Chapter 5 — Statistics & Probability

What is statistics? Statistics is the scientific process of collecting, organising, summarising and interpreting numerical information so we can understand patterns, compare groups and make decisions. It is used to turn raw observations into concise summaries and visual displays that communicate essential features of the data. In everyday life and in many subjects — science, economics, medicine and social studies — statistics helps answer questions such as ‘what is typical?’, ‘how variable are things?’ and ‘is there an effect or relationship?’.

Steps in a statistical investigation A typical sequence begins with a clear question or objective, followed by deciding what data are needed. Next comes data collection using an appropriate method, cleaning and checking the data for errors, choosing suitable tables and graphs to display the data, calculating numerical summaries (means, medians, measures of spread) and interpreting results in the context of the original question. At each step document assumptions and potential sources of bias.

Types of data and measurement scales Data are broadly of two kinds: qualitative (categorical) and quantitative (numerical). Qualitative data describe qualities or categories — for example blood group, favourite fruit or grade categories — and are measured on nominal or ordinal scales. Quantitative data measure amounts and come as discrete (countable integers like number of books) or continuous (measurements like height or time) and are on interval or ratio scales. Recognising the scale of measurement guides which graphs and summary measures are appropriate: means require numerical data while mode is meaningful for categories.

Population and sample The population is the full set of units about which we want to draw conclusions. A sample is a subset selected for study. A well-chosen random sample can allow reliable conclusions about the population; a biased or non-random sample may mislead. Sampling design, including sample size and method, determines how precise and generalisable our conclusions are.

Data quality and cleaning Always check data for missing values, entry mistakes, impossible values or outliers. Decide how to treat missing observations — omit, impute, or flag — and justify the choice. Outliers may be genuine or errors; investigate and report their effect on summaries. Label variables with units and keep a clear data dictionary to avoid confusion. Good practice in data handling is as important as correct formulas.

Practical tips When presenting results, always include the context: what was measured, sample size, units and any limitations. Use multiple views — both graphical and numeric — to give a balanced picture, and avoid over-interpretation of small samples or chance fluctuations. Thinking critically about data collection and representation is the first, and essential, step in doing good statistics.

Why group data? When raw data has many observations, working directly with every value becomes cumbersome. Grouping values into class intervals simplifies the dataset and helps reveal general patterns such as central tendency and spread. A frequency distribution lists the classes (intervals) and the number of observations (frequency) in each class. This condensed form is the starting point for many graphical displays and summary measures.

Choosing classes and class width Key choices when grouping are: the number of classes, the size (width) of each class, and class boundaries. Good practice aims for 5–15 classes depending on the number of observations; classes should be mutually exclusive and exhaustive. The class width may be chosen by computing the range = max − min, then dividing by the desired number of classes and rounding to a convenient number. Prefer equal-width classes unless there is a clear reason to use varying widths (in which case use frequency density to plot a histogram).

Class limits and boundaries For grouped data define lower and upper class limits clearly (for example 10–19, 20–29). When data are continuous, use class boundaries that avoid gaps, e.g., 9.5–19.5, 19.5–29.5. When counting integers, integer limits are fine. Ensure conventions are stated so that counts are assigned consistently.

Constructing the frequency table Steps: (1) Find min and max and compute range. (2) Decide on number of classes m. (3) Compute class width h ≈ range / m and round to a convenient value. (4) Choose the lower limit of the first class and form subsequent classes of width h. (5) Tally observations into classes and compute frequencies. Include columns for cumulative frequency (running total) and relative frequency (frequency divided by total n). Relative frequencies are useful for comparing distributions of different sample sizes.

Grouped vs ungrouped Grouping is an approximation: class midpoint is used to represent all values in a class when computing means or variances. Narrower classes reduce approximation error. Avoid grouping when the dataset is small or when exact values are required. For large datasets grouped summaries are practical and usually sufficient for understanding the main features of the distribution.

Reporting and interpretation When presenting grouped results always report the number of observations n and the class scheme chosen so others can reproduce or critique your choices. Mention that results for measures computed from grouped data are approximate and note any open-ended classes (like ‘50 and above’) which may need special handling. Finally, consider using both a table and a graph to make patterns immediately visible to the reader.

Overview of graphical tools Graphs are essential to communicate distributional features quickly. For numerical grouped data the common graphs are histograms, frequency polygons and ogives. For categorical or proportional data bar charts and pie charts are typical. Choosing the correct graph depends on the scale of measurement and the message you want to convey.

Histograms in detail A histogram represents a grouped numerical distribution by adjacent rectangular bars. For equal class widths, each bar's height is the class frequency and bars touch each other to emphasise continuity. For unequal widths bars must be drawn so that the area of each bar, rather than the height, is proportional to frequency; compute frequency density = frequency / class width and use density as the bar height. Histograms allow you to visualise central tendency, spread, skewness (left or right), and the presence of multiple peaks (modes).

How to draw a histogram Choose class boundaries and label axes. On the horizontal axis mark class intervals; on the vertical axis mark frequency or frequency density. Draw contiguous rectangles spanning each class whose heights equal frequencies (or densities). Always include axis labels, units and an appropriate scale — avoid truncating axes which can mislead.

Frequency polygon and its uses A frequency polygon is formed by plotting class mid-points against class frequencies and connecting these points with straight lines. It provides the same information as a histogram but is easier to compare between groups by overlaying multiple polygons on the same axes. To close the polygon, add points at the mid-points of adjacent outside classes with frequency zero.

Ogive (cumulative frequency curve) An ogive plots cumulative frequency versus class boundary (usually the upper class boundary for the less-than ogive). It helps find medians, quartiles and percentiles graphically: draw a horizontal line at n/2 and read across to the curve, then down to the x-axis to estimate the median. Two ogives (less-than and greater-than) can be drawn; their intersection gives the median as well.

Pie charts and bar charts Pie charts display categorical data as sectors of a circle where sector angles are proportional to category frequencies (angle = frequency/total × 360°). Use pie charts for showing proportions but limit the number of categories for clarity. Bar charts are preferred for categorical comparisons because they can easily accommodate many categories and allow direct visual comparison of heights.

Best practice and interpretation Always choose the graph that suits the data and the audience. Label axes and include units, avoid misleading scales, and complement graphical displays with numerical summaries. Use multiple graphs when they add insight — for example, an ogive for medians and a histogram for shape. Finally, practise sketching these graphs by hand from frequency tables so you can interpret exam questions and perform quick checks in practical situations.

Definition and interpretation The arithmetic mean of n observations x1, x2, ..., xn is the sum of the observations divided by n. It is often called the average and is a measure of central location representing the balance point of the data. If the data are thought of as weights placed on a number line, the mean is the point where the weighted moments balance.

Calculation methods For ungrouped data compute Σ xi and divide by n to obtain x̄. When numbers are large or inconvenient, use a deviation method: choose an arbitrary origin A (often a value near the data), calculate di = xi − A, sum the deviations Σ di, then x̄ = A + (Σ di)/n. For data involving equal spacing, step-deviation where you scale deviations by a factor may be useful. These shortcut techniques reduce arithmetic work and rounding errors on paper.

Properties of the arithmetic mean The arithmetic mean is unique and lies between the smallest and largest observations. It satisfies linearity: mean(a + b xi) = a + b mean(xi). Combined means are computed using sizes and subgroup means: x̄_combined = (n1 x̄1 + n2 x̄2)/(n1 + n2). The mean minimises the sum of squared deviations: x̄ is the value that minimises Σ (xi − c)^2 over c. This optimality explains why variance uses squared deviations about the mean.

Sensitivity to outliers The arithmetic mean uses all data values and so is sensitive to extreme values or outliers. A single very large or small observation can shift the mean substantially, making it less representative of a typical observation in skewed distributions. For such data consider median or trimmed means as robust alternatives.

When to use mean Use the arithmetic mean for quantitative measurements on an interval or ratio scale when values are reasonably symmetric and outliers do not dominate. For financial averages, test scores, heights, weights and many scientific measures the mean is appropriate. Always state units and sample size when reporting the mean, and complement it with a measure of spread such as standard deviation.

Worked calculation tips Keep intermediate sums, Σ xi and Σ (xi − A), recorded and check by recomputing with a calculator. For grouped data replace xi by class mid-points and use frequencies; remember this yields an approximation dependent on class widths. Report approximate signs when grouping was used and avoid using the mean for nominal data where arithmetic operations have no meaning.

Median explained The median is the middle value of an ordered dataset and gives a central point that splits the data into two equal halves. For an odd number of observations the median is the middle item; for an even number it is the average of the two middle items. The median is particularly useful when the distribution is skewed or when outliers are present because it depends only on the order of values, not their magnitudes.

Finding median for grouped data For grouped frequency distributions the median is estimated by locating the class where the cumulative frequency reaches n/2. A linear interpolation assumes values are uniformly distributed within the median class. The grouped median formula is: median = L + [(n/2 − cfb)/f] × h where L is the lower boundary of the median class, cfb the cumulative frequency before the median class, f the class frequency and h the class width. This gives an approximate but useful value for the median when raw data are not available.

Mode and modal class Mode is the most frequently occurring value in a data set. In ungrouped data it is the value with highest frequency; there can be more than one mode if two or more values share the maximum frequency. For grouped data the modal class is the class with the highest frequency, and a formula provides an estimate of the mode within that class: mode ≈ L + [(fm − f1)/(2fm − f1 − f2)] × h where fm is frequency of modal class, f1 and f2 are frequencies of preceding and succeeding classes, L is lower boundary and h is class width. This interpolation gives a location of the peak inside the modal class.

Comparing median and mode Median gives a central location with resistance to outliers, while mode identifies the most common outcome. Use mode for categorical data to identify the most popular category; median for skewed numerical data. In symmetric unimodal distributions mean = median = mode; when skewed these measures differ in predictable ways and their relative positions indicate direction of skewness.

Practical considerations Report the method used to compute median (grouped or ungrouped) and note any approximation. For small datasets calculate exact median and mode from ordered values. When data are categorical with natural order (ordinal) medians and percentiles may still be meaningful. For multimodal distributions describe all modes or state that multiple peaks exist; the presence of multiple modes suggests a mixture of sub-populations.

Geometric mean — concept and use The geometric mean (G.M.) of n positive numbers x1, x2, ..., xn is the nth root of their product, (Π xi)^{1/n}. It is most appropriate when data are multiplicative in nature, such as growth rates, relative changes, or index numbers. For example, average annual growth rate of an investment is best computed using the geometric mean of annual multipliers (1 + r). Geometric mean gives a central tendency that dampens the effect of very large values compared to the arithmetic mean.

Computing G.M. with logs Direct multiplication of many numbers can be impractical; use logarithms: ln(G.M.) = (1/n) Σ ln(xi) and then exponentiate. This method reduces rounding errors and simplifies calculations on a calculator or by hand when values cover several orders of magnitude. For grouped frequency data, treat each midpoint x_i with frequency f_i and compute G.M. = (Π x_i^{f_i})^{1/N} = exp[(1/N) Σ f_i ln x_i].

Harmonic mean — concept and use The harmonic mean (H.M.) of n positive numbers is defined as n divided by the sum of reciprocals: H.M. = n / (Σ 1/xi). It is appropriate for averaging rates when the quantity of interest is formed by combining reciprocals, for example average speed when equal distances are covered at different speeds. For equal distances the harmonic mean gives the correct overall average speed. Harmonic mean is dominated by small values, so a single small observation can greatly reduce the H.M.

Relations among means For any positive dataset the inequality H.M. ≤ G.M. ≤ A.M. holds, with equality only if all values are equal. These relations show that the arithmetic mean is largest and harmonic mean smallest for the same positive data. This ordering helps decide which measure to use: arithmetic for additive contexts, geometric for multiplicative contexts, harmonic for rate-based contexts.

Grouped data and frequencies For grouped data you can compute approximate G.M. and H.M. using class mid-points and frequencies. G.M. = (Π x_i^{f_i})^{1/N} and H.M. = N / Σ (f_i / x_i) where N = Σ f_i. Remember these are approximations, dependent on using mid-points and on class widths, so report results as estimates.

Practical examples and interpretation Use geometric mean for average growth: if a population grows by 10%, 5% and 15% over three years, convert to multipliers 1.10,1.05,1.15 and compute G.M. to find average annual multiplier and subtract 1 to get average percentage growth. Use harmonic mean when averaging speeds for equal distances. Always state why a particular mean is chosen based on the context to avoid misuse.

Range — simplest spread measure The range is the difference between the maximum and minimum values and gives a quick measure of total spread. While easy to compute, the range is highly sensitive to outliers and does not reflect how data are distributed between extremes. Use range for simple checks but not as the sole descriptor of spread for important decisions.

Quartiles — dividing data into four parts Quartiles divide ordered data into four equal parts. The first quartile Q1 is the value below which 25% of observations lie; the second quartile Q2 is the median (50%); the third quartile Q3 is the value below which 75% of observations lie. Quartiles are robust and summarise the location of the central part and tails of a distribution. Different conventions exist for computing quartiles when sample sizes do not divide evenly; be consistent and state the convention if needed.

Interquartile range (IQR) — robust spread measure IQR = Q3 − Q1 measures the spread of the middle 50% of data and is resistant to extreme values. Because it ignores the outer 25% on both sides, it gives a stable sense of typical variability for skewed or heavy-tailed distributions. IQR is widely used in exploratory data analysis and as the basis for identifying potential outliers.

Outlier detection using IQR A common rule of thumb: observations below Q1 − 1.5×IQR or above Q3 + 1.5×IQR are considered potential outliers. This rule is heuristic and useful for flagging values that merit further investigation, but context should guide the final decision on whether to exclude or keep such points.

Quartiles in grouped data For grouped frequency distributions locate positions kN/4 for Qk (k=1,3) using cumulative frequencies. Apply linear interpolation within the class containing the quartile position: Qk = L + [(kN/4 − cfb)/f] × h where L is lower class boundary of the quartile class, cfb cumulative before that class, f class frequency and h class width. This produces an approximate quartile value suitable for large grouped datasets.

Visual display and interpretation Box-and-whisker plots (boxplots) display minimum, Q1, median, Q3 and maximum; outliers are plotted individually. Boxplots are compact and effective for comparing distributions across groups. When reporting quartiles and IQR, always include sample size and note that quartiles are positional statistics which provide different information from mean and standard deviation.

Mean deviation — what and why Mean deviation (often called average absolute deviation) from a central value measures the average absolute distance of observations from that central value. About the mean it is MD = (Σ |xi − x̄|)/n. Because absolute values remove sign, MD expresses typical deviation in the original units and is easily understood. However, because of the absolute values it lacks simple algebraic properties, making it less convenient for theoretical work than variance.

Choosing centre for MD Mean deviation can be computed about the mean, the median or any other reference point. For skewed distributions MD about the median is often smaller and more representative. MD is particularly useful in descriptive contexts where robustness and interpretability matter more than algebraic simplicity.

Variance and standard deviation — precise spread measures Variance measures average squared deviation from the mean. For a population variance σ^2 = (Σ (xi − μ)^2)/N; for a sample the usual sample variance is s^2 = (Σ (xi − x̄)^2)/(n − 1) which corrects bias in estimating the population variance. Standard deviation is the positive square root of variance and is expressed in the same units as data, making it easier to interpret. Standard deviation and variance are the backbone of many statistical methods because of their favourable mathematical properties.

Why square deviations? Squaring deviations penalises larger deviations more strongly and produces a differentiable function of the centre; this leads to the property that the sum of squared deviations is minimised at the arithmetic mean. This algebraic convenience yields important theoretical results and simplifies work with sums of independent variables.

Computational formula and tips Use the shortcut (computational) formula to compute variance efficiently: Σ (xi − x̄)^2 = Σ xi^2 − (Σ xi)^2/n. This avoids subtracting the mean from every value repeatedly and is handy for manual calculations. When using calculators keep intermediate sums Σ xi and Σ xi^2 to maintain accuracy. For small samples be careful whether your context requires population or sample variance.

Interpretation and rule-of-thumb Standard deviation gives a measure of spread: for distributions approximately normal, about 68% of observations lie within ±1σ of the mean and about 95% within ±2σ. These are guidelines, not exact rules for non-normal data. Always report both a measure of central tendency and spread together: e.g., mean = 50, s = 6, n = 30, to give a clear picture of the data.

Why grouped formulas differ When observations are grouped into classes, exact values are not available so we approximate by representing all values in a class by its midpoint x_i. Using midpoints and frequencies f_i we obtain approximate sums Σ f_i x_i and Σ f_i x_i^2 necessary for mean and variance. Though approximate, this method is practical for large datasets where raw lists are unavailable.

Formulas for grouped data Let N = Σ f_i and x̄ = (Σ f_i x_i)/N where x_i are midpoints. Then population variance approximation is σ^2 ≈ [Σ f_i x_i^2 / N] − (x̄)^2. For sample variance use s^2 ≈ [Σ f_i x_i^2 − (Σ f_i x_i)^2 / N] / (N − 1). These formulas come from substituting grouped sums into the usual variance formulas and are standard in manual calculation.

Assumed mean and step deviation methods To reduce arithmetic and rounding errors, use an assumed mean A close to x̄ and compute deviations d_i = x_i − A. Then compute Σ f_i d_i and Σ f_i d_i^2 which often have smaller numbers. Alternatively use step-deviation dividing d_i by a common class width and work with smaller integers. After computing variance in terms of deviations, convert back to obtain the final variance. These methods are especially helpful in exams where time is limited.

Accuracy and class width The approximation improves as class widths become smaller and the distribution of values within classes is more uniform. Very wide classes or open-ended classes (like ‘50 and above’) reduce precision and complicate midpoint choice. If possible refine classes or obtain raw data for accurate computation. Always mention that grouped results are approximate and may differ slightly from computations on raw data.

Practical worked approach Step 1: find midpoints x_i and write down frequencies f_i. Step 2: compute f_i x_i and f_i x_i^2 and sum these columns. Step 3: compute grouped mean x̄ = Σ f_i x_i / N and grouped variance by the stated formula. Step 4: take square root for standard deviation and report units, rounding reasonably. Check arithmetic by recomputing sums or using assumed mean method for verification.

Interpreting grouped standard deviation The standard deviation for grouped data represents approximate spread about the grouped mean. Use it to compare variability between similar datasets or to inform further analysis. When comparing datasets with different units, convert to coefficient of variation to compare relative dispersion reliably.

Need for relative measures Absolute measures of spread such as standard deviation are useful, but when comparing variability across datasets with different units or widely different means a relative measure is required. The coefficient of variation (C.V.) expresses spread relative to the mean, giving a unitless percentage that allows direct comparison between disparate datasets.

Definition and interpretation The coefficient of variation for a sample is C.V. = (s / x̄) × 100%. For a population use σ/μ × 100%. A lower C.V. indicates less relative variability; a higher C.V. indicates more variability compared to the typical size of observations. C.V. is especially useful in fields such as finance (risk per unit return), biology (variation relative to mean), and engineering (consistency of measurements).

When C.V. is appropriate Use C.V. when data are measured on a ratio scale (mean is positive and meaningful). Do not use C.V. when mean is zero or close to zero because it becomes unstable or undefined. Also avoid using C.V. for data on interval scales where zero does not represent absence of the measured quantity.

Alternative relative measures Relative mean deviation = (MD / x̄) × 100% gives a robust relative spread measure using absolute deviations. Quartile deviation QD = (Q3 − Q1)/2 can be expressed relative to median or mean to give insight into middle spread. Choose the relative measure based on the nature of the data and whether robustness or algebraic properties are more important.

Examples and comparison If two datasets have identical C.V., they have the same ratio of standard deviation to mean even if absolute means differ. This means they have similar proportional variability. When comparing experimental methods or instruments, a lower C.V. signals more precise measurements relative to the mean reading.

Reporting and limitations When reporting C.V. give sample size and state whether sample or population values are used. Remember that C.V. conveys relative variability but not shape or outliers; combine it with other summaries and graphical checks to give a full picture of the data. Contextual standards are needed to judge whether a particular C.V. should be considered large or small.

Random experiment and outcomes A random experiment is an action or process with more than one possible result, such as tossing a coin, rolling a die, or measuring daily rainfall. The set of all possible outcomes is the sample space, denoted S. An event is any subset of S; for example, in a die roll the event ‘even number’ is {2,4,6}. Clearly specifying S and events is the first step to computing probabilities.

Classical (equally likely) probability When outcomes in S are equally likely, the probability of an event A is P(A) = number of outcomes in A divided by total number of outcomes in S. This simple formula applies to fair coins, symmetric dice and many combinatorial problems. Counting carefully is essential: ensure outcomes are listed properly without omission or duplication.

Axioms of probability Modern probability is built on three axioms: (1) For any event A, 0 ≤ P(A) ≤ 1. (2) P(S) = 1. (3) For mutually exclusive events A and B (A ∩ B = ∅), P(A ∪ B) = P(A) + P(B). From these axioms follow useful results such as P(∅) = 0 and P(A') = 1 − P(A) where A' denotes the complement of A.

Frequentist interpretation The frequentist view treats probability as a long-run relative frequency: if an experiment is repeated many times under identical conditions, the proportion of times event A occurs approaches P(A). This empirical approach underlies simulation and estimation of probabilities when theoretical calculation is difficult or when trials produce approximate random behaviour.

Common mistakes to avoid Be precise about the sample space and ensure events are correctly defined. Avoid treating dependent outcomes as independent (for example drawing without replacement). Remember that probabilities add for mutually exclusive events but require subtraction of overlap otherwise. Ensure that probabilities assigned to outcomes sum to 1 in discrete models.

Practical applications Probability provides a formal language to model uncertainty, evaluate risks and make decisions. It is used in games of chance, reliability studies, weather forecasts, quality control and many branches of science. Practise both theoretical counting-based problems and empirical estimation through experiments or simulations to build intuition about probability values and their interpretation.

Addition rule — avoiding double counting The addition rule gives the probability that at least one of two events occurs. For any events A and B, P(A ∪ B) = P(A) + P(B) − P(A ∩ B). The subtraction of P(A ∩ B) corrects for counting the intersection twice when adding P(A) and P(B). When A and B are mutually exclusive (cannot both occur simultaneously), P(A ∩ B) = 0 and the rule reduces to P(A ∪ B) = P(A) + P(B). This rule extends to more than two events using the inclusion–exclusion principle, but computations become more elaborate as the number of events increases.

Multiplication rule — joint probabilities and conditionality The multiplication rule computes the probability that both events A and B occur. It states P(A ∩ B) = P(A) × P(B|A). Here P(B|A) is the probability that B occurs given that A has occurred. This formulation explicitly handles dependence between events. If A and B are independent — meaning occurrence of one does not change the probability of the other — then P(B|A) = P(B) and P(A ∩ B) = P(A)P(B). Independence must be established by reasoning or calculation, not assumed.

Extension to sequences and tree diagrams For several sequential events, use conditional probabilities step by step. Tree diagrams are an effective visual tool: list possible outcomes at each stage with branch probabilities, multiply along branches to get joint probabilities, and sum appropriate branches to obtain probabilities of composite events. This method is especially helpful for multi-step sampling without replacement or for multi-stage processes.

Special cases and combinatorial checks Use combinations and permutations to count favourable outcomes when outcomes are equally likely; then apply classical probability. For sampling without replacement from finite populations, compute probabilities using changing denominators to reflect reduced population. Always check whether independence applies: with sampling without replacement events are usually dependent unless the sample fraction is negligible.

Practical examples Drawing two cards from a deck without replacement: probability both are aces = (4/52) × (3/51). Two independent coin tosses: probability both heads = (1/2) × (1/2) = 1/4. When combining rules, be careful to identify intersections and condition on appropriate events to avoid mistakes.

Conditional probability — updating the sample space Conditional probability P(A|B) quantifies the chance of event A happening given that event B has already occurred. It is defined by P(A|B) = P(A ∩ B)/P(B), provided P(B) > 0. Conceptually it restricts the sample space to B and measures how many of those outcomes also satisfy A. Conditioning is essential when information reduces uncertainty and is widely used in sequential experiments and decision problems.

Interpretation with examples Consider drawing from a deck: the probability of drawing a queen (A) given the card is a face card (B) is P(A|B) = P(queen ∩ face)/P(face) = (4/52)/(12/52) = 4/12 = 1/3. In real contexts conditional probabilities appear in medical testing (probability of disease given a positive test) and reliability (probability of component failure given stress).

Independence defined precisely Two events A and B are independent if P(A|B) = P(A), which is equivalent to P(A ∩ B) = P(A)P(B). Independence means knowledge that B occurred gives no information about A. Note independence is a probabilistic property, not a causal statement; independent events can be unconnected or may be related in complex ways where probabilities nonetheless factorise.

Law of total probability When events B1, B2, ..., Bk partition the sample space and P(Bi) > 0, any event A can be decomposed as A ∩ (B1 ∪ ... ∪ Bk). The law of total probability states P(A) = Σ P(A|Bi) P(Bi). This is practical when direct calculation of P(A) is hard but conditional probabilities given each Bi are available. The law is the basis for Bayes' theorem and many applied calculations.

Common pitfalls Do not interchange conditioning direction without applying Bayes' theorem; P(A|B) and P(B|A) are generally different. Verify that denominators in conditional probabilities are non-zero. When sampling without replacement events are dependent; use conditional probabilities to account for changing composition. Use tree diagrams to keep track of sequential conditioning when required.

Applications and practice Solve problems involving sequential draws, multi-stage processes and diagnostic testing to build skill. Practice rewriting joint probabilities as products using conditional probabilities and check independence by verifying whether P(A ∩ B) equals P(A)P(B). These techniques form the backbone of applied probability work.

What Bayes' theorem does Bayes' theorem reverses conditional probabilities: it tells us how to compute the probability of a cause given that an effect has been observed, using prior probabilities of causes and likelihoods of observing the effect under each cause. It formalises learning from evidence and is widely used in diagnostics, decision-making and any situation where information updates beliefs.

Formal statement Suppose B1, B2, ..., Bk partition the sample space with P(Bi) > 0 and A is an event with P(A) > 0. Bayes' theorem states that for any j, P(Bj|A) = [P(A|Bj) P(Bj)] / Σ_i P(A|Bi) P(Bi). The numerator is the prior probability of Bj multiplied by the likelihood of A under Bj; the denominator, by the law of total probability, equals P(A). The result is the posterior probability of Bj after observing A.

Step-by-step application 1) Identify the mutually exclusive, exhaustive causes Bi and their prior probabilities P(Bi). 2) For each cause compute the likelihood P(A|Bi). 3) Compute P(A) = Σ P(A|Bi)P(Bi). 4) Compute the posterior P(Bj|A) using the formula. Present results as probabilities (or percentages) and interpret carefully. Bayes' theorem requires accurate priors and likelihoods; errors in these inputs affect the posterior significantly.

Interpretation and counter-intuition Bayes' results can seem surprising: even with a highly accurate test (high sensitivity), if the prior probability (prevalence) is very low, the posterior probability after a positive test may still be low. This emphasises the importance of the prior and explains why screening tests in low-prevalence populations often produce many false positives relative to true positives.

Practical examples and tree diagrams A standard example: medical testing. Suppose disease prevalence is 1%, sensitivity 95% and false positive rate 5%. Use Bayes' theorem or a tree diagram to compute posterior probability of disease given a positive test. Tree diagrams help visualise priors as branch probabilities and likelihoods along the next level, with leaf joint probabilities summing to P(A).

Limitations and responsibility Bayes' theorem is mathematically exact but relies on correct models for priors and likelihoods. Priors may come from past data or expert judgement; different priors lead to different posteriors. In practice be explicit about the priors used and check sensitivity of results to reasonable changes in priors. Teaching students to apply Bayes' theorem carefully helps avoid misinterpretation of diagnostic and probabilistic claims.

Random variable and pmf A discrete random variable X assigns a numerical value to each outcome of an experiment and takes on countable values x1, x2, ... For such X the probability mass function (pmf) gives probabilities p_i = P(X = x_i). The pmf must satisfy p_i ≥ 0 for all i and Σ p_i = 1. The pmf summarises the distribution fully and is the starting point for computing probabilities and expectations.

Expectation — meaning and calculation The expectation or mean E(X) = Σ x_i p_i is the long-run average value of X if the experiment is repeated many times. Expectation is linear and provides a central value for the distribution. It is often called the theoretical mean and should be distinguished from the sample mean computed from observed data.

Variance and spread Variance Var(X) = E[(X − E(X))^2] measures the spread of X around its mean. It can be computed using Var(X) = E(X^2) − [E(X)]^2 where E(X^2) = Σ x_i^2 p_i. The standard deviation is the square root of variance and provides a spread measure in the same units as X. For many problems computing E(X) and Var(X) is sufficient to describe the first two moments of the distribution.

Common discrete distributions Class 11 focuses primarily on Bernoulli and binomial distributions. Bernoulli describes a single trial with two outcomes (success with probability p and failure with q = 1 − p). Binomial counts the number of successes in n independent Bernoulli trials and has pmf P(X = k) = C(n,k) p^k q^{n−k}. Another simple discrete model is the uniform discrete distribution where each of a finite set of outcomes is equally likely.

Expectation of functions and linearity Expectation generalises to functions g(X): E[g(X)] = Σ g(x_i) p_i. A key property is linearity: E(aX + bY + c) = aE(X) + bE(Y) + c without needing independence. This property simplifies calculations for sums and transformations of random variables.

Practical interpretation Use discrete distributions to model counts, number of successes, dice outcomes and categorical-coded numeric data. Compute probabilities to assess chance of particular outcomes and use E(X) and Var(X) to summarise expected behaviour and variability. Always check that probabilities sum to 1 and that support of the pmf matches the problem context.

Definition and assumptions The binomial distribution models the number of successes in n independent trials where each trial has exactly two outcomes (success with probability p and failure with probability q = 1 − p). Key assumptions are identical trials, fixed number n, binary outcomes and independence. Under these conditions the random variable X, the number of successes, follows the binomial distribution denoted Bin(n, p).

Probability mass function The probability of exactly k successes in n trials is P(X = k) = C(n, k) p^k q^{n−k}, where C(n,k) is the binomial coefficient n!/(k!(n−k)!). The coefficient counts the number of sequences of length n containing k successes. The factor p^k q^{n−k} gives the probability of any specific sequence with k successes and n−k failures.

Mean and variance and interpretation For X ~ Bin(n, p) the expected number of successes is E(X) = np and variance Var(X) = npq. Standard deviation is sqrt(npq). The mean np provides the central tendency while npq shows how much variation around that mean to anticipate. These formulas follow by viewing X as the sum of n independent Bernoulli( p ) indicator variables and applying linearity of expectation and additivity of variance for independent terms.

Using the binomial in practice Binomial models are used in quality control (number of defective items in a sample), surveys (number of positive responses), biological experiments (number of successes in trials), and reliability tests. To compute probabilities for moderate n use calculators or binomial tables; for small n compute directly. For manual exam problems small values of n and p are chosen to keep arithmetic manageable.

Approximations For large n with p not too close to 0 or 1, the normal approximation to the binomial can be used with continuity correction: approximate X by N(np, npq). For large n and small p with np moderate, Poisson approximation (with parameter λ = np) is sometimes appropriate (introduced later). Always check approximation conditions before using them.

Computational tips For successive probabilities use ratio P(k+1)/P(k) = [(n − k)/(k + 1)](p/q) to compute probabilities recursively. When asked for cumulative probabilities compute sums of pmf values or use tables/technology. Interpret numerical results in context and remember boundary cases: P(0) = q^n, P(n) = p^n.

Expectation of a function If X is a discrete random variable with pmf p(x), the expectation of any function g(X) is E[g(X)] = Σ g(x) p(x). This general result lets us compute expected values of transformations such as squares, indicators, payoffs and other derived quantities without finding the distribution of g(X) itself. It is a powerful and flexible tool in probability and statistics.

Linearity of expectation A central and extremely useful property is linearity: for any random variables X and Y (not necessarily independent) and constants a, b, c, E[aX + bY + c] = aE(X) + bE(Y) + c. Linearity allows easy computation of expected sums, even when variables are dependent, and greatly simplifies analysis of sums of random variables in sampling, games and experiments.

Variance and the role of independence Variance does not satisfy the same simple linearity: Var(aX + b) = a^2 Var(X), but for two variables Var(X + Y) = Var(X) + Var(Y) + 2Cov(X, Y) where Cov(X, Y) = E[(X − E(X))(Y − E(Y))]. If X and Y are independent then Cov = 0 and variances add: Var(X + Y) = Var(X) + Var(Y). Independence gives simpler variance computations, but linearity of expectation holds even without independence, making expectation easier to handle in many settings.

Applications to sums and repeated trials For n independent and identically distributed (i.i.d.) random variables X1, X2,... with common expectation μ and variance σ^2, the sum S_n = Σ Xi has expectation E(S_n) = nμ and variance Var(S_n) = nσ^2. This result is the basis for many sampling formulas and predictions about aggregate behaviour such as total counts or cumulative returns.

Worked techniques To compute E[g(X)] list the possible x values, compute g(x) and multiply by p(x), then sum. Use linearity to break complicated expectations into simpler parts: E(a + bX + cX^2) = a + bE(X) + cE(X^2). For problems involving indicators use E(I_A) = P(A) to translate event probabilities into expectations and vice versa. Practise with examples to see how expectations simplify calculations in probability models.

Interpretation Expectation provides a central, predictive value — the average outcome over many repeated trials — and linearity allows building expectations of complex expressions from simpler components. Understanding when variances add and when covariances matter is essential for correct assessment of uncertainty in sums and for designing experiments and samples.

Population versus sample In practical statistics we are often interested in properties of a large population (mean income in a city, average height of students) but cannot measure every unit. A sample is a subset selected for study, and sample statistics (sample mean, sample proportion) are used to estimate population parameters. The quality of these estimates depends on how the sample was chosen and its size.

Sampling methods — basics Several basic sampling methods are used in practice. Simple random sampling gives every subset of a given size equal chance; it is conceptually clean but may be hard to implement without a list of population members. Stratified sampling divides the population into homogeneous groups (strata) and samples within each stratum, improving precision especially when strata differ. Systematic sampling selects every kth unit from an ordered list — practical but sensitive to hidden patterns in the list. Convenience or voluntary samples are easy but can be biased and unreliable.

Sampling error and bias Sampling error is the random variation that results because a sample is only a subset of the population; it decreases as sample size increases. Bias is systematic error due to the sampling method (e.g., selecting volunteers, using poor response rates), measurement error or poor question design. While sampling error can be reduced by larger samples, bias must be addressed by better design and careful selection procedures.

Law of large numbers (informal) The law of large numbers states informally that as the sample size increases, the sample mean tends to get closer to the population mean. This justifies using larger samples for more reliable estimates. However, large size does not correct for biased sampling; representative design remains essential.

Point estimates and properties A point estimate provides a single value (e.g., sample mean x̄) to estimate a population parameter. Important properties include bias (difference between expected estimate and true parameter), consistency (estimator tends to the true value as sample size grows) and efficiency (smaller variance among unbiased estimators). These theoretical ideas appear later in detail but the intuition helps evaluate estimators in practice.

Practical guidance for surveys Before collecting data define population, sampling frame and method, choose sample size considering resources, and plan for non-response. Record how the sample was drawn when reporting results. When stating estimates give context and limitations; always present the sample size, method and a measure of variability where possible so readers can judge reliability.

Statistics describes; probability models Statistics summarises observed data and provides descriptive measures such as means and variances. Probability supplies models for randomness and uncertainty and is used to predict or infer what might happen under repeated trials. Linking the two allows us to make inferences from sample data about population parameters, but only when assumptions and sampling design are valid.

Common misuses and pitfalls Several errors occur frequently: confusing correlation with causation (two variables may move together due to a third factor), drawing strong conclusions from small or biased samples, misreading p-values or over-interpreting insignificant differences, and manipulating graphs or scales to give misleading impressions. Always ask how the data were collected and whether the sample represents the population of interest.

Averages can be misleading Different averages tell different stories. The arithmetic mean can be affected by outliers; the median might be a better summary for skewed distributions. Reporting just the mean without a measure of spread (like standard deviation or IQR) hides variability. For example, two classes with similar means may differ greatly in consistency; checking standard deviations or boxplots reveals that difference.

Role of probability in inference Probability underlies the idea that observed sample results may differ from population parameters due to randomness. Formal inference (confidence intervals, hypothesis tests) quantifies uncertainty, but even informal reasoning should consider sampling variability. Unusual-looking results in a single sample could be due to chance rather than a systematic effect; probability helps judge how surprising an observation is under a given model.

Responsible reporting and graphical honesty Present data with clear labels, units, sample size and method. Use honest scales (avoid truncating axes that exaggerate differences) and prefer percentages or rates when comparing groups of different sizes. When choosing graphs, select ones that truthfully represent the underlying data and avoid decorative features that obscure interpretation.

Critical thinking with examples Examine claims based on data: ask who collected the data, what the sampling method was, whether confounding factors were considered, and whether reported statistics are appropriate. Practise critically evaluating real-world news stories, reports and graphs to build judgement. Understanding limitations and communicating uncertainty are as important as computing correct numbers.