Sampling Techniques: Methods, Types & Applications

TAKEAWAYS:

• Sampling is the process of selecting a subset of a population to represent the whole for cost-effective and accurate research.
• Probability sampling ensures every population member has a known chance of selection (e.g., simple random, stratified), while non-probability sampling relies on convenience or judgment (e.g., quota, snowball).
• Stratified sampling divides populations into homogeneous subgroups (strata) to improve precision, commonly used in market research (e.g., Daraz customer surveys).
• Cluster sampling groups populations into clusters (e.g., villages in Nepal) and samples entire clusters, reducing logistical costs.
• Convenience sampling (e.g., surveying Pathao drivers at a hub) is fast but biased; purposive sampling targets specific groups (e.g., NEPSE traders).
• Sampling error (difference between sample and population stats) is minimized in probability methods but unavoidable in non-probability; sample size affects precision (larger = more accurate).

1. Introduction to Sampling

Sampling is the backbone of efficient research. A population is the entire group of interest (e.g., all Ncell users in Kathmandu), while a sample is a manageable subset studied to infer population characteristics. Why sample?

• Cost: Surveying 100,000 NEPSE investors is impractical; 1,000 is enough.
• Time: Analyzing customer feedback for a Daraz order queue of 50,000 takes weeks; 500 orders suffice.
• Feasibility: Studying every Pathao driver’s route is logistically impossible; 200 drivers represent the group.

Key Terms:

• Sampling Frame: The list of population members (e.g., NTC’s registered subscribers).
• Sampling Unit: The entity selected (e.g., an individual, a household, or a business like a Khalti agent).
• Sampling Error: The difference between sample and population statistics (e.g., a survey overestimating eSewa users by 5%).

2. Types of Sampling: Probability vs. Non-Probability

A. Probability Sampling (Every member has a known chance of selection)

flowchart TD
    A["Probability Sampling"] --> B["Simple Random"]
    A --> C["Stratified"]
    A --> D["Cluster"]
    A --> E["Systematic"]
    B -->|"Every member has equal probability"| F["Example: Lottery for Ncell survey"]
    C -->|"Population divided into strata (e.g., income groups)"| G["Example: Daraz surveys by customer spending tiers"]
    D -->|"Population divided into clusters (e.g., wards)"| H["Example: NTC coverage study by district"]
    E -->|"Select every *k*-th member from a list"| I["Example: Polling NEPSE traders at fixed intervals"]

Worked Example: Stratified Sampling for eSewa Users Problem: Estimate eSewa’s market share in Kathmandu, where users vary by age (18–30, 31–50, 50+). Solution:

1. Stratum 1: 18–30 (40% of population).
2. Stratum 2: 31–50 (50%).
3. Stratum 3: 50+ (10%).
4. Take 100 samples per stratum (total 300).
   • Stratum 1: 40 users → 80% eSewa adoption.
   • Stratum 2: 50 users → 60% adoption.
   • Stratum 3: 10 users → 30% adoption.
5. Weighted average: $(0.4\times 0.8)+(0.5\times 0.6)+(0.1\times 0.3)=0.61$ → 61% adoption.

B. Non-Probability Sampling (Selection based on convenience/judgment)

flowchart TD
    A["Non-Probability Sampling"] --> B["Convenience"]
    A --> C["Purposive"]
    A --> D["Snowball"]
    A --> E["Quota"]
    B -->|"Easiest to access"| F["Example: Surveying Pathao drivers at a hub"]
    C -->|"Target specific groups"| G["Example: Interviewing NEPSE brokers"]
    D -->|"Participants recruit others"| H["Example: Studying rare diseases via patient networks"]
    E -->|"Fill quotas (e.g., 50 males, 50 females)"| I["Example: Daraz’s gender-based product feedback"]

Worked Example: Convenience Sampling for Kathmandu Traffic Problem: Estimate average delay at Thapathali Chowk during peak hours. Solution:

• Method: Stop 50 drivers at random (convenience).
• Bias: Overrepresents taxi drivers (who wait longer) and underrepresents motorcyclists (who zip through).
• Fix: Use quota sampling (e.g., 20 cars, 20 motorcycles, 10 buses).

3. Determining Sample Size

Sample size depends on:

1. Population size (N): Larger populations need bigger samples.
2. Confidence level (Z): 95% (Z = 1.96) is standard.
3. Margin of error (E): Typically 5%.
4. Variability (σ): Higher variability → larger sample.

Formula for Simple Random Sampling: $$n=\frac{Z^2\cdot {\sigma }^2}{E^2}$$ Where:

• $n$ = sample size,
• $Z$ = confidence level (1.96 for 95%),
• $\sigma$ = standard deviation,
• $E$ = margin of error (0.05 for 5%).

Worked Example: Ncell Customer Survey

• Population (N): 500,000 users.
• Variability (σ): 20% (from pilot study).
• Margin of error (E): 5%.
• Calculation: $$n=\frac{1.96^2\times 0.2^2}{0.05^2}=\frac{3.8416\times 0.04}{0.0025}=615$$
• Result: 615 users needed for 95% confidence.

4. Sampling Error and Bias

How to Reduce Bias:

• Use probability sampling (e.g., stratified for Daraz orders).
• Pilot test questions (e.g., pre-test Pathao driver surveys).
• Incentivize responses (e.g., Ncell vouchers for participation).

5. Choosing the Right Sampling Technique

mindmap
  root((Sampling Technique Selection))
    Simple Random
      - Population is homogeneous.
      - Sampling frame available.
    Stratified
      - Population has distinct subgroups.
      - Need to represent all strata.
    Cluster
      - Large, geographically spread population.
      - Cost-effective for remote areas (e.g., NTC towers).
    Convenience
      - Quick, exploratory research.
      - Not for generalizable results.
    Purposive
      - Targeting specific experts (e.g., NEPSE analysts).
    Snowball
      - Hard-to-reach groups (e.g., migrant workers).
    Quota
      - Need balanced subgroups (e.g., gender in Daraz ads).

Case Study: Nabil Bank’s Loan Approval Study Problem: Assess why 30% of loan applicants are rejected. Solution:

1. Population: 5,000 loan applicants in Kathmandu.
2. Method: Stratified sampling by income (low, medium, high).
   • Stratum 1: Low income (2,000 applicants) → 200 samples.
   • Stratum 2: Medium income (2,500) → 250 samples.
   • Stratum 3: High income (500) → 50 samples.
3. Findings:
   • Low-income applicants: 40% rejection (credit score issue).
   • High-income applicants: 10% rejection (documentation).
4. Action: Simplify documentation for high-income applicants.

In the Real World

1. eSewa’s User Feedback Loop

   • Idea: Stratified sampling by transaction frequency (daily, weekly, monthly users).
   • How: eSewa sends surveys to 100 users per stratum monthly.
   • Why: Ensures feedback reflects all user types, not just heavy users.

2. Daraz’s Inventory Management

   • Idea: Cluster sampling by warehouse location (Kathmandu, Pokhara, Biratnagar).
   • How: Daraz samples entire orders from 5 warehouses (10% of total).
   • Why: Reduces cost of tracking every order while maintaining accuracy.

3. Pathao’s Driver Performance Tracking

   • Idea: Systematic sampling of driver routes (every 10th trip logged).
   • How: Pathao’s app records data for drivers numbered 10, 20, 30, etc.
   • Why: Balances workload and ensures randomness without manual selection.

Exam Tip

1. Compare Methods: Always contrast probability (e.g., stratified) and non-probability (e.g., convenience) sampling in your answers. Use a table (like above) to score full marks.

   • Example: "While simple random sampling ensures unbiased selection, convenience sampling is faster but prone to non-response bias."

2. Apply to Real Scenarios:

   • For probability sampling, link to NEPSE (stratified by stock type) or Ncell (systematic by region).
   • For non-probability, use Pathao (purposive for experienced drivers) or Khalti (convenience for agent feedback).

3. Calculate Sample Size:

   • If the question provides σ and E, plug into the formula. If not, explain the trade-off between sample size and accuracy.
   • Example: "A larger sample reduces sampling error but increases cost and time."

4. Highlight Bias:

   • Mention systematic error (e.g., surveying only TU students for Kathmandu opinions) and how to mitigate it (e.g., stratified sampling).

5. Case Study Analysis:

   • For ABC Bank (past exam), identify:
     • Population: Bank customers in Kathmandu.
     • Sampling method: Likely stratified (by account type: savings, current).
     • Bias risk: If convenience sampling was used (e.g., only branch visitors surveyed).

Visual Summary:

mindmap
  root((Sampling Techniques Summary))
    Probability Sampling
      - Simple Random
      - Stratified
      - Cluster
      - Systematic
    Non-Probability Sampling
      - Convenience
      - Purposive
      - Snowball
      - Quota
    Key Considerations
      - Sample Size Formula
      - Bias vs. Accuracy
      - Cost vs. Feasibility
    Real-World Links
      - eSewa (Stratified)
      - Daraz (Cluster)
      - Pathao (Systematic)