Hypothesis Testing: Types, Steps & Statistical Analysis

Key Concepts & Definitions

1. What is a Hypothesis?

A hypothesis is an educated guess or tentative explanation for a research problem, derived from theory or prior research. It must be:

• Testable (can be verified or disproved).
• Specific (clearly defined).
• Logical (based on evidence).

Example in Tourism: "Increasing online reviews on TripAdvisor by 20% will boost hotel bookings in Kathmandu by 15% within six months." This hypothesis can be tested using data from booking platforms.

2. Types of Hypotheses

Hypotheses are classified based on their nature and purpose:

Steps in Hypothesis Testing

Hypothesis testing follows a structured process:

Step-by-Step Worked Example: NEPSE Stock Performance

Research Question: "Does the NEPSE index increase significantly during the Dashain festival season compared to other months?"

1. State Hypotheses:

   • H₀: The NEPSE index does not increase during Dashain (μ = 0).
   • H₁: The NEPSE index increases during Dashain (μ > 0).

2. Choose Significance Level (α):

   • Commonly α = 0.05 (5% chance of rejecting H₀ when it’s true).

3. Select Test Statistic:

   • Use a one-sample t-test (since we compare a sample mean to a known value).

4. Collect Data:

   • Gather NEPSE index values for 5 years of Dashain months (October–November) and 5 years of non-Dashain months.

5. Calculate Test Statistic:

   • Compute the mean increase during Dashain: Mean₁ = 1200 points.
   • Compute the mean increase in other months: Mean₂ = 800 points.
   • Use the formula for a two-sample t-test: $$t=\frac{{\bar{X}}_1-{\bar{X}}_2}{\sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}}}$$ (Assume standard deviations $s_1=150$, $s_2=120$, $n_1=n_2=5$). $$t=\frac{1200-800}{\sqrt{\frac{150^2}{5}+\frac{120^2}{5}}}=\frac{400}{\sqrt{4500+2880}}=\frac{400}{96.95}\approx 4.13$$

6. Compare p-value with α:

   • For df = 8 (n₁ + n₂ – 2), the critical t-value at α = 0.05 (one-tailed) is 1.86.
   • Since 4.13 > 1.86, we reject H₀.

7. Decision:

   • Reject H₀: There is statistically significant evidence that the NEPSE index increases during Dashain.

8. Interpretation:

   • "The Dashain festival season has a positive impact on NEPSE stock performance, suggesting investors should consider timing their trades around this period."

Types of Errors in Hypothesis Testing

No test is perfect—errors can occur:

Trade-off:

• Reducing α (e.g., from 0.05 to 0.01) lowers Type I errors but increases Type II errors.
• Power of a test = 1 – β (ability to detect a true effect).

Common Statistical Tests for Hypothesis Testing

Choose the right test based on data type and research question:

In the Real World

1. eSewa & Digital Payment Adoption (Nepal)

Idea Used: Hypothesis Testing for Market Impact

• Research Question: "Does the introduction of eSewa’s ‘eSewa Pay’ feature increase online transactions by 30% in a year?"
• Hypothesis Testing Applied:
  • H₀: No significant increase in transactions (μ ≤ 5%).
  • H₁: Significant increase (μ > 30%).
• Method: Compare transaction volumes before and after the feature’s launch using a paired t-test.
• Result: eSewa’s data showed a 42% increase, leading to expanded digital payment services.

2. Pathao’s Ride-Hailing Demand (Nepal)

Idea Used: ANOVA for Regional Differences

• Research Question: "Does ride demand vary significantly across Kathmandu, Pokhara, and Chitwan?"
• Hypothesis Testing Applied:
  • H₀: No difference in demand across regions.
  • H₁: Demand differs by region.
• Method: One-way ANOVA on daily ride requests.
• Result: Kathmandu had 2.5x higher demand than Chitwan, guiding Pathao’s fleet allocation.

3. NTC’s Internet Speed Complaints (Nepal)

Idea Used: Chi-Square Test for Customer Satisfaction

• Research Question: "Is there a relationship between internet speed complaints and time of day?"
• Hypothesis Testing Applied:
  • H₀: Complaints are uniformly distributed.
  • H₁: Complaints peak during evening hours (5–9 PM).
• Method: Chi-Square goodness-of-fit test on complaint logs.
• Result: 70% of complaints came between 5–9 PM, leading NTC to optimize server loads during peak hours.

Visualizing Hypothesis Testing: The Decision Rule

flowchart TD
    A["p-value ≤ α"] -->|"Reject H₀"| B["Significant Result"]
    A -->|"p-value > α"| C["Fail to Reject H₀"]
    C --> D["No Significant Evidence"]

Key Takeaway:

• If p ≤ 0.05, reject H₀ (evidence supports H₁).
• If p > 0.05, fail to reject H₀ (insufficient evidence).

Worked Example: Kathmandu Traffic Congestion Study

Research Question: "Does the introduction of a new bus rapid transit system (BRTS) reduce traffic congestion in Kathmandu by 20%?"

Step 1: Define Hypotheses

• H₀: BRTS does not reduce congestion (μ ≤ 0%).
• H₁: BRTS reduces congestion by ≥20% (μ ≥ 20%).

Step 2: Data Collection

• Measure average travel time before and after BRTS implementation:
  • Before BRTS: Mean travel time = 45 minutes.
  • After BRTS: Mean travel time = 36 minutes.

Step 3: Calculate Test Statistic (Paired t-test)

$$t=\frac{{\bar{X}}_{\text{before}}-{\bar{X}}_{\text{after}}}{s_{\text{diff}}/\sqrt{n}}$$ Where:

• ${\bar{X}}_{\text{diff}}=45-36=9$ minutes.
• $s_{\text{diff}}=5$ minutes (standard deviation of differences).
• $n=30$ (sample size).

$$t=\frac{9}{5/\sqrt{30}}=\frac{9}{0.91}\approx 9.89$$

Step 4: Compare with Critical Value

• df = 29, α = 0.05 (one-tailed).
• Critical t-value ≈ 1.699.
• Since 9.89 > 1.699, reject H₀.

Conclusion:

"The BRTS system significantly reduced traffic congestion in Kathmandu by 20%, supporting its expansion."

Exam Tip

How to Score Full Marks in Hypothesis Testing Questions

1. Always state H₀ and H₁ clearly (1 mark each).
2. Justify your test choice (e.g., "t-test for two means" or "ANOVA for three groups").
3. Show calculations step-by-step (even if simplified).
4. Interpret results in context (e.g., "Reject H₀ means the new marketing strategy works").
5. Discuss limitations (e.g., small sample size, non-random sampling).

Common Mistakes to Avoid:

• Confusing Type I and Type II errors.
• Forgetting to mention significance level (α).
• Misinterpreting p-value (e.g., saying "p = 0.05 means 5% chance H₁ is true"—wrong! It means 5% chance of a false positive).

Final Checklist for Hypothesis Testing