APPM1022A - Mathematical Statistics for Construction: Chapter 1 Sampling Distribution

Sampling Distributions and the Central Limit Theorem

• Fundamental Concepts of Sampling:

  • Population vs. Sample: In construction and engineering environments, measuring every single element produced—such as every concrete cylinder or weld—is practically impossible. Instead, an observable subset of size $n$ is extracted from the broader population of size $N$.

  • Sample: A collection of observable random variables $X_1, X_2, \dots, X_n$, where the integer $n$ defines the sample size.

  • Random Sample: A set of $n$ independent and identically distributed ($iid$) random variables $X_1, X_2, \dots, X_n$. Identically distributed signifies that every individual random variable $X_i$ possesses the identical probability distribution.

  • Statistic: Any mathematical function $T$ of the observable random variables $X_1, X_2, \dots, X_n$ that does not depend on any unknown population parameters.

  • Sampling Distribution: The probability distribution of a given sample statistic (such as the sample mean $\bar{X}$ or sample proportion $\hat{p}$) generated by repeatedly drawing samples of size $n$ from the target population.

• Common Sampling Methodologies:

  • Stratified Sampling ("Guaranteed Coverage" Method): The entire population is divided into distinct, non-overlapping subgroups or strata; every group is tested, but only a few representative elements are drawn from each subgroup.

  • Cluster Sampling ("Energy-Saver" Method): The population is partitioned into clusters; only a few clusters are selected for testing, but many or all individuals inside those chosen clusters are evaluated.

  • Systematic Sampling ("Assembly-Line" Method): Elements are sampled in a steady, predictable rhythm or fixed intervals (for instance, testing every $k$-th unit on a production conveyor).

• The Central Limit Theorem (CLT):

  • Formal Theorem: Let $X_1, X_2, \dots, X_n$ denote a random sample ($iid$ random variables) extracted from a population characterized by mean $\mu$ and finite variance $\sigma^2 > 0$. Defining the sample mean as:     $\bar{X}_n = \frac{1}{n} \sum_{i=1}^{n} X_i$     As $n \to \infty$, the standardized sample mean converges in distribution to a standard normal random variable, regardless of the shape or skew of the underlying population distribution:     $\frac{\bar{X}_n - \mu}{\sigma / \sqrt{n}} \xrightarrow{d} N(0, 1)$

  • Finite-Sample Approximation: For sufficiently large finite samples (typically $n \ge 30$ or when the underlying population is inherently normal), the distribution of the sample mean is approximated by:     $\bar{X}_n \dot{\sim} N\left(\mu, \frac{\sigma^2}{n}\right)$

  • Equivalent Statement for the Sum: Defining the sum of observations as $S_n = X_1 + \dots + X_n = n\bar{X}_n$, the standardized sum satisfies:     $\frac{S_n - n\mu}{\sigma\sqrt{n}} \xrightarrow{d} N(0, 1)$

  • Center and Spread: The expected value of all possible sample averages equals the true population mean ($\mu_{\bar{x}} = \mu$). The variability of sample averages is governed by the Standard Error ($SE$):     $SE = \frac{\sigma}{\sqrt{n}}$

  • Finite Population Correction (FPC): When sampling without replacement from a finite population of size $N$, the mean and standard error of the sample mean are governed by:     $\mu_{\bar{x}} = \mu$     $\sigma_{\bar{x}} = \sigma \sqrt{\frac{N - n}{Nn - n}}$

• Standardized $Z$-Score for Sample Means:

  • Transition from Individual to Mean: Standardizing an individual measurement converts $z = \frac{x - \mu}{\sigma}$ into a metric measuring the dispersion of sample averages:     $z = \frac{\bar{x} - \mu}{\sigma / \sqrt{n}}$

  • Structural Components of the $Z$-Formula:

  • Numerator ($\bar{x} - \mu$): Quantifies the absolute distance between the observed sample average and the expected population center.

  • Denominator ($\sigma / \sqrt{n}$): Represents the expected chance variation ("wobble") of a sample mean of size $n$.

  • Quotient ($z$): Expresses how many standard errors the observed sample mean lies away from the hypothesized target.

• Sampling Distributions of Proportions and Differences:

  • Sampling Distribution of a Proportion ($\hat{p}$):

  • Applied to binary engineering outcomes (e.g., whether a weld passes inspection or a road segment cracks within 5 years).

  • For population success parameter $p$ and failure parameter $q = 1 - p$, with sample proportion $\hat{p} = \frac{x}{n}$ ($x$ being the count of successes):       $\mu_p = p$       $\sigma_p = \sqrt{\frac{pq}{n}}$       $z = \frac{\hat{p} - p}{\sqrt{\frac{pq}{n}}}$

  • Sampling Distribution of Differences in Means:

  • Given two populations with parameters $(\mu_1, \sigma_1)$ and $(\mu_2, \sigma_2)$, with independent samples $n_1$ and $n_2$:       $\mu_{(\bar{x}_1 - \bar{x}_2)} = \mu_1 - \mu_2$       $\sigma_{(\bar{x}_1 - \bar{x}_2)} = \sqrt{\frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}}$

  • Sampling Distribution of Differences in Proportions:

  • For independent binomial samples with parameters $p_1, q_1, n_1$ and $p_2, q_2, n_2$ where $n_1, n_2 \ge 30$:       $\mu_{(p_1 - p_2)} = p_1 - p_2$       $\sigma_{(p_1 - p_2)} = \sqrt{\sigma_{p_1}^2 + \sigma_{p_2}^2} = \sqrt{\frac{p_1 q_1}{n_1} + \frac{p_2 q_2}{n_2}}$       $z = \frac{\hat{p}_1 - \hat{p}_2 - (p_1 - p_2)}{\sqrt{\frac{p_1 q_1}{n_1} + \frac{p_2 q_2}{n_2}}}$

• Worked Examples and Problem Scenarios:

  • Discrete Population Sampling Without Replacement: For a population $\{1, 2, 3, 4, 5\}$ evaluating all possible samples of size $n = 3$, or population $\{2, 4, 6, 8\}$ evaluating samples of size $n = 2$, calculate the mean of each sample combination to construct the discrete sampling distribution of $\bar{X}$ and verify that the grand mean of sample means equals the population mean $\mu$.

  • Soft Drink Can Fill Volume: A beverage canning line dispenses cans with mean volume $\mu = 350\text{ ml}$ and $\sigma = 25\text{ ml}$. For a random sample of $n = 10$ cans:     $SE = \frac{25}{\sqrt{10}} \approx 7.9057\text{ ml}$     Evaluating the probability that the sample mean falls within $15\text{ ml}$ of $350\text{ ml}$ involves finding $P(335 \le \bar{X} \le 365)$, which translates to:     $z = \frac{\pm 15}{7.9057} \approx \pm 1.90$     $P(-1.90 \le Z \le 1.90) = 2 \times 0.4713 = 0.9426$

  • Community Intelligence Scores: For normally distributed IQ scores with $\mu = 100$ and $\sigma = 15$, the probability of selecting an individual with an IQ of 140 or higher is computed via:     $z = \frac{140 - 100}{15} = 2.67$     $P(X \ge 140) = P(Z \ge 2.67) = 0.5 - 0.4962 = 0.0038$

  • Tool Production Defect Rates: A machine generates $2\%$ defective tools ($p = 0.02, q = 0.98$). In a sample of $n = 400$ tools:     $\sigma_p = \sqrt{\frac{0.02 \times 0.98}{400}} = \sqrt{\frac{0.0196}{400}} = 0.007$

  • (a) Probability that $3\%$ or more are defective ($\hat{p} \ge 0.03$):       $z = \frac{0.03 - 0.02}{0.007} \approx 1.43$       $P(\hat{p} \ge 0.03) = P(Z \ge 1.43) = 0.5 - 0.4236 = 0.0764$

  • (b) Probability that $2\%$ or fewer are defective ($\hat{p} \le 0.02$):       $z = \frac{0.02 - 0.02}{0.007} = 0$       $P(\hat{p} \le 0.02) = 0.5000$

  • Manufacturer Light Bulb Lifespan Comparison: Manufacturer A produces bulbs with $\mu_A = 1400\text{ h}$ and $\sigma_A = 200\text{ h}$; Manufacturer B produces bulbs with $\mu_B = 1200\text{ h}$ and $\sigma_B = 100\text{ h}$. Testing independent samples of $n_A = n_B = 125$ bulbs yields:     $\mu_{(\bar{X}_A - \bar{X}_B)} = 1400 - 1200 = 200\text{ h}$     $\sigma_{(\bar{X}_A - \bar{X}_B)} = \sqrt{\frac{200^2}{125} + \frac{100^2}{125}} = \sqrt{\frac{40000 + 10000}{125}} = \sqrt{400} = 20\text{ h}$

  • (a) Probability of mean lifetime at least $160\text{ h}$ greater ($\bar{X}_A - \bar{X}_B \ge 160$):       $z = \frac{160 - 200}{20} = -2.00$       $P(Z \ge -2.00) = 0.5 + 0.4772 = 0.9772$

  • (b) Probability of mean lifetime at least $250\text{ h}$ greater ($\bar{X}_A - \bar{X}_B \ge 250$):       $z = \frac{250 - 200}{20} = 2.50$       $P(Z \ge 2.50) = 0.5 - 0.4938 = 0.0062$

  • Continuous Population Difference: For continuous random variables $x \sim N(62, 10^2)$ and $y \sim N(21, 5^2)$, drawing one random element from each yields difference distribution parameters:     $\mu_{x-y} = 62 - 21 = 41$     $\sigma_{x-y} = \sqrt{10^2 + 5^2} = \sqrt{125} \approx 11.18$     The probability that $x - y > 50$ is:     $z = \frac{50 - 41}{11.18} = \frac{9}{11.18} \approx 0.81$     $P(x - y > 50) = P(Z > 0.81) = 0.5 - 0.2910 = 0.2090$

  • University Female Enrolment Proportions: University A has $50\%$ women ($p_A = 0.50, n_A = 40$); University B has $25\%$ women ($p_B = 0.25, n_B = 60$). The sampling distribution of difference in sample proportions satisfies:     $\mu_{(\hat{p}_A - \hat{p}_B)} = 0.50 - 0.25 = 0.25$     $\sigma_{(\hat{p}_A - \hat{p}_B)} = \sqrt{\frac{0.50 \times 0.50}{40} + \frac{0.25 \times 0.75}{60}} = \sqrt{\frac{0.25}{40} + \frac{0.1875}{60}} = \sqrt{0.00625 + 0.003125} = \sqrt{0.009375} \approx 0.0968$     The probability of finding more women in sample A than sample B ($\hat{p}_A - \hat{p}_B > 0$) is:     $z = \frac{0 - 0.25}{0.0968} \approx -2.58$     $P(Z > -2.58) = 0.5 + 0.4951 = 0.9951$

Estimation Theory: Point and Interval Estimation

• Point Estimation vs. Interval Estimation:

  • Point Estimation: Utilizing a single sample statistic (e.g., sample mean $\bar{x} = 42.3\text{ MPa}$) as the single best guess for an unknown true population parameter (e.g., $\mu$).

• Comparison Between Inference Methods:

  • Estimation: Inquires "Exactly how strong is it?" producing a single point estimate (e.g., "Average yield strength is $42.3\text{ MPa}$") or an interval estimate.

  • Hypothesis Testing: Compares empirical data to historical specifications (e.g., "Is it stronger than the old supplier?") producing a qualitative decision (e.g., "Yes, it is significantly stronger").

• Dart vs. Net Conceptual Model:

  • A point estimate resembles throwing a single dart at a bullseye; there is substantial risk of missing the true parameter due to sampling sway.

  • An interval estimate resembles throwing a wide net over the target, establishing a bounded numerical range ($\bar{x} \pm \text{Margin of Error}$) that captures the true parameter with specified probability.

• Properties of Desirable Estimators:

  • Unbiasedness: An estimator is unbiased if the expected value (mean) of its sampling distribution equals the true parameter being estimated ($E[\hat{\theta}] = \theta$). Systematic over- or under-estimation denotes bias.

  • Minimum Variance: Among unbiased estimators, the preferred estimator has the smallest sampling variance, ensuring tighter clustering around the target.

  • Reliability and Sway: Governed directly by the Standard Error ($SE = \frac{\sigma}{\sqrt{n}}$). A small $SE$ indicates high precision and trustworthy point estimates; an inflated $SE$ renders point estimates unreliable.

• Constructing Confidence Intervals:

  • General Structure:     $\text{Confidence Interval} = \text{Point Estimate} \pm \text{Margin of Error}$     $\text{Confidence Interval} = \text{Estimator} \pm (z \times SE)$

• Confidence Level Multipliers and Trade-Offs:

  • 90% Confidence Level: Critical value $z_{\alpha/2} = 1.645$. Used for low-risk, rapid exploratory estimates ("tight net").

  • 95% Confidence Level: Critical value $z_{\alpha/2} = 1.960$. Industry standard default setting ("standard net").

  • 99% Confidence Level: Critical value $z_{\alpha/2} = 2.576$. Applied in life-safety-critical, high-liability, or high-cost structural engineering evaluations ("wide net").

  • Core Trade-Off: Greater statistical confidence necessitates a broader critical multiplier, producing a wider, less precise interval.

• Standard Confidence Interval Formulas:

  • Single Population Mean ($\mu$):       $\bar{x} \pm z_{\alpha/2} \frac{\sigma}{\sqrt{n}}$

  • Single Population Proportion ($p$):       $\hat{p} \pm z_{\alpha/2} \sqrt{\frac{\hat{p}\hat{q}}{n}}$

  • Difference Between Two Population Means ($\mu_1 - \mu_2$):       $(\bar{x}_1 - \bar{x}_2) \pm z_{\alpha/2} \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}$

  • Difference Between Two Population Proportions ($p_1 - p_2$):       $(\hat{p}_1 - \hat{p}_2) \pm z_{\alpha/2} \sqrt{\frac{\hat{p}_1\hat{q}_1}{n_1} + \frac{\hat{p}_2\hat{q}_2}{n_2}}$

  • The "Zero Rule" for Interval Comparisons:

• When assessing difference intervals ($\mu_1 - \mu_2$ or $p_1 - p_2$):

  • If the confidence interval captures zero within its lower and upper bounds, conclude there is no statistically significant difference between the two populations at the chosen significance level.

  • If the interval is strictly positive (entirely above 0), conclude Group 1 is significantly higher than Group 2.

  • If the interval is strictly negative (entirely below 0), conclude Group 2 is significantly higher than Group 1.

  • Sample Size Determination Formulas:

• To achieve a specified maximum tolerable margin of error ($B$), invert the margin of error equation $B = z \times \frac{\sigma}{\sqrt{n}}$, yielding:     $n = \left(\frac{z \times \sigma}{B}\right)^2$

• For proportions, where $B = z \times \sqrt{\frac{pq}{n}}$, inverting gives:     $n = \frac{z^2 \, p(1-p)}{B^2}$

• Conservative Rule for Proportions: When the prior population proportion $p$ is unknown, assign $p = 0.5$, maximizing the product $pq = 0.25$ and guaranteeing an adequate sample size.

  • Practical Estimation Case Studies:

• Client Protein Intake: In a sample of $n = 50$ nutrition clients, mean intake is $\bar{x} = 75\text{ g}$ with $s = 6\text{ g}$. Constructing a $95\%$ confidence interval:     $SE = \frac{6}{\sqrt{50}} \approx 0.8485\text{ g}$     $\text{Margin of Error} = 1.96 \times 0.8485 \approx 1.663\text{ g}$     $95\%\text{ CI} = 75 \pm 1.663 = (73.34\text{ g}, 76.66\text{ g})$

• University Large Language Model Usage: A survey of $n = 200$ Wits university students revealed $x = 143$ use LLMs regularly ($\hat{p} = \frac{143}{200} = 0.715, \hat{q} = 0.285$). For a $98\%$ confidence interval ($z_{\alpha/2} = 2.326$):     $SE = \sqrt{\frac{0.715 \times 0.285}{200}} = \sqrt{\frac{0.203775}{200}} \approx 0.03192$     $\text{Margin of Error} = 2.326 \times 0.03192 \approx 0.0742$     $98\%\text{ CI} = 0.715 \pm 0.0742 = (0.6408, 0.7892) \implies (64.08\%, 78.92\%$

• Engineering Industry Gender Pay Gap: Stats SA sampled $n_1 = 100$ male engineers ($\bar{x}_1 = 398000\text{ ZAR}, s_1 = 80000\text{ ZAR}$) and $n_2 = 100$ female engineers ($\bar{x}_2 = 358000\text{ ZAR}, s_2 = 75000\text{ ZAR}$):     $\bar{x}_1 - \bar{x}_2 = 398000 - 358000 = 40000\text{ ZAR}$     $SE = \sqrt{\frac{80000^2}{100} + \frac{75000^2}{100}} = \sqrt{64000000 + 56250000} = \sqrt{120250000} \approx 10965.86\text{ ZAR}$     $\text{Margin of Error} = 1.96 \times 10965.86 \approx 21493.08\text{ ZAR}$     $95\%\text{ CI} = 40000 \pm 21493.08 = (18506.92\text{ ZAR}, 61493.08\text{ ZAR})$     Because the interval lies entirely above 0 ($\text{R}18507$ to $\text{R}61493$), the data demonstrates statistically significant gender disparity favoring male salaries.

• University Matriculant Enrolment by School Quintile: A sample evaluates $n_1 = 150$ Quintile 5 matriculants ($x_1 = 60, \hat{p}_1 = 0.40$) and $n_2 = 150$ Quintile 1–3 matriculants ($x_2 = 15, \hat{p}_2 = 0.10$). Evaluating at $99\%$ confidence ($z_{\alpha/2} = 2.576$):     $\hat{p}_1 - \hat{p}_2 = 0.40 - 0.10 = 0.30$     $SE = \sqrt{\frac{0.40 \times 0.60}{150} + \frac{0.10 \times 0.90}{150}} = \sqrt{\frac{0.24}{150} + \frac{0.09}{150}} = \sqrt{0.0016 + 0.0006} = \sqrt{0.0022} \approx 0.0469$     $\text{Margin of Error} = 2.576 \times 0.0469 \approx 0.1208$     $99\%\text{ CI} = 0.30 \pm 0.1208 = (0.1792, 0.4208) \implies (17.92\%, 42.08\%$     Zero is excluded from the interval, verifying substantially higher enrolment rates from affluent schools.

• Youth Unemployment Sample Sizing: In surveying municipal youth (ages 15–34), national QLFS benchmarks establish $p = 0.46$. For $95\%$ confidence ($z = 1.96$) and a margin of error of $3\%$ ($B = 0.03$):     $n = \frac{1.96^2 \times 0.46 \times 0.54}{0.03^2} = \frac{3.8416 \times 0.2484}{0.0009} \approx 1060.29 \implies n = 1061$     Applying the conservative assumption $p = 0.50$:     $n = \frac{1.96^2 \times 0.50 \times 0.50}{0.03^2} = \frac{3.8416 \times 0.25}{0.0009} \approx 1067.11 \implies n = 1068$     If the required precision is relaxed to a $5\%$ margin of error ($B = 0.05$) with $p = 0.46$:     $n = \frac{1.96^2 \times 0.46 \times 0.54}{0.05^2} = \frac{0.95425}{0.0025} \approx 381.7 \implies n = 382$

Hypothesis Testing Principles and Large-Sample Z-Tests

• Core Philosophy and The Courtroom Analogy:

  • Hypothesis testing operates under legal jurisprudence principles: a defendant is presumed innocent until proven guilty beyond a reasonable doubt.

  • Null Hypothesis ($H_0$): Represents the status quo, the manufacturer's claim, or baseline assumption. Assumed true until empirical evidence demonstrates extreme implausibility.

  • Alternative Hypothesis ($H_a$): Represents the research suspicion, potential defect, or assertion being tested. Research does not prove $H_a$ directly; rather, sample evidence discredits $H_0$.

• Mechanics of Hypothesis Testing:

  • Test Statistic ($Z$-score as "The Bouncer"): Quantifies how many standard errors the sample estimate deviates from the null hypothesis baseline:     $Z = \frac{\text{Sample Estimate} - \text{Claimed Value}}{\text{Standard Error}}$

  • $P$-Value: The exact probability of obtaining a test statistic at least as extreme as the observed value, assuming $H_0$ is strictly true.

  • Significance Level ($\alpha$ as "The Red Rope"): The threshold risk tolerance for committing an error in rejecting the status quo. Standard thresholds are fixed at $0.05$ or $0.01$.

  • Decision Criterion:

  • If $P\text{-value} < \alpha$: Reject $H_0$. The evidence against the null claim is overwhelming.

  • If $P\text{-value} \ge \alpha$: Do not reject $H_0$. Insufficient evidence exists to warrant abandoning the status quo.

• Directionality: One-Tailed vs. Two-Tailed Tests:

  • Right-Tailed Test: Formulated when testing for an increase or surplus ($H_0: \mu \le \mu_0$ vs. $H_a: \mu > \mu_0$). The entire critical rejection region $\alpha$ is concentrated in the upper tail.

  • Left-Tailed Test: Formulated when testing for a deficit or underperformance ($H_0: \mu \ge \mu_0$ vs. $H_a: \mu < \mu_0$). Rejection region $\alpha$ is concentrated entirely in the lower tail.

  • Two-Tailed Test: Formulated when testing for any deviation or non-conformance ($H_0: \mu = \mu_0$ vs. $H_a: \mu \ne \mu_0$). The significance level is divided evenly, allocating $\frac{\alpha}{2}$ to each tail.

• Type I and Type II Errors:

  • Type I Error ($\alpha$, False Positive / False Alarm): Rejecting a null hypothesis $H_0$ that is actually true (e.g., sounding a fire alarm when no fire exists).

  • Type II Error ($\beta$, False Negative / Missed Defect): Failing to reject a null hypothesis $H_0$ when it is actually false (e.g., passing an inspection when a structural crack is physically present).

• Controlling Errors: The investigator directly dictates $\alpha$. Calculating $\beta$ is mathematically complex, which enforces the strict operational convention of stating "Do not reject $H_0$" instead of "Accept $H_0$".

  • The 5-Step Hypothesis Testing Protocol:

• Set the Stage: Explicitly declare $H_0$ and $H_a$.

• Measure Evidence: Compute the sample point statistic ($\bar{x}$ or \hat{p}$).\n 3. *Measure Sway*: Determine the standard error (SE).\n 4. *Calculate Test Statistic and P-Value*: Compute Z$and establish the exact associated$P-value.\n 5. *Deliver the Verdict*: Compare P$-value to$\alpha and issue an engineering decision, keeping in mind that statistical significance does not automatically guarantee practical significance.\n\n* **Large-Sample Hypothesis Testing Case Studies**:\n * **Municipal Standpipe Water Pressure**:\n * *Specification*: Legal minimum water pressure is \ge 25\text{ kPa}.\n * *Data*: Sample n = 36$, average pressure$\bar{x} = 23.5\text{ kPa}$, standard deviation$s = 4.5\text{ kPa}$, significance level$\alpha = 0.01.\n * *Hypotheses*: H_0: \mu \ge 25\text{ kPa}$vs.$H_a: \mu < 25\text{ kPa} (Left-tailed test).\n * *Calculations*:\n      SE = \frac{4.5}{\sqrt{36}} = 0.75\text{ kPa}\n      Z = \frac{23.5 - 25}{0.75} = -2.00\n * *Decision*: For \alpha = 0.01$, the critical threshold is$Z_{\text{crit}} = -2.33$($P\text{-value} = 0.0228$). Because$-2.00 > -2.33$($P\text{-value} > 0.01$), do not reject$H_0$. There is insufficient evidence at the$1\% level to prove legal under-pressure.\n * **Construction Site Monthly Wages**:\n * *Claim*: Management asserts average monthly site worker wages equal \text{R}25000.\n * *Data*: Sample n = 50$,$\bar{x} = \text{R}24000$,$s = \text{R}3000$,$\alpha = 0.01.\n * *Hypotheses*: H_0: \mu \ge 25000$vs.$H_a: \mu < 25000 (Left-tailed test).\n * *Calculations*:\n      SE = \frac{3000}{\sqrt{50}} \approx 424.26\text{ ZAR}\n      Z = \frac{24000 - 25000}{424.26} \approx -2.36\n * *Decision*: The critical value is Z_{\text{crit}} = -2.33$($P\text{-value} = 0.0091$). Because$-2.36 < -2.33$($P < 0.01$), reject$H_0. There is statistically significant evidence of underpayment.\n * **Road Asphalt Layer Thickness**:\n * *Specification*: Contract thickness must equal exactly 120\text{ mm}.\n * *Data*: Sample n = 40$cores,$\bar{x} = 118\text{ mm}$,$s = 6\text{ mm}$,$\alpha = 0.05.\n * *Hypotheses*: H_0: \mu = 120\text{ mm}$vs.$H_a: \mu e 120\text{ mm} (Two-tailed test).\n * *Calculations*:\n    SE = \frac{6}{\sqrt{40}} \approx 0.9487\text{ mm}\n    Z = \frac{118 - 120}{0.9487} \approx -2.11\n * *Decision*: For two-tailed \alpha = 0.05$, critical thresholds are$Z_{\text{crit}} = \pm 1.96$. Because$|-2.11| > 1.96$($P\text{-value} = 0.0348 < 0.05$), reject$H_0. The asphalt layer deviates significantly from specification.\n* **National Building Code Inspection Compliance**:\n * *Specification*: Regulation requires \ge 85\%$($p_0 = 0.85) of new residential units to pass initial inspection.\n * *Data*: An inspector evaluates n = 200$units, passing$x = 160$($\hat{p} = 0.80$), at$\alpha = 0.05.\n * *Hypotheses*: H_0: p \ge 0.85$vs.$H_a: p < 0.85 (Left-tailed test).\n * *Calculations*:\n    SE = \sqrt{\frac{0.85 \times 0.15}{200}} = \sqrt{\frac{0.1275}{200}} \approx 0.02525\n    Z = \frac{0.80 - 0.85}{0.02525} \approx -1.98\n * *Decision*: Lower critical bound is Z_{\text{crit}} = -1.645$($P\text{-value} = 0.0239$). Because$-1.98 < -1.645$, reject$H_0. The developer is failing building regulation targets.\n* **Provincial Sanitation Inequality Comparison**:\n * *Data*: Independent surveys of n_1 = 150$households in Western Cape ($x_1 = 143, \hat{p}1 = 0.953$) and$n_2 = 150$in Limpopo ($x_2 = 45, \hat{p}_2 = 0.300).\n * *Hypotheses*: H_0: p_1 - p_2 = 0$vs.$H_a: p_1 - p_2 e 0.\n* *Calculations*:\n  SE = \sqrt{\frac{0.953(0.047)}{150} + \frac{0.300(0.700)}{150}} = \sqrt{0.0002986 + 0.001400} = \sqrt{0.0016986} \approx 0.0412\n  Z = \frac{0.953 - 0.300}{0.0412} = \frac{0.653}{0.0412} \approx 15.85\n* *Decision*: The calculated test statistic Z = 15.85 far exceeds any standard critical threshold, confirming massive, statistically significant disparity in sanitation access between the provinces.\n\n\n# Small-Sample Inference: Student's t-Distribution\n\n* **Breakdown of the Z-Test for Small Samples**:\n * When sample sizes fall below 30 (n < 30$) and the true population standard deviation$\sigma$is unknown, the sample standard deviation$s must be used as a proxy.\n * The resulting standardized ratio does not follow a normal distribution:\n    \frac{\bar{x} - \mu}{s / \sqrt{n}} ot\sim N(0, 1)\n * Replacing \sigma$with the sample estimate$s$introduces substantial additional variability because$s varies from sample to sample.\n\n* **Properties of Student's t-Distribution**:\n * Symmetrical and bell-shaped with a center located precisely at 0.\n * Characterized by "heavier tails" (greater area in the tails) and a flatter peak than the standard normal distribution, reflecting heightened uncertainty.\n * The exact geometric profile is dictated entirely by its Degrees of Freedom (df = n - 1).\n * As sample size increases (n \to \infty$or$df \to \infty$), the$t$-distribution converges toward the standard normal$Z-distribution.\n\n\n\n* **Mathematical Meaning of Degrees of Freedom (df = n - 1)**:\n * Calculating sample variance requires prior calculation of the sample mean \bar{x}.\n * Imposing the restriction \sum{i=1}^n (x_i - \bar{x}) = 0$consumes one piece of mathematical information, leaving$n - 1 independent observations free to vary.\n\n* **Reading the t-Distribution Table**:\n * Critical values (t_{\alpha, df}) across key degrees of freedom and two-sided confidence intervals:\n * df = 1$: 90$6.314$; 95$12.706$; 98$31.821$; 99$63.657\n * df = 2$: 90$2.920$; 95$4.303$; 98$6.965$; 99$9.925\n * df = 3$: 90$2.353$; 95$3.182$; 98$4.541$; 99$5.841\n * df = 4$: 90$2.132$; 95$2.776$; 98$3.747$; 99$4.604\n * df = 5$: 90$2.015$; 95$2.571$; 98$3.365$; 99$4.032\n * df = 10$: 90$1.812$; 95$2.228$; 98$2.764$; 99$3.169\n * df = 15$: 90$1.753$; 95$2.131$; 98$2.602$; 99$2.947\n * df = 20$: 90$1.725$; 95$2.086$; 98$2.528$; 99$2.845\n * df = 30$: 90$1.697$; 95$2.042$; 98$2.457$; 99$2.750\n * df = \infty$: 90$1.645$; 95$1.960$; 98$2.326$; 99$2.576\n\n* **One-Sample t-Confidence Interval and Safety Analysis**:\n * **Formula**: For a 100(1 - \alpha)\% confidence interval:\n    \bar{x} \pm t_{\alpha/2, df} \times \frac{s}{\sqrt{n}}\n * **Concrete Batch Cylinder Strength Scenario**:\n * A site engineer tests n = 5$cylinders, yielding strengths:$28, 30, 29, 27, 31\text{ MPa}.\n * Mean: \bar{x} = 29.0\text{ MPa}$; Standard deviation:$s = 1.581\text{ MPa}$; Standard error:$SE = \frac{1.581}{\sqrt{5}} = 0.707\text{ MPa}.\n * Degrees of freedom: df = 5 - 1 = 4$. From the table,$t_{0.025, 4} = 2.776.\n * t$-Interval Margin of Error:$2.776 \times 0.707 = 1.963\text{ MPa}.\n * Correct 95\%$CI:$(27.04\text{ MPa}, 30.96\text{ MPa}).\n * **Direct Comparison with Erroneous Z-Calculation**:\n * Erroneous Z$-Interval:$1.96 \times 0.707 = 1.386\text{ MPa} \implies (27.61\text{ MPa}, 30.39\text{ MPa}).\n * Discrepancy: The Z$-interval is$0.575\text{ MPa} narrower than reality demands.\n * **Practical Engineering Risk**: If the contract specifies a minimum threshold of 27.5\text{ MPa}$, the erroneous$Z$-test asserts a pass ($27.61 > 27.5$), whereas the valid$t$-test warns that true strength could be as low as$27.04\text{ MPa}, failing compliance and creating structural danger.\n\n* **One-Sample t-Hypothesis Test**:\n * **Test Statistic**:\n    t = \frac{\bar{x} - \mu_0}{s / \sqrt{n}}\n * **RDP House Floor Area Audit Example**:\n * *Problem*: Legal minimum floor area is 40\text{ m}^2$. An auditor measures$n = 8$completed homes:$36, 38, 35, 39, 37, 34, 38, 36\text{ m}^2.\n * *Sample Metrics*: \bar{x} = 36.625\text{ m}^2, s = 1.685\text{ m}^2, df = 7.\n * *Hypotheses*: H_0: \mu \ge 40$vs.$H_a: \mu < 40.\n * *Calculations*:\n      SE = \frac{1.685}{\sqrt{8}} \approx 0.5957\text{ m}^2\n      t = \frac{36.625 - 40}{0.5957} \approx -5.66\n * *Decision*: Critical threshold for df = 7, \alpha = 0.05$is$t_{\text{crit}} = -1.895$. The observed$t = -5.66$falls deep within the rejection region ($P = P(T_7 \le -5.66) \approx 0.00038 < 0.05$). Reject$H_0; the contractor is constructing substandard homes.\n\n\n\n* **Two Independent Samples: Pooled Variance t-Test**:\n * **Underlying Assumptions**:\n 1. Both populations are normally distributed.\n 2. Samples are drawn independently.\n 3. Population variances are equal (Homoscedasticity: \sigma_1^2 = \sigma_2^2 = \sigma^2).\n * **Variance Ratio Rule of Thumb**:\n    \text{Ratio} = \frac{\text{larger } s^2}{\text{smaller } s^2}\n * If \text{Ratio} \le 3: Variances are approximately equal; pool the variance.\n * If \text{Ratio} > 3: Variances are unequal; employ separate-variance alternatives (Welch's / Satterthwaite approximation).\n * **Pooled Formulas**:\n * Pooled Sample Variance:\n      s_p^2 = \frac{(n_1 - 1)s_1^2 + (n_2 - 1)s_2^2}{n_1 + n_2 - 2}\n * Test Statistic (df = n_1 + n_2 - 2):\n      t = \frac{(\bar{x}1 - \bar{x}_2) - D_0}{\sqrt{s_p^2 \left(\frac{1}{n_1} + \frac{1}{n_2}\right)}}\n * Confidence Interval:\n      (\bar{x}_1 - \bar{x}_2) \pm t{\alpha/2, df} \sqrt{s_p^2 \left(\frac{1}{n_1} + \frac{1}{n_2}\right)}\n * **Steel Rebar Reinforcing Quality Example**:\n * Supplier A (n_1 = 8$):$1.58, 1.56, 1.60, 1.55, 1.59, 1.57, 1.61, 1.54\text{ kg/m}.\n * Supplier B (n_2 = 7$):$1.52, 1.54, 1.53, 1.50, 1.55, 1.51, 1.53\text{ kg/m}.\n * Total degrees of freedom: df = 8 + 7 - 2 = 13.\n * At \alpha = 0.05$, two-tailed critical cutoff is$t_{\text{crit}} = \pm 2.16.\n * The calculated test statistic yields t = 4.44$. Because$4.44 > 2.16$, reject$H_0. Supplier steel quality differs significantly.\n\n\n\n * **Skills Training Program Productivity**:\n * New Method (n_1 = 8$):$\bar{x}1 = 42\text{ units/day}, s_1 = 5.2.\n * Old Method (n_2 = 8$):$\bar{x}_2 = 38\text{ units/day}, s_2 = 4.8.\n * Variance Check: \text{Ratio} = \frac{5.2^2}{4.8^2} = \frac{27.04}{23.04} = 1.174 \le 3 (equal variance assumption valid).\n * Pooled Variance:\n      s_p^2 = \frac{7(27.04) + 7(23.04)}{14} = 25.04 \implies s_p = 5.004\n * Standard Error:\n      SE = 5.004 \sqrt{\frac{1}{8} + \frac{1}{8}} = 5.004 \sqrt{0.25} = 2.502\n * Test Statistic (df = 14):\n      t = \frac{42 - 38}{2.502} = \frac{4}{2.502} = 1.599\n * At \alpha = 0.05$, critical value is$t{0.025, 14} = 2.145$. Because$1.599 < 2.145$, fail to reject$H_0. Productivity improvement is not statistically significant.\n * 95\% Confidence Interval for difference:\n      4 \pm 2.145(2.502) = 4 \pm 5.367 = (-1.367\text{ units}, 9.367\text{ units})\n      Because zero is within the interval, significant improvement cannot be claimed.\n\n* **Paired t-Test for Dependent Samples**:\n * Applied when observations are inherently matched, paired, or repeated across the same experimental units (e.g., pre-treatment vs. post-treatment).\n * Compute paired differences for each unit: d_i = x_{1i} - x_{2i}.\n * Test Statistic (df = n - 1):\n    t = \frac{\bar{d} - 0}{s_d / \sqrt{n}}\n * **RDP Settlement Ceiling-Retrofit Thermal Evaluation**:\n * Indoor temperatures measured at 7 a.m. in n = 8 homes before and after ceiling installation:\n * House 1: Before = 9^\circ\text{C}$, After =$14^\circ\text{C}$, Difference$d_1 = 5\n * House 2: Before = 8^\circ\text{C}$, After =$12^\circ\text{C}$, Difference$d_2 = 4\n * House 3: Before = 10^\circ\text{C}$, After =$13^\circ\text{C}$, Difference$d_3 = 3\n * House 4: Before = 7^\circ\text{C}$, After =$12^\circ\text{C}$, Difference$d_4 = 5\n * House 5: Before = 9^\circ\text{C}$, After =$14^\circ\text{C}$, Difference$d_5 = 5\n * House 6: Before = 8^\circ\text{C}$, After =$11^\circ\text{C}$, Difference$d_6 = 3\n * House 7: Before = 11^\circ\text{C}$, After =$16^\circ\text{C}$, Difference$d_7 = 5\n * House 8: Before = 9^\circ\text{C}$, After =$12^\circ\text{C}$, Difference$d_8 = 3\n * Mean difference: \bar{d} = \frac{5+4+3+5+5+3+5+3}{8} = \frac{33}{8} = 4.125^\circ\text{C}.\n * Differences squared: \sum d^2 = 25+16+9+25+25+9+25+9 = 143.\n * Sample Variance of differences:\n      s_d^2 = \frac{143 - \frac{33^2}{8}}{7} = \frac{143 - 136.125}{7} = \frac{6.875}{7} \approx 0.9821 \implies s_d \approx 0.991\n * Test Statistic (df = 7):\n      t = \frac{4.125}{0.991 / \sqrt{8}} = \frac{4.125}{0.3504} \approx 11.77\n * With t_{0.025, 7} = 2.365$, reject$H_0$($11.77 > 2.365); the ceiling retrofits generate highly significant thermal insulation improvements.\n\n\n# The Chi-Square Distribution and Categorical Data Analysis\n\n* **Mathematical Foundations of the Chi-Square (\chi^2) Distribution**:\n * Arises when measuring squared distances from a center rather than directional deviations. Squaring eliminates negative signs, tracking pure dispersion.\n * Defined formally as the sum of independent squared standard normal variables:\n    \chi^2 = \sum_{i=1}^k Z_i^2\n * If one Z$-score is squared ($Z^2$), the resulting variable follows a$\chi^2$-distribution with$df = 1.\n * Summing five squared independent Z$-scores ($Z_1^2 + \dots + Z_5^2$) yields a$\chi^2$-distribution with$df = 5.\n * Connection to sample variance: Because s^2 = \frac{\sum (x - \bar{x})^2}{n - 1}, standardizing the observations establishes that:\n    \frac{(n - 1)s^2}{\sigma^2} \sim \chi^2_{n - 1}\n * **Key Geometric and Structural Features**:\n * Support is strictly non-negative: Range lies on (0, \infty).\n * Skewed to the right (asymmetrical); right-skewness attenuates as df grows.\n * Mean of \chi^2$distribution equals$df$; variance equals$2 \times df.\n * While degrees of freedom adjust tail thickness in Z$and$t$, in$\chi^2 they dictate the location of the distribution's peak.\n\n\n\n* **Inference for a Single Population Variance (\sigma^2)**:\n * **Confidence Interval Formulas**:\n * For population variance \sigma^2:\n      \frac{(n - 1)s^2}{\chi^2_{\text{right}}} < \sigma^2 < \frac{(n - 1)s^2}{\chi^2_{\text{left}}}\n * For population standard deviation \sigma:\n      \sqrt{\frac{(n - 1)s^2}{\chi^2_{\text{right}}}} < \sigma < \sqrt{\frac{(n - 1)s^2}{\chi^2_{\text{left}}}}\n * Where \chi^2_{\text{right}} = \chi^2_{\alpha/2, n-1}$and$\chi^2_{\text{left}} = \chi^2_{1 - \alpha/2, n-1}.\n * **Concrete Compressive Strength Variance Case Study**:\n * A quality engineer takes n = 25$concrete cylinders, measuring sample standard deviation$s = 4.2\text{ MPa}.\n * Degrees of freedom: df = 25 - 1 = 24.\n * For a 95\%$confidence level ($\alpha = 0.05), the critical values are:\n      \chi^2_{\text{right}} = \chi^2_{0.025, 24} = 36.415\n      \chi^2_{\text{left}} = \chi^2_{0.975, 24} = 13.848\n * Numerator computation: (n - 1)s^2 = 24 \times 4.2^2 = 24 \times 17.64 = 423.36.\n * Variance Interval:\n      \frac{423.36}{36.415} < \sigma^2 < \frac{423.36}{13.848} \implies 11.626\text{ MPa}^2 < \sigma^2 < 30.572\text{ MPa}^2\n * Standard Deviation Interval:\n      \sqrt{11.626} < \sigma < \sqrt{30.572} \implies 3.41\text{ MPa} < \sigma < 5.53\text{ MPa}\n\n\n\n* **Categorical Data Analysis: The Three Fundamental \chi^2 Tests**:\n * **General Test Statistic**:\n    \chi^2 = \sum \frac{(O - E)^2}{E}\n    where O$is observed cell frequency and$E$is expected cell frequency under$H_0.\n * **Core Assumptions**:\n 1. *Random Sampling*: Data represents an unbiased random sample from the relevant population.\n 2. *Mutually Exclusive Categories*: Each observation falls into exactly one cell.\n 3. *Adequate Cell Counts*: Expected frequencies must satisfy E \ge 5$in all cells. If any$E < 5, combine adjacent categories.\n\n* **Goodness-of-Fit Test (Single Categorical Variable)**:\n * Evaluates whether observed sample frequencies match expected theoretical distributions.\n * Degrees of Freedom: df = k - 1$, where$k is the number of categories.\n * **Fatal Road Crash Day-of-Week Case Study**:\n * Urban planners test if n = 84$fatal crashes occur uniformly across all 7 days ($E = \frac{84}{7} = 12\text{ crashes/day}):\n * Monday: O = 10\n * Tuesday: O = 8\n * Wednesday: O = 9\n * Thursday: O = 11\n * Friday: O = 12\n * Saturday: O = 18\n * Sunday: O = 16\n * Calculation:\n      \chi^2 = \frac{(10-12)^2}{12} + \frac{(8-12)^2}{12} + \frac{(9-12)^2}{12} + \frac{(11-12)^2}{12} + \frac{(12-12)^2}{12} + \frac{(18-12)^2}{12} + \frac{(16-12)^2}{12}\n      \chi^2 = \frac{4 + 16 + 9 + 1 + 0 + 36 + 16}{12} = \frac{82}{12} \approx 6.833\n * Degrees of Freedom: df = 7 - 1 = 6$. At$\alpha = 0.05$, critical threshold is$\chi^2_{0.05, 6} = 12.592.\n * *Decision*: Because 6.833 < 12.592$, do not reject$H_0. Although weekend counts are elevated, a sample of 84 crashes is insufficient to prove non-uniformity beyond chance.\n\n* **Test of Independence (Two Categorical Variables, Single Sample)**:\n * Assesses whether two categorical variables evaluated in a single sample are statistically dependent.\n * Expected Frequency Formula:\n    E = \frac{(\text{Row Total}) \times (\text{Column Total})}{\text{Grand Total}}\n * Degrees of Freedom: df = (r - 1)(c - 1).\n * **Structural Defect vs. Geographic Location Case Study**:\n * Single random sample of N = 150 defective housing units categorized by Location and Defect Type:\n * Material Failure: Coastal = 25, Inland = 15, High-altitude = 10; Row Total = 50\n * Design Flaw: Coastal = 10, Inland = 20, High-altitude = 15; Row Total = 45\n * Workmanship Error: Coastal = 15, Inland = 15, High-altitude = 25; Row Total = 55\n * Column Totals: Coastal = 50, Inland = 50, High-altitude = 50; Grand Total = 150\n * Expected Frequencies (E = \frac{\text{Row Total} \times 50}{150} = \frac{\text{Row Total}}{3}):\n * Material Failure: E = \frac{50}{3} = 16.67\n * Design Flaw: E = \frac{45}{3} = 15.00\n * Workmanship Error: E = \frac{55}{3} = 18.33\n * Cell Chi-Square Contributions:\n * Coastal Material Failure: \frac{(25 - 16.67)^2}{16.67} = \frac{69.39}{16.67} \approx 4.16\n * Coastal Design Flaw: \frac{(10 - 15)^2}{15} = \frac{25}{15} \approx 1.67\n * High-altitude Workmanship: \frac{(25 - 18.33)^2}{18.33} = \frac{44.49}{18.33} \approx 2.43\n * Degrees of Freedom: df = (3 - 1)(3 - 1) = 4$. At$\alpha = 0.05$,$\chi^2_{0.05, 4} = 9.488$.

  • Decision: The overall sum $\chi^2$ exceeds $9.488$, rejecting independence. Material failures dominate coastal zones (saltwater corrosion), while workmanship errors dominate high-altitude sites (thermal setting difficulties).

• Standardized Residuals Analysis:     $\text{Standardized Residual} = \frac{O - E}{\sqrt{E}}$     Values outside $[-2, +2]$ identify cells driving the statistical association (e.g., Coastal Material Failure has residual $\frac{25 - 16.67}{\sqrt{16.67}} = +2.04$).

  • Test of Homogeneity (Multiple Independent Samples):

• Evaluates whether multiple distinct populations share identical category proportions. Column totals are fixed by research design prior to sampling.

• Provincial Employment Status Distribution Case Study:

  • Independent samples of $n = 150$ adults drawn from Gauteng, KZN, and Western Cape ($N = 450$):

    • Employed: Gauteng = 88, KZN = 72, Western Cape = 80; Row Total = 240

    • Unemployed: Gauteng = 42, KZN = 55, Western Cape = 33; Row Total = 130

    • Not Economically Active (NEA): Gauteng = 20, KZN = 23, Western Cape = 37; Row Total = 80

    • Column Totals: 150 in each province

  • Expected Frequencies:

    • Employed: $E = \frac{240 \times 150}{450} = 80.0$

    • Unemployed: $E = \frac{130 \times 150}{450} = 43.33$

    • NEA: $E = \frac{80 \times 150}{450} = 26.67$

  • Degrees of Freedom: $df = (3 - 1)(3 - 1) = 4$. Critical value at $\alpha = 0.05$ is $9.488$.

  • Decision: Calculated $\chi^2 > 9.488$, rejecting homogeneity. KZN displays significantly elevated unemployment ($55$ observed vs. $43.33$ expected), while the Western Cape displays elevated NEA individuals ($37$ observed vs. $26.67$ expected).

The F-Distribution and Comparing Process Variances

• Mathematical Definition and Principles:

  • Defined as the ratio of two independent chi-square distributed random variables, each divided by its respective degrees of freedom:     $F = \frac{s_1^2 / \sigma_1^2}{s_2^2 / \sigma_2^2}$

  • Under the null hypothesis of equal population variances ($H_0: \sigma_1^2 = \sigma_2^2$), the parameter variances cancel, reducing the statistic to:     $F = \frac{s_1^2}{s_2^2}$

  • Degrees of Freedom: Numerator degrees of freedom $df_1 = n_1 - 1$; Denominator degrees of freedom $df_2 = n_2 - 1$.

  • Computational Convention: Always place the larger sample variance in the numerator ($s_1^2 \ge s_2^2$). This forces the calculated $F$-ratio to be $\ge 1$, allowing the test to utilize upper-tail tables.

• Primary Engineering Functions of the $F$-Distribution:

  1. Variance Comparison: Comparing the variability, consistency, or precision between two production processes, material suppliers, or testing techniques.

  2. Statistical Gatekeeper: Serving as an obligatory preliminary test prior to conducting two-sample pooled $t$-tests.

  3. Analysis of Variance (ANOVA): Testing for equivalence across three or more group means simultaneously.

• Hypothesis Testing for Two Population Variances:

  • Hypotheses:

  • Two-tailed test: $H_0: \sigma_1^2 = \sigma_2^2$ vs. $H_a: \sigma_1^2 \ne \sigma_2^2$ (critical value evaluated at $\frac{\alpha}{2}$

  • Right-tailed test: $H_0: \sigma_1^2 \le \sigma_2^2$ vs. $H_a: \sigma_1^2 > \sigma_2^2$ (critical value evaluated at $\alpha$

  • Confidence Interval for the Ratio of Two Variances ($\sigma_1^2 / \sigma_2^2$):     $\left[ \frac{s_1^2 / s_2^2}{F_{\alpha/2, df_1, df_2}}, \; \left(\frac{s_1^2}{s_2^2}\right) \times F_{\alpha/2, df_2, df_1} \right]$

• Engineering Case Studies in Variance Testing:

  • Cement Strength Consistency (Supplier A vs. Supplier B):

  • Data: Supplier A has $n_1 = 11$ batches, $s_1 = 5.8\text{ MPa}$; Supplier B has $n_2 = 10$ batches, $s_2 = 2.3\text{ MPa}$.

  • Test at $\alpha = 0.05$ (two-tailed; $\frac{\alpha}{2} = 0.025$).

  • Degrees of Freedom: $df_1 = 11 - 1 = 10$ (numerator); $df_2 = 10 - 1 = 9$ (denominator).

  • Calculated $F$-statistic:       $F = \frac{s_1^2}{s_2^2} = \frac{5.8^2}{2.3^2} = \frac{33.64}{5.29} \approx 6.36$

  • Critical Value from table: $F_{0.025, 10, 9} = 3.96$.

  • Decision: Because $6.36 > 3.96$, reject $H_0$. There is statistically significant evidence that the two cement suppliers exhibit unequal variance.

  • Engineering Implications:

    • Supplier A exhibits high variance, leading to unpredictable compressive strengths, risk of batch failure, and structural overdesign waste.

    • Supplier B exhibits tightly controlled quality and predictable strength.

    • Because equal variance is violated, a pooled two-sample $t$-test cannot be used to compare their means; Welch's separate-variance $t$-test must be employed.

  • Confidence Interval on Ratio of Cement Variances:

  • Sample variance ratio: $\frac{s_1^2}{s_2^2} = 6.36$.

  • Upper-tail critical values: $F_{0.025, 10, 9} = 3.96$ and $F_{0.025, 9, 10} = 3.78$.

  • Lower bound: $\frac{6.36}{3.96} \approx 1.61$.

  • Upper bound: $6.36 \times 3.78 \approx 24.04$.

  • $95\%$ CI for $\frac{\sigma_1^2}{\sigma_2^2}$: $(1.61, 24.04)$. Because the interval does not capture 1, unequal variance is confirmed.

  • Steel Rebar Tensile Strength Variance:

  • Sample X: $n_1 = 9, s_1 = 4.2\text{ MPa} \implies df_1 = 8$.

  • Sample Y: $n_2 = 7, s_2 = 2.8\text{ MPa} \implies df_2 = 6$.

  • Test at $\alpha = 0.05$ (two-tailed, $\frac{\alpha}{2} = 0.025$):       $F = \frac{4.2^2}{2.8^2} = \frac{17.64}{7.84} = 2.25$

  • From $F$-tables, $F_{0.025, 8, 6} = 5.60$.

  • Decision: Because $2.25 < 5.60$, do not reject $H_0$. The evidence is insufficient to conclude that tensile strength variances differ.

• Critical Assumptions and Diagnostics:

  • Normality: The $F$-test is sensitive to departures from normality in either parent population.

  • Independence: Observations within and between groups must be independent.

  • Violation Action: When parent distributions depart from normality, the parametric $F$-test loses validity; non-parametric tests such as Levene's test should be substituted.

Correlation and Simple Linear Regression

• The Analytical Shift: From Differences to Relationships:

  • Previous analytical modules addressed differences in location ($\mu$) or dispersion ($\sigma^2$).

  • Correlation and regression address relationship structures between paired quantitative variables $(x, y)$.

  • Correlation: Quantifies the direction and strength of the linear association between two variables.

  • Regression: Models the mathematical relationship, expressing $y$ as a functional equation of $x$ to enable predictive estimation.

• Pearson Correlation Coefficient ($r$):

  • Measures the degree of linear association between two variables, bounded strictly within $-1.0 \le r \le +1.0$.

  • Algebraic Formula:     $r = \frac{n \sum xy - \sum x \sum y}{\sqrt{\left[n \sum x^2 - (\sum x)^2\right] \left[n \sum y^2 - (\sum y)^2\right]}}$