STATISTICAL

METHODS
Comprehensive Study Notes

10 Key Statistical Topics

Curve Fitting • Time Series • Probability • Hypothesis Testing

Distributions • ANOVA • Chi-Square • F-Test

1. Least Squares Parabola Fitting 6. Markov Chain

2. Moving Average Method 7. One-Way ANOVA
3. Exponential Curve Fitting 8. Student's t-Test
4. Joint Probability Distribution 9. Chi-Square Test
5. Binomial Distribution 10. F-Test (Variance Ratio Test)

Prepared for Exam Reference • Statistics & Probability

STATISTICAL METHODS Detailed Study Notes

Table of Contents
No. Topic Page*

1 Least Squares Parabola Fitting —

2 Moving Average Method —

3 Exponential Curve Fitting —

4 Joint Probability Distribution —

5 Binomial Distribution —

6 Markov Chain —

7 One-Way ANOVA —

8 Student's t-Test —

9 Chi-Square Test —

10 F-Test (Variance Ratio Test) —

Quick Revision Summary —

* Page numbers are approximate; refer to section headers.

Statistical Methods — Comprehensive Study Notes Page 2

STATISTICAL METHODS Detailed Study Notes

1
Least Squares Parabola Fitting

Overview
When data follows a curved pattern rather than a straight line, we use a quadratic (parabolic) model. The
method of least squares finds the coefficients a, b, and c that minimise the total squared error between the fitted
curve and the actual data points. This is an extension of linear regression into second-degree polynomials.

Key Formulas

Model: y = ax² + bx + c

Normal Equation 1: Σy = aΣx² + bΣx + cn

Normal Equation 2: Σxy = aΣx³ + bΣx² + cΣx

Normal Equation 3: Σx²y = aΣx■ + bΣx³ + cΣx²

Worked Example

For the dataset x = {1, 2, 3, 4, 5} and y = {14, 27, 40, 55, 68}, compute the sums Σx, Σx², Σx³, Σx■, Σy, Σxy,
Σx²y and solve the 3×3 normal equation system. The resulting fitted parabola is: y ≈ 0.1429x² + 12.7429x + 1
This curve closely follows the increasing curvature in the data.

Use Cases

Production growth curves, population forecasting, projectile motion analysis, engineering stress-strain
relationships, and any process exhibiting accelerating or decelerating change.

Statistical Methods — Comprehensive Study Notes Page 3

STATISTICAL METHODS Detailed Study Notes

2
Moving Average Method

Overview
A moving average is a time-series smoothing technique that replaces each data value with the average of a
surrounding window of values. By doing so, short-term random fluctuations are dampened and the underlying
long-term trend becomes visible. The width of the window determines how much smoothing occurs.

Key Formulas

3-period MA: MA_t = (x_{t-1} + x_t + x_{t+1}) / 3

4-period MA: MA_t = (x_t + x_{t+1} + x_{t+2} + x_{t+3}) / 4

Even-period: use centered moving average to align with original time axis

Trend value = centred average of two consecutive even-period averages

Worked Example

Monthly sales data: 75, 60, 55, 60, 65, 70, 80, 75, ... 4-period MA at period 1: (75 + 60 + 55 + 60) / 4 = 62.5
4-period MA at period 2: (60 + 55 + 60 + 65) / 4 = 60.0 Centred MA (between positions 1 and 2): (62.5 + 60.0)
/ 2 = 61.25 This centred value aligns with period 2.5 and is placed at period 3 for even windows.

Use Cases

Retail sales analysis, stock market trend identification, weather pattern smoothing, economic indicator
analysis, and inventory forecasting in supply chain management.

Statistical Methods — Comprehensive Study Notes Page 4

STATISTICAL METHODS Detailed Study Notes

3
Exponential Curve Fitting

Overview
When a quantity increases or decreases by a constant percentage per unit time, it follows an exponential model.
Taking the natural logarithm of both sides linearises the relationship, allowing standard linear regression
techniques to estimate the parameters. This is one of the most widely used models in natural and social
sciences.

Key Formulas

Model: y = ae^{bx}

Linearise: ln(y) = ln(a) + b·x

Let Y = ln(y) and A = ln(a); then Y = A + bx

Estimate b and A by least squares on (x, Y) pairs

Recover: a = e^A

Worked Example

For rapidly increasing data (e.g. bacterial colony count over time), after computing (x, ln y) pairs and applying
linear regression, suppose we obtain: ln(y) = 3.090 + 0.2983x Then a = e^{3.090} ≈ 22.03 and b = 0.2983
Fitted model: y = 22.03 e^{0.2983x} At x = 5: y ≈ 22.03 × e^{1.49} ≈ 98.6 (count units)

Use Cases

Bacterial and viral growth, radioactive decay, compound interest, population dynamics, epidemic spread
modelling, and cooling/heating processes (Newton's law of cooling).

Statistical Methods — Comprehensive Study Notes Page 5

STATISTICAL METHODS Detailed Study Notes

4
Joint Probability Distribution

Overview
A joint probability distribution characterises the simultaneous behaviour of two or more random variables. From
it we can derive marginal distributions for each variable separately, compute expected values, measure
covariance (how variables move together), and calculate the Pearson correlation coefficient to judge the
strength and direction of the relationship.

Key Formulas

E(X) = Σ■ x · P(X = x) [marginal expectation of X]

E(Y) = Σ■ y · P(Y = y) [marginal expectation of Y]

E(XY) = Σ■ Σ■ xy · P(X=x, Y=y)

Cov(X,Y) = E(XY) − E(X)·E(Y)

ρ = Cov(X,Y) / (σ_X · σ_Y) [−1 ≤ ρ ≤ 1]

Worked Example

Let X ∈ {1, 5} and Y ∈ {−4, 2, 7} with a known joint table. Summing rows/columns gives marginal probabilities.
Suppose E(X) = 2.6, E(Y) = 1.4, E(XY) = 1.8. Cov(X,Y) = 1.8 − (2.6)(1.4) = 1.8 − 3.64 = −1.84 A negative
covariance indicates that when X is large, Y tends to be small (negative association). After finding σ_X and
σ_Y, compute ρ to get the standardised correlation.

Use Cases

Studying height vs weight, price vs demand, temperature vs energy consumption, insurance risk modelling,
financial portfolio analysis, and bivariate quality control.

Statistical Methods — Comprehensive Study Notes Page 6

STATISTICAL METHODS Detailed Study Notes

5
Binomial Distribution

Overview
The binomial distribution models the number of successes in n independent, identical Bernoulli trials, each with
success probability p. It assumes: (i) fixed number of trials n, (ii) each trial has exactly two outcomes, (iii) trials
are independent, and (iv) p is constant across all trials. It is the cornerstone of discrete probability models.

Key Formulas

P(X = x) = C(n,x) · p^x · (1−p)^{n−x} for x = 0, 1, 2, ..., n

C(n,x) = n! / [x!(n−x)!] (binomial coefficient)

Mean: E(X) = np

Variance: Var(X) = np(1−p)

Std Dev: σ = √[np(1−p)]

Worked Example

A production line has defect probability p = 0.4 per item. In a batch of n = 5: P(X=0) = C(5,0)(0.4)■(0.6)■ ≈
0.0778 P(X=2) = C(5,2)(0.4)²(0.6)³ = 10 × 0.16 × 0.216 ≈ 0.3456 E(X) = 5 × 0.4 = 2.0 (expected defects per
batch) Var(X) = 5 × 0.4 × 0.6 = 1.2

Use Cases

Quality control and acceptance sampling, clinical trial pass/fail analysis, election polling, genetics (allele
inheritance), insurance modelling, and reliability engineering.

Statistical Methods — Comprehensive Study Notes Page 7

STATISTICAL METHODS Detailed Study Notes

6
Markov Chain

Overview
A Markov chain is a stochastic process satisfying the Markov property: the probability of transitioning to any
future state depends only on the current state, not on the history of past states. The process is fully described by
a transition probability matrix P whose rows sum to 1. Steady-state (long-run) probabilities give the limiting
behaviour.

Key Formulas

State vector after k steps: π_k = π_0 · P^k

Transition matrix entry: P_{ij} = P(next state = j | current state = i)

Row sums: Σ_j P_{ij} = 1 for every state i

Steady-state condition: π · P = π with Σ_i π_i = 1

Worked Example

Weather model: states = {Sunny, Rainy}. Transition matrix: P = [[0.8, 0.2], [0.4, 0.6]] If today is Sunny (π_0 =
[1, 0]): After 1 day: π_1 = [0.8, 0.2] After 2 days: π_2 = [0.8²+0.2×0.4, ...] = [0.72, 0.28] Steady state: π = [2/3,
1/3] ≈ [0.667, 0.333]

Use Cases

Weather prediction, Google PageRank algorithm, customer brand switching models, queueing theory (M/M/1
queues), speech recognition (HMMs), and game theory.

Statistical Methods — Comprehensive Study Notes Page 8

STATISTICAL METHODS Detailed Study Notes

7
One-Way ANOVA

Overview
Analysis of Variance (ANOVA) tests whether the means of three or more independent groups are equal. Rather
than running multiple t-tests (which inflates Type I error), ANOVA partitions total variance into between-group
variance (explained by group differences) and within-group variance (random error). The F-ratio compares these
two sources.

Key Formulas

Grand mean: X■■ = (Σ all observations) / N

SSB (between): Σ n_i (X■_i − X■■)² df = k − 1

SSW (within): Σ■ Σ_j (X_{ij} − X■_i)² df = N − k

MSB = SSB / (k−1); MSW = SSW / (N−k)

F = MSB / MSW ~ F(k−1, N−k) under H■

Worked Example

Three machines A1, A2, A3 produce items per hour: A1={45,47,50,52}, A2={38,40,41,43}, A3={55,57,60,62}.
Compute group means: X■_1=48.5, X■_2=40.5, X■_3=58.5; Grand mean X■■=49.17 SSB =
4[(48.5−49.17)²+(40.5−49.17)²+(58.5−49.17)²] = 641.5 SSW = sum of squared deviations within each group =
48.5 F = (641.5/2) / (48.5/9) = 320.75 / 5.39 = 59.5 >> F_crit(2,9) = 4.26 Decision: Reject H■ — machine
means differ significantly.

Use Cases

Comparing teaching methods, drug efficacy across patient groups, fertiliser effects on crop yield, machine
performance comparison, and marketing channel effectiveness.

Statistical Methods — Comprehensive Study Notes Page 9

STATISTICAL METHODS Detailed Study Notes

8
Student's t-Test

Overview
The t-test is used when the sample size is small (typically n < 30) and the population standard deviation is
unknown. It tests whether a sample mean significantly differs from a hypothesised population mean. The test
statistic follows a t-distribution with n−1 degrees of freedom, which has heavier tails than the normal to account
for sample uncertainty.

Key Formulas

t = (x■ − µ) / (s / √n)

x■ = sample mean; µ = hypothesised population mean

s = sample std dev = √[Σ(x_i − x■)² / (n−1)]

Degrees of freedom: df = n − 1

Reject H■ if |t_calc| > t_{crit}(α, df)

Worked Example

Sample heights (cm): 63, 63, 66, 67, 68, 69, 70, 70, 71, 71 (n=10) H■: µ = 66 cm H■: µ ≠ 66 cm (two-tailed,
α=0.05) x■ = 67.8 cm; s = 3.01 cm t = (67.8 − 66) / (3.01 / √10) = 1.8 / 0.952 = 1.890 t_crit(0.05, 9) = 2.262
Since |1.890| < 2.262, we fail to reject H■: Insufficient evidence that mean height differs from 66 cm.

Use Cases

Comparing a sample's average exam score against a national benchmark, clinical trial small-group drug
testing, quality control of batch output, and A/B testing with limited sample sizes.

Statistical Methods — Comprehensive Study Notes Page 10

STATISTICAL METHODS Detailed Study Notes

9
Chi-Square Test

Overview
The chi-square (χ²) test measures the discrepancy between observed frequencies and expected frequencies
derived from a theoretical distribution or null hypothesis. A large χ² indicates poor fit. It is also used as a test of
independence between two categorical variables in a contingency table.

Key Formulas

χ² = Σ (O_i − E_i)² / E_i

O_i = observed frequency in class i

E_i = expected frequency in class i (from theory or marginal totals)

df = (number of classes − 1 − estimated parameters) [goodness of fit]

df = (r−1)(c−1) [contingency table independence test]

Worked Example

A die is rolled 264 times. Under fairness, E_i = 264/6 = 44 for each face. Observed: {40, 48, 42, 50, 36, 48} χ²
= (40−44)²/44 + (48−44)²/44 + (42−44)²/44 + (50−44)²/44 + (36−44)²/44 + (48−44)²/44 =
0.364+0.364+0.091+0.818+1.455+0.364 = 3.456 χ²_crit(0.05, 5) = 11.07 Since 3.456 < 11.07, the die is not
significantly biased.

Use Cases

Testing fairness of dice or random number generators, checking if data follows a normal/Poisson distribution,
testing association between gender and product preference, and genetics (Mendelian ratios).

Statistical Methods — Comprehensive Study Notes Page 11

STATISTICAL METHODS Detailed Study Notes

10
F-Test (Variance Ratio Test)

Overview
The F-test compares the variances of two independent samples to determine whether they come from
populations with equal variability. A ratio close to 1 suggests equal variances; a large ratio indicates significantly
different spreads. It is also the test statistic underlying ANOVA and regression significance testing.

Key Formulas

F = s■² / s■² where s■² ≥ s■² (place larger variance in numerator)

s² = Σ(x_i − x■)² / (n−1) [unbiased sample variance]

df■ = n■ − 1 (numerator degrees of freedom)

df■ = n■ − 1 (denominator degrees of freedom)

Reject H■: σ■² = σ■² if F_calc > F_crit(α, df■, df■)

Worked Example

Worker A output: {45, 50, 55, 48, 52} (n■=5, s■² = 16.0) Worker B output: {32, 60, 45, 55, 38} (n■=5, s■² =
126.5) F = 126.5 / 16.0 = 7.91 F_crit(0.05, 4, 4) = 6.39 Since 7.91 > 6.39, reject H■: Worker B's output is
significantly more variable than Worker A's.

Use Cases

Comparing machine consistency in manufacturing, checking homogeneity of variance before a two-sample

t-test (Levene's test), comparing financial return volatility, and regression model adequacy testing.

Statistical Methods — Comprehensive Study Notes Page 12

STATISTICAL METHODS Detailed Study Notes

Quick Revision Summary

The table below provides a concise reference for all ten statistical methods covered in these notes.

# Topic Key Concept When to Use

1 Parabola Fitting y = ax² + bx + c via normal equations Curved, non-linear trend data

2 Moving Average Rolling window average to smooth series Time series trend extraction

3 Exponential Fitting y = ae^{bx}; linearise with ln Constant-rate growth or decay

4 Joint Probability Cov & correlation from joint table Two-variable relationship study

5 Binomial Dist. C(n,x) p^x (1-p)^{n-x} Fixed n, two-outcome trials

6 Markov Chain π_k = π_0 P^k; steady state πP = π Sequential state-transition modelling

7 One-Way ANOVA F = MSB / MSW Compare 3+ group means

8 t-Test t = (x■ - µ)/(s/√n) Small-sample mean vs hypothesis

9 Chi-Square χ² = Σ(O-E)²/E Observed vs expected frequencies

10 F-Test F = s■² / s■² Compare two sample variances

Exam Tips & Common Pitfalls

• Curve Fitting (Topics 1–3): Always verify fit by plotting residuals.

• Probability (Topics 4–5): Check that all probabilities sum to 1 before computing moments.

• Markov Chains (Topic 6): Ensure the transition matrix rows each sum to exactly 1.

• Hypothesis Tests (Topics 7–10): State H■ and H■ clearly; choose α before seeing data.

• ANOVA (Topic 7): Assumes normality and homogeneity of variance within groups.

• t-Test (Topic 8): Use Welch's t-test if two-sample variances are unequal.

• Chi-Square (Topic 9): Each expected frequency should be ≥ 5 for reliable results.

• F-Test (Topic 10): Always place larger variance in numerator to use upper-tail critical value.

Statistical Methods — Comprehensive Study Notes Page 13