Statistical Tests z,t,f,chi square , ANOVA Master Guide for M.Tech

Statistical Tests Master Guide for M.Tech

Engineering Statistics, Quality & DOE

1. Fundamentals of Hypothesis Testing

Hypothesis testing is a formal procedure for investigating our ideas about the world using statistics. It is a structured framework used to validate engineering assumptions, analyze quality metrics, and interpret designed experiments.

1.1 The Null (

$$

) and Alternative (

$$

) Hypotheses

• Null Hypothesis (
  

  $$
  ): The default assumption that there is no effect, no difference, or no relationship. It acts as the status quo in quality control (e.g., the new machine part has the exact same diameter as the old one).
• Alternative Hypothesis (
  

  $$
  ): The claim we are trying to prove. It indicates an effect, a difference, or a relationship (e.g., the new machine part has a different diameter than the old one).

1.2 The Concept of 

$$

-Value

The 

$$

-value represents the probability of obtaining test results at least as extreme as the ones observed, assuming the Null Hypothesis (

$$

) is true.

• Low 
  

  $$
  -value (
  
  
  ): Strong evidence against 
  

  $$
  . We reject 
  

  $$
  .
• High 
  

  $$
  -value (
  
  
  ): Weak evidence against 
  

  $$
  . We fail to reject 
  

  $$
  .

1.3 Level of Significance (

$$

) and Errors

The probability of making a wrong decision depends on the chosen significance level: 

$$

.

Decision	$$  is True	$$  is False
Fail to Reject  $$	Correct Decision (Confidence  )	Type II Error ( $$ )
Reject  $$	Type I Error ( $$ , False Positive)	Correct Decision (Power  )

• Type I Error (
  

  $$
  ): Concluding there is a difference when there is none. (e.g., stopping a production line for a false alarm).
• Type II Error (
  

  $$
  ): Concluding there is no difference when a real difference exists. (e.g., letting a batch of defective parts ship to customers).

---

2. Z-Test

Used to determine whether two population means are different when the variances are known and the sample size is large.

2.1 Formula

$$

Where:

• 

  $$
   = sample mean
• 

  $$
   = population mean
• 

  $$
   = population standard deviation
• 

  $$
   = sample size

2.2 Assumptions

• Data must be continuous.
• Samples must be randomly selected.
• Data must be approximately normally distributed.
• 
  
   (Central Limit Theorem applies).
• Population standard deviation (
  

  $$
  ) must be known.

2.3 Degrees of Freedom (DF)

Not applicable (uses the standard normal 

$$

-distribution).

2.4 Effect Size

Cohen's 

2.5 Interpretation

If the calculated 

$$

-value falls outside the critical range (e.g., beyond 

$$

 for a 95% confidence level), reject 

$$

.

2.6 Example

A bearing manufacturer claims their steel balls have a mean diameter of 

. A sample of 

$$

 balls yields a mean of 

. Historically, the process standard deviation is 

. Test if the mean differs at 

.

$$

Interpretation: Since 

, we reject 

$$

. The mean diameter significantly differs from 

.

2.7 When MUST You Use Z-Test?

• Large sample (
  
  
  ) and population 
  

  $$
   is known (e.g., from historical process data or standards).
• In quality control when the process is stable and 
  

  $$
   is well-established from long-term data.
• Testing proportions (where 
  
  
   and 
  
  
  
  
  
  
  ) which are often approximated by 
  

  $$
  .

---

3. Student’s t-Test

Used to compare means when the population standard deviation (

$$

) is unknown and the sample size is relatively small.

3.1 One-Sample t-Test

Formula:

$$

Where 

$$

 is the sample standard deviation.

Assumptions:

• Normally distributed population.
• Unknown population standard deviation.

Degrees of Freedom:

$$

Effect Size:

$$

Example:
A new composite material has a target tensile strength (

$$

) of 

. A sample of 

$$

 batches gives 

 and 

.

$$

3.2 Independent Two-Sample t-Test (Pooled vs. Welch's)

Compares the means of two independent groups.

Formula (Pooled, assuming equal variances):

$$

Where 

Formula (Welch's, assuming unequal variances - The Default):

$$

Degrees of Freedom (Welch's):

$$

3.3 Paired Sample t-Test

Compares means from the same group at different times (e.g., before and after a treatment).

Formula:

$$

Where 

$$

 is the mean of the differences, 

$$

 is the standard deviation of the differences, and 

$$

 is usually 

$$

.

Degrees of Freedom:

$$

3.4 t-Test Engineering Context

The 

$$

-test is vital in manufacturing for checking if a supplier change, a new operator, or a new batch of raw materials causes a significant difference in product dimensions or properties.

---

4. F-Test (Variance)

Used to compare the variances of two independent populations or to evaluate the overall significance in regression models.

4.1 Formula

$$

(By convention, 

$$

 is typically the larger variance, making 

)

4.2 Assumptions

• Both populations are approximately normally distributed.
• Samples are independent.

4.3 Degrees of Freedom

• 
  
  
• 
  
  

4.4 Engineering Application

Used to test if two different machines or operators exhibit the same level of precision (consistency).

---

5. Chi-Square Test

Used for categorical data and evaluating frequency counts.

5.1 Goodness-of-Fit Test

Determines if a single categorical variable matches an expected theoretical distribution.

Formula:

$$

Where 

$$

 is the observed frequency and 

$$

 is the expected frequency.

5.2 Test of Independence

Determines if there is a significant association between two categorical variables.

Expected Frequency Formula:

$$

5.3 Assumptions

• Data must be randomly sampled counts.
• All individual expected frequencies (
  

  $$
  ) must be 
  

  $$
  .

5.4 Degrees of Freedom

• Goodness-of-Fit: 
  
  
  
   (where 
  

  $$
   is the number of categories)
• Independence: 
  
  
  
  
  
  
  
  
  
  
   (where 
  

  $$
   is rows, 
  

  $$
   is columns)

5.5 Yates' Continuity Correction

Applied when 

 (a 

 contingency table) to prevent overestimating the chi-square value.

Formula:

$$

---

6. One-Way ANOVA + Post-hoc

Used to determine whether there are any statistically significant differences between the means of three or more independent (unrelated) groups.

6.1 Formula (Sums of Squares)

$$

• Total Sum of Squares (
  

  $$
  ): 
  
  
  
  
  
  
• Treatment/Between Sum of Squares (
  

  $$
  ): 
  
  
  
  
  
  
• Error/Within Sum of Squares (
  

  $$
  ): 
  
  
  
  
  
  

6.2 Mean Squares

• 
  
• 
  

6.3 F-Statistic

$$

6.4 Degrees of Freedom

• Numerator (Between): 
  
  
  
• Denominator (Within): 
  
  
  

6.5 Assumptions

• Normality: Residuals are normally distributed.
• Independence: Observations are independent.
• Homogeneity of Variances (Homoscedasticity): Variances across groups are equal (often checked via Levene's Test).

6.6 Post-hoc Tests (If ANOVA is Significant)

ANOVA tells us that at least one group differs, but not which one. Post-hoc tests pinpoint the differences.

• Tukey's HSD: Controls the Type I error rate across all pairwise comparisons. Best for equal sample sizes.
• Bonferroni: Highly conservative; adjusts the 
  

  $$
   level directly (
  
  
  ).
• Games-Howell: Used when the assumption of equal variances is violated.

---

7. Design of Experiments (Basic Factorial)

Engineering statistics relies heavily on Factorial Designs to evaluate how multiple factors affect a process simultaneously.

7.1 

$$

 Factorial Design

This design evaluates 2 factors, each at 2 levels (Low and High, coded as 

$$

 and 

$$

).

Main Effects Calculation:

• 
  
  
• 
  
  

Interaction Effect Calculation:

• 
  

7.2 Sum of Squares for Effects

$$

Where 

$$

 is replicates, 

$$

 is the number of factors, and Contrast 

.

---

8. Parametric vs Non-Parametric

Parametric tests assume underlying statistical distributions (like the Normal distribution). When assumptions are severely violated, engineers switch to non-parametric tests, which make no assumptions about the underlying population distribution.

8.1 Parametric vs Non-Parametric Counterparts

Statistical Task	Parametric Test	Non-Parametric Equivalent
2 Independent Means	Independent  $$ -test	Mann-Whitney U test
2 Dependent Means	Paired  $$ -test	Wilcoxon Signed-Rank test
$$  Independent Means	One-Way ANOVA	Kruskal-Wallis test
Correlation	Pearson Correlation	Spearman Rank Correlation

8.2 Testing Flowchart

Start Analysis
 │
 ├──> Is data continuous?
 │     ├── NO  ──> Categorical (Use Chi-Square)
 │     └── YES ──> Continue
 │
 ├──> Are assumptions met (Normality, Homogeneity)?
       ├── NO  ──> Use Non-Parametric Equivalents
       └── YES ──> Use Parametric Tests

---

9. Test Selection Decision Tree & Matrix

Choosing the right statistical test depends on the type of data being analyzed and the number of groups being evaluated.

9.1 Test Selection Matrix

Data Type / Objective	1 Group	2 Independent Groups	2 Dependent Groups	3+ Independent Groups
Mean (Parametric, Normal)	One-Sample  $$ -test	Independent  $$ -test	Paired  $$ -test	One-Way ANOVA
Mean (Non-Parametric)	Wilcoxon Signed-Rank	Mann-Whitney U	Wilcoxon Signed-Rank	Kruskal-Wallis
Variance	Chi-Square Variance Test	$$ -Test		Bartlett's / Levene's
Proportion / Frequency	Chi-Square Goodness of Fit	Chi-Square Test of Independence	McNemar's Test	Chi-Square Test

---

10. Software Implementation

10.1 R

R

# One-Way ANOVA and Tukey's Test
model <- aov(response ~ factor_group, data = df)
summary(model)
TukeyHSD(model)

Use code with caution.

10.2 Python (SciPy & StatsModels)

python

import scipy.stats as stats
import statsmodels.api as sm
from statsmodels.formula.api import ols

# Independent t-test
t_stat, p_val = stats.ttest_ind(group1, group2)

# One-Way ANOVA
model = ols('response ~ C(group)', data=df).fit()
anova_table = sm.stats.anova_lm(model, typ=2)

Use code with caution.

10.3 Excel

• To run a t-test or ANOVA: Go to Data > Data Analysis and select "t-Test: Two-Sample Assuming Equal Variances" or "Anova: Single Factor".

10.4 Minitab

• To run DOE: Go to Stat > DOE > Factorial > Create Factorial Design, then analyze via Stat > DOE > Factorial > Analyze Factorial Design.

---

11. Viva Q&A Bank

Q1: What is the fundamental difference between a Z-test and a t-test?
Answer: The 

$$

-test is used when the population standard deviation (

$$

) is known, typically with large samples (

). The 

$$

-test is used when the population standard deviation is unknown and is estimated using the sample standard deviation (

$$

), which is more common with small samples.

Q2: What are the consequences of a Type I vs. a Type II error in an engineering process?
Answer: A Type I error (

$$

) occurs when we incorrectly reject a true null hypothesis (e.g., halting a compliant production line and causing unnecessary downtime). A Type II error (

$$

) occurs when we incorrectly fail to reject a false null hypothesis (e.g., allowing a defective batch of products to ship to customers).

Q3: Explain Degrees of Freedom (DF) in your own words.
Answer: Degrees of Freedom represent the number of independent values in a dataset that have the freedom to vary when estimating a statistical parameter. For example, if we have a sample of 

$$

 values that must sum to a known total, 

 values can be anything, but the last value is fixed to make the sum correct.

Q4: How do you verify the assumption of normality before running an ANOVA?
Answer: We analyze the residuals of the model. This can be done by generating a Normal Probability Plot of the residuals or by conducting a normality test such as the Anderson-Darling, Shapiro-Wilk, or Kolmogorov-Smirnov test.

Q5: What is a Post-hoc test, and why is it required after an ANOVA?
Answer: An ANOVA only indicates whether there is a statistically significant difference among three or more group means. It does not specify which groups differ from each other. Post-hoc tests (like Tukey's HSD) are designed to compare all possible group pairs while managing the cumulative risk of a Type I error.

Q6: What is the Central Limit Theorem (CLT) and why is it important?
Answer: The Central Limit Theorem states that if you have a sufficiently large sample size (

) with a finite variance, the sampling means of any independent, non-normally distributed population will approximate a normal distribution. This allows engineers to use parametric tests like the 

$$

-test even when the raw data is not normally distributed.

Q7: Why do we use Welch's t-test over the Student's t-test?
Answer: The Student's 

$$

-test assumes that the two independent populations have equal variances. If this assumption is violated, it increases the risk of false positives. Welch's 

$$

-test is a robust alternative that adjusts for unequal variances, protecting the validity of the test.

Q8: What is an interaction effect in a Designed Experiment (DOE)?
Answer: An interaction effect occurs when the effect of one independent variable on the response depends on the level of another independent variable. When an interaction is present, the main effects cannot be interpreted independently without misrepresenting the process.

Q9: How do you know when to use a non-parametric test?
Answer: Non-parametric tests are used when the data fails to meet parametric assumptions, such as when it is heavily skewed, measured on an ordinal scale, or when the sample size is too small to accurately assess normality

Here is the complete, production-ready document for Experiment 6: F-Test for Equality of Variances with Engineering Machine Precision Applications.

This document is meticulously designed for an M.Tech lab manual, featuring strict post-graduate engineering notation ($\text{K\text{a}\text{T\text{e}X}}$), distinct manual calculation workbooks, and programmatic verification.

---

Experiment 6: F-Test for Equality of Variances with Engineering Machine Precision Applications

6.1 Objective

To evaluate and compare the process precision, repeatability, and structural variability of two independent engineering populations using the Variance Ratio ($F$-test); to mathematically verify the prerequisite assumption of homoscedasticity for subsequent parametric testing; and to interpret statistical boundaries within manufacturing tolerances.

6.2 Theoretical Background & Engineering Application

In quality engineering and manufacturing automation, checking process mean targets is rarely enough. A machine tool can hit a dimensional target on average but still produce a high rate of scrap if its variance is out of control. The $F$-test evaluates whether the variances of two independent populations are equal ($\sigma_1^2 = \sigma_2^2$).

Engineers use the $F$-test for two main purposes:

1. Machine/Process Selection: Comparing the structural repeatability of an aging CNC lathe against a newly commissioned machining center to see if the new machine delivers a significant upgrade in precision.
2. Parametric Validation: Serving as a mathematical gatekeeper before running a standard independent two-sample $t$-test or an Analysis of Variance (ANOVA), both of which require equal variances (homoscedasticity).

---

6.3 Mathematical Formulations & Derivations

The $F$-test statistic is the direct ratio of two sample variances. By statistical convention, to keep the analysis clean, the larger sample variance is placed in the numerator. This sets up a right-tailed or upper-tailed critical boundary framework.

Test Statistic ($F_{calc}$):

$$F_{calc} = \frac{s_1^2}{s_2^2}$$

Where:

• $s_1^2$ is the sample variance of Group 1, calculated using Bessel's correction: $s_1^2 = \frac{\sum (x_{1i} - \bar{x}_1)^2}{n_1 - 1}$
• $s_2^2$ is the sample variance of Group 2, calculated using Bessel's correction: $s_2^2 = \frac{\sum (x_{2i} - \bar{x}_2)^2}{n_2 - 1}$
• Strict Mathematical Constraint: $s_1^2 \ge s_2^2$

Degrees of Freedom ($\nu_1, \nu_2$):

The sampling distribution of this variance ratio follows Snedecor's $F$-distribution, defined by two distinct degrees of freedom:

• Numerator Degrees of Freedom ($\nu_1$): $\nu_1 = n_1 - 1$
• Denominator Degrees of Freedom ($\nu_2$): $\nu_2 = n_2 - 1$

Two-Tailed Alpha Adjustment ($\alpha_{adj}$):

When testing the non-directional hypothesis $H_0: \sigma_1^2 = \sigma_2^2$ versus $H_1: \sigma_1^2 \neq \sigma_2^2$, forcing $s_1^2 \ge s_2^2$ means you are evaluating only the upper tail. To keep the test accurate at your target significance level ($\alpha$), you must compare $F_{calc}$ against the critical value evaluated at a sliced alpha level:
$$F_{crit} = F_{\left(\frac{\alpha}{2}, \, \nu_1, \, \nu_2\right)}$$

---

6.4 Core Assumptions & Diagnostic Testing

Before executing an $F$-test, the data must satisfy these critical prerequisites:

1. Strict Normality: The $F$-test is highly sensitive to departures from normality. If the underlying data distributions are skewed or have heavy tails, the Type I error rate inflates drastically. Normality must be confirmed via Shapiro-Wilk tests or Quantile-Quantile (Q-Q) plots.
2. Independence: Sample groups must be completely independent of one another. There can be no overlapping elements or cross-contamination between the data pipelines.
3. Continuous Metric: The data must be measured on a continuous interval or ratio scale (e.g., millimeters, Rockwell hardness numbers, surface roughness in microns).

Robust Alternative: If the normality check fails, the $F$-test should be discarded in favor of Levene's Test or the Brown-Forsythe Test, which evaluate variance equality using medians or trimmed means to remain robust against non-normal data.

---

6.5 Worked Engineering Example: CNC Spindle Runout Comparison

A reliability engineer is evaluating two multi-axis CNC milling machines to find out if a newly installed spindle (Machine B) has significantly better dimensional precision (lower variance) than an older spindle (Machine A).

• Machine A (Older Spindle): $n_1 = 11$ shafts measured, $s_1^2 = 24.5 \ \mu\text{m}^2$
• Machine B (New Spindle): $n_2 = 16$ shafts measured, $s_2^2 = 8.2 \ \mu\text{m}^2$
• Significance Level ($\alpha$): $0.05$ (Two-tailed evaluation)

Step-by-Step Manual Solution:

1. Formulate Hypotheses:
   • $H_0: \sigma_1^2 = \sigma_2^2$ (Both spindles operate with identical precision)
   • $H_1: \sigma_1^2 \neq \sigma_2^2$ (The spindles exhibit a true difference in precision)
2. Compute Test Statistic ($F_{calc}$):
   • Since $s_1^2 = 24.5$ is greater than $s_2^2 = 8.2$, Machine A acts as the numerator group.
     $$F_{calc} = \frac{24.5}{8.2} = 2.9878$$
3. Determine Degrees of Freedom:
   • Numerator degrees of freedom: $\nu_1 = n_1 - 1 = 11 - 1 = 10$
   • Denominator degrees of freedom: $\nu_2 = n_2 - 1 = 16 - 1 = 15$
4. Determine Critical Boundary Value:
   • Slicing alpha for a two-tailed test: $\frac{\alpha}{2} = \frac{0.05}{2} = 0.025$
   • Looking up the standard statistical F-table for $F_{(0.025, \, 10, \, 15)}$ yields: $F_{crit} = 3.06$
5. Statistical Decision Framework:
   • Compare values: $F_{calc} = 2.9878$ and $F_{crit} = 3.06$.
   • Because $F_{calc} = 2.9878 < 3.06$, the test statistic falls just short of the critical rejection zone.
   • Decision: Fail to reject the null hypothesis ($H_0$).
6. Engineering Interpretation:
   At a 95% confidence level, there is not enough evidence to prove that the new spindle has significantly better precision than the old one. The observed difference in sample variances can still be attributed to random sampling error. The engineer should maintain the assumption of equal variance if performing further multi-sample testing.

---

6.6 Data Sheets & Lab Exercise (To be filled by student)

Exercise Background

The table below records the tensile yield strength variations (MPa) of structural aluminum samples sourced from two automated extrusion production lines.

Sample ID	Line 1 Yield Strength ($X_1$)	$(X_1 - \bar{X}_1)^2$	Line 2 Yield Strength ($X_2$)	$(X_2 - \bar{X}_2)^2$
S01	312.4		305.2
S02	318.6		309.4
S03	308.2		307.1
S04	322.1		304.8
S05	315.7		308.5
S06	325.4		306.2
S07	310.9		—	—
S08	319.3		—	—

---

6.7 Step-by-Step Calculation Workbook

Step 1: Hypothesis Formulation

• $H_0$: $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• $H_1$: $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$

Step 2: Compute Group Sample Means

• Line 1 Sample Size ($n_1$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$ ; Mean ($\bar{X}_1$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• Line 2 Sample Size ($n_2$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$ ; Mean ($\bar{X}_2$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$

Step 3: Compute Sample Variances

• Line 1 Sample Variance ($s_1^2 = \frac{\sum(X_{1i}-\bar{X}_1)^2}{n_1-1}$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• Line 2 Sample Variance ($s_2^2 = \frac{\sum(X_{2i}-\bar{X}_2)^2}{n_2-1}$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$

Step 4: Calculate the Variance Ratio ($F_{calc}$)

• Assign the larger variance to the numerator: $s_{max}^2 =$ $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$ ; $s_{min}^2 =$ $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• $F_{calc} = \frac{s_{max}^2}{s_{min}^2} =$ $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$

Step 5: Critical Value Extraction & Final Decision

• Numerator df ($\nu_{num}$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$ ; Denominator df ($\nu_{den}$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• Target Significance ($\alpha$) = $0.05 \rightarrow$ Sliced Value Matrix ($F_{(0.025, \, \nu_{num}, \, \nu_{den})}$) = $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$
• Statistical Decision: Reject / Fail to Reject $H_0$ because: $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$

---

6.8 Software Verification Guide (Python Syntax)

Run this script to verify your hand-calculated variance components and test statistics:

import numpy as np
import scipy.stats as stats

# Input data sheets from aluminum extrusion lines
line1 = np.array([312.4, 318.6, 308.2, 322.1, 315.7, 325.4, 310.9, 319.3])
line2 = np.array([305.2, 309.4, 307.1, 304.8, 308.5, 306.2])

# Compute raw sample variances
var1 = np.var(line1, ddof=1)
var2 = np.var(line2, ddof=1)

# Format structural F-ratio
f_calc = var1 / var2 if var1 >= var2 else var2 / var1
df_num = len(line1) - 1 if var1 >= var2 else len(line2) - 1
df_den = len(line2) - 1 if var1 >= var2 else len(line1) - 1

# Extract p-value (multiply by 2 for a two-tailed test)
p_value = 2 * (1 - stats.f.cdf(f_calc, df_num, df_den))

print(f"Line 1 Variance: {var1:.4f} | Line 2 Variance: {var2:.4f}")
print(f"Calculated F-Statistic: {f_calc:.4f}")
print(f"Degrees of Freedom: ({df_num}, {df_den})")
print(f"Two-tailed p-value: {p_value:.4f}")

---

6.9 Lab Evaluation & Deliverables

Results and Discussion Field

(Detail the variance behavior of the two production lines, confirm if they meet the requirements for further parametric tests, and discuss how process variability affects structural consistency).
$$\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$$
$$\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$$

Signature of Lab Evaluator: $\text{\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_}$ Date: $\text{\_\_\_\_\_\_\_\_\_\_\_}$

---

6.10 Viva Target Questions for Experiment 6

1. Why is the $F$-test highly vulnerable to variations in data normality, and what occurs if normality is violated?
   • Model Answer: The mathematical derivation of the $F$-distribution relies directly on the ratios of independent Chi-Square variables, which are themselves sums of squared standard normal distributions. If the data is even slightly non-normal, the tail areas change dramatically. This distorts the true Type I error rate ($\alpha$), rendering the table's critical boundaries unreliable.
2. How do you perform a directional (one-tailed) $F$-test versus a non-directional (two-tailed) $F$-test?
   • Model Answer: For a one-tailed test, your alternative hypothesis targets a specific direction (e.g., $H_1: \sigma_1^2 > \sigma_2^2$), and you look up the critical value using the full value of $\alpha$ (e.g., $F_{(\alpha, \, \nu_1, \, \nu_2)}$). For a two-tailed test ($H_1: \sigma_1^2 \neq \sigma_2^2$), we still place the larger variance on top to look only at the upper tail, but we must use a split alpha value ($F_{\left(\frac{\alpha}{2}, \, \nu_1, \, \nu_2\right)}$) to account for both sides of the distribution.
3. What does an $F$-statistic value exactly equal to $1.0$ indicate?
   • Model Answer: An $F$-statistic of exactly $1.0$ shows that the two sample variances are perfectly identical ($s_1^2 = s_2^2$). In this scenario, the test statistic falls dead center in the acceptance region of the null hypothesis, meaning there is zero evidence of any difference in variance between the two populations.

---

To advance your master notes or complete your lab manual compilation, please indicate your next priority. Either request critical details to proceed with Experiment 7 (Chi-Square Tests) or Experiment 8 (One-Way ANOVA), or tell me how you would like to structure the next section.

Oops, something went wrong.