Hypothesis Testing for One and Two Population Means

Foundations of Hypothesis Testing for Means

  • Transition from Proportions to Means:

    • Statistical inference shifts focus from categorical parameters to quantitative parameters.

    • Parameter of Interest: The population mean is denoted by μ\mu (pronounced "mew"), contrasting with the population proportion pp.

    • Sample Statistic: The sample mean is denoted by xˉ\bar{x} (or yˉ\bar{y}), calculated as ∑xn\frac{\sum x}{n}, replacing the sample proportion p^=xn\hat{p} = \frac{x}{n}.

    • Typical Inferential Questions: Instead of categorical proportions (such as "What percentage of the city smokes?"), mean inference addresses continuous measures (such as "What is the average GPA?" or "What is the average household income?").

  • The Four-Step Hypothesis Testing Framework:

    • The core decision engine remains unchanged across all hypothesis tests:

    • Step 1: State Hypotheses: Establish the null hypothesis (H0H_0) and alternative hypothesis (HaH_a).

    • Step 2: Check Assumptions and Calculate Test Statistic: Validate data collection standards and compute the standardized yardstick (zz or tt).

    • Step 3: Determine the P-Value: Compute the probability, assuming H0H_0 is true, of obtaining data at least as extreme as the observed sample.

    • Step 4: State the Decision and Conclusion: Compare the P-value against the significance level (α\alpha); reject H0H_0 if the P-value is less than or equal to α\alpha, and fail to reject H0H_0 if the P-value is greater than α\alpha. Never declare that the null hypothesis is "accepted".

  • Formulating Hypotheses for a Single Mean:

    • The Null Hypothesis (H0H_0): Pins the parameter to a specific null value: H0:μ=μ0H_0: \mu = \mu_0. It always contains an exact equality sign (==).

    • The Alternative Hypothesis (HaH_a): Established strictly prior to data collection based entirely on the research question. The data must never be used to choose the direction of the alternative hypothesis.

    • Lower-tail (Left-sided): Ha:μ<μ0H_a: \mu < \mu_0

    • Two-tailed (Two-sided): Ha:μ≠μ0H_a: \mu \neq \mu_0

  • Upper-tail (Right-sided): Ha:μ>μ0H_a: \mu > \mu_0

Rejection regions for one-sided and two-sided alternative hypotheses

The One-Sample Decision Fork: z-Test vs. t-Test

  • The Central Criterion (σ\sigma Known vs. σ\sigma Unknown):

    • The choice of test statistic hinges entirely on whether the true population standard deviation (σ\sigma) is known or whether it must be estimated from sample data using the sample standard deviation (ss).

  • The Population Standard Deviation Known Fork (zz-Test):

    • When σ\sigma is known, the sampling distribution of the sample mean follows a Normal model.

    • Standard Error (Yardstick):     SD(xˉ)=σn\text{SD}(\bar{x}) = \frac{\sigma}{\sqrt{n}}

    • Test Statistic Formula:     z=xˉ−μ0σnz = \frac{\bar{x} - \mu_0}{\frac{\sigma}{\sqrt{n}}}

    • Practical Applicability: Extremely rare in real-world contexts, occasionally occurring in industrial quality control settings with extensive long-run process histories.

    • Critical values (z∗z^*) are derived directly from the standard Normal model via invNorm.

  • The Population Standard Deviation Unknown Fork (tt-Test):

    • When σ\sigma is unknown, it must be estimated using the sample standard deviation ss.

    • Estimated Standard Error:     SE(xˉ)=sn\text{SE}(\bar{x}) = \frac{s}{\sqrt{n}}

    • Test Statistic Formula:     t=xˉ−μ0snt = \frac{\bar{x} - \mu_0}{\frac{s}{\sqrt{n}}}

    • Degrees of Freedom: Governed by Student's tt-distribution with degrees of freedom:     df=n−1df = n - 1

    • Practical Applicability: The universal standard for empirical data where population parameters are unknown.

Student's t-Distribution and Degrees of Freedom

  • Origin and Theoretical Need for a New Curve:

    • Substituting the sample statistic ss in place of the parameter σ\sigma introduces extra random sampling variability into the denominator of the test statistic.

    • Because ss wobbles from sample to sample, particularly in small samples, the test statistic exhibits more dispersion than predicted by the standard Normal distribution.

    • Historical Development: Formulated in 1908 by William Sealy Gosset, a chemist and statistician conducting quality control at the Guinness Brewery in Dublin. Gosset published the distribution under the pen name "Student" due to proprietary restrictions imposed by his employer.

  • Properties of the tt-Distribution:

    • Symmetry and Center: Like the standard Normal distribution, Student's tt-distribution is symmetric, bell-shaped (mound-shaped), and centered at zero (μ=0\mu = 0).

    • Heavier / Fatter Tails: The tt-distribution possesses fatter tails than the standard Normal curve, allocating more probability area far from zero. This reflects the greater likelihood of obtaining extreme test statistics when estimating σ\sigma with ss.

    • Degrees of Freedom (dfdf): Student's tt is not a single distribution but an entire family of continuous probability distributions indexed by degrees of freedom (df=n−1df = n - 1).

    • Asymptotic Convergence: As the sample size increases (n→∞n \rightarrow \infty, df→∞df \rightarrow \infty), the estimate ss stabilizes and approaches σ\sigma. Consequently, the tt-distribution converges to the standard Normal distribution. In practice, when df>40df > 40, the difference between the tt-distribution and the standard Normal distribution becomes negligible.

Comparison of the standard normal distribution and Student's t-distributions with varying degrees of freedom

Assumptions and the Sample-Size Ladder

  • Sampling Honesty Assumptions (Step 0):

    • Independence Assumption: Individual data observations must be independent of one another. The measurement of one subject must provide no information regarding the value of another subject.

    • Randomization Condition: The data must originate from a representative random sample (ideally a Simple Random Sample, or SRS) or a properly randomized comparative experiment. Without randomization, inferential procedures lack validity.

    • 10% Condition: When sampling without replacement from a finite population, the sample size must not exceed 10%10\% of the total population (n≤0.10Nn \le 0.10 N) to prevent compromising independence.

  • The Normality Requirement and the Nearly Normal Condition:

    • Because population distributions cannot be directly observed, the population shape is evaluated through the sample.

    • Nearly Normal Condition: The sample data must come from a population distribution that is unimodal and reasonably symmetric.

    • Diagnostic Tools: Assessed by inspecting a histogram, stem-and-leaf display, or Normal probability plot (Q-Q plot) of the sample. On a Normal probability plot, data following a straight linear pattern indicate consistency with a Normal distribution.

  • The Sample-Size Ladder for the tt-Test:

    • Small Samples (n<15n < 15): The strictest threshold. The data must closely adhere to a Normal model without visible skewness, gaps, or outliers. The presence of any outlier or notable skewness entirely invalidates the tt-test.

    • Moderate Samples (15≤n≤4015 \le n \le 40): The tt-procedures perform reliably provided the sample distribution is unimodal and reasonably symmetric, without extreme outliers.

    • Large Samples (n>40n > 40 to 5050): The Central Limit Theorem ensures that the sampling distribution of the sample mean is approximately Normal regardless of underlying population shape, rendering the tt-method robust against non-Normality unless the data exhibit severe skewness or extreme outliers.

Computational Procedures: TI-84 and R

  • TI-84 Graphing Calculator Commands:

    • Hypothesis Testing for One Mean:

    • When σ\sigma is known: STAT →\rightarrow TESTS →\rightarrow 1: Z-Test. Input: μ0\mu_0, σ\sigma, xˉ\bar{x}, nn, and the direction of HaH_a.

    • When σ\sigma is unknown: STAT →\rightarrow TESTS →\rightarrow 2: T-Test. Input: μ0\mu_0, ss, xˉ\bar{x}, nn, and the direction of HaH_a.

    • Output: Standardized test statistic (zz or tt) and the P-value.

    • Calculating Critical Values:

    • For standard Normal: 2nd VARS (DISTR) →\rightarrow 3: invNorm(area).

    • For Student's tt: 2nd VARS (DISTR) →\rightarrow 4: invT(area, df), where df=n−1df = n - 1.

    • Left-Tail Probability Entry Rule: The calculator command requires the cumulative area to the left of the critical value:

      • Left-tailed test: area=α\text{area} = \alpha

      • Right-tailed test: area=1−α\text{area} = 1 - \alpha

      • Two-tailed test: area=α2\text{area} = \frac{\alpha}{2} (yielding −t∗-t^*, mirrored symmetrically to +t∗+t^*)

  • R Statistical Computing Syntax:

    • Standardizing from Summary Statistics:

    • Two-sided zz-test: 2 * pnorm((x_bar - mu_0) / (sigma / sqrt(n)))

    • One-sided tt-test (left tail): pt((x_bar - mu_0) / (s / sqrt(n)), df = n - 1)

    • Direct Raw Data Input:

    • Execute: t.test(x, mu = mu_0, alternative = "less")

    • Parameter options for alternative include "less", "greater", or "two.sided".

    • P-Value Doubling Rule: One-tailed tests use the direct tail area; two-tailed tests require doubling the tail probability (P=2×tail areaP = 2 \times \text{tail area}).

Worked Examples: One-Sample Inference

  • Example 1: Disposable Cup Usage (zz-Test):

    • Context: The average college student nationally uses 500 cups500\,\text{cups} annually. A random sample of n=50n = 50 students from a large university demonstrates a mean usage of xˉ=476 cups\bar{x} = 476\,\text{cups}. Historical research provides a known population standard deviation of σ=42 cups\sigma = 42\,\text{cups}. Test at α=0.10\alpha = 0.10 whether the mean cup consumption differs from 500 cups500\,\text{cups}.

    • Parameter: μ=\mu = true mean number of disposable cups used per year by students attending this campus.

    • Hypotheses:     H0:μ=500H_0: \mu = 500     Ha:μ≠500H_a: \mu \neq 500

  • Condition Verification:

    • Sampling is random (n=50n = 50).

    • 5050 students constitute less than 10%10\% of a large university campus population.

    • Sample size n=50n = 50 sits on the top rung of the sample-size ladder (n>40−50n > 40-50), ensuring the sampling distribution of xˉ\bar{x} is comfortably Normal.

    • Known σ=42\sigma = 42 dictates a one-sample zz-test.

  • Mathematical Mechanics:

    • Standard Error:     SD(xˉ)=σn=4250≈5.9397 cups\text{SD}(\bar{x}) = \frac{\sigma}{\sqrt{n}} = \frac{42}{\sqrt{50}} \approx 5.9397\,\text{cups}

    • Test Statistic:     z=476−5005.9397=−245.9397≈−4.04z = \frac{476 - 500}{5.9397} = \frac{-24}{5.9397} \approx -4.04

    • P-Value Computation:     P=2⋅P(Z<−4.04)=0.00005333P = 2 \cdot P(Z < -4.04) = 0.00005333

    • Critical Value Cutoff:     z∗=±1.645z^* = \pm 1.645

  • Decision and Contextual Conclusion:

    • Because P=0.00005333<α=0.10P = 0.00005333 < \alpha = 0.10 (and z=−4.04z = -4.04 lies beyond the lower critical cutoff −1.645-1.645), reject H0H_0.

    • There is strong statistical evidence to conclude that the mean annual disposable cup usage by students at this university differs from 500 cups500\,\text{cups}.

    • Calculator Readout Alert: Calculator outputs often display this result in scientific notation as 5.33E-5, which denotes 5.33×10−5=0.00005335.33 \times 10^{-5} = 0.0000533, not 5.335.33.

Rejection region and test statistic for two-sided disposable cups z-test
  • Example 2: Commercial Wine Bottling Volume (tt-Test):

    • Context: Regulatory standards require wine bottles to contain a mean volume of exactly 0.7 liter0.7\,\text{liter}. A random sample of n=6n = 6 bottles yielded a sample mean of xˉ=0.697 liter\bar{x} = 0.697\,\text{liter} and a sample standard deviation of s=0.01 liters = 0.01\,\text{liter}. Determine at α=0.10\alpha = 0.10 whether the importer is under-filling bottles.

    • Parameter: μ=\mu = true mean fill volume of all bottles packaged by this importer.

    • Hypotheses:     H0:μ=0.7H_0: \mu = 0.7     Ha:μ<0.7H_a: \mu < 0.7

    • Condition Verification:

    • The 66 bottles represent an independent random sample.

    • Six bottles represent less than 10%10\% of commercial production.

    • Sample size n=6<15n = 6 < 15 occupies the strictest rung of the ladder; a histogram or Normal probability plot must confirm no skewness or outliers. Automated bottling volumes are typically Normal.

    • Unknown σ\sigma requires estimating spread using s=0.01s = 0.01, mandating a Student's tt-test with degrees of freedom:       df=n−1=6−1=5df = n - 1 = 6 - 1 = 5

    • Mathematical Mechanics:

    • Standard Error:       SE(xˉ)=sn=0.016≈0.004082 liter\text{SE}(\bar{x}) = \frac{s}{\sqrt{n}} = \frac{0.01}{\sqrt{6}} \approx 0.004082\,\text{liter}

    • Test Statistic:       t=0.697−0.70.004082=−0.0030.004082≈−0.735t = \frac{0.697 - 0.7}{0.004082} = \frac{-0.003}{0.004082} \approx -0.735

    • Critical Value Cutoff:       t∗=invT(0.10,5)=−1.476t^* = \text{invT}(0.10, 5) = -1.476

    • P-Value Computation:       P=P(t5<−0.735)=0.2477P = P(t_5 < -0.735) = 0.2477

    • Decision and Contextual Conclusion:

    • Because P=0.2477>α=0.10P = 0.2477 > \alpha = 0.10 (and t=−0.735>−1.476t = -0.735 > -1.476, failing to enter the critical region), fail to reject H0H_0.

    • The data do not provide statistically significant evidence to conclude that the importer is under-filling wine bottles.

    • Failing to reject does not establish that the bottles contain exactly 0.7 liter0.7\,\text{liter}; an undersized sample of n=6n = 6 cannot prove the null hypothesis.

One-sided t-test rejection region and test statistic for wine bottle fill volume

Comparing Two Population Means: The Difference Model

  • Parameter and Sampling Distribution of the Difference:

    • Inference shifts to comparing two separate populations with means μ1\mu_1 and μ2\mu_2.

    • Parameter of Interest: The difference between the two population means: μ1−μ2\mu_1 - \mu_2.

    • Sample Statistic: The difference between two independent sample means: yˉ1−yˉ2\bar{y}_1 - \bar{y}_2.

  • The Additivity of Variances Principle:

    • For independent random variables, variances always add, regardless of whether calculating a sum or a difference:     Var(Yˉ1−Yˉ2)=Var(Yˉ1)+Var(Yˉ2)=σ12n1+σ22n2\text{Var}(\bar{Y}_1 - \bar{Y}_2) = \text{Var}(\bar{Y}_1) + \text{Var}(\bar{Y}_2) = \frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}

    • True Standard Deviation (Both σ1,σ2\sigma_1, \sigma_2 Known):     SD(yˉ1−yˉ2)=σ12n1+σ22n2\text{SD}(\bar{y}_1 - \bar{y}_2) = \sqrt{\frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}}

    • Estimated Standard Error (Both σ1,σ2\sigma_1, \sigma_2 Unknown):     SE(yˉ1−yˉ2)=s12n1+s22n2\text{SE}(\bar{y}_1 - \bar{y}_2) = \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}

  • Formulating Two-Sample Hypotheses:

    • Null Hypothesis (H0H_0): Posits that the two population means are identical:     H0:μ1=μ2⟺H0:μ1−μ2=0H_0: \mu_1 = \mu_2 \quad \Longleftrightarrow \quad H_0: \mu_1 - \mu_2 = 0

    • Alternative Hypotheses (HaH_a): Established prior to observation based on the research objective:

    • Two-sided: Ha:μ1≠μ2H_a: \mu_1 \neq \mu_2

    • Upper-tail (Right-sided): Ha:μ1>μ2H_a: \mu_1 > \mu_2

    • Lower-tail (Left-sided): Ha:μ1<μ2H_a: \mu_1 < \mu_2

  • The Two-Sample Test Statistic Formulas:

    • Two-Sample zz-Test (σ1,σ2\sigma_1, \sigma_2 Known):     z=(yˉ1−yˉ2)−(μ1−μ2)σ12n1+σ22n2z = \frac{(\bar{y}_1 - \bar{y}_2) - (\mu_1 - \mu_2)}{\sqrt{\frac{\sigma_1^2}{n_1} + \frac{\sigma_2^2}{n_2}}}

    • Two-Sample tt-Test (σ1,σ2\sigma_1, \sigma_2 Unknown):     t=(yˉ1−yˉ2)−(μ1−μ2)SE(yˉ1−yˉ2)=yˉ1−yˉ2s12n1+s22n2t = \frac{(\bar{y}_1 - \bar{y}_2) - (\mu_1 - \mu_2)}{\text{SE}(\bar{y}_1 - \bar{y}_2)} = \frac{\bar{y}_1 - \bar{y}_2}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}     (assuming under H0H_0 that μ1−μ2=0\mu_1 - \mu_2 = 0).

Two-Sample Assumptions and Pre-Analysis Exploration

  • Exploratory Data Analysis via Side-by-Side Boxplots:

    • Prior to formal statistical calculations, data must be plotted using side-by-side boxplots sharing a common scale.

    • Centers: Evaluate differences between group medians.

    • Spreads: Compare interquartile ranges (box widths) to assess whether variances are comparable.

    • Outliers: Spot isolated data points plotted beyond the whiskers. Outliers exert substantial leverage on means and standard deviations, potentially distorting inferential outcomes.

Side-by-side boxplots comparing battery duration between brand name and generic batteries
  • Comprehensive Two-Sample Assumption Checklist:

    • Every assumption must be verified independently for both sample groups.

    • Within-Group Independence: Observations within Sample 1 must be independent of one another; observations within Sample 2 must be independent of one another.

    • Randomization Condition: Both groups must originate from independent random samples or participants must be assigned to treatment conditions via a randomized experiment.

    • 10% Condition: Generally omitted in two-sample difference of means analysis, unless evaluating unusually small populations or extremely large sampling fractions.

    • Nearly Normal Condition: Both groups must be unimodal and reasonably symmetric, verified by evaluating histograms or Normal probability plots for each sample. A violation in either group compromises the test.

    • Independent Groups Assumption: The groups themselves must be independent of each other (separate subjects, no matching, no overlap). If subjects are matched, paired, or tracked over a pre-test/post-test interval, independent two-sample methods are invalid and paired-sample methodology must be employed.

Unpooled vs. Pooled Two-Sample t-Tests

  • Unpooled Two-Sample tt-Test (Welch’s tt-Test):

    • Core Assumption: Does not assume population variances are equal (σ12≠σ22\sigma_1^2 \neq \sigma_2^2).

    • Standard Error:     SE(yˉ1−yˉ2)=s12n1+s22n2\text{SE}(\bar{y}_1 - \bar{y}_2) = \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}

    • Degrees of Freedom: Calculated using the Welch-Satterthwaite approximation. This value is typically a non-integer containing decimal places.

    • Practical Standard: Serves as the reliable default setting because it avoids making unverified assumptions regarding population variances.

  • Pooled Two-Sample tt-Test:

    • Core Assumption: Requires assuming equal population variances (σ12=σ22=σ2\sigma_1^2 = \sigma_2^2 = \sigma^2).

    • Pooled Sample Variance (spooled2s_{\text{pooled}}^2): Combines the sum of squared deviations from both samples weighted by their respective degrees of freedom:     spooled2=(n1−1)s12+(n2−1)s22(n1−1)+(n2−1)s_{\text{pooled}}^2 = \frac{(n_1 - 1)s_1^2 + (n_2 - 1)s_2^2}{(n_1 - 1) + (n_2 - 1)}

    • Pooled Standard Error:     SEpooled(yˉ1−yˉ2)=spooled1n1+1n2\text{SE}_{\text{pooled}}(\bar{y}_1 - \bar{y}_2) = s_{\text{pooled}} \sqrt{\frac{1}{n_1} + \frac{1}{n_2}}

    • Degrees of Freedom: Exactly equals an integer value:     df=n1+n2−2df = n_1 + n_2 - 2

    • Technology Setting: Controlled on the TI-84 via the prompt Pooled?: No (Unpooled) or Pooled?: Yes (Pooled).

  • Two-Sample Critical Value Execution:

    • Critical cutoffs require executing invT(area, df) using the non-integer degrees of freedom generated by the unpooled software output (for instance, entering 17.553 rather than rounding to 17).

    • Left area inputs for invT:

    • Right-tailed test: area=1−α\text{area} = 1 - \alpha

    • Left-tailed test: area=α\text{area} = \alpha

    • Two-tailed test: area=α2\text{area} = \frac{\alpha}{2}

Worked Examples: Two-Sample Inference

  • Example 1: Comparative Reading Instruction Methods (One-Sided Unpooled tt-Test):

    • Context: A study evaluates whether a new instructional method outperforms the standard reading method for slower learners over a six-month period. Twenty children are randomly assigned to two groups: Group 1 (New Method: n1=8n_1 = 8, xˉ1=76.9\bar{x}_1 = 76.9, s1=4.85s_1 = 4.85) and Group 2 (Standard Method: n2=12n_2 = 12, xˉ2=72.7\bar{x}_2 = 72.7, s2=6.35s_2 = 6.35). Test at α=0.05\alpha = 0.05 whether the new method produces higher mean scores.

    • Hypotheses:     H0:μ1=μ2⟺μ1−μ2=0H_0: \mu_1 = \mu_2 \quad \Longleftrightarrow \quad \mu_1 - \mu_2 = 0     Ha:μ1>μ2⟺μ1−μ2>0H_a: \mu_1 > \mu_2 \quad \Longleftrightarrow \quad \mu_1 - \mu_2 > 0

    • Assumption Check:

    • Random sampling and random allocation across instructional methods verified.

    • Small sample sizes (n1=8n_1 = 8, n2=12n_2 = 12) require checking histograms for each group; test scores typically follow a unimodal, symmetric pattern.

    • Independent groups verified: distinct, non-overlapping student sets.

    • Variances are not assumed equal; apply unpooled tt-test (Pooled: No).

    • Mathematical Mechanics:

    • Unpooled Standard Error:       SE(yˉ1−yˉ2)=4.8528+6.35212=2.9403+3.3602=6.3005≈2.5101\text{SE}(\bar{y}_1 - \bar{y}_2) = \sqrt{\frac{4.85^2}{8} + \frac{6.35^2}{12}} = \sqrt{2.9403 + 3.3602} = \sqrt{6.3005} \approx 2.5101

    • Difference of Means:       yˉ1−yˉ2=76.9−72.7=4.2\bar{y}_1 - \bar{y}_2 = 76.9 - 72.7 = 4.2

    • Test Statistic:       t=4.22.5101≈1.673t = \frac{4.2}{2.5101} \approx 1.673

    • Software Degrees of Freedom: df=17.55307df = 17.55307

    • P-Value: P=P(t17.553>1.673)=0.0560P = P(t_{17.553} > 1.673) = 0.0560

    • Critical Cutoff: t∗=invT(0.95,17.553)=1.736t^* = \text{invT}(0.95, 17.553) = 1.736

    • Decision and Conclusion:

    • Because P=0.0560>α=0.05P = 0.0560 > \alpha = 0.05 (and t=1.673<1.736t = 1.673 < 1.736), fail to reject H0H_0.

    • The data do not provide statistically significant evidence at the α=0.05\alpha = 0.05 level to support the claim that the new instructional method improves reading test performance.

    • Comparison with Pooled Method: If pooled, the test yields t=1.583t = 1.583, df=8+12−2=18df = 8 + 12 - 2 = 18, and P=0.0654P = 0.0654, leading to the same substantive decision.

One-sided two-sample t-test rejection region and test statistic for reading methods
  • Example 2: Weekly Study Hours of Athletes vs. Non-Athletes (Two-Sided Unpooled tt-Test):

    • Context: An institutional study assesses whether student athletes and non-athletes differ in weekly study hours. Sample data: Group 1 (Athletes: n1=55n_1 = 55, xˉ1=20.6 hours\bar{x}_1 = 20.6\,\text{hours}, s1=5.3 hourss_1 = 5.3\,\text{hours}) and Group 2 (Non-athletes: n2=200n_2 = 200, xˉ2=23.5 hours\bar{x}_2 = 23.5\,\text{hours}, s2=4.1 hourss_2 = 4.1\,\text{hours}). Conduct a two-sided test at α=0.01\alpha = 0.01.

    • Hypotheses:     H0:μ1=μ2H_0: \mu_1 = \mu_2     Ha:μ1≠μ2H_a: \mu_1 \neq \mu_2

  • Assumption Check:

    • Independent random samples collected from both student populations.

    • Large sample sizes (n1=55n_1 = 55, n2=200n_2 = 200) place both groups on the top rung of the sample-size ladder, ensuring robustness against non-Normality.

    • Two distinct student cohorts satisfy the independent groups assumption.

  • Mathematical Mechanics:

    • Unpooled Standard Error:     SE(yˉ1−yˉ2)=5.3255+4.12200=0.5107+0.0841=0.5948≈0.7712\text{SE}(\bar{y}_1 - \bar{y}_2) = \sqrt{\frac{5.3^2}{55} + \frac{4.1^2}{200}} = \sqrt{0.5107 + 0.0841} = \sqrt{0.5948} \approx 0.7712

    • Difference of Means:     yˉ1−yˉ2=20.6−23.5=−2.9\bar{y}_1 - \bar{y}_2 = 20.6 - 23.5 = -2.9

    • Test Statistic:     t=−2.90.7712≈−3.76t = \frac{-2.9}{0.7712} \approx -3.76

    • Software Degrees of Freedom: df=72.70167df = 72.70167

    • P-Value: P=2⋅P(t72.70<−3.76)=3.40513×10−4=0.00034P = 2 \cdot P(t_{72.70} < -3.76) = 3.40513 \times 10^{-4} = 0.00034

    • Critical Cutoff (α/2=0.005\alpha/2 = 0.005): t∗=±2.645t^* = \pm 2.645

  • Decision and Conclusion:

    • Because P=0.00034<α=0.01P = 0.00034 < \alpha = 0.01 (and observed t=−3.76<−2.645t = -3.76 < -2.645 falls deep within the rejection region), reject H0H_0.

    • There is strong statistical evidence to conclude that the mean weekly study hours differ significantly between student athletes and non-athletes.

    • Scope of Inference: The test establishes that a significant difference exists; it does not demonstrate causal attribution. Observed sample averages show athletes studied less (20.620.6 vs. 23.5 hours23.5\,\text{hours}), but causality cannot be inferred from observational sampling.

    • Comparison with Pooled Method: If pooled, the test produces t=−4.344t = -4.344, df=55+200−2=253df = 55 + 200 - 2 = 253, and P≈2.0×10−5P \approx 2.0 \times 10^{-5}, arriving at the same rejection decision.

Two-sided two-sample t-test rejection region and test statistic for study hours

Conceptual Review and Practice Problems

  • Problem Set 1 (One-Sample Inference):

    • Question 1: A food label claims an average content of 5 oz5\,\text{oz} per box. A random sample of n=25n = 25 boxes reveals xˉ=4.9 oz\bar{x} = 4.9\,\text{oz} and s=0.4 ozs = 0.4\,\text{oz}. Which test should be performed?

    • A: zz-test — the SD is known

    • B: tt-test with df=24df = 24

    • C: tt-test with df=25df = 25

    • Correct Answer: B

    • Rationale: The standard deviation s=0.4s = 0.4 was calculated from the sample, meaning σ\sigma is unknown. This requires a tt-test with degrees of freedom df=n−1=25−1=24df = n - 1 = 25 - 1 = 24.

    • Question 2: A one-sample tt-test utilizes n=15n = 15 observations. The degrees of freedom are:

    • A: 1515

    • B: 1414

    • C: 1616

    • Correct Answer: B

    • Rationale: Degrees of freedom are defined by df=n−1=15−1=14df = n - 1 = 15 - 1 = 14.

    • Question 3: Compared directly to the standard Normal distribution, Student's tt-distribution has:

    • A: fatter tails — extreme values are more likely

    • B: thinner tails — extreme values are less likely

    • C: a different center

    • Correct Answer: A

    • Rationale: Both distributions share the same center (μ=0\mu = 0), but estimating σ\sigma with ss adds variability, spreading out probability into fatter tails.

    • Question 4: A one-sided left-tailed tt-test at α=0.05\alpha = 0.05 with df=24df = 24 has a critical cutoff of t∗=−1.711t^* = -1.711. The calculated test statistic is t=−1.2t = -1.2. The proper statistical decision is:

    • A: reject H0H_0

    • B: fail to reject H0H_0

    • C: accept H0H_0

    • Correct Answer: B

    • Rationale: The test statistic t=−1.2t = -1.2 is greater than −1.711-1.711, failing to enter the critical rejection region. In hypothesis testing, we never "accept" the null hypothesis.

  • Problem Set 2 (Two-Sample Inference):

    • Question 1: Two independent samples yield xˉ1,s1,n1\bar{x}_1, s_1, n_1 and xˉ2,s2,n2\bar{x}_2, s_2, n_2 without population standard deviations provided. Which inferential test is appropriate to compare means?

    • A: 2-Sample zz-test — the σ\sigma's are known

    • B: 2-Sample tt-test — the σ\sigma's are unknown

    • C: A one-sample tt-test on each group

    • Correct Answer: B

    • Rationale: When only sample standard deviations s1s_1 and s2s_2 are available, population standard deviations are unknown, requiring a two-sample tt-test.

    • Question 2: A two-sample tt-test conducted with unpooled variances (Pooled: No) yields degrees of freedom df=21.87df = 21.87. Why is the degrees of freedom value not an integer?

    • A: The calculator made a rounding error

    • B: Unpooled df is computed by software — decimals are expected

    • C: df is always n1+n2−2n_1 + n_2 - 2

    • Correct Answer: B

    • Rationale: The Welch-Satterthwaite approximation evaluates variance inequality using a complex formula resulting in a non-integer degrees of freedom value.

    • Question 3: A pooled two-sample tt-test with sample sizes n1=8n_1 = 8 and n2=12n_2 = 12 has degrees of freedom equal to:

    • A: 1818

    • B: 2020

    • C: 17.55317.553

    • Correct Answer: A

    • Rationale: When pooling variances, degrees of freedom are defined by the integer formula df=n1+n2−2=8+12−2=18df = n_1 + n_2 - 2 = 8 + 12 - 2 = 18. (Note that 17.55317.553 represents the unpooled degrees of freedom for these sample sizes).

    • Question 4: A graphing calculator returns a P-value of p = 4.2E-3. The decimal equivalent of this P-value is:

    • A: 0.00420.0042

    • B: 4.24.2

    • C: 0.420.42

    • Correct Answer: A

    • Rationale: Scientific notation 4.2E-3 represents 4.2×10−3=0.00424.2 \times 10^{-3} = 0.0042. The negative exponent requires shifting the decimal point three places to the left.