6.5 Hypothesis Testing Fundamentals

6.5 Hypothesis Testing Fundamentals

In quality engineering, data-driven decision making requires transitioning from subjective observations to rigorous statistical inference. Whether evaluating whether a new machining process alters shaft diameters, determining if a vendor component batch meets purity specifications, or validating an engineering modification, hypothesis testing provides the objective mathematical framework required to make risk-managed decisions.

---

1. Fundamentals of Hypothesis Testing

A hypothesis test evaluates two mutually exclusive statements about a population parameter (such as a mean $\mu$, variance $\sigma^2$, or proportion $p$) using sample data. The process follows a structured six-step sequence:

1. State the Null ($H_0$) and Alternative ($H_1$) Hypotheses based on the engineering question.
2. Select the Significance Level ($\alpha$) based on acceptable Type I risk.
3. Identify the Appropriate Test Statistic and verify distribution assumptions.
4. Define the Decision Rule establishing critical rejection boundaries.
5. Calculate the Test Statistic from sample measurements.
6. State the Statistical Decision and Practical Conclusion (reject or fail to reject $H_0$).

---

2. Hypothesis Formulation: $H_0$ and $H_1$

The formulation of statistical hypotheses requires absolute clarity regarding equality statements:

• Null Hypothesis ($H_0$): The status quo statement assuming no effect, no difference, or no change. $H_0$ must always contain an equality sign ($=$, $\ge$, or $\le$). It is assumed true until sample evidence overwhelmingly contradicts it.
• Alternative Hypothesis ($H_1$ or $H_a$): The operational hypothesis representing the claim, modification, or shift under investigation. $H_1$ never includes equality ($\neq$, $>$, or $<$).

Directional vs. Non-Directional Tests

• Two-Tailed Test (Non-Directional): Used when any departure from the parameter target is critical (e.g., shaft diameter too large or too small).

  $$H_0: \mu = \mu_0 \quad \text{vs.} \quad H_1: \mu \neq \mu_0$$

  The significance area $\alpha$ is divided equally into two tail rejection regions of $\alpha/2$.

• One-Tailed Test (Directional): Used when concern is strictly restricted to one direction (e.g., tensile strength below minimum, or defect rate above maximum).

  • Upper-tail test: $H_0: \mu \le \mu_0 \quad \text{vs.} \quad H_1: \mu > \mu_0$ (Rejection region in upper tail $\alpha$)
  • Lower-tail test: $H_0: \mu \ge \mu_0 \quad \text{vs.} \quad H_1: \mu < \mu_0$ (Rejection region in lower tail $\alpha$)

---

3. Decision Errors, Significance Level ($\alpha$), and Power ($1-\beta$)

Because conclusions are drawn from finite samples rather than entire populations, statistical testing inherently entails risk. Two distinct types of errors can occur.

Decision Matrix & Error Definitions

• Type I Error ($\alpha$): Rejecting $H_0$ when $H_0$ is actually true. In quality engineering, this represents a producer's risk or false alarm—scrapping good product or unnecessarily stopping a running process. The probability $\alpha$ is preset by the engineer (typically 0.05, 0.01, or 0.10).
• Type II Error ($\beta$): Failing to reject $H_0$ when $H_0$ is actually false. This represents a consumer's risk or missed detection—shipping non-conforming material or failing to detect a critical process shift.
• Statistical Power ($1 - \beta$): The probability of correctly rejecting a false null hypothesis when a true process shift of magnitude $\delta = |\mu_1 - \mu_0|$ exists.

Factors Influencing Statistical Power

Quality engineers deliberately design test procedures to achieve adequate power (typically $1 - \beta \ge 0.80$ or $80\%$):

1. Sample Size ($n$): Increasing $n$ reduces standard error ($\sigma / \sqrt{n}$), narrowing sample distributions and dramatically boosting power.
2. Effect Size ($\delta = |\mu_1 - \mu_0|$): Larger physical shifts are much easier for a test to detect, resulting in higher power.
3. Significance Level ($\alpha$): Increasing $\alpha$ (e.g., from 0.01 to 0.05) expands the rejection region, decreasing $\beta$ and increasing power, but increases Type I error risk.
4. Population Variability ($\sigma$): Reducing inherent process noise (variance $\sigma^2$) sharpens statistical discrimination and increases power.

---

4. Decision Rules: $p$-Values and Critical Value Rejection Regions

Two mathematically equivalent methods exist to decide whether sample evidence warrants rejecting $H_0$.

Critical Value Rejection Regions

The critical value ($Z_{crit}$ or $t_{crit}$) divides the sampling distribution into an acceptance region and a rejection region. If the calculated test statistic falls into the rejection region, $H_0$ is rejected.

For a two-tailed $Z$-test at $\alpha = 0.05$, the critical values are $Z_{\alpha/2} = \pm 1.96$. If $|Z_{calc}| > 1.96$, $H_0$ is rejected.

$p$-Value Logic and Interpretation

The $p$-value (probability value) is the exact conditional probability of obtaining a test statistic at least as extreme as the observed sample value, assuming $H_0$ is true.

$$\text{Decision Rule:} \quad \begin{cases} \text{If } p \le \alpha, \& \text{Reject } H_0 \quad (\text{Statistically significant result}) \\ \text{If } p > \alpha, \& \text{Fail to reject } H_0 \quad (\text{Insufficient evidence of shift}) \end{cases}$$

Critical CQE Insight: A $p$-value is NOT the probability that $H_0$ is true. It is $P(\text{Data this extreme} \mid H_0 \text{ is true})$. A small $p$-value indicates that the observed sample data would be extremely rare if the null hypothesis were true.

---

5. Worked Example: One-Sample $Z$-Test for Cable Tensile Strength

Scenario: A cable manufacturing process produces steel wire with a historically known standard deviation of $\sigma = 15\text{ MPa}$. The target mean tensile strength is $\mu_0 = 500\text{ MPa}$. A structural improvement is introduced, and a random sample of $n = 36$ cables is tested, yielding a sample mean of $\bar{x} = 506\text{ MPa}$. Test at $\alpha = 0.05$ whether the new process significantly increases tensile strength.

Step-by-Step Solution:

1. Formulate Hypotheses: $$H_0: \mu \le 500\text{ MPa} \quad \text{vs.} \quad H_1: \mu > 500\text{ MPa} \quad (\text{Upper-tailed test})$$
2. Identify Significance Level: $\alpha = 0.05$.
3. Calculate Test Statistic: Since population $\sigma$ is known, use the one-sample $Z$-test: $$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}} = \frac{15}{\sqrt{36}} = \frac{15}{6} = 2.5\text{ MPa}$$ $$Z_{calc} = \frac{\bar{x} - \mu_0}{\sigma / \sqrt{n}} = \frac{506 - 500}{2.5} = \frac{6}{2.5} = 2.40$$
4. Determine Critical Value & $p$-Value:
   • For an upper-tailed $Z$-test at $\alpha = 0.05$, $Z_{crit} = +1.645$.
   • $p\text{-value} = P(Z \ge 2.40) = 1 - 0.9918 = 0.0082$.
5. Statistical Decision:
   • Critical value method: $Z_{calc} = 2.40 > 1.645 \implies$ Reject $H_0$.
   • $p$-value method: $p = 0.0082 \le 0.05 \implies$ Reject $H_0$.
6. Engineering Conclusion: There is strong statistical evidence ($p = 0.0082$) that the structural improvement significantly increases the mean tensile strength beyond $500\text{ MPa}$.