What would you like to understand?

Popular destinationsCalculatorsCoursesPracticeFormulas

Unlock Premium Help

Explore every tutor and study workspace for seven days, with no daily AI cap during the trial.

Choose a billing period
Ask a parent to pay

Nothing is charged during the first seven days. Unless you cancel, the plan you selected begins when the trial ends and renews on its stated schedule. Manage cancellation from your plan page. Stripe handles the payment details.

University mathematics

Free Mathematical Statistics Course: Learn Every Unit

A 7-stage path beginning with Sampling distributions and ending with Bayesian foundations, with recall checks between major ideas.

Question, model, interpret, and challenge the evidence

Short cycles create evidence that the idea is becoming usable rather than merely familiar.

Question

Name the population, variable, and decision the analysis must support.

Model

Check how the data and assumptions permit the chosen method.

Interpret

Return every numerical result to its units and context.

Challenge

Inspect uncertainty, alternatives, and conclusions the evidence cannot support.

Follow the investigation from data to conclusion

Later conclusions depend on how the question, data, assumptions, and model were established earlier.

Questions, models, and interpretations by unit

Each unit names the actual skills, demonstrates one complete case, and identifies the mistake most likely to change the result.

Sampling distributions

You can describe how a statistic bounces around from sample to sample, and give its centre and spread. You can also keep a population parameter and a sample statistic apart.

Decisions to interpret

  • State the mean and standard error of the sample mean.
  • Recognise a chi-square or t distribution from the setting it came from.
  • Explain why bigger samples give a narrower sampling distribution.

Worked example: Population mean 50, standard deviation 12, samples of size 36. Describe the sample mean.

  1. Its centre is the population mean, 50
  2. Standard error = 12 / 6
  3. Centred at 50 with a standard error of 2

Interpretation risk: Quoting the population standard deviation as the sampling spread: using 12 instead of 2 makes every interval six times wider than it should be.

Point estimation

You can build a single best guess for an unknown quantity and judge whether it is biased or wasteful. You can also compare two competing estimators.

Decisions to interpret

  • Show the sample mean is unbiased for the population mean.
  • Explain why the sample variance divides by n - 1.
  • Compare estimators using mean squared error.

Worked example: Sample: 4, 8, 9, 11. Estimate the population variance.

  1. Mean = 32/4 = 8
  2. Squared gaps: 16, 0, 1, 9, adding to 26
  3. 26/3, about 8.67

Interpretation risk: Dividing by n instead of n - 1: reporting 26/4 = 6.5, an estimate that runs systematically low sample after sample.

Interval estimation

You can build a range for an unknown quantity at a chosen confidence level, with z or with t. You can also say which of the two your sample calls for.

Decisions to interpret

  • Use t when the population standard deviation is unknown and n is small.
  • Find the degrees of freedom as n - 1.
  • Work out the sample size needed to hit a target margin of error.

Worked example: n = 25, sample mean 60, sample sd 10, and t* = 2.064 for 95%.

  1. Standard error = 10/5 = 2
  2. Margin = 2.064 x 2 = 4.13
  3. 55.87 to 64.13

Interpretation risk: Using z* = 1.96 with a small sample and an unknown standard deviation: the interval comes out 56.08 to 63.92, too narrow to really carry 95% confidence.

Hypothesis tests

You can state a null and an alternative, compute a test statistic and p-value, and reach a decision. You can also name what a Type I and a Type II error would cost.

Decisions to interpret

  • Write both hypotheses before you look at the data.
  • Compare a p-value with alpha and state the conclusion in context.
  • Explain what a false alarm would mean in the situation at hand.

Worked example: Test H0: the mean is 100. Sample mean 106, sigma 15, n = 25.

  1. Standard error = 15/5 = 3
  2. z = (106 - 100)/3 = 2
  3. Two-sided p is about 0.046, so reject H0 at the 0.05 level

Interpretation risk: Reading p = 0.046 as 'a 4.6% chance the null is true': it is the chance of data this extreme assuming the null already holds.

Likelihood

You can write down how likely your data is for each possible parameter and pick the value that makes it most likely. You can also take logs to make the algebra easy.

Decisions to interpret

  • Write the likelihood for a binomial sample.
  • Take the log, differentiate, and solve for the maximum.
  • Run a likelihood ratio test between two nested models.

Worked example: 7 heads in 10 flips. Find the most likely value of p.

  1. Likelihood is proportional to p to the 7 times (1 - P) cubed
  2. Take logs and differentiate: 7/p - 3/(1 - P) = 0
  3. p = 0.7

Interpretation risk: Pushing p to 1 because a bigger p raises p to the 7: the (1 - P) cubed factor collapses to zero, killing the whole likelihood.

Regression

You can fit a least-squares line, test whether its slope is real, and check the assumptions behind it. You can also tell a confidence interval from a prediction interval.

Decisions to interpret

  • Test whether the slope differs from zero with a t statistic.
  • Read a residual plot for even spread and no pattern.
  • Separate an interval for the mean response from one for a single new case.

Worked example: Fitted slope 2.4, standard error 0.6, n = 20. Is the slope real?

  1. t = 2.4 / 0.6
  2. t = 4.0 with 18 degrees of freedom
  3. p is under 0.001, so the slope is significantly different from zero

Interpretation risk: Judging the slope by its size alone: a slope of 2.4 with a standard error of 3.0 gives t = 0.8 and no evidence at all, even though 2.4 still looks big.

Bayesian foundations

You can combine a prior belief with new data to get an updated belief, and read the result as a whole distribution. You can also report a credible interval.

Decisions to interpret

  • Update a Beta prior with binomial data.
  • Report a 95% credible interval and say plainly what it means.
  • Compare the Bayesian answer with the frequentist one on the same data.

Worked example: Prior Beta(2, 2), then 7 heads and 3 tails.

  1. Add successes to the first parameter and failures to the second
  2. Beta(2 + 7, 2 + 3) = Beta(9, 5)
  3. Posterior mean 9/14, about 0.643

Interpretation risk: Adding the counts to the wrong parameters: Beta(5, 9) gives a mean of about 0.357, which is the chance of tails rather than heads.

Questions about studying Mathematical Statistics

Where should I begin in this course?

Start with the earliest unit whose definitions or notation you cannot explain from memory. Later units often depend on that foundation even when the current assignment looks different.

When should I move to the next unit?

Move on after you can choose the method on a changed problem, complete it without copying a model, and check the answer independently.

How should I use the practice workspace?

Begin without a hint, record the first uncertain decision, repair only that connection, and then answer a fresh question to test whether the repair transferred.