Two-Sample Hypothesis Testing and Confidence Intervals

Statistical Distribution Selection: Z Versus T Distribution

• Decision Criteria for Choosing Between Z and T Distributions:
  • The choice between a standard normal ($Z$) distribution and a Student's $t$ ($T$) distribution for hypothesis tests and confidence intervals concerning population means depends on knowledge of the population standard deviation ($\sigma$) and the sample size ($n$).
  • Knowledge of Population Standard Deviation ($\sigma$):
    • If the population standard deviation ($\sigma$) is known:
      • Evaluate whether the sample size satisfies $n \ge 30$.
      • If $n \ge 30$: Use the $Z$ distribution.
      • If $n < 30$: Use the $T$ distribution.
    • If the population standard deviation ($\sigma$) is unknown:
      • Always use the $T$ distribution, regardless of sample size.

College Cost of Attendance Case Study

• Benchmark Economic Context:
  • An April 2012 analysis published in Money evaluated the relationship between the annual Cost of Attendance (COA)—comprising tuition, room, board, books, and required fees—and the median annual household income of families supporting college-aged dependents:
    • Private Higher Education Institutions: Average COA accounted for approximately $60\%$ of total household income.
    • Public Higher Education Institutions: Average COA accounted for nearly $\frac{1}{3}$ (33.3\%$) of total household income.\n* **Dataset Structure (`CollegeCosts.xlsx`):**\n * **Sample Characteristics:** Contains a random sample of n = 22$higher education institutions, divided into$n_1 = 10$private institutions and$n_2 = 12 public institutions.\n * **Unit of Measurement:** COA values are recorded in thousands of US dollars (1000\,\text{USD}.\n * **Worksheet Organization:**\n * `CollegeCosts_Raw`: Presents raw observations in standard tabular structure with categorical identifiers.\n * `CollegeCosts_Edited`: Reorganizes the data by separating institution categories into independent columns.\n * `CollegeCosts_Paired`: Contains matched observations across multiple measurement periods.\n* **Descriptive Statistics via PivotTable Analysis:**\n * **Average COA by Institutional Sector:**\n * Private Institutions: \$42.5\,\text{thousand}$($\$42,500\n * Public Institutions: \$22.3\,\text{thousand}$($\$22,300\n * Grand Total Average: \$31.48181818\,\text{thousand}$($\$31,481.82\n * **Observed Sample Mean Difference:**\n * \bar{x}{\text{Private}} - \bar{x}{\text{Public}} = 42.5 - 22.3 = 20.2\,\text{thousand}$($\$20,200\n\n\n\n# Confidence Intervals for the Difference Between Two Independent Means\n\n* **Two-Sample T-Confidence Interval Specification (Unequal Variance):**\n * Used when comparing independent groups where either sample size is small (n_1 < 30$or$n_2 < 30$) or the population standard deviations ($\sigma_1, \sigma_2) are unknown and not assumed equal.\n * **Mathematical Formula:**\n * (\bar{x}1 - \bar{x}_2) \pm t{\alpha/2} \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}\n * **Margin of Error (E):**\n * E = t_{\alpha/2} \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}\n * **Component Definitions:**\n * \bar{x}1, \bar{x}_2: Sample means of group 1 and group 2.\n * s_1^2, s_2^2: Sample variances of group 1 and group 2.\n * n_1, n_2: Respective sample sizes of group 1 and group 2.\n * t{\alpha/2}$: Critical value derived from the$T$distribution for significance level$\alpha.\n * **Welch–Satterthwaite Degrees of Freedom (df):**\n * df = \frac{\left(\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}\right)^2}{\frac{1}{n_1 - 1}\left(\frac{s_1^2}{n_1}\right)^2 + \frac{1}{n_2 - 1}\left(\frac{s_2^2}{n_2}\right)^2}\n * **Excel Implementation:** Compute the critical value using `=T.INV.2T(alpha, df)`.\n* **Empirical Interpretation of the 90% Confidence Interval:**\n * Using the `CollegeCosts_Edited` data at a 90\%$confidence level ($\alpha = 0.10), the true population difference in mean COA between private and public institutions lies within:\n * [\15.7\,\text{thousand}, \24.7\,\text{thousand}]\n* **Indirect Hypothesis Testing via Confidence Intervals:**\n * A confidence interval constructed for the difference between two population means (\mu_1 - \mu_2$) serves as an indirect two-tailed hypothesis test at significance level$\alpha = 1 - \text{Confidence Level}:\n * **Null Hypothesis Inclusion Rule:** If the interval contains the hypothesized difference of 0$, fail to reject the null hypothesis ($H_0: \mu_1 - \mu_2 = 0).\n * **Null Hypothesis Exclusion Rule:** If the interval does not contain 0$, reject the null hypothesis ($H_0: \mu_1 - \mu_2 = 0$) at$\alpha = 1 - \text{Confidence Level}.\n * Because the 90\%$confidence interval of$[15.7, 24.7]$strictly excludes$0$, the null hypothesis stating that average attendance costs are identical across private and public sectors is rejected at$\alpha = 0.10.\n\n# Framework of Two-Sample Hypothesis Tests\n\n* **Core Purpose:** Evaluates whether differences observed between two sample statistics reflect statistically significant differences in their respective underlying population parameters.\n* **Four Methodological Variations:**\n 1. **Two-Sample Z$Test:** Applied when independent samples are large ($n_1, n_2 \ge 30$) and both population variances ($\sigma_1^2, \sigma_2^2) are known.\n 2. **Two-Sample T$Test (Assuming Equal Variance):** Applied when independent samples are drawn from populations with unknown but equal variances ($\sigma_1^2 = \sigma_2^2).\n 3. **Two-Sample T Test (Assuming Unequal Variance):** Applied when independent samples have unknown and unequal population variances, or small sample sizes.\n 4. **Paired Sample T Test (Matched Pairs):** Applied when observations across the two datasets are dependent or linked to the same subject/entity.\n\n# Two-Sample T-Test Assuming Unequal Variance\n\n* **Operational Conditions:**\n * Used when samples are independent, sample sizes are small (n < 30), or population standard deviations are unknown and cannot be assumed equal.\n* **Hypothesis Formulation for Sector Costs:**\n * H_0: \mu_1 = \mu_2$(or$\mu_1 - \mu_2 = 0; the population mean COA is identical across sectors)\n * H_A: \mu_1 eq \mu_2$(or$\mu_1 - \mu_2 eq 0; the population mean COA differs significantly)\n * Significance level: \alpha = 0.10\n* **Test Statistic Formula:**\n * t = \frac{(\bar{x}1 - \bar{x}_2) - \Delta}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}\n * Where \Delta$is the hypothesized difference between population means ($\Delta = 0 unless specified otherwise).\n * Degrees of freedom (df) are calculated via the Welch–Satterthwaite equation:\n * df = \frac{\left(\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}\right)^2}{\frac{1}{n_1 - 1}\left(\frac{s_1^2}{n_1}\right)^2 + \frac{1}{n_2 - 1}\left(\frac{s_2^2}{n_2}\right)^2}\n* **Computational Methods:**\n * **Method 1 (Manual Calculation):** Calculate t$and evaluate using `=T.DIST(x, df, cumulative)` or `=T.DIST.2T(x, df)` to obtain exact$p-values.\n * **Method 2 (Excel Analysis Toolpak):** Execute via the `t-Test: Two-Sample Assuming Unequal Variances` tool using columns from `CollegeCosts_Edited`.\n\n# Rigorous Protocol for Reporting Statistical Decisions\n\n* **Standardized Academic Reporting Template:**\n * All hypothesis tests must be reported using the syntax:\n * "P = [\text{value}]$. At$\alpha = [\text{threshold}], we [reject / fail to reject] the null hypothesis."\n* **Mathematical Decision Rules:**\n * If p < \alpha$: **Reject$H_0**.\n * If p \ge \alpha$: **Fail to reject$H_0**.\n* **Strict Terminology Constraints:**\n * Permissible conclusions are strictly limited to **"reject"** or **"fail to reject"** the null hypothesis.\n * Never use the terms **"accept"**, **"prove"**, or **"disprove"** in formal statistical reporting.\n\n# Paired Sample T-Test (Dependent Observations)\n\n* **Applicability:** Employed when observations consist of natural pairs or repeated measurements on the exact same unit or entity across two conditions or time periods.\n* **Directional Hypotheses for Longitudinal College Costs (2012 to 2022):**\n * Research question: Did college costs significantly increase from 2012 to 2022?\n * **Pairwise Difference Definition:**\n * d = \text{Measurement}_1 - \text{Measurement}_2 = \text{COA}{2012} - \text{COA}{2022}\n * A positive difference (d > 0) denotes that 2012 costs exceeded 2022 costs.\n * A negative difference (d < 0) denotes that 2012 costs were lower than 2022 costs (indicating an increase over time).\n * **Hypothesis Structure at \alpha = 0.05:**\n * H_0: \mu_d \ge 0 (Costs did not increase; mean pairwise difference is zero or positive)\n * H_A: \mu_d < 0 (Costs increased significantly; mean pairwise difference is negative)\n* **Test Statistic and Parameters:**\n * t = \frac{\bar{d} - \Delta}{\frac{s_d}{\sqrt{n}}}\n * **Component Definitions:**\n * \bar{d}$: Sample mean of pairwise differences ($\bar{d} = \frac{1}{n}\sum d_i\n * \Delta$: Hypothesized mean difference ($\Delta = 0 unless stated otherwise).\n * s_d: Sample standard deviation of the pairwise differences.\n * n: Total number of paired observations.\n * Degrees of freedom: df = n - 1\n* **Implementation Methods:**\n * **Method 1:** Manually create a difference column (d_i = x{1i} - x_{2i}$), calculate$\bar{d}$and$s_d$, evaluate the$t$-statistic, and determine$p-values via `=T.DIST(t, df, TRUE)`.\n * **Method 2:** Utilize the Excel Analysis Toolpak module `t-Test: Paired Two Sample for Means`.\n\n# Supplementary Formulations and Alternative Test Procedures\n\n* **Two-Sample Z Confidence Interval (Independent Samples):**\n * **Conditions:** Both sample sizes are large (n_1 \ge 30$and$n_2 \ge 30$) and population standard deviations ($\sigma_1, \sigma_2) are known.\n * **Interval Formula:**\n * (\bar{x}1 - \bar{x}_2) \pm z{\alpha/2} \sqrt{\frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}}\n * **Excel Critical Value:** `=NORM.S.INV(1 - alpha/2)`.\n* **Two-Sample Z Hypothesis Test:**\n * **Test Statistic Formula:**\n * z = \frac{(\bar{x}1 - \bar{x}_2) - \Delta}{\sqrt{\frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}}}\n * **Excel Implementation:** Compute lower-tail probabilities via `=NORM.S.DIST(z, TRUE)` or employ the Analysis Toolpak `z-Test: Two Sample for Means`.\n* **Two-Sample T Test Assuming Equal Variance (Pooled Variance):**\n * **Conditions:** Small samples (n < 30$) or unknown population variances that are assumed equal ($\sigma_1^2 = \sigma_2^2 = \sigma^2).\n * **Degrees of Freedom:**\n * df = n_1 + n_2 - 2\n * **Pooled Sample Variance (s_p^2):**\n * s_p^2 = \frac{(n_1 - 1)s_1^2 + (n_2 - 1)s_2^2}{n_1 + n_2 - 2}\n * **Pooled Test Statistic:**\n * t = \frac{(\bar{x}_1 - \bar{x}_2) - \Delta}{\sqrt{s_p^2 \left(\frac{1}{n_1} + \frac{1}{n_2}\right)}}\n * **Excel Implementation:** Analysis Toolpak `t-Test: Two-Sample Assuming Equal Variances`.\n* **Confidence Interval for Paired Differences:**\n * **Interval Formula:**\n * \bar{d} \pm t{\alpha/2} \frac{s_d}{\sqrt{n}}\n * **Degrees of Freedom:** df = n - 1\n * **Excel Implementation:**\n * Method 1: Calculate critical t$using `=T.INV.2T(alpha, df)` and multiply by standard error$\frac{s_d}{\sqrt{n}}.\n * Method 2: Directly obtain margin of error using `=CONFIDENCE.T(alpha, s_d, n)`.\n\n# Applied Analytical Scenarios and Hypothesis Formulations\n\n* **Scenario 1: Testing Center Differences with Known Population Variances:**\n * Dataset: `TestScores_Edited` stores standardized test scores from two testing centers.\n * Parameters: Known population standard deviation for each center is \sigma_1 = \sigma_2 = 10$. Significance level$\alpha = 0.01.\n * Tool: Analysis Toolpak `z-Test: Two Sample for Means`.\n * Hypotheses:\n * H_0: \mu_1 \le \mu_2\n * H_A: \mu_1 > \mu_2\n* **Scenario 2: Testing Center Differences Assuming Equal Variance:**\n * Dataset: `TestScores_Edited`.\n * Parameters: Unknown population standard deviations assumed equal (\sigma_1^2 = \sigma_2^2$). Significance level$\alpha = 0.01.\n * Tool: Analysis Toolpak `t-Test: Two-Sample Assuming Equal Variances`.\n * Hypotheses:\n * H_0: \mu_1 = \mu_2\n * H_A: \mu_1 eq \mu_2\n* **Scenario 3: Longitudinal College Cost Comparison (Two-Tailed Paired Test):**\n * Dataset: `CollegeCosts_Paired`.\n * Parameters: Significance level \alpha = 0.05.\n * Tool: Analysis Toolpak `t-Test: Paired Two Sample for Means`.\n * Hypotheses:\n * H_0: \mu_d = 0\n * H_A: \mu_d eq 0\n * Complementary Estimation: Calculate a 95\%$confidence interval around the average paired difference$\bar{d}.\n* **Scenario 4: College Cost Differential Exceeding a Specific Threshold:**\n * Dataset: `CollegeCosts_Edited`.\n * Parameters: Test if the mean difference exceeds \$10\,\text{thousand}$at$\alpha = 0.05.\n * Tool: Analysis Toolpak `t-Test: Two-Sample Assuming Unequal Variances` with hypothesized difference \Delta = 10.\n * Hypotheses:\n * H_0: \mu_1 - \mu_2 \le 10\n * H_A: \mu_1 - \mu_2 > 10$$