Find your next learning step
Choose the question closest to what you need today, or read the guide from the beginning.
If a trial has a success probability of 0.6, then ten such trials have an expected success count of six when each trial has that same probability. The word expected does not mean guaranteed. It describes a model average, not a promise about the next ten outcomes.
Under an independent-trial model, the next ten trials can produce five successes, seven successes, or any count from zero to ten. Six is a useful centre for planning and comparison, but it is only one possible observed count. Randomness has not made an arithmetic mistake when the realised count differs from six.
This guide explains that distinction using original classroom examples, clear assumptions and fully answered practice. It is written for G3 readers, but the central distinction between model and observation is useful across Mathematics pathways. Optional distribution calculations are labelled as extensions rather than presented as a requirement for every G3 learner.
Contents
Sections 1–5
Sections 6–10
Sections 11–15
Sections 16–20
Sections 21–25
Expected frequency equals number of trials multiplied by the success probability when each trial has that probability. Thus 10 × 0.6 = 6 is the expected count. Observed frequency is the number of successes actually recorded. These quantities can differ in a finite experiment.
Independence is not needed merely to add individual expected counts. If ten trials each have marginal success probability 0.6, their total expected successes is six. Independence becomes important when calculating the usual binomial probabilities for particular counts, such as the probability of exactly six successes.
A probability of 0.6 also does not mean that after four failures the process owes you six successes. Under independent trials with unchanged probability, the next trial still has success probability 0.6. Past outcomes do not force the future to repair a shortfall. Keep three questions separate: what the model says, what has happened, and what follows from the assumptions about future trials.
Imagine a fictional spinner designed so that a success region covers 60% of its equally likely angular positions. Under a suitable fair-spinning model, the probability of success on one spin is 0.6. That probability describes the chance of the event before the outcome is known.
It does not describe 60% of an individual outcome. A particular spin is either a success or not a success under the stated definition. Probability is a number attached to uncertainty about that outcome, not a fractional piece of the outcome itself.
The model matters. The spinner must operate in a way consistent with the assumed outcome probabilities. A decorative region covering 60% of the surface does not alone prove that every real spinning mechanism gives success probability 0.6. Classroom problems often supply the probability directly so that the mathematical reasoning can focus on its consequences.
State what counts as success before collecting data. If the definition changes halfway through an experiment, the resulting count no longer describes repetitions of one consistently defined event. Clear event definitions are as important as clear calculations.
Frequency is a count of how often an event occurs in a specified collection of trials. If success occurs seven times in ten spins, the observed success frequency is seven. The observed relative frequency is 7/10 = 0.7. The count and the proportion are related but different quantities.
The theoretical probability in the model can remain 0.6 while the observed relative frequency is 0.7. One is a model input; the other is a summary of data. They do not have to be numerically identical after a small experiment.
Always report the number of trials with the observed count. Seven successes out of ten and seven successes out of one hundred describe very different records. A bare count leaves out the scale needed to interpret it as evidence about a probability.
Similarly, “the frequency is 60%” usually means relative frequency rather than a count. In clear school working, label the quantities explicitly: observed count seven; total trials ten; observed relative frequency 0.7. This vocabulary prevents an expected count from being confused with a probability or an observed proportion.
Expected frequency is the average count predicted by a probability model across repetitions of the whole experiment. If the experiment consists of ten trials, expectation refers to the count within each ten-trial experiment, averaged in the model over possible repeated experiments.
For constant success probability p over n trials, the expected count is np. With n = 10 and p = 0.6, the result is six. This number summarises the possible counts together with their probabilities; it does not select which count will happen next.
An average need not be an outcome that occurs. For three trials with success probability one half each, expected successes is 1.5. No individual three-trial experiment can contain one and a half successes. The decimal expectation is still meaningful because it describes a model average of whole-number counts.
That example is often the quickest way to break the mistaken link between expected and guaranteed. If expectation were a compulsory outcome, 1.5 successes would be impossible to interpret. Once expectation is understood as an average, both integer and noninteger expectations fit the same definition.
For each trial, imagine a success indicator: one if success occurs and zero otherwise. If the success probability is p, the expected value of this indicator is 1 × p + 0 × (1 − p) = p. This simply weights its two possible values by their probabilities.
The total count is the sum of these indicators. Expected values add, so the expected total is p + p + … + p across n trials, giving np. This addition rule does not require the indicators to be independent.
If the individual probabilities differ, add those probabilities instead: p₁ + p₂ + … + pₙ. For example, success probabilities 0.2, 0.5 and 0.8 give expected total 1.5. Multiplying three by one of those individual probabilities would not be justified.
The constant-probability formula is therefore a convenient special case of a broader addition principle. You do not need advanced notation to use it, but understanding the reason stops you from attaching independence to the wrong claim. Independence controls how outcomes relate to one another, which affects the distribution of the total even when its expectation stays the same.
The question is: a process has success probability 0.6 on each of ten trials. Find the expected number of successes and explain whether six successes are guaranteed. The expected count is 10 × 0.6 = 6. Its unit is successes, not a probability of six.
Six successes are not guaranteed from this information. Even when the trials are independent and the probability remains 0.6, the count is random. A possible sequence is success, failure, success, success, failure, success, failure, success, failure, failure. That sequence contains five successes.
Under the independent constant-probability model, the probability of that specific ordered sequence is 0.6⁵ × 0.4⁵, which is positive. Its existence therefore directly contradicts a claim that every ten-trial sequence must contain exactly six successes. You do not need to calculate the full distribution to disprove the guarantee.
The final answer should say: “The expected count is six, but the observed count may differ.” A complete explanation identifies both the multiplication and its interpretation. Reporting only the number six leaves the central misunderstanding unresolved.
For the same constant probability, the expected count scales with the number of trials. The actual count remains a result to be observed, not a value fixed by the multiplication.
| Trials | Success probability | Expected count | Example possible observed count under independence |
|---|---|---|---|
| 5 | 0.6 | 3 | 2 |
| 10 | 0.6 | 6 | 7 |
| 25 | 0.6 | 15 | 14 |
| 100 | 0.6 | 60 | 63 |
The observed counts in the last column are invented possible examples, not data from a real experiment and not a forecast of what a reader will obtain. Other counts are possible under the stated independent model.
Dividing expected count by trial count returns 0.6 in every row. Dividing each example observed count by its trial count gives a potentially different relative frequency. For instance, seven successes in ten trials gives 0.7, while sixty-three in one hundred gives 0.63.
This comparison also shows why a count difference alone can mislead. Being three above expectation has a different proportional meaning in ten trials and in one hundred trials. Report both the count and the total before judging the size of a departure.
A quota would require exactly six successes in a block of ten. A probability model with independent success probability 0.6 imposes no such fixed block total. Every next trial remains uncertain, and different ten-trial blocks can have different success counts.
To see the distinction, imagine a bag containing exactly six red counters and four blue counters. Drawing all ten without replacement gives exactly six reds. That is a fixed-composition situation. It is not the same experiment as independently drawing with replacement from the bag ten times.
With replacement and thorough mixing, each draw can have red probability 0.6 and the count varies. Without replacement, the probabilities conditional on previous draws change, and drawing the entire bag gives a fixed total. Both situations involve six out of ten counters initially, but they produce different models for repeated draws.
Read the mechanism rather than treating the decimal 0.6 as a complete description of every experiment. A probability for one trial alone does not specify all the dependence between trials. The structure of repetition determines which further conclusions are justified.
Under independent trials, learning the outcomes of earlier trials does not change the success probability of the next one. If that probability is 0.6, it remains 0.6 after four failures, after four successes, or after a mixed record, as long as the stated model remains appropriate.
The idea that success is “due” after several failures is a common reasoning error. It treats the expected proportion as a short-term target the process must actively restore. Independent randomness contains no mechanism that remembers a shortfall and compensates for it.
The opposite error is assuming success must continue because a run has occurred. A run can be an ordinary possible outcome of an unchanged independent process. Whether a real process has changing conditions is a separate modelling question, not something settled by the word streak.
Keep the assumptions explicit. If the process is not independent, past observations can affect conditional probabilities. Drawing without replacement is a familiar example. The correct lesson is not that history never matters. It is that history does not alter the next-trial probability in the particular independent model being used.
Suppose ten independent trials were planned, each with success probability 0.6, and the first four have all failed. Before any trials, expected total successes was six. After observing the four failures, the expected count among the remaining six trials is 6 × 0.6 = 3.6.
The expected final total conditional on those observed failures is therefore 0 + 3.6 = 3.6. It is no longer six because some uncertain outcomes have become known zeros. The process has not changed its probability; our information about the total has changed.
There is also no guarantee that the remaining six will all succeed to make the original expected count come true. Under independence, the probability of six successes in those six remaining trials is 0.6⁶, approximately 0.046656. That is a possible outcome, not a required repair.
This example separates prior expectation from an updated expectation based on observed outcomes. The first was correct before the observations. The second is correct after conditioning on them. A model can support both statements at their respective stages without inconsistency.
Imagine recording the success count from each of many independent ten-trial experiments using the same 0.6 model. One experiment might give four, another seven, another six, and another eight. The average of these counts is the quantity associated with the expectation six.
This does not require the counts to rotate through a tidy compensating pattern. A low-count experiment need not be followed by a high-count one. The sequence of experiments can fluctuate, while their average becomes more stable under the usual independent repeated-experiment assumptions.
Do not describe a finite list of invented counts as proof that the average must equal six. A classroom list can illustrate the distinction between an individual count and an average; it cannot establish a long-run theorem by itself. The probability model supplies the mathematical expectation.
If you run an actual experiment, label its data honestly and keep it separate from the theoretical calculation. A spreadsheet average of observed counts estimates or illustrates the model quantity. It is not automatically identical to the theoretical expectation, even after a sizeable finite number of repetitions.
Under an independent constant-probability model, the observed relative frequency tends toward the probability in the appropriate long-run sense. This does not mean it gets closer after every single trial. A new success or failure can temporarily move it farther from 0.6.
For example, six successes in ten trials gives exactly 0.6. If the next trial succeeds, the relative frequency becomes 7/11, approximately 0.636. It has moved away from 0.6 despite the increased sample size. A later failure would move the fraction again.
Also distinguish proportional discrepancy from count discrepancy. A count of 610 successes in 1000 trials is ten above the expected 600, but its relative frequency is only 0.01 above 0.6. A count of eight in ten is only two above expectation, but its relative frequency is 0.2 above 0.6.
Larger samples can have larger absolute count differences while having smaller relative differences. This is why “the gap must shrink” needs a clear statement of which gap is meant. Avoid turning a long-run proportion principle into an incorrect claim about every finite count.
An expected value is a probability-weighted average. A most likely value is one with the largest individual probability. They can coincide in some models, but they are different concepts. Do not define expectation as “the result most likely to happen”.
Consider one trial with success probability 0.6. Its success count is either zero or one. The most likely count is one, while the expected count is 0.6. This simple example already shows why the definitions cannot be identical.
For ten independent trials with probability 0.6, six is a most likely count as well as the expectation. Even then, being the most likely single count does not mean it has probability above one half or that it is guaranteed. The remaining probability is spread across other counts.
The optional calculation later in this guide shows that exactly six successes has probability about 0.251 under this binomial model. Other counts collectively have probability about 0.749. A single count can be the most likely category while still being much less likely than all other categories combined.
For ten independent trials with success probability 0.6 each, the count follows the usual binomial model. Exactly six successes can occur in 210 different positions among the ten trials. Each such ordered sequence has probability 0.6⁶ × 0.4⁴.
Multiplying gives P(exactly six) = 210 × 0.6⁶ × 0.4⁴ = 0.250822656. This is about 25.08%, not 60% and not 100%. The 0.6 is the probability of success on one trial, not the probability of achieving the expected count in ten trials.
The factor 210 counts arrangements. For example, successes in the first six positions and failures in the last four is only one arrangement. Omitting the arrangement count calculates the probability of that one ordered pattern, not of every pattern with six successes.
This distribution calculation is an extension for learners who have met the relevant counting or binomial ideas. It is not needed to understand why expectation is not a guarantee. The earlier positive-probability example of a five-success sequence already establishes that simpler point.
Imagine a different experiment: choose one random switch that is on with probability 0.6. If it is on, all ten trials are labelled successes; if it is off, all ten are failures. Each individual trial still has marginal success probability 0.6.
The expected total is still six, because the total is ten with probability 0.6 and zero with probability 0.4. Its expectation is 10 × 0.6 + 0 × 0.4 = 6. Yet exactly six successes is impossible in this model. The only possible totals are zero and ten.
Compare this with the independent model, where every count from zero to ten has positive probability. The same individual probability and the same expectation can coexist with very different distributions of the total. This is why independence must be stated before using binomial count probabilities.
The example is deliberately artificial to isolate the logic. It does not describe a real classroom experiment unless its mechanism actually matches that shared-switch structure. Its purpose is to demonstrate that knowing an average is not enough to know how variable the possible outcomes are.
A spinner has success probability 0.3 on each of twenty trials. Find the expected success count. Then suppose the experiment records eight successes. Find the observed relative frequency and explain whether the count disproves the model by itself.
Expected count is 20 × 0.3 = 6. Observed relative frequency is 8/20 = 0.4. The observed count exceeds expectation by two. That departure alone does not logically disprove a probability-0.3 model under independent trials, because eight successes is one possible count with positive probability.
A second question uses five trials with success probability 0.5 each. Expected count is 2.5. The observed count must be an integer from zero to five; it cannot literally be 2.5. That does not invalidate the expectation, which is an average rather than a required observation.
For both questions, write the labels beside the numbers. Six is an expected count, 0.4 is an observed relative frequency, and 0.3 is the supplied single-trial model probability. Clear labels stop different quantities from being compared as though they had the same meaning.
A process has independent success probability 0.6 on every trial. The first three trials fail. What is the probability of success on the fourth? It remains 0.6 under the stated independent unchanged-probability model. The earlier failures do not create a compensating probability increase.
If a total of ten trials was planned, what is the expected number of successes in the remaining seven? It is 7 × 0.6 = 4.2. Since the observed first three contributed zero successes, the conditional expected final total is also 4.2.
Now suppose instead the first three trials all succeeded. The remaining expected count is still 4.2, while the conditional expected final total is 3 + 4.2 = 7.2. Our information changes the expected final total, not the unchanged probability of each future independent trial.
These answers are a useful check against the idea of a fixed ten-trial quota. If the final total were forced to six, future probabilities would need to respond to the record. That would be a different model with dependence or other constraints, not the independent model specified here.
A bag contains six red and four blue counters. In the first experiment, one counter is drawn, replaced and mixed before each of ten draws. Under the usual equally likely selection model, each draw has red probability 0.6. If draws are independent, expected reds is six but the actual count can differ.
In the second experiment, all ten counters are drawn without replacement. The final count is exactly six reds because every counter is eventually drawn. That guarantee comes from the bag’s fixed contents and the decision to draw all ten, not merely from the initial probability 0.6.
After a red first draw without replacement, the next red probability is 5/9. After a blue first draw, it is 6/9. These conditional probabilities differ because the remaining bag changes. Using 0.6 for every conditional stage would ignore the mechanism.
The first draw in both experiments has the same red probability, yet the whole experiments differ. Before applying a repeated-trial formula, ask whether outcomes are replaced, whether conditions remain stable and whether earlier outcomes alter later possibilities. The story’s mechanism supplies the assumptions that the decimal alone cannot.
Sometimes a question supplies no theoretical probability. Instead, a trial process produces 84 successes in 140 recorded trials. The observed relative frequency is 84/140 = 0.6. It can be used as a data-based estimate of a stable underlying probability when the sampling conditions support that interpretation.
This estimated 0.6 is not an exact theorem about the process. It summarises a finite record and carries uncertainty. More representative data can improve the estimate, while biased sampling or changed conditions can make even a large record misleading.
If you use that estimate to predict successes in fifty similar future trials, the estimated expected count is 50 × 0.6 = 30. Label it as an estimate based on the observed proportion. Do not describe thirty as a guaranteed future count or describe 0.6 as a known exact probability unless additional information justifies that claim.
This is a two-stage reasoning task: first estimate the probability from data, then use the estimate in an expectation calculation. Keeping those stages separate makes it easier to see where uncertainty enters and which assumptions connect the old observations to the new trials.
Suppose a class estimates a spinner’s success probability using only spins selected because they looked successful. Increasing the number of such selected records does not turn the procedure into an unbiased sample of all spins. The selection rule has already distorted the data.
A good experiment defines the event, records every planned trial, uses a consistent mechanism and avoids quietly stopping when a preferred result appears. It also records the total number of trials, not just the successes. Without the denominator, an observed proportion cannot be calculated.
These are design principles rather than a promise that every carefully run finite experiment will match theory closely. Good design makes the data interpretable; random variation still remains. Poor design introduces additional problems that cannot be explained away as ordinary fluctuation.
The ecosystem guide on testing a mathematical model provides a wider route for this habit. Ask what the model assumes, how observations were gathered, and whether the calculation answers the intended question. The result of multiplication is only as meaningful as the model and data to which it is attached.
One observed count different from expectation does not automatically invalidate a model. Equally, the phrase “anything can happen” is not a good reason to ignore every mismatch. Some results are much less plausible than others under a specified model, especially over many well-recorded trials.
For example, seeing a few more successes than expected in ten trials is not the same evidential situation as seeing a very large persistent departure across many independently run trials under stable conditions. Formal judgement requires the appropriate probability model and, when relevant, statistical methods beyond this introductory lesson.
Before blaming the theory, check recording mistakes, event definitions, apparatus and changing conditions. Before declaring the apparatus unfair, check whether the apparent departure could reasonably arise through the specified random process. Good reasoning avoids both automatic rejection and automatic acceptance.
This guide does not set a universal numerical threshold for deciding that a model has failed. Such a threshold depends on the assumptions and the question being asked. Its narrower message is to compare data with a model thoughtfully rather than demanding exact equality with an expected count.
A computer simulation can generate independent success indicators with probability 0.6 and count successes in blocks of ten. Repeating that procedure many times can illustrate the variety of block counts and the stability of their average. It is useful for seeing that six is not compulsory.
However, a simulation produces outcomes from its programmed assumptions. It does not establish that a real spinner or process actually follows those assumptions. If dependence, changing probabilities or biased selection matter in reality, a simple independent simulation may omit them.
When presenting simulated data, label them as simulated, state the probability and repetition structure, and avoid implying that an invented display is a measured classroom record. Distinguish a simulation result from an exact calculation such as np = 6.
No simulated results are being reported in this article. The numerical examples are exact calculations or explicitly invented possible observations. That distinction is deliberate: readers should be able to tell whether a number is a model input, a theoretical result, a possible illustration or an actual measurement without having to guess from its appearance.
For seventeen trials with success probability 0.6, expected count is 10.2. If asked for the expectation, 10.2 is a valid exact answer. Rounding it to ten merely because observed counts are whole numbers changes the quantity requested.
A planning task may separately require a whole-number decision. That decision can involve rounding, capacity, risk or another rule stated by the problem. It is not the definition of expected frequency. Keep the calculated expectation and any subsequent decision distinct.
For example, an expected 10.2 successes does not mean ten are guaranteed and does not mean eleven are impossible. Under independent trials with probability 0.6, both counts are possible, along with other counts. Rounding the average does not remove the distribution around it.
Use the question’s wording to determine the answer format. If it asks for an expected number to one decimal place, retain the decimal. If it asks for a practical whole-number choice under specified conditions, explain the separate decision rule. Do not silently convert an average into an asserted future outcome.
When a learner says, “It should be six because the probability is 0.6,” ask what “should” means. Do they mean the expected count, the most likely count, or a guaranteed count? Those three interpretations lead to different claims.
Next ask whether 1.5 successes can occur in three coin tosses. The physical count cannot be 1.5, but the expectation can. This gives a simple route into the average interpretation without requiring advanced distribution notation.
Then compare replacement and no replacement using a fixed bag. The contrast reveals why some experiments have a fixed final count while others do not. It also helps the learner see that probability alone does not specify the dependence between trials.
Finish with a new question using a different probability and trial count. Ask for an expected count, a possible observed count and a sentence about whether they must match. A correct multiplication without the explanatory sentence is not enough evidence that the misconception has changed. The skill is interpreting the result, not merely multiplying two familiar numbers.
The core of this guide is expected frequency, observed relative frequency and the assumptions behind repeated trials. Indicator arguments, conditional expectation and binomial calculations are optional explanations for readers ready for them. They are not presented as content every G3 learner must already know.
For official course context, consult the 2027 G3 Mathematics syllabus. Match practice difficulty to school guidance and the learner’s actual topic sequence. An explanatory article should make reasoning clearer without overstating assessment requirements.
For the wider concept, read Experimental and Theoretical Probability: Why They Need Not Match Exactly. This article gives a longer focused route through the expected-count misunderstanding rather than replacing that broader probability lesson.
The How Mathematics Works probability guide connects the topic to sample spaces and sequential events. Return to the Mathematics explainer directory when the next difficulty concerns reading a table, defining an event or interpreting a numerical answer.
Decide whether each statement is correct under independent trials with unchanged success probability 0.6. “Ten trials have expected count six” is correct. “Every block of ten contains six successes” is false. “After three failures, the next trial has success probability 0.6” is correct. “Exactly six successes has probability 0.6” is false.
The last statement confuses a single-trial event with a count event across ten trials. Under the binomial extension, the probability of exactly six is about 0.251. You do not need that exact number to see that the two questions refer to different events.
Now consider twenty-five trials with probability 0.6 each. Expected count is fifteen. If seventeen successes are observed, relative frequency is 17/25 = 0.68. The model probability, expected count, observed count and observed proportion are four distinct quantities: 0.6, 15, 17 and 0.68.
If you can label all four without mixing their units or roles, the main distinction is secure. If not, return to the definitions before doing harder count-probability questions. More advanced formulas will not repair a basic confusion about what their outputs mean.
When a probability question asks for a count, first decide whether it wants an expected count, an observed count or the probability of a particular count. These are different tasks. Expected count uses the relevant individual probabilities; observed count comes from data; a count probability requires a fuller model of the repeated trials.
For ten trials each having success probability 0.6, expectation is six. That statement alone does not guarantee the outcome. Under independence, all counts from zero to ten are possible. Under other dependence structures, the possible counts and their probabilities can change while the expectation remains six.
Do not give the process a memory it has not been specified to possess. Independent trials do not compensate for past failures, and larger samples do not approach a target smoothly after every observation. Those mistaken ideas turn an average into a fictional control mechanism.
The central habit is simple: keep model, average and observation separate. That habit makes probability more honest and more useful. It allows you to predict a centre, acknowledge variation and examine evidence without pretending that uncertainty has disappeared because a multiplication produced a neat number.
These short answers summarise the main distinctions. Read the assumptions in each answer, especially whether the trials are independent and whether their success probabilities remain unchanged.
Does 0.6 mean six successes out of every ten?
No. It gives an expected count of six in ten trials when each trial has that probability. Under independence, the observed count can be any integer from zero to ten. A fixed six-out-of-ten quota would be a different model.
Can an expected frequency be a decimal?
Yes. It is an average, not a required individual observation. Three trials with probability one half each have expected count 1.5, although every observed count is a whole number.
Is independence needed for expected count np?
Not when each of the n trials has marginal success probability p. Expectations add without independence. Independence is needed for the standard binomial distribution formula used to calculate probabilities of particular counts under that model.
Should a run of failures make success more likely next?
Not under independent unchanged-probability trials. The next-trial probability remains the stated probability. If the mechanism changes or draws occur without replacement, conditional probabilities can change, but that follows from a different model rather than a debt owed to the expected frequency.