IB Maths AI HL: How to Choose Statistical Tests and Models

In IB Maths AI HL, choose a statistical method by identifying what the question asks, what each variable measures and how the observations were collected. Prediction suggests a model, comparison of a population mean suggests a mean test, and association between categorical variables suggests a chi-squared test of independence.

Start with the population and the observational unit: one student, one measurement or one experimental trial. Then distinguish numerical measurements from categories and identify whether observations are independent or linked; two measurements from the same student form a pair, so treating them as unrelated observations loses essential information about the investigation.

The table gives starting points, not automatic permissions to run a test. A suitable variable type does not establish random sampling, independence or a distributional assumption; those conditions need evidence from the question, and a large sample cannot repair a systematically biased selection process.

For broader IB mathematics difficulties, use /insights/ib-math-challenges alongside this decision process. The central habit is to state the purpose before selecting a calculator menu: a test assesses evidence against a hypothesis, whereas a fitted model describes a relationship and may support predictions within a justified range.

In IB Maths AI HL, choose a model whose mathematical behaviour matches both the data pattern and the situation being described. Check the scatter diagram, interpret the parameters and inspect residuals before trusting predictions; a strong numerical fit alone does not establish that the model is appropriate outside the observed range.

Suppose a measured quantity has values $100$, $80$ and $64$ at times $t=0$, $t=1$ and $t=2$. Successive ratios are $\frac{80}{100}=\frac{64}{80}=0.8$, suggesting the exponential model $Q(t)=100(0.8)^t$; equal absolute changes would instead suggest a linear model.

The initial value is $Q(0)=100$, and the quantity decreases by $20\%$ per time unit. The prediction at $t=3$ is $Q(3)=100(0.8)^3=51.2$; if the observed value is $55$, the residual is $e=y-\hat y=55-51.2=3.8$, so the model underpredicts that observation.

A residual plot with curvature suggests the model misses a systematic feature, while an apparently patternless plot supports its adequacy over the measured range. Extrapolation still needs a contextual argument: exponential decline assumes a constant proportional change, and a real process may change mechanism or approach a nonzero background level.

A one-sample t-test in IB Maths AI HL assesses whether sample data provide evidence against a stated population mean when the population standard deviation is unknown. Identify the population parameter, justify the sampling and distributional conditions, choose the alternative hypothesis before calculation, and interpret the p-value against the stated significance level.

Suppose a random sample of $n=16$ independent measurements comes from a population assumed to be normally distributed, with sample mean $\bar x=52$ and sample standard deviation $s=4$. To investigate whether the population mean differs from $50$, state $H_0:\mu=50$ and $H_1:\mu\ne50$, with $\alpha=0.05$.

The test statistic is $$t=\frac{\bar x-\mu_0}{s/\sqrt n}=\frac{52-50}{4/\sqrt{16}}=2.$$$$\begin{aligned}t &= \frac{\bar x-\mu_0}{s/\sqrt n} \\ &= \frac{52-50}{4/\sqrt{16}} \\ &= 2.\end{aligned}$$ With $15$ degrees of freedom, the two-sided p-value is approximately $p=0.064$, so $p>0.05$ and we do not reject the null hypothesis.

The contextual conclusion is that the sample does not provide sufficient evidence, at the stated significance level, that the population mean differs from $50$. This does not prove equality: the p-value describes how unusual a result at least this extreme would be under the null hypothesis and the test assumptions, not the probability that the null hypothesis is true.

Use a chi-squared test of independence in IB Maths AI HL when the question asks whether two categorical variables are associated in a population. Use a goodness-of-fit test when the question instead compares observed category counts with a specified distribution; both methods require counts and sufficiently large expected frequencies.

Consider a hypothetical random sample of $100$ students classified by revision format and whether they completed a task. Among $50$ using format A, $30$ completed it and $20$ did not; among $50$ using format B, $20$ completed it and $30$ did not, giving four mutually exclusive cells.

Under $H_0$, revision format and completion are independent. Each expected count is $E=\frac{50\times50}{100}=25$, so all expected frequencies exceed $5$, and the statistic is $$\chi^2=\sum\frac{(O-E)^2}{E}=4\left(\frac{5^2}{25}\right)=4.$$$$\begin{aligned}\chi^2 &= \sum\frac{(O-E)^2}{E} \\ &= 4\left(\frac{5^2}{25}\right) \\ &= 4.\end{aligned}$$

The degrees of freedom are $(2-1)(2-1)=1$, giving approximately $p=0.0455$. At $\alpha=0.05$, reject independence and conclude there is evidence of an association between revision format and completion in the population represented by the sample; without random assignment and appropriate controls, this does not establish that the format caused completion.

IB Maths AI HL exam answers need to connect the chosen method, relevant assumptions, numerical output and contextual conclusion in a visible chain of reasoning. All three AI HL examination papers require technology, but calculator access does not replace mathematical communication; the command word determines how much explanation a particular answer needs.

For a hypothesis test, record the hypotheses, the relevant test output, the comparison with the significance level and the conclusion in context. For a model, give the equation, interpret relevant parameters and answer the requested prediction or evaluation; retain full calculator precision during working and follow any stated rounding instruction.

“State” calls for a brief answer, “calculate” requires a numerical answer with relevant working, and “justify” requires supporting reasons or evidence. An instruction to interpret a result needs its contextual meaning: reporting $p=0.064$ alone does not explain what the investigation established about the population mean.

Paper 3 uses extended problems, so later parts may require you to question assumptions or evaluate earlier results. Practise complete responses using /insights/ib-past-papers-where-to-find, then organise corrections with /insights/ib-exam-preparation-strategy; check whether your missing marks came from calculation, method selection or interpretation before repeating the exercise.

Parents can recognise secure understanding of IB Maths AI HL statistics when their child can explain why a method fits, name a relevant condition and translate the result into ordinary language. Support this by asking for a short explanation of one completed problem, then encouraging correction of the specific gap revealed.

Ask what was measured, why the chosen test or model was appropriate and what the result cannot establish. A student who can explain an assumption or limitation without reading calculator output is demonstrating understanding; if the explanation breaks down, return to that decision rather than assigning another long set of similar calculations.