Review & Practical Applications: Statistics in Business

TAKEAWAYS:

• Synthesize concepts: Combine measures of central tendency, dispersion, and probability to analyze business data holistically.
• Model selection: Choose the right statistical tool (e.g., regression vs. ANOVA) based on data type and research question.
• Real-world applications: Apply statistical techniques to financial forecasting, quality control, and market research.
• Interpret results: Translate statistical outputs (e.g., p-values, R²) into actionable business insights.
• Critical evaluation: Assess the limitations of statistical methods (e.g., assumptions of normality, sample size constraints).
• Exam focus: Expect integrated problems requiring multi-step reasoning across units (e.g., testing hypotheses using sample data).

1. Integration of Statistical Concepts

This unit consolidates all prior topics into cohesive frameworks for solving business problems. Below is a roadmap of how concepts interconnect:

graph TD
    A["Data Collection"] --> B["Descriptive Stats"]
    B --> C["Measures of Central Tendency"]
    B --> D["Measures of Dispersion"]
    C --> E["Probability Theory"]
    D --> E
    E --> F["Inference: Hypothesis Testing"]
    F --> G["Regression/Correlation"]
    G --> H["Time Series & Index Numbers"]
    H --> I["ANOVA/Chi-Square"]
    I --> J["Decision-Making"]

Key Connections

• Descriptive → Inferential: Use mean/median (Unit 2) to summarize data, then apply probability (Unit 4) to make inferences (Unit 9).
• Dispersion → Normality: Standard deviation (Unit 3) informs whether to use parametric tests (e.g., t-tests) or non-parametric alternatives.
• Correlation → Causation: Regression (Unit 7) helps distinguish correlation (e.g., ice cream sales vs. temperature) from causation (e.g., marketing spend vs. revenue).

2. Practical Applications in Business

A. Financial Forecasting (Time Series & Index Numbers)

Example: Nepal Rastra Bank (NRB) inflation reports

• Tool: Moving averages (smoothing time series data) and Consumer Price Index (CPI) calculations.
• How it works:
  1. Collect monthly CPI data (e.g., 2020–2023).
  2. Compute weighted index numbers to adjust for base-year changes.
  3. Use trend analysis (linear regression) to predict future inflation.
• Visual:
• Worked Example: Calculate the inflation rate for Q2 2023 if the CPI in Q1 2023 = 145 and Q2 2023 = 150. Solution: $$\text{Inflation Rate}=\left(\frac{150-145}{145}\right)\times 100=3.45\%$$

B. Quality Control (ANOVA & Chi-Square)

Example: Daraz customer satisfaction surveys

• Tool: One-way ANOVA to compare mean ratings across product categories (e.g., electronics vs. groceries).
• How it works:
  1. Collect survey ratings (1–5) from 300 customers per category.
  2. Compute F-statistic to test if category means differ significantly.
  3. Use Chi-Square to check if customer complaints follow expected distributions (e.g., delivery delays vs. product defects).
• Visual:
• Worked Example: Test if the mean rating for electronics (${\bar{x}}_1=4.1$) differs from groceries (${\bar{x}}_2=3.2$) at $\alpha =0.05$. Steps:
  1. State hypotheses: $H_0:{\mu }_1={\mu }_2$ vs. $H_1:{\mu }_1\ne {\mu }_2$.
  2. Compute pooled variance and F-statistic (assume $s_p^2=1.2$, $n_1=n_2=300$): $$F=\frac{({\bar{x}}_1-{\bar{x}}_2{)}^2}{2s_p^2/n}=\frac{(4.1-3.2{)}^2}{2\times 1.2/300}=112.5$$
  3. Compare to critical F-value ($F_{0.05,2,598}\approx 3.85$). Since $112.5>3.85$, reject $H_0$.

C. Market Research (Correlation & Regression)

Example: Pathao driver earnings vs. ride demand

• Tool: Linear regression to model earnings ($Y$) as a function of rides taken ($X$).
• How it works:
  1. Collect data: 50 drivers’ monthly earnings (₹50,000–₹200,000) and rides (500–2000).
  2. Fit $Y={\beta }_0+{\beta }_{1}X+ϵ$.
  3. Interpret $R^2$ (e.g., 75% of earnings variance explained by rides).
• Visual:
• Worked Example: Given ${\beta }_0=15,000$, ${\beta }_1=70$, predict earnings for 1,000 rides. Solution: $$Y=15,000+70\times 1,000=₹85,000$$

D. Hypothesis Testing in Business Decisions

Example: Ncell customer churn prediction

• Tool: Chi-Square test to compare observed vs. expected churn rates by demographic.
• How it works:
  1. Categorize churners by age (18–30, 31–45, 46+).
  2. Test if churn rates differ from expected (e.g., 30% uniform).
• Visual:
• Worked Example: Test if churn rates differ from expected (${\chi }^2$ critical value = 7.815 at $\alpha =0.05$). Steps:
  1. Compute ${\chi }^2$: $${\chi }^2=\sum \frac{(O-E{)}^2}{E}=\frac{(45-30{)}^2}{30}+\frac{(25-30{)}^2}{30}+\frac{(10-30{)}^2}{30}=20$$
  2. Since $20>7.815$, reject $H_0$: churn rates are not uniform.

3. Choosing the Right Statistical Tool

Use this decision tree to select methods based on research questions:

flowchart TD
    A["Research Question"] --> B{"Is data quantitative or categorical?"}
    B -->|"Quantitative"| C{"Compare groups?"}
    C -->|"Yes"| D{"One or two groups?"}
    D -->|"One group"| E["ANOVA"]
    D -->|"Two groups"| F["t-test"]
    C -->|"No"| G{"Relationship?"}
    G -->|"Yes"| H["Regression"]
    G -->|"No"| I["Descriptive Stats"]
    B -->|"Categorical"| J{"Test independence?"}
    J -->|"Yes"| K["Chi-Square"]
    J -->|"No"| L["Proportions Test"]

Comparison Table: Key Methods

4. Common Pitfalls and Limitations

• Overfitting: A regression model with too many predictors may fit noise (e.g., predicting stock prices using 20 variables).
• Non-normality: Violating normality assumptions invalidates t-tests/ANOVA (use Mann-Whitney U or Kruskal-Wallis instead).
• Correlation ≠ Causation: Example: Ice cream sales and drowning incidents both rise in summer, but neither causes the other.
• Small Sample Bias: Hypothesis tests lose power with $n<30$ (e.g., testing a new product with only 20 users).

Visual:

5. Worked Example: Integrated Problem

Scenario: Khalti wants to test if a new UI reduces transaction time.

1. Data: 60 users tested; mean time = 45 sec, $s=10$ sec.
2. Hypothesis: $H_0:\mu \ge 50$ sec vs. $H_1:\mu <50$ sec ($\alpha =0.05$).
3. Test: One-sample t-test.
4. Calculation: $$t=\frac{45-50}{10/\sqrt{60}}=-2.45$$ Critical $t_{0.05,59}=-1.67$. Since $-2.45<-1.67$, reject $H_0$.
5. Conclusion: The new UI significantly reduces transaction time.

Visual:

6. Exam Tips

• Integrated Problems: Expect questions combining multiple units (e.g., "Given a sample, compute mean, standard deviation, and test if it differs from a population mean").
• Interpretation: Always explain what results mean in plain language (e.g., "The p-value of 0.02 means there’s a 2% chance the observed difference is due to randomness").
• Assumptions: State assumptions explicitly (e.g., "We assume the data is normally distributed for the t-test").
• Calculations: Show all steps, even if partial credit is given. Use the formula sheet provided in exams.
• Real-World Tie-Ins: Relate answers to business contexts (e.g., "This ANOVA result suggests Pathao should invest in driver training for the low-rated age group").

Common Exam Questions:

1. "A bank’s loan default rates are 5% for urban and 8% for rural areas. Test if the difference is significant."
2. "Given a time series of NEPSE stock prices, forecast the next quarter’s value using a moving average."
3. "A factory’s production data shows 90% pass rate. After a machine upgrade, 95 out of 100 samples pass. Test if the upgrade improved quality."