Statistics and probability cheat sheet

BOOKLET given in the examLEARN you must remember it · Each block links to its lesson.

Statistics and probability

IB MATHS AI · HL
pastpaperprep.com/revision-notes

4.1Sampling, data quality and outliers

Population: the whole group a question is about. Sample: the part you collect data from. Counted data is discrete; measured data is continuous, even when rounded.LEARN
Sampling bias comes from how a sample is chosen; a bigger sample chosen the same way doesn't remove it.LEARN
Leave a missing value out and say how many are missing; never record it as 00. Correct a recording error only if you know the true value; otherwise remove it and say so.LEARN
IQR=Q3−Q1\mathrm{IQR} = Q_{3} - Q_{1} (in the formula booklet).BOOKLET
outlier: x<Q1−1.5×IQRorx>Q3+1.5×IQR\text{outlier: } x < Q_{1} - 1.5 \times \mathrm{IQR} \quad \text{or} \quad x > Q_{3} + 1.5 \times \mathrm{IQR}LEARN
A value exactly on a fence is not an outlier. Keep an outlier that is possible in context; correct or remove only an error.LEARN
Simple random: every sample of that size equally likely. Systematic: every kkth of an ordered list, k=Nnk = \frac{N}{n} (rounded down if not whole), random start in the first kk, stop at nn (if nn doesn't divide NN, the end of the list is never chosen). Stratified: random within groups, in proportion. Quota: fixed numbers per group, collector chooses. Convenience: easiest to reach.LEARN
number from a stratum=size of stratumsize of population×sample size\text{number from a stratum} = \frac{\text{size of stratum}}{\text{size of population}} \times \text{sample size}LEARN

4.2Data presentation

Class intervals are inequalities with no gaps and no overlaps, such as 150≤m<175150 \le m < 175 and 175≤m<200175 \le m < 200. A boundary value goes in the class whose inequality it satisfies.LEARN
In a histogram with equal class widths, each bar's height is its frequency and neighbouring bars touch.LEARN
Plot cumulative frequency at each upper class boundary, starting from 00 at the first lower boundary. It counts values at or below a point, so for “more than” subtract the reading from nn.LEARN
On a cumulative frequency graph of nn values, read across from n4\frac{n}{4} for Q1Q_{1}, n2\frac{n}{2} for the median, 3n4\frac{3n}{4} for Q3Q_{3} and k100×n\frac{k}{100} \times n for the kkth percentile.LEARN
IQR=Q3−Q1\mathrm{IQR} = Q_{3} - Q_{1}BOOKLET
Box plot: box from Q1Q_{1} to Q3Q_{3}, a line at the median, whiskers to the most extreme values that are not outliers, and a cross for each outlier.LEARN
Compare two box plots by their medians and their IQRs or ranges, with numbers and in context. A symmetric box with equal whiskers means the data may be normal; a skewed one is unlikely to be.LEARN

4.3Central tendency and dispersion

xˉ=∑i=1kfixin,where n=∑i=1kfi\bar{x} = \frac{\sum_{i=1}^{k} f_{i}x_{i}}{n}, \quad \text{where } n = \sum_{i=1}^{k} f_{i}BOOKLET
Median: the n+12\frac{n+1}{2}th value in order; for even nn, the mean of the middle two. Mode: the most frequent value.LEARN
Grouped data: estimate the mean with mid-interval values a+b2\frac{a+b}{2}. With equal class widths, the modal class is the class with the highest frequency.LEARN
IQR=Q3−Q1\mathrm{IQR} = Q_{3} - Q_{1}BOOKLET
Quartiles of discrete data: use the GDC. Methods differ, so a by-hand value can differ from the GDC's.LEARN
Standard deviation of a data set: GDC only, using σx\sigma_{x}, not sxs_{x}. Variance =σ2= \sigma^{2}: square the unrounded value.LEARN
Every value x→ax+bx \to ax + b: mean →axˉ+b\to a\bar{x} + b, standard deviation →∣a∣σ\to \lvert a\rvert\sigma, variance →a2σ2\to a^{2}\sigma^{2}. Adding a constant leaves σ\sigma unchanged.LEARN

4.4Correlation and regression (y on x)

Describe correlation by direction (positive, negative or zero) and strength (strong, weak or none), in context.LEARN
−1≤r≤1-1 \le r \le 1LEARN
Pearson's rr measures linear correlation only: rr near 00 means no straight-line link, but there may be a curve.LEARN
If ∣r∣\lvert r\rvert is greater than the critical value for that number of pairs and significance level, the linear correlation is significant at that level.LEARN
A line of best fit by eye and the regression line of yy on xx both pass through the mean point (xˉ,yˉ)(\bar{x}, \bar{y}).LEARN
Correlation does not imply causation: a third variable may drive both.LEARN
In the regression line y=ax+by = ax + b, aa is the predicted change in yy for each increase of 11 in xx, and bb is the predicted yy when x=0x = 0, which only means something if x=0x = 0 makes sense and is near the data.LEARN
Use the yy on xx line only to predict yy, for xx inside the data range and when the correlation is strong; it cannot reliably predict xx from yy.LEARN

4.5Probability foundations and expected value

A trial has outcomes; the sample space UU lists them all; an event is a set of outcomes. For two stages, list pairs or draw a table.LEARN
P(A)=n(A)n(U)when the outcomes in U are equally likely\mathrm{P}(A) = \frac{n(A)}{n(U)} \quad \text{when the outcomes in } U \text{ are equally likely}BOOKLET
Every probability is between 00 and 11. From a two-way table with one person chosen at random from everyone in it, divide by the grand total.LEARN
Relative frequency =times the event happensnumber of trials= \frac{\text{times the event happens}}{\text{number of trials}} is an experimental estimate of a probability. More trials usually give a better estimate.LEARN
P(A)+P(A′)=1\mathrm{P}(A) + \mathrm{P}(A') = 1BOOKLET
For "at least one", use 1−P(none)1 - \mathrm{P}(\text{none}).LEARN
Expected number of occurrences =n×P(A)= n \times \mathrm{P}(A) for nn trials. It is a long-run average, so it need not be a whole number.LEARN

4.6Probability diagrams, rules and independence

Fill a Venn diagram from the middle out: the overlap, then the rest of each circle, then the outside. "Or" includes both: A∪BA \cup B means AA, BB or both.LEARN
P(A∪B)=P(A)+P(B)−P(A∩B)\mathrm{P}(A \cup B) = \mathrm{P}(A) + \mathrm{P}(B) - \mathrm{P}(A \cap B)BOOKLET
Mutually exclusive events can't both happen, so P(A∩B)=0\mathrm{P}(A \cap B) = 0. That is not the same as independent.LEARN
P(A∪B)=P(A)+P(B)for mutually exclusive A,B\mathrm{P}(A \cup B) = \mathrm{P}(A) + \mathrm{P}(B) \quad \text{for mutually exclusive } A, BBOOKLET
P(A∣B)=P(A∩B)P(B),P(B)>0\mathrm{P}(A \mid B) = \frac{\mathrm{P}(A \cap B)}{\mathrm{P}(B)}, \quad \mathrm{P}(B) > 0BOOKLET
P(A∩B)=P(B) P(A∣B)\mathrm{P}(A \cap B) = \mathrm{P}(B)\,\mathrm{P}(A \mid B)LEARN
Tree diagrams: multiply along a path, add across paths. Without replacement, the totals and second-stage probabilities change.LEARN
P(A∩B)=P(A) P(B)for independent A,B\mathrm{P}(A \cap B) = \mathrm{P}(A)\,\mathrm{P}(B) \quad \text{for independent } A, BBOOKLET

4.7Discrete random variables and expected value

A discrete random variable XX has a list of possible values xx. Its probability distribution gives P(X=x)\mathrm{P}(X = x) for each one, as a table or a formula.LEARN
∑P(X=x)=1,0≤P(X=x)≤1\sum \mathrm{P}(X = x) = 1, \quad 0 \le \mathrm{P}(X = x) \le 1LEARN
E(X)=∑x P(X=x)\mathrm{E}(X) = \sum x\,\mathrm{P}(X = x)BOOKLET
E(X)\mathrm{E}(X) is the long-run mean of XX. It need not be a value XX can take. Over nn independent repeats of the trial, expect a total of n×E(X)n \times \mathrm{E}(X).LEARN
GDC: one-variable statistics with the values as data and the probabilities as frequencies gives xˉ=E(X)\bar{x} = \mathrm{E}(X); with the probabilities entered as they are, the total frequency should be 11.LEARN
fair game: E(X)=0,where X is the player’s gain (prize minus cost)\text{fair game: } \mathrm{E}(X) = 0, \quad \text{where } X \text{ is the player's gain (prize minus cost)}LEARN

4.8Binomial distribution

X∼B(n,p)X \sim \mathrm{B}(n, p) counts the successes in a fixed number nn of independent trials, each with two outcomes and the same probability pp of success.LEARN
On a GDC, binomial pdf gives P(X=r)\mathrm{P}(X = r) and binomial cdf gives P(X≤k)\mathrm{P}(X \le k).LEARN
For binomial XX (whole numbers only): P(X≥k)=1−P(X≤k−1)\mathrm{P}(X \ge k) = 1 - \mathrm{P}(X \le k - 1) and P(X>k)=1−P(X≤k)\mathrm{P}(X > k) = 1 - \mathrm{P}(X \le k).LEARN
E(X)=np\mathrm{E}(X) = npBOOKLET
Var(X)=np(1−p)\mathrm{Var}(X) = np(1-p)BOOKLET
For X∼B(n,p)X \sim \mathrm{B}(n, p), npnp is the expected number of successes: a long-run average that need not be a whole number. The standard deviation is np(1−p)\sqrt{np(1-p)}.LEARN

4.9Normal distribution

X∼N(μ,σ2):P(X<μ)=P(X>μ)=0.5,P(X<μ−d)=P(X>μ+d)X \sim \mathrm{N}(\mu, \sigma^{2}): \quad \mathrm{P}(X < \mu) = \mathrm{P}(X > \mu) = 0.5, \qquad \mathrm{P}(X < \mu - d) = \mathrm{P}(X > \mu + d)LEARN
X∼N(μ,σ2):P(μ−σ<X<μ+σ)≈0.68P(μ−2σ<X<μ+2σ)≈0.95P(μ−3σ<X<μ+3σ)≈0.997X \sim \mathrm{N}(\mu, \sigma^{2}): \quad \begin{aligned} \mathrm{P}(\mu - \sigma < X < \mu + \sigma) &\approx 0.68 \\ \mathrm{P}(\mu - 2\sigma < X < \mu + 2\sigma) &\approx 0.95 \\ \mathrm{P}(\mu - 3\sigma < X < \mu + 3\sigma) &\approx 0.997 \end{aligned}LEARN
P(X>k)=p  ⟺  P(X<k)=1−p(X continuous)\mathrm{P}(X > k) = p \iff \mathrm{P}(X < k) = 1 - p \quad (X \text{ continuous})LEARN
X∼N(μ,σ2)X \sim \mathrm{N}(\mu, \sigma^{2}): a symmetric bell-shaped curve about μ\mu, with total area 11. Mean, median and mode all equal μ\mu. The second number is the variance.LEARN
For a continuous random variable, P(X=a)=0\mathrm{P}(X = a) = 0, so << and ≤\le give the same probability.LEARN
Normal cdf takes lower bound, upper bound, μ\mu and σ\sigma. Use −1099-10^{99} as the lower bound for less than, and 109910^{99} as the upper bound for more than.LEARN
Inverse normal, with the tail set to left, returns kk from P(X<k)=p\mathrm{P}(X < k) = p. For a middle region of probability pp centred on μ\mu, each tail holds 1−p2\frac{1 - p}{2}.LEARN
Expected number of nn independent items in a region =n×p= n \times p, using the unrounded probability. Give it to 3 s.f.: it need not be a whole number.LEARN

4.10Spearman's rank correlation

Rank each variable separately, both in the same direction. Tied values each get the mean of the ranks they share.LEARN
rsr_s is Pearson's rr calculated on the ranks. On the GDC: enter the two lists of ranks, run linear regression, read rr.LEARN
−1≤rs≤1-1 \le r_s \le 1. rs=1r_s = 1: the rankings are identical. rs=−1r_s = -1: one is the exact reverse of the other.LEARN
Pearson's rr measures only linear correlation. rsr_s measures any monotonic relationship, straight or curved.LEARN
rsr_s is less sensitive to outliers than rr: an extreme value can only take the top or bottom rank.LEARN
A link that rises then falls can give rr and rsr_s near 00, and neither coefficient shows causation.LEARN

4.11Chi-squared tests and the t-test

fe=row total×column totalgrand totalf_e = \frac{\text{row total} \times \text{column total}}{\text{grand total}}LEARN
H0\mathrm{H}_0 is the no-effect claim: independent, fits the claimed distribution, or μ1=μ2\mu_1 = \mu_2 for population means.LEARN
Reject H0\mathrm{H}_0 when the p-value is less than the significance level, or χcalc2\chi^2_{\text{calc}} is greater than the critical value.LEARN
Expected frequencies are what H0\mathrm{H}_0 predicts. The test needs every expected frequency greater than 55: SL exam questions always meet this; at HL you may need to merge classes first.LEARN
χ² tests use the upper tail only. df =(r−1)(c−1)= (r - 1)(c - 1) for an r×cr \times c contingency table; df =n−1= n - 1 for goodness of fit at SL (nn categories).LEARN
SL t-test: pooled two-sample, unpaired samples, normal populations, equal population variances. ≠\neq is two-tailed; >> or << is one-tailed.LEARN
Every conclusion gives the comparison, the decision and its meaning in context.LEARN

4.12 ◆Data collection, reliability and validity

Good survey questions are unbiased (neutral wording) and precise (one meaning, a stated time period). Structured answer choices don't overlap and cover every possible answer. Ask personal questions only if needed, and anonymously.LEARN
Choose variables linked to the question. For a linear model, compare candidates by the size of rr, not its sign. Use only data from the population and conditions the question is about, and say what you left out.LEARN
In IB χ² questions every expected frequency must be greater than 55. Merge a small class with an adjacent class, observed and expected together, until every expected frequency is greater than 55, then count the categories again.LEARN
Goodness of fit: df=k−1−m\text{df} = k - 1 - m, with kk categories after merging and mm parameters estimated from the data. Binomial pp estimated: m=1m = 1; normal μ\mu and σ\sigma estimated: m=2m = 2; parameters given in the question: m=0m = 0.LEARN
Reliability: the same results when the method is repeated under the same conditions. Validity: the method measures what it claims to. Reliable does not mean valid.LEARN
Reliability checks: test–retest (same test, same people, two times) and parallel forms (two equivalent versions, same people). A strong positive correlation between the two sets of scores suggests reliability.LEARN
Validity checks: content (the test covers every part of what it should) and criterion-related (a strong correlation, in the expected direction, with an established measure of the same thing).LEARN

4.13 ◆Non-linear regression and R²

Choose the model type from the context and the scatter diagram (or the form the question names); the GDC fits the parameters. Give them to 3 s.f.LEARN
Residual =y−y^= y - \hat{y}, observed minus predicted: positive above the curve, negative below.LEARN
SSres=∑i=1n(yi−y^i)2SS_{\text{res}} = \sum_{i=1}^{n} \left(y_{i} - \hat{y}_{i}\right)^{2}LEARN
For models fitted to the same data, a smaller SSresSS_{\text{res}} means a closer fit; the least squares curve of a type has the smallest SSresSS_{\text{res}} of that type (the GDC's exponential and power regressions minimise it for ln⁡y\ln y instead).LEARN
R2R^{2} is the proportion of the variation in yy that the model accounts for. For a linear least squares model, R2=r2R^{2} = r^{2}.LEARN
R2R^{2} alone does not choose a model: check the context, the shape and any extrapolation. On the same data, a cubic regression's R2R^{2} is never lower than a quadratic regression's.LEARN

4.14 ◆Linear combinations and unbiased estimates

E(aX+b)=aE(X)+b\mathrm{E}(aX + b) = a\mathrm{E}(X) + bBOOKLET
Var(aX+b)=a2 Var(X)\mathrm{Var}(aX + b) = a^{2}\,\mathrm{Var}(X)BOOKLET
The standard deviation of aX+baX + b is ∣a∣σ\lvert a\rvert\sigma. Adding a constant moves the mean but not the spread.LEARN
Expected values add for any random variables, independent or not: E(a1X1±⋯±anXn)=a1E(X1)±⋯±anE(Xn)\mathrm{E}(a_{1}X_{1} \pm \dots \pm a_{n}X_{n}) = a_{1}\mathrm{E}(X_{1}) \pm \dots \pm a_{n}\mathrm{E}(X_{n}).BOOKLET
Only for independent X1,…,XnX_{1}, \dots, X_{n}: Var(a1X1±⋯±anXn)=a12Var(X1)+⋯+an2Var(Xn)\mathrm{Var}(a_{1}X_{1} \pm \dots \pm a_{n}X_{n}) = a_{1}^{2}\mathrm{Var}(X_{1}) + \dots + a_{n}^{2}\mathrm{Var}(X_{n}). Every term is added, even for a difference. Add variances, never standard deviations.BOOKLET
The total of nn independent items, each with mean μ\mu and variance σ2\sigma^{2}, has mean nμn\mu and variance nσ2n\sigma^{2}. One item scaled by nn, nXnX, has variance n2σ2n^{2}\sigma^{2}.LEARN
From a random sample, xˉ\bar{x} is an unbiased estimate of μ\mu and sn−12=nn−1sn2s_{n-1}^{2} = \frac{n}{n-1}s_{n}^{2} is an unbiased estimate of σ2\sigma^{2}. On the GDC, sn−1s_{n-1} is Sx (sx on the Casio and Nspire).BOOKLET

4.15 ◆Sample mean and the central limit theorem

If X1,…,XnX_{1}, \dots, X_{n} are independent and each is normal, then a1X1±⋯±anXna_{1}X_{1} \pm \dots \pm a_{n}X_{n} is normal. Find its mean and variance with the booklet rules, then enter σ=variance\sigma = \sqrt{\text{variance}} in normal cdf.LEARN
X∼N(μ,σ2)  ⇒  Xˉ∼N(μ,σ2n)for a random sample of size nX \sim \mathrm{N}(\mu, \sigma^{2}) \;\Rightarrow\; \bar{X} \sim \mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right) \quad \text{for a random sample of size } nLEARN
For a random sample of size nn from a population with variance σ2\sigma^{2}, Var(Xˉ)=σ2n\mathrm{Var}(\bar{X}) = \frac{\sigma^{2}}{n} and the standard deviation of Xˉ\bar{X} is σn\frac{\sigma}{\sqrt{n}}: four times the sample size halves the spread.LEARN
Central limit theorem: for a random sample of size nn from any population with mean μ\mu and variance σ2\sigma^{2}, Xˉ\bar{X} is approximately N(μ,σ2n)\mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right) when nn is large. How large depends on the population; in IB exams n>30n > 30 is enough.LEARN
A normal population gives an exactly normal Xˉ\bar{X} for every nn. A population that is not normal needs a large sample before Xˉ\bar{X} can be treated as normal.LEARN

4.16 ◆Confidence intervals for the mean

xˉ±z∗σn(normal population, σ known)\bar{x} \pm z^{*}\frac{\sigma}{\sqrt{n}} \qquad (\text{normal population, } \sigma \text{ known})LEARN
xˉ±t∗sn−1n(normal population, σ unknown, ν=n−1)\bar{x} \pm t^{*}\frac{s_{n-1}}{\sqrt{n}} \qquad (\text{normal population, } \sigma \text{ unknown, } \nu = n - 1)LEARN
For the mean of a normal population: σ\sigma known, use the normal distribution; σ\sigma unknown, use the t-distribution with n−1n - 1 degrees of freedom and sn−1s_{n-1}, whatever the sample size.LEARN
Normal multipliers (σ\sigma known): z∗=1.645z^{*} = 1.645 for 90%90\%, 1.9601.960 for 95%95\%, 2.5762.576 for 99%99\%.LEARN
A 95%95\% confidence interval comes from a method that captures μ\mu for about 95%95\% of random samples. One interval either contains μ\mu or not.LEARN
A claimed mean inside the interval is plausible (no evidence against it); one outside the interval is evidence against the claim.LEARN
From the same sample, a higher confidence level gives a wider interval. At the same confidence level and standard deviation, a larger sample gives a narrower one.LEARN

4.17 ◆Poisson distribution

X∼Po(m)X \sim \mathrm{Po}(m) models the number of events in a fixed interval of time or space when the events happen independently at a uniform average rate of mm per interval. XX can be 0,1,2,…0, 1, 2, \ldots with no upper limit.LEARN
E(X)=m\mathrm{E}(X) = mBOOKLET
Var(X)=m\mathrm{Var}(X) = mBOOKLET
Because the rate is uniform, a Poisson mean of mm per unit interval becomes mtmt for an interval tt times as long.LEARN
If X∼Po(m1)X \sim \mathrm{Po}(m_{1}) and Y∼Po(m2)Y \sim \mathrm{Po}(m_{2}) are independent, then X+Y∼Po(m1+m2)X + Y \sim \mathrm{Po}(m_{1} + m_{2}). Multiples such as 2X2X and differences such as X−YX - Y are not Poisson.LEARN
Choosing a model: a continuous measurement with a symmetric bell-shaped spread is normal; successes in a fixed number of independent trials with constant pp are binomial; independent events in an interval at a uniform average rate, with no upper limit, are Poisson.LEARN

4.18 ◆Further hypothesis tests and errors

Xˉ∼N(μ0,σ2n) under H0: μ=μ0(normal population, σ known)\bar{X} \sim \mathrm{N}\left(\mu_0, \frac{\sigma^{2}}{n}\right) \text{ under } \mathrm{H}_0\text{: } \mu = \mu_0 \quad (\text{normal population, } \sigma \text{ known})LEARN
Mean of a normal population: z-test when σ\sigma is known, t-test when σ\sigma is unknown, whatever the sample size. IB questions don't ask for a critical region in a t-test: decide with the p-value.LEARN
Matched pairs: take differences dd in a fixed order, then a one-sample t-test with H0\mathrm{H}_0: μd=0\mu_d = 0, assuming the differences are normally distributed.LEARN
In IB AI, binomial and Poisson tests are one-tailed. The critical region has the largest probability under H0\mathrm{H}_0 that is still less than the significance level.LEARN
Correlation: test H0\mathrm{H}_0: ρ=0\rho = 0 with the GDC, assuming a bivariate normal distribution.LEARN
Type I error: rejecting a true H0\mathrm{H}_0. Its probability is the significance level for a normal test with known σ\sigma, and the critical region's probability under H0\mathrm{H}_0 for a binomial or Poisson test.LEARN
Type II error: not rejecting a false H0\mathrm{H}_0. Its probability is P(not in the critical region)\mathrm{P}(\text{not in the critical region}), using a stated true parameter value.LEARN

4.19 ◆Transition matrices and Markov chains

Tns0=sn,where s0 is the initial stateT^{n}\boldsymbol{s}_0 = \boldsymbol{s}_n, \quad \text{where } \boldsymbol{s}_0 \text{ is the initial state}BOOKLET
Transition matrix: TijT_{ij} is the chance of going to state ii at the next step, given the chain is in state jj now, so each column adds up to 11. Columns are “from”, rows are “to”.LEARN
In a transition diagram the probabilities on the arrows leaving any one state, including its loop, add up to 11.LEARN
(Tn)ij(T^{n})_{ij} is the probability of being in state ii after nn steps, starting in state jj. A probability along one fixed path is a product of single-step probabilities instead.LEARN
A Markov chain is regular when some power of TT has all its entries positive. Then the columns of TnT^{n} all approach the steady state, whatever the start.LEARN
Steady state s\boldsymbol{s}: Ts=sT\boldsymbol{s} = \boldsymbol{s} with entries adding up to 11. For a regular chain it is unique: find it from a column of a high power of TT, or exactly by solving Ts=sT\boldsymbol{s} = \boldsymbol{s} with one equation replaced by the sum =1= 1.LEARN
A steady state is an eigenvector of TT for the eigenvalue 11, divided by the sum of its entries. For a regular chain there is only one.LEARN

Which method when

'Is this value an outlier?' → GDC one-variable stats for Q1Q_{1}, Q3Q_{3}; outlier if x<Q1−1.5×IQRx < Q_{1} - 1.5 \times \mathrm{IQR} or x>Q3+1.5×IQRx > Q_{3} + 1.5 \times \mathrm{IQR} (a value on a fence is not one) (4.1)LEARN
'Estimate the median, quartiles or a percentile' from grouped data → cumulative frequency graph, plotted at upper class boundaries; read across from n4\frac{n}{4}, n2\frac{n}{2}, 3n4\frac{3n}{4}, k100×n\frac{k}{100} \times n (4.2)LEARN
'Find the mean and standard deviation' of data or a frequency table → GDC one-variable stats with the frequency list; quote σx\sigma_{x}, not sxs_{x}; grouped data uses mid-interval values, so the mean is an estimate (4.3)LEARN
'Is there linear correlation? Predict yy' → GDC linear regression for rr and y=ax+by = ax + b; significant if ∣r∣\lvert r\rvert is above the given critical value; predict yy only, for xx inside the data, when the correlation is strong (4.4)LEARN
'Given that', or 'or' / 'and' with an overlap → Venn diagram, two-way table or tree; P(A∣B)=P(A∩B)P(B)\mathrm{P}(A \mid B) = \frac{\mathrm{P}(A \cap B)}{\mathrm{P}(B)}; multiply P(A) P(B)\mathrm{P}(A)\,\mathrm{P}(B) only for independent events (4.6)LEARN
A table of values with probabilities, 'find kk', 'is the game fair?' → ∑P(X=x)=1\sum \mathrm{P}(X = x) = 1 and E(X)=∑x P(X=x)\mathrm{E}(X) = \sum x\,\mathrm{P}(X = x); fair when E(X)=0\mathrm{E}(X) = 0 for the player's gain (4.7)LEARN
'Number of successes in nn trials', fixed nn, same pp, independent trials → X∼B(n,p)X \sim \mathrm{B}(n, p); binomial pdf for P(X=r)\mathrm{P}(X = r), binomial cdf for P(X≤k)\mathrm{P}(X \le k); P(X≥k)=1−P(X≤k−1)\mathrm{P}(X \ge k) = 1 - \mathrm{P}(X \le k - 1) (4.8)LEARN
A measurement modelled as X∼N(μ,σ2)X \sim \mathrm{N}(\mu, \sigma^{2}), 'find the probability' → normal cdf with σ\sigma, not σ2\sigma^{2}; 'find the value the top 10%10\% exceed' → inverse normal with left area 0.90.9 (or area 0.10.1 with the tail set to right) (4.9)LEARN
Ranked data, a curved but always rising or falling link, or an outlier → Spearman's rsr_s: rank both variables the same way, tied values take the mean rank, then GDC linear regression on the ranks (4.10)LEARN
Counts in categories → χ² test on the GDC: two categorical variables in a table, 'is X independent of Y?' → test for independence, df =(r−1)(c−1)= (r - 1)(c - 1); one variable against claimed proportions ('equally likely', stated percentages) → goodness of fit, expected == total × claimed proportion, df =n−1= n - 1 for nn categories; every expected frequency must be greater than 55 (4.11)LEARN
Two independent (unpaired) groups of measurements, 'is the mean of A different from / greater than B?' → pooled two-sample t-test on the GDC, H0\mathrm{H}_0: μ1=μ2\mu_1 = \mu_2, assuming normal populations with equal variances; ≠\neq is two-tailed, >> or << one-tailed (4.11)LEARN

◆Which method when

Goodness of fit to a binomial or normal model whose parameters are estimated from the data → merge adjacent classes (observed and expected) until every expected frequency is greater than 55, then df =k−1−m= k - 1 - m (kk classes after merging, mm parameters estimated) (4.12)LEARN
Curved scatter, or a named model such as y=abxy = ab^{x}, y=axby = ax^{b}, quadratic, cubic or sine → GDC regression of that type, parameters to 3 s.f.; compare models on the same data by SSresSS_{\text{res}} or R2R^{2}, together with the context and any extrapolation, never R2R^{2} alone (4.13)LEARN
Probability for a total or difference of independent normal quantities → mean by the E\mathrm{E} rules, variance ∑ai2Var(Xi)\sum a_{i}^{2}\mathrm{Var}(X_{i}) (added even for a difference), then normal cdf with σ=variance\sigma = \sqrt{\text{variance}}; for the mean of a random sample → Xˉ∼N(μ,σ2n)\bar{X} \sim \mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right), exactly for a normal population, approximately by the central limit theorem when n>30n > 30 (4.14, 4.15)LEARN
Events in a fixed interval of time or space, happening independently at a uniform average rate, 'how many in…' → X∼Po(m)X \sim \mathrm{Po}(m) with mm scaled to the interval, independent Poisson counts add; a fixed number nn of independent trials with constant pp → binomial instead (4.17)LEARN
Estimating or testing the mean of one normal population → σ\sigma given: Z interval or z-test; σ\sigma unknown: T interval or t-test with sn−1s_{n-1} and n−1n - 1 degrees of freedom, whatever nn; the same subjects measured twice → one-sample t-test on the differences, H0\mathrm{H}_0: μd=0\mu_{d} = 0 (4.16, 4.18)LEARN
A count tested against a claimed proportion or rate → one-tailed binomial B(n,p)\mathrm{B}(n, p) or Poisson Po(m)\mathrm{Po}(m) test, p-value from the tail H1\mathrm{H}_1 points to; 'is there correlation in the population?' → GDC test of H0\mathrm{H}_0: ρ=0\rho = 0, assuming bivariate normal data (4.18)LEARN
States that change step by step, where the next depends only on the current state ('after 3 days', 'in the long run') → transition matrix with columns 'from', each adding up to 11, and sn=Tns0\boldsymbol{s}_n = T^{n}\boldsymbol{s}_0; long run → for a regular chain, the unique steady state Ts=sT\boldsymbol{s} = \boldsymbol{s} with entries adding up to 11 (an eigenvector for eigenvalue 11, scaled), which sn\boldsymbol{s}_n approaches from any start (4.19)LEARN
Follows the IB Maths AI guide (first assessment 2021). Not affiliated with the IB.