Research Design & Variables: Types, Models & Applications

TAKEAWAYS:

• Research design determines how you answer a research question—choose exploratory (qualitative), descriptive (quantitative), or causal (experimental) based on your goal.
• Variables are the building blocks of research: independent variables (IVs) drive change, dependent variables (DVs) measure outcomes, and extraneous variables must be controlled.
• Mixed-methods design combines qualitative and quantitative approaches (e.g., Nabil Bank’s customer interviews + survey data).
• Field experiments (e.g., Pathao’s ride-sharing pricing tests) vs. laboratory experiments (e.g., NTC’s call-center script trials) differ in realism and control.
• Moderating variables (e.g., age affecting WhatsApp usage) and mediating variables (e.g., trust mediating eSewa adoption) explain why relationships exist.
• Pilot testing (e.g., Daraz’s beta launch for new product categories) ensures your design works before full-scale data collection.

1. What Is Research Design?

Research design is the blueprint for your study—it outlines:

• Purpose: Exploratory (discover ideas), descriptive (describe phenomena), or causal (test cause-effect).
• Approach: Qualitative (words, themes), quantitative (numbers, statistics), or mixed.
• Strategy: Survey, experiment, case study, or ethnography.
• Timeframe: Cross-sectional (one-time) or longitudinal (over time).

Why Does Design Matter?

A poor design leads to invalid conclusions. For example:

• If Ncell designs a survey without random sampling, its "customer satisfaction" results may only reflect urban users, ignoring rural areas.
• If a bank (e.g., Global IME) tests a new loan approval algorithm only on existing customers, it may miss how first-time borrowers behave.

2. Types of Research Design

Use this table to match your research question to the right design:

3. Variables: The Heart of Research

Variables are measurable traits that change or are manipulated in a study. Classify them correctly to avoid flawed conclusions.

A. Independent vs. Dependent Variables

graph LR
    A["Independent Variable (IV)"] -->|"Affects"| B["Dependent Variable (DV)"]
    A -->|"Example: Marketing spend (IV)"| C["Sales growth (DV)"]
    A -->|"Example: Training hours (IV)"| D["Employee productivity (DV)"]

• Independent Variable (IV): The cause (e.g., Daraz’s discount percentage).
• Dependent Variable (DV): The effect (e.g., order volume).
• Example: If NEPSE studies whether "investor confidence" (IV) affects stock prices (DV), it must measure both accurately.

B. Extraneous Variables: The Silent Saboteurs

These unwanted variables can distort your results. Control them or randomize them:

• Confounding Variables: Overlap with IV/DV (e.g., in a study on "social media use and stress," age could be a confounder).
• Moderating Variables: Affect the IV-DV relationship (e.g., "gender moderates the impact of ads on purchasing").
• Mediating Variables: Explain how IV affects DV (e.g., "trust mediates the relationship between eSewa ads and usage").

Real-World Example: Ncell’s "Does 4G speed increase customer retention?" study must control for:

• Extraneous: Competitor promotions (confounding).
• Moderator: Urban vs. rural users (different needs).
• Mediator: Customer service quality (why faster speed might not always help).

4. Research Design in Action: Case Studies

Case 1: Daraz’s A/B Testing for Promotions

Problem: Daraz wants to know if "limited-time discounts" (IV) increase sales (DV) more than "free shipping" (IV). Design: Causal (Experimental)

• Method: Randomly assign users to two groups:
  • Group A: See "20% off, ends soon" banner.
  • Group B: See "Free shipping on orders above Rs. 2000."
• Control: Same product, same time, same user demographics.
• Result: Group A had 30% higher conversion—Daraz scaled this strategy.

Visual:

Case 2: Nabil Bank’s Loan Default Prediction

Problem: Predict which borrowers will default (DV) based on income (IV), credit score (IV), and loan amount (IV). Design: Descriptive + Predictive Modeling

• Method: Survey 1000 borrowers, collect data on IVs, use regression analysis.
• Extraneous Variables Controlled: Economic downturns (external shocks), bank policies (internal).
• Finding: Borrowers with <Rs. 50K income and >Rs. 200K loans had 40% default rate—Nabil adjusted loan limits.

5. Qualitative vs. Quantitative Designs

When to Mix Both?

• Example: Himalayan Java wanted to know why customers switched to Starbucks. They:
  1. Qualitative: Interviewed 20 customers (found themes like "lack of loyalty program").
  2. Quantitative: Surveyed 500 customers (quantified "45% cite price as reason").

6. Ethical Considerations in Design

• Informed Consent: Ncell must tell users their data is used for research (e.g., call-duration studies).
• Anonymity: Daraz’s user surveys must not link responses to individual accounts.
• Deception: Avoid in experiments (e.g., don’t fake "product shortages" to test panic buying).
• Conflict of Interest: Nabil Bank’s research on loan defaults shouldn’t favor its own products over competitors’.

7. Common Pitfalls and How to Avoid Them

In the Real World

1. eSewa’s Fraud Detection System

   • Design: Causal (experimental)
   • IV: New fraud-alert algorithm vs. old system.
   • DV: Number of fraudulent transactions.
   • Result: Reduced fraud by 25%—now used by all fintech apps in Nepal.

2. NTC’s Network Optimization

   • Design: Mixed methods
   • Qualitative: Interviews with engineers to identify bottlenecks.
   • Quantitative: Data on call drops in different regions.
   • Outcome: Targeted tower upgrades in high-drop zones (e.g., Bhaktapur).

3. Pathao’s Surge Pricing

   • Design: Field experiment
   • IV: Price increase during peak hours.
   • DV: Rider demand and driver availability.
   • Finding: Surge pricing reduced wait times by 40%—now standard in ride-hailing.

Exam Tip: How to Score Full Marks

1. Define Clearly: Always start with precise definitions (e.g., "Research design is a framework that structures the methods and techniques used to collect, analyze, and interpret data...").
2. Use Diagrams: Draw flowcharts for processes (e.g., experimental design steps) or tables for comparisons (e.g., qualitative vs. quantitative).
3. Link to Nepal: Examiners love local examples. For variables:
   • IV: "Nepal Rastra Bank’s repo rate changes."
   • DV: "Inflation rate."
   • Extraneous: "Global oil prices."
4. Critique Designs: If asked to evaluate a study (e.g., "How valid is Daraz’s survey on customer satisfaction?"), discuss:
   • Sampling bias (e.g., only urban users).
   • Measurement issues (e.g., vague questions like "How satisfied are you?").
   • Ethical concerns (e.g., lack of consent).
5. Practical Application: End answers with a real-world fix. Example:

   "To improve Ncell’s customer satisfaction study, the researcher should use stratified random sampling to include rural users and pilot-test questions to avoid ambiguity."

Worked Example: Designing a Study for Kathmandu Traffic Congestion

Research Question: Does the introduction of dedicated bus lanes reduce travel time for private vehicles?

1. Design: Causal (Quasi-experimental) – No random assignment (can’t force bus lanes everywhere).
2. IV: Presence of bus lanes (introduced in Thapathali vs. no lanes in Kalanki).
3. DV: Average travel time during peak hours (measured via GPS tracking).
4. Extraneous Variables to Control:
   • Weather (same season for both areas).
   • Road construction (none during study period).
5. Data Collection:
   • Quantitative: GPS data from 1000 vehicles in each area.
   • Qualitative: Driver interviews on perceived congestion.
6. Ethical Consideration: Inform drivers their data is used for research (NTC could partner for this).

Expected Result:

• If Thapathali shows 20% faster travel times, the design proves bus lanes work—policy recommendation for other cities.

Key Formulas and Checklists

Checklist for Valid Research Design

Variable Relationships