Review & Practical Applications of Business Statistics

TAKEAWAYS:

• Integrated Analysis: Combines all statistical tools (mean/median, standard deviation, probability, regression, hypothesis testing) to solve complex business problems.
• Real-World Mapping: Shows how eSewa (fraud detection via standard deviation), Daraz (demand forecasting via time series), and Ncell (customer segmentation via ANOVA) apply statistical methods.
• Decision-Making Framework: Teaches how to choose the right statistical test (e.g., t-test vs. ANOVA) based on data type and business question.
• Error Avoidance: Highlights common pitfalls (e.g., misusing correlation as causation, ignoring sample size in hypothesis tests).
• Exam Strategy: Focuses on applied questions (e.g., "Calculate the optimal inventory level for Daraz using mean and standard deviation") over rote memorization.

1. Synthesis of Statistical Concepts: A Unified Framework

Business statistics is not a collection of isolated tools but a cohesive system where each concept builds on others. Below is a decision tree for selecting the right statistical method based on the business problem and data type:

flowchart TD
    A["Business Problem"] --> B["Descriptive?\n(Summarize data)"]
    A --> C["Inferential?\n(Test hypotheses)"]
    A --> D["Predictive?\n(Forecast trends)"]

    B --> E["Central Tendency\n(Mean/Median/Mode)"]
    B --> F["Dispersion\n(Standard Deviation/Variance)"]
    B --> G["Distribution\n(Normal/Skewed)"]

    C --> H["One Sample?\n(*t*-test/Z-test)"]
    C --> I["Two Samples?\n(*t*-test/ANOVA)"]
    C --> J["Categorical Data?\n(Chi-Square)"]

    D --> K["Linear Relationship?\n(Regression)"]
    D --> L["Time-Based Trends?\n(Time Series)"]

Key Insight:

• Descriptive statistics answer "What is happening?" (e.g., average customer spending at Khalti).
• Inferential statistics answer "Why is it happening?" (e.g., testing if Ncell’s new tariff increases churn rate).
• Predictive statistics answer "What will happen?" (e.g., forecasting NEPSE stock trends using moving averages).

2. Practical Applications: How Companies Use Statistics

A. eSewa: Fraud Detection via Standard Deviation

Real-World Example: eSewa uses standard deviation to flag unusual transactions. If a user’s typical monthly spending is ₹5,000 (±₹1,000), a sudden ₹20,000 payment triggers a fraud alert because it lies 3 standard deviations above the mean.

Worked Example: Calculating Fraud Threshold Given:

• Mean transaction (μ) = ₹5,000
• Standard deviation (σ) = ₹1,000
• Threshold = μ + 3σ

Calculation: Threshold = 5,000 + 3(1,000) = ₹8,000 Any transaction >₹8,000 is flagged for review.

Why It Works:

• Outlier detection relies on dispersion (σ) to identify anomalies.
• Advantage: Reduces false positives by setting dynamic thresholds.
• Disadvantage: Requires clean data; noisy data (e.g., one-time large purchases) may trigger false alarms.

B. Daraz: Demand Forecasting with Time Series

Real-World Example: Daraz uses moving averages and seasonal decomposition to predict product demand. For example, air conditioner sales spike in June (pre-monsoon) and October (post-monsoon).

Worked Example: 3-Month Moving Average for Inventory Given monthly sales (units) for ACs:

Step 1: Calculate 3-month moving average (MA₃):

• MA₃(Jun) = (Jan + Feb + Mar)/3 = (500 + 480 + 600)/3 = 526.67
• MA₃(Jul) = (Feb + Mar + Apr)/3 = (480 + 600 + 550)/3 = 543.33

Step 2: Forecast July sales using MA₃(Jun) = 526.67 units. Business Impact:

• Daraz stocks 530 units in July to avoid stockouts.
• Advantage: Smooths short-term fluctuations.
• Disadvantage: Lags behind sudden trends (e.g., viral products).

C. Ncell: Customer Segmentation with ANOVA

Real-World Example: Ncell uses Analysis of Variance (ANOVA) to compare call duration across customer segments (prepaid vs. postpaid vs. corporate).

Worked Example: ANOVA Table for Call Duration Given:

• Prepaid: Mean = 180 sec, n = 50
• Postpaid: Mean = 240 sec, n = 40
• Corporate: Mean = 300 sec, n = 30
• Total Mean (μ) = 220 sec

Step 1: Calculate Between-group variance (SSB) and Within-group variance (SSW). Assume:

• Sum of Squares Between (SSB) = 10,000
• Sum of Squares Within (SSW) = 15,000
• Degrees of freedom:
  • Between (df₁) = 3 – 1 = 2
  • Within (df₂) = 120 – 3 = 117

ANOVA Table:

Step 2: Compare F-critical (F₀.₀₅,₂,₁₁₇) ≈ 3.05. Since 38.73 > 3.05, reject H₀: Call duration differs significantly across segments.

Business Impact:

• Ncell offers discounted data to prepaid users to increase call duration.
• Advantage: Identifies high-value segments.
• Disadvantage: Requires large sample sizes for accuracy.

3. Comparative Analysis: When to Use Which Tool

4. Common Pitfalls and How to Avoid Them

A. Misusing Correlation as Causation

Example:

• Claim: "Ice cream sales rise with drowning incidents → Ice cream causes drowning."
• Reality: Both are correlated with temperature (a confounding variable).

Solution:

• Use regression analysis to control for confounders.
• Rule of Thumb: Correlation (r) does not imply causation unless:
  1. Theory supports the link.
  2. Experimentation isolates variables.

B. Ignoring Sample Size in Hypothesis Testing

Example:

• Testing if Khalti’s new UI reduces login time with n = 5 users.
• Problem: Small n → high variance → unreliable p-values.

Solution:

• Use central limit theorem: n ≥ 30 ensures normal distribution of sample means.
• Power Analysis: Calculate required n to detect effect size (e.g., 80% power, α = 0.05).

C. Overfitting in Regression

Example:

• Fitting a 10th-degree polynomial to NEPSE stock data to "explain" past trends.
• Problem: Model fits noise, not signal → poor predictions.

Solution:

• Use adjusted R² (penalizes extra predictors).
• Cross-validation: Split data into training/test sets.

5. Integrated Case Study: Optimizing Daraz’s Warehouse Inventory

Scenario: Daraz wants to minimize stockouts and overstock for a product with:

• Demand (X): Normally distributed, μ = 1,000 units/month, σ = 200.
• Lead time: 2 weeks (1/4 month).
• Costs:
  • Holding cost = ₹50/unit/month.
  • Stockout cost = ₹200/unit.

Step 1: Calculate Safety Stock (SS) Use Z-score for 95% service level (Z = 1.645): SS = Z × σ × √L = 1.645 × 200 × √(1/4) = 164.5 units

Step 2: Optimal Order Quantity (Q) Use Economic Order Quantity (EOQ) model: Q = √(2DS / H) Where:

• D = Demand = 1,000 units
• S = Ordering cost = ₹500/order
• H = Holding cost = ₹50/unit

Q = √(2 × 1,000 × 500 / 50) = √20,000 = 141.42 → 141 units

Step 3: Reorder Point (ROP) ROP = (Demand × Lead time) + SS = (1,000 × 1/4) + 164.5 = 414.5 units

Business Impact:

• Reduces holding costs by 30% vs. ordering 1,000 units at once.
• Minimizes stockouts with 95% confidence.

6. Exam Tip: How to Score Full Marks

A. Question Patterns

1. Applied Problems (60% weight):

   • "Calculate the optimal inventory level for Daraz using mean and standard deviation."
   • Solution Structure:
     • State assumptions (e.g., normal demand).
     • Show all formulas (e.g., SS = Zσ√L).
     • Draw a demand curve with μ and σ.
     • Interpret results (e.g., "Order 141 units every month").

2. Interpretation (20% weight):

   • "Why might ANOVA not be suitable for comparing customer satisfaction scores across 5 regions?"
   • Key Points:
     • Assumes normality and homogeneity of variance.
     • Small sample sizes violate assumptions.
     • Alternative: Kruskal-Wallis test (non-parametric).

3. Critical Thinking (20% weight):

   • "A company claims their new product increases sales by 20%. How would you verify this using statistics?"
   • Expected Answer:
     • Hypothesis: H₀: μ₁ = μ₂ (no difference).
     • Test: Two-sample t-test (if normal) or Mann-Whitney U (if skewed).
     • Check: p-value < 0.05 → reject H₀.
     • Caution: Control for confounders (e.g., marketing spend).

B. Marking Scheme Example

C. Pro Tips

• Always draw diagrams for:
  • Normal distributions (label μ, σ, and critical values).
  • ANOVA tables (show degrees of freedom).
  • Time series (plot actual vs. forecasted).
• Use real numbers from examples (e.g., "Ncell’s data shows...").
• Justify your choice of test (e.g., "ANOVA is used because we have 3+ groups and continuous data.").

7. Summary: The Big Picture

Business statistics is about translating data into decisions. Here’s how the tools connect:

mindmap
  root((Business Statistics))
    Descriptive
      Central Tendency
      Dispersion
    Inferential
      Hypothesis Testing
        *t*-test
        ANOVA
        Chi-Square
    Predictive
      Regression
      Time Series
    Applications
      Fraud Detection (eSewa)
      Inventory (Daraz)
      Customer Segmentation (Ncell)

Final Takeaway:

• Descriptive stats tell you what’s happening.
• Inferential stats tell you why it’s happening.
• Predictive stats tell you what will happen. Master all three, and you’ll ace the exam—and impress employers.