented volume if the exposures remain 0.001 and 0.003 mL.
The pooled estimate N/(e₁ + e₂) is then 12/0.004 = 3,000 count units per millilitre for every one of those pairs. That equality does not make the records interchangeable. It shows exactly which information the pooled estimate omits.
For learners, list several pairs with the same total before introducing probability. Ask what changes and what stays fixed. The total describes overall scale, while the allocation describes the distribution between known exposures. This simple decomposition makes the later conditional calculation less mysterious: we hold the observed total fixed and examine the remaining information about its split. We are changing the question deliberately, not pretending that the discarded scale information never existed.
CHAPTER 6 OF 23 · Derive the allocation
6. Derive the conditional allocation in words
Under the common-concentration model, a larger exposure has proportionally more expected count. Once the total N is fixed, the fraction assigned to the first record is governed by its share of the combined exposure. That share is p = e₁/(e₁ + e₂).
For exposures 0.001 and 0.003 mL, p = one quarter. Conditional on N = 12, the first count follows a binomial distribution with twelve allocations and probability one quarter. Its expected value is therefore three, and the expected second count is nine.
The unknown concentration λ has disappeared from this allocation question. Increasing λ changes how likely different total counts are, but it does not change the conditional exposure share when the same λ multiplies both exposures. That cancellation is what makes the method useful for checking a common concentration without first knowing its value.
This is a result of the particular independent Poisson model. It should not be turned into a general rule that any two counts become binomial once they are added. Dependence, different recovery processes or a different distribution can change the result. The derivation below shows where the assumptions enter, so that a learner can recognise which step no longer follows after the model changes.
For nonnegative integers j and N − j, independence makes the joint probability the product of two Poisson probabilities. Its numerator contains exp[−λ(e₁ + e₂)], λ raised to N, e₁ raised to j and e₂ raised to N − j. Its denominator is j! times (N − j)!.
The total N has a Poisson distribution with mean λ(e₁ + e₂). Dividing the joint probability by the probability of that total cancels the exponential and λ-to-N factors. The remaining expression is N!/[j!(N − j)!] times pʲ(1 − p)ᴺ⁻ʲ, where p is the first exposure share.
That is the binomial probability formula. One can also verify the total distribution by summing the joint probabilities across all allocations and using the binomial expansion of (e₁ + e₂)ᴺ. The cancellation is therefore tied to a normalised probability account, rather than an informal assertion that λ should disappear.
The derivation assumes a positive combined exposure and a model under which the observed total can occur. If λ were exactly zero, a positive observed total would have zero probability under that model. For ordinary positive-total exercises we consider positive λ, then condition on the observed N. Domain conditions keep even a familiar formula attached to a meaningful probability question.
CHAPTER 8 OF 23 · Derive the allocation
8. Compare two original records with one pooled estimate
Our first invented pair is C₁ = 3 and C₂ = 9, with exposures 0.001 and 0.003 mL. The second pair is C₁ = 6 and C₂ = 6 with exactly the same exposures. Both have N = 12 and pooled estimate 3,000 units/mL.
The first pair matches the conditional expected allocation. The second places half the count in an exposure representing only one quarter of the combined original-water volume. That makes it a useful example for a prespecified question about an excess in the first record.
| Fictional pair | First count | Second count | Total | Pooled estimate |
|---|---|---|---|---|
| A | 3 | 9 | 12 | 3,000 units/mL |
| B | 6 | 6 | 12 | 3,000 units/mL |
We should not jump from that description to a causal verdict. Random variation can produce unequal allocations. The next task is to calculate the probability of an outcome at least as extreme in the direction chosen before inspecting the counts.
Matching an expectation exactly is also not proof of the model. Many incorrect biological or measurement explanations could generate the same two numbers. Pair A is useful for understanding the expected split; it does not certify homogeneity, independence or method validity. Both pairs need the assumptions stated beside their pooled estimate.
CHAPTER 9 OF 23 · Derive the allocation
9. Define the tail question before calculating it
For this exercise, specify in advance that the concern is an excess of counts in the first exposure. With N = 12 and p = 0.25, the relevant upper-tail question for observing six is P(J ≥ 6). It includes six, seven and every higher possible first count through twelve.
The probability of exactly six is a different quantity. Using only that single outcome would omit more extreme outcomes in the selected direction. A tail probability describes a set of outcomes, so the set needs to be stated before its numerical result is meaningful.
Choosing the direction after seeing which record looks high changes the procedure. A real analysis may need a two-sided test, a specified ordering of allocations or a broader review across several records. There is no universal instruction to double whichever probability looks smallest without defining the intended test.
Our prespecified direction is a teaching condition, not a claim about how every laboratory should investigate a count disagreement. It lets us calculate one exact finite sum transparently. A complete explanation names the model, the observed total, the exposure share and the tail event. Those four items identify the probability being calculated much more clearly than the bare label statistically significant.
CHAPTER 10 OF 23 · Calculate and challenge
10. Calculate the small-count tail exactly
For J distributed as Binomial(12, 1/4), add the terms for j = 6 through 12. Each term is choose(12, j) multiplied by (1/4)ʲ and (3/4)¹²⁻ʲ. The exact sum is 456,359/8,388,608, approximately 0.0544022.
This means that, under the declared common-concentration model and conditional on total twelve, the specified upper-tail event has probability about 5.44 per cent. It is not the probability that the model is true. It is not the probability that the water is safe. It is also not a direct probability that a particular laboratory error occurred.
A useful independent arithmetic check computes all thirteen allocation probabilities, confirms that they sum to one, and then sums the selected upper tail. This catches an omitted endpoint or a binomial coefficient entered incorrectly. No normal approximation is required for this small finite calculation.
If a classroom exercise supplies a five per cent decision threshold in advance, this tail lies just above it. That procedural outcome should not become a claim that no issue exists. A nearby probability and a rigid label do not erase the observed mismatch or its uncertainties. The more useful result is the transparent probability account and the next evidence question it motivates.
CHAPTER 11 OF 23 · Calculate and challenge
11. Explain why the same proportions can carry different evidence
Now multiply both counts in pair B by ten, keeping their equal split: C₁ = 60 and C₂ = 60, so N = 120. Keep the same one-to-three exposure ratio. Conditional expectation for the first count is now thirty, while the observed first count is sixty.
The upper-tail probability P[J ≥ 60] under Binomial(120, 1/4) is approximately 3.42 × 10⁻⁹. The observed first-count fraction is still one half, but the amount of count information is different. A fixed departure in proportion becomes much harder to attribute to this model’s random allocation when the total is much larger.
This is an original mathematical contrast, not a recommendation to reinterpret any invalid or overcrowded laboratory count as a valid large sample. The exercise assumes that all supplied counts are legitimate observations for the declared model. Real method restrictions must be checked before the probability calculation is applied.
The comparison also prevents the opposite mistake: a small dataset with an inconclusive tail does not establish that the underlying proportions are equal. Limited information can leave substantial alternatives compatible with observation. The total count and the exposure ratio both matter. Reporting only the two normalised concentrations hides the quantity needed to understand this change in evidence.
CHAPTER 12 OF 23 · Calculate and challenge
12. Keep estimation and model checking in their proper roles
Under our common-rate independent Poisson model, the pooled concentration estimate is λ̂ = N/(e₁ + e₂). The likelihood contains λᴺ times exp[−λ(e₁ + e₂)], multiplied by factors not involving λ. Maximising that λ-dependent part gives the pooled estimate when N is positive.
The conditional allocation probability asks a different question. It removes λ to examine whether the split between exposures fits the shared-rate assumption. A pooled estimate can therefore be straightforward to calculate even when the allocation gives reason to question the model behind pooling.
This distinction is why our two twelve-count pairs share the same estimate but have different allocations. The pooled number has not made a numerical error. It is simply not a complete description of the data or of their compatibility with the assumed model.
A finished report can contain both pieces: the estimate under the stated model and the evidence relevant to that model’s adequacy. It should not present the estimate as an unconditional biological fact after ignoring a material incompatibility. Nor should a model-checking concern automatically justify replacing the data with a more convenient pair. Preserve the records, state the assumptions and identify the next review question rather than allowing one summary number to perform every task.
CHAPTER 13 OF 23 · Calculate and challenge
13. Introduce known recovery differences explicitly
Suppose the exercise additionally supplies known independent detection or recovery multipliers r₁ and r₂. Under a modified model, expected counts are λr₁e₁ and λr₂e₂. The conditional first share becomes r₁e₁/(r₁e₁ + r₂e₂).
For e₁ = 0.001 and e₂ = 0.003 mL, let the supplied multipliers be one and one half. The effective exposures are then 0.001 and 0.0015 mL, giving p = 0.4 instead of 0.25. A calculation that ignores this declared difference tests a different null model.
This extension is mathematical bookkeeping for supplied known multipliers. It is not a claim that real recovery can be reliably corrected with an arbitrary constant. Unknown, variable or dependent recovery introduces additional uncertainty and may require a different analysis.
The example shows what the conditional method actually identifies: compatibility with a common λ after the stated exposure and recovery structure has been applied. A strange allocation may reflect concentration differences, a wrong exposure, a wrong recovery assumption or another departure. It does not uniquely identify which component failed. Naming a candidate mechanism requires evidence that distinguishes it from the alternatives, rather than selecting the most familiar explanation after seeing a small tail probability.
CHAPTER 14 OF 23 · Calculate and challenge
14. Examine dependence before using the product rule
The derivation multiplied the two Poisson probabilities because the records were assumed independent. Merely assigning two labels does not establish independence. Shared preparation uncertainty, a fluctuating common factor or repeated readings of the same underlying observation can connect the records.
Consider a simple paper example in which the second reported count is copied from the first. The pair can still be written as two numbers, but it does not contain two independent outcomes. A probability calculation built from an independent product would describe a fictional mechanism different from the one that generated the record.
More subtle dependence can arise without literal duplication. If both records share an uncertain dilution factor, their exposure errors are linked. Conditioning on the total does not automatically remove every kind of shared uncertainty. Our algebra removes the common concentration parameter under one exact model, not all possible common causes.
For learners, draw the provenance of each number: which part is separately observed, which is shared and which is inferred. Then point to the independence step in the derivation and ask whether the record supports it. This is a meaningful evidence check because it connects an assumption with a specific operation in the mathematics. A generic request for more data would not make that connection as clearly.
CHAPTER 15 OF 23 · Change the evidence
15. Handle zero counts without inventing certainty
If both counts are zero, N = 0. The conditional allocation is degenerate: there are no observed units to split. It provides no allocation evidence for distinguishing the two positive exposures. That does not mean the underlying concentration is known to be zero.
A separate total-count model can address what zero observations imply about a concentration under its assumptions. That is a different inference from the conditional compatibility question. Mixing the two can make the empty allocation look more informative than it is.
If the first count is zero and the second positive, the upper-tail question about an excess in the first record is not the appropriate way to investigate a shortage there. The direction and outcome set still need to match the prespecified question. Zeros are ordinary possible count values, but they do not make the choice of test irrelevant.
A practical classroom check supplies the same zero observation with two different total exposures. Ask which part of the information changes. The conditional allocation with N = 0 remains uninformative in both cases, while the total-count evidence about λ can differ. The lesson is to keep the inferential target visible. No-detection statements, concentration estimates and allocation checks answer related but distinct questions.
For independent Poisson counts Cᵢ with means λeᵢ across several positive exposures, conditioning on their total gives a multinomial allocation. Each exposure receives probability pᵢ = eᵢ divided by the total exposure. The probabilities sum to one.
As an original example, take exposures in ratio one to two to three and total count twelve. The conditional expected allocation is two, four and six. A different realised vector can share the same total and pooled concentration while departing from that pattern.
There are several possible ways to define a discrepancy across a multinomial vector. The analyst must specify which statistic or ordered set of outcomes forms the test. Selecting the most surprising individual pair after searching many pairs introduces another procedural choice that needs to be acknowledged.
For an introductory learner, the important extension is the probability account rather than a new memorised test name. The total controls the number of observed units, while the exposure shares control their expected allocation. The same cancellation of a common λ follows from the independent Poisson structure. Before calculating a complex tail, have students verify that every pᵢ is positive and their sum is one. Those checks preserve the model’s basic meaning as the number of records grows.
CHAPTER 17 OF 23 · Change the evidence
17. Solve an independently changed exposure case
Use a new fictional pair with represented exposures 0.002 and 0.003 mL. Let the reported counts be eight and two. Their total is ten, and the common-rate conditional first share is p = 2/5, not one quarter.
The pooled estimate under the model is ten divided by 0.005, or 2,000 units/mL. Conditional expected counts are four and six. For a prespecified concern about excess in the first record, calculate P[J ≥ 8] under Binomial(10, 0.4).
The exact sum includes j = 8, 9 and 10. It is 120,064/9,765,625, or 0.0122945536. That is about 1.23 per cent for the specified event under this new model. The calculation uses the changed exposure share and changed total rather than recycling the first example’s probability.
A complete answer then returns to scope. This small tail is evidence against the joint declared assumptions in the selected direction. It does not identify a pathogen, a safe concentration or a unique error. The changed case tests both the arithmetic and the restraint needed to explain what an exact probability does and does not establish about a real water record.
CHAPTER 18 OF 23 · Change the evidence
18. Ask what evidence could distinguish the causes
Suppose the allocation looks incompatible with the common-rate model. Several explanations remain possible: the source represented by the records may not be comparable, the reported exposure may be wrong, recovery may differ, dependence may have been ignored or the Poisson variation model may be unsuitable.
A useful next question identifies evidence that would separate these possibilities. A record review might establish whether the dilution fraction was transcribed consistently, whether the reported method was the same or whether the counts refer to distinct observations. These are questions for the responsible laboratory and appropriate professional process, not instructions for a learner to repeat unknown-water work.
The distinction between detecting disagreement and explaining it is central. A probability calculation can flag tension within a model without deciding which assumption should be revised. Choosing a cause requires additional information with a relevant connection to that cause.
For classroom practice, provide two fictional provenance notes that produce the same count pair: one has an explicitly changed exposure, the other a copied count. Ask how each note changes the model before recalculating. The exercise makes evidence actionable. Learners can point to the assumption affected by a record, rather than treating every unusual pair as the same generic problem with the same generic remedy.
CHAPTER 19 OF 23 · Practise and continue
19. Report the result in a bounded sentence
A useful result sentence might say: under the supplied independent Poisson common-concentration model, with known exposures in ratio one to three and total count twelve, the probability of six or more counts in the first record is approximately 0.0544. That sentence identifies the event and the conditions without overclaiming.
Add the observed pair, exposure units, method identifier where available and the reason the tail direction was selected. If a decision threshold was used, state that it was chosen in advance and report the actual probability as well as any procedural label.
Keep the pooled estimate on a separate line with its model qualification. A reader should be able to see that concentration estimation and allocation checking have different purposes. Missing exposure provenance or uncertain dependence should remain visible rather than being buried after an unconditional headline.
For an independent check, ask another learner to reconstruct the probability question from the prose alone. If they choose a different N, p or tail, the report needs clarification even if the original calculator output was correct. Clear statistical writing is part of verification: it allows a second person to calculate the same quantity and to recognise when the evidence does not support the assumptions that quantity requires.
CHAPTER 20 OF 23 · Practise and continue
20. Repair and stabilise the learner's reasoning
If exposure is the first gap, begin with one already reported assessed volume and one stated original-water fraction. Have the learner multiply them and explain the unit before introducing counts. This repairs the denominator that every later step depends on.
If pooling is mistaken for compatibility, use the two twelve-count pairs. Their equal pooled estimate and different allocations provide a direct counterexample. Ask what information was lost, then restore it with N and the first count.
If an expected ratio is treated as compulsory, list several possible binomial outcomes with positive probability. Randomness permits departures. The question concerns how the selected departure compares with the model, rather than whether it matches an expectation exactly.
For stabilisation, change the exposure ratio to two-to-three and use the independent worked case. For extension, introduce known recovery factors or an explicitly shared error. Learners should revise the model before revising a probability. Progress appears when they can name which assumption supports each mathematical step and which missing record prevents a justified inference. That is a more transferable skill than memorising that one particular count pair gives a small or large probability.
CHAPTER 21 OF 23 · Practise and continue
21. Check which rescalings preserve the allocation question
Convert both exposures from millilitres to litres. The values 0.001 and 0.003 mL become 0.000001 and 0.000003 L. Their share remains one quarter, so the conditional allocation probabilities are unchanged. The concentration estimate changes its numerical unit: 3,000 units/mL is 3,000,000 units/L.
This is a useful independent check because it separates a dimensionless probability from a concentration carrying volume units. A unit conversion must not alter the binomial share when it is applied consistently to every exposure. If it does, the conversion or denominator is wrong.
Now multiply both physical exposures by ten while keeping the observed counts fixed. The conditional share again remains one quarter, but the pooled concentration estimate becomes one tenth as large. These are different records with the same conditional allocation question. That does not mean they provide identical information about the underlying concentration.
For learners, compare unit conversion with a physical change explicitly. In the first operation, the same represented water is described in another unit. In the second, more represented water is assumed for the same count. The conditional ratio can remain unchanged in both while the interpretation of the concentration differs. This check reinforces why removing a common scale from one probability question should not be confused with proving that scale never matters.
Does a small tail prove that concentrations differ? It indicates tension with the declared joint assumptions in the selected direction. Exposure errors, recovery differences, dependence and other departures can also matter.
Does a large tail prove the common model? No. The observed allocation may be compatible with the model while other models remain possible, especially with few counts.
Is the pooled concentration wrong when allocation looks unusual? It is still the pooled estimate under the specified common-rate model. The concern is whether that model provides an adequate interpretation of the records.
Why condition on the total? It removes the unknown common scale λ from this allocation question. It does not make the total irrelevant to every other question about concentration.
Can this establish drinking-water safety? No. The exercise supplies neither the full required evidence nor a regulatory interpretation. Use the responsible official and laboratory routes for actual water decisions.
Can I simply double the upper tail? A two-sided procedure needs an explicit definition. Choosing a direction or ordering after viewing the data changes the question.
Should learners create their own microbial records experimentally? This guide is designed for supplied paper records and probability calculations. It provides no collection, culture or handling activity involving unknown microorganisms.
Check the probability and method context
NIST's Poisson distribution reference and binomial distribution reference support the distribution definitions. The conditional derivation, datasets, tail calculations and learning tasks here are original. The USGS indicator-bacteria field-manual record identifies professional enumeration and reporting guidance. These primary links were checked on 23 October 2026; they do not certify our fictional model for a real sample.
Preserve the biological meaning
Return to Water Microbiology for organisms, indicators, sampling and interpretation. Read Yeast: cell multiplication, cluster breakup and colony counts for the count-unit and pooling questions. Genetics: fluctuation tests, mutants and mutations examines a different inferential target: growth history, events and descendants. Shared probability vocabulary should not conceal those different biological questions.
Microbiology provides the broader subject context. Water treatment engineering connects measurement with a larger system of barriers and accountable outcomes. Follow those routes when the task extends beyond interpreting our supplied count allocation.
Transfer the evidence habit
Continue through How X Works and Learning That Transfers into Life. A useful final explanation identifies what the pooled estimate preserves, what it discards and why the allocation model needs independent assumptions. That connection between a summary and its limits travels well beyond microbial counts.