Sampling Techniques: Types, Methods & Applications

TAKEAWAYS:

• Sampling reduces costs/time by studying a subset instead of the entire population, but must ensure representativeness to avoid bias.
• Probability sampling (random, systematic, stratified, cluster) guarantees statistical validity, while non-probability sampling (convenience, purposive, snowball) is faster but less generalizable.
• Sampling error (difference between sample stats and population parameters) can be minimized by increasing sample size or using stratified methods.
• Real-world tie: Ncell uses stratified sampling to survey customers across age groups, while Daraz relies on convenience sampling for quick feedback from active users.
• Exam focus: Classify sampling types, compare methods in tables, and justify choices for hypothetical research (e.g., "How would you sample for a study on Kathmandu traffic congestion?").

1. What Is Sampling?

Sampling is the process of selecting a subset (sample) from a larger group (population) to study, analyze, or make inferences about the whole. It is essential in business research to:

• Save time and money (e.g., surveying 500 Daraz customers instead of all 5 million).
• Ensure practicality (e.g., NTC cannot test every vehicle’s fuel efficiency).
• Maintain representativeness (the sample must reflect the population’s key traits).

Worked Example: Kathmandu Traffic Study Population: All vehicles in Kathmandu. Sample: 1,000 vehicles selected via systematic sampling (every 100th vehicle at checkpoints). Why? Random stops would miss rush-hour patterns; systematic sampling covers peak times uniformly.

2. Probability vs. Non-Probability Sampling

mindmap
  root((Sampling Techniques))
    Probability Sampling
      Simple Random
      Systematic
      Stratified
      Cluster
    Non-Probability Sampling
      Convenience
      Purposive
      Snowball
      Quota

Stratified sampling divides the population into homogeneous subgroups (strata) before random selection. (Image: Dan Kernler, CC BY-SA 4.0, via Wikimedia Commons)

3. Probability Sampling Techniques

A. Simple Random Sampling

• Every member of the population has an equal chance of being selected.
• Method: Use random number tables or software (e.g., Excel’s RAND()).
• Example: NEPSE selects 100 random listed companies to analyze stock trends.
• Advantages:
  • Unbiased, easy to analyze statistically.
  • No need for population stratification.
• Disadvantages:
  • Requires a complete sampling frame (e.g., Ncell needs all customer phone numbers).
  • May miss subgroups (e.g., rural vs. urban users).

Worked Example: Nabil Bank Loan Study Population: All 50,000 loan applicants in 2023. Sample: 500 selected via random number generator. Analysis: 95% confidence interval for default rates = 8% ± 3%.

B. Systematic Sampling

• Selects every k-th member from a list after a random start.
• Formula: $k=\frac{\text{Population size}}{\text{Sample size}}$
• Example: NTC inspects every 50th vehicle exiting Kathmandu for emissions.
• Advantages:
  • Simpler than random sampling.
  • Ensures even coverage (e.g., Daraz surveys every 100th order).
• Disadvantages:
  • Periodicity bias: If the population has hidden patterns (e.g., vehicles sorted by engine size), results may skew.

Worked Example: Pathao Driver Survey Population: 20,000 drivers. Sample: 200 drivers selected via $k=100$ (start at driver #42, then #142, #242, etc.). Risk: If drivers are listed by region, urban drivers may dominate.

C. Stratified Sampling

• Divides the population into homogeneous subgroups (strata) and samples from each.
• Proportional allocation: Sample size per stratum = $\frac{\text{Stratum size}}{\text{Population size}}\times \text{Total sample}$.
• Example: Khalti surveys strata by age (18–30, 31–45, 46+) to ensure all groups are represented.
• Advantages:
  • More precise estimates for subgroups.
  • Reduces sampling error for heterogeneous populations.
• Disadvantages:
  • Requires prior knowledge of strata (e.g., income levels for a bank study).

Worked Example: eSewa User Study Population: 5 million users. Strata:

• Urban (60%), Rural (30%), Suburban (10%). Sample: 600 urban, 300 rural, 100 suburban users. Why? Rural users may have different payment habits (e.g., lower mobile money usage).

D. Cluster Sampling

• Divides the population into heterogeneous clusters, randomly selects clusters, and samples all members within them.
• Example: NTC tests 5 random districts and surveys all vehicles in those districts.
• Advantages:
  • Cost-effective (no need for full population list).
  • Useful for geographically dispersed populations (e.g., Daraz sellers across Nepal).
• Disadvantages:
  • Higher sampling error if clusters are not representative.

Worked Example: Himalayan Java Tea Farmer Survey Population: 10,000 tea farmers in 50 districts. Clusters: Randomly select 5 districts → survey all 200 farmers in those districts. Risk: If selected districts are all high-altitude, results may not apply to terai farmers.

4. Non-Probability Sampling Techniques

A. Convenience Sampling

• Selects the easiest accessible members.
• Example: Daraz asks recent buyers (who already visited the site) for feedback.
• Advantages:
  • Fast and cheap.
  • Useful for pilot studies (e.g., testing a new Ncell app feature).
• Disadvantages:
  • Highly biased (e.g., only tech-savvy users respond to eSewa surveys).

Worked Example: Kathmandu University Café Study Population: All students. Sample: 50 students already sitting in the café at 3 PM. Bias: Excludes night-shift students or those who study elsewhere.

B. Purposive Sampling

• Researcher intentionally selects members based on specific criteria.
• Example: Nabil Bank studies 5 high-risk loan defaulters to understand patterns.
• Advantages:
  • Deep insights into rare cases (e.g., fraud in Khalti transactions).
• Disadvantages:
  • Not generalizable (cannot apply findings to all customers).

C. Snowball Sampling

• Initial subjects refer others who fit the criteria.
• Example: Studying undocumented migrant workers in Nepal (hard to locate directly).
• Advantages:
  • Useful for hidden populations.
• Disadvantages:
  • Referral bias (friends may share similar traits).

D. Quota Sampling

• Divides the population into strata and fills quotas via convenience.
• Example: A market research firm for Coca-Cola in Nepal needs 200 males, 200 females, 100 under 25, 100 over 50—selected via street interviews.
• Advantages:
  • Ensures proportional representation without random selection.
• Disadvantages:
  • Researcher bias in selecting quota members.

5. Sampling Error and How to Minimize It

Sampling error = Difference between sample statistic (e.g., mean income) and population parameter. Factors affecting error:

• Sample size: Larger samples reduce error (e.g., Ncell’s 1,000 vs. 100 surveys).
• Sampling method: Probability methods (e.g., stratified) yield lower error than convenience sampling.
• Population homogeneity: More variation → larger required sample.

Formula for Sample Size (Simple Random Sampling): $$n=\frac{N\times Z^2\times p(1-p)}{(N-1)\times e^2+Z^2\times p(1-p)}$$

• $N$ = Population size
• $Z$ = Z-score (1.96 for 95% confidence)
• $p$ = Expected proportion (e.g., 0.5 for maximum variability)
• $e$ = Margin of error (e.g., 5% = 0.05)

Worked Example: NEPSE Stock Index Study Population: 150 listed companies. Desired margin of error: 4%. Sample size: $$n=\frac{150\times 1.96^2\times 0.5\times 0.5}{(150-1)\times 0.04^2+1.96^2\times 0.25}\approx 85\text{ companies}$$

6. Real-World Applications in Nepal

Case Study 1: Ncell Customer Satisfaction Survey

• Method: Stratified random sampling (urban/rural, age groups).
• Why? Ncell needs to tailor services (e.g., rural data packs vs. urban 5G).
• Outcome: Identified that 60% of rural users prefer cash recharge, leading to more agent outlets in villages.

Case Study 2: Daraz Order Fulfillment Study

• Method: Convenience sampling (surveying customers who just placed orders).
• Risk: Biases toward repeat buyers (not first-time users).
• Fix: Combined with purposive sampling of 50 first-time buyers for deeper insights.

Case Study 3: NTC Vehicle Emission Test

• Method: Systematic sampling (every 20th vehicle at checkpoints).
• Challenge: Periodicity bias if vehicles are sorted by age (older vehicles may cluster).
• Solution: Randomize checkpoint times (morning vs. evening).

7. Ethical Considerations in Sampling

• Informed consent: Participants must know they’re part of a study (e.g., Khalti’s survey pop-ups).
• Avoiding coercion: No pressure to respond (e.g., Daraz shouldn’t offer discounts for feedback).
• Anonymity: Ensure data is confidential (e.g., Nabil Bank’s loan defaulter study).
• Bias mitigation: Disclose sampling methods to interpret results correctly.

Exam Tip

1. Classification Questions:

   • Draw a mindmap (like above) or table to classify sampling techniques.
   • Example answer for "Explain the classification of sampling technique":

     Sampling techniques are divided into probability (random, systematic, stratified, cluster) and non-probability (convenience, purposive, snowball, quota) methods. Probability methods ensure generalizability via random selection, while non-probability methods are faster but biased.

2. Justify Your Choice:

   • For hypothetical studies (e.g., "How would you sample for a study on student stress at TU?"), explain:
     • Method: Stratified sampling (by faculty: Management, Engineering, etc.).
     • Why? Ensures each faculty’s unique stress factors (e.g., exam pressure vs. project workload) are represented.
     • Sample size: 300 students (10% of population) for 95% confidence.

3. Calculate Sample Size:

   • Memorize the sample size formula and apply it to exam questions (e.g., "Calculate the sample size for a population of 5,000 with a 3% margin of error").

4. Real-World Links:

   • Connect sampling to Nepali businesses:
     • Ncell: Uses stratified sampling for regional coverage.
     • Daraz: Uses convenience sampling for quick feedback but supplements with purposive sampling for edge cases.
     • NTC: Uses systematic sampling for vehicle checks.

5. Avoid Common Mistakes:

   • ❌ Saying "random sampling is always best" (ignore cost/time constraints).
   • ❌ Confusing stratified (subgroups sampled separately) with cluster (whole clusters sampled).
   • ❌ Ignoring sampling error in non-probability methods.

flowchart TD
    A["Research Question"] --> B{"Probability or Non-Probability?"}
    B -->|"Probability"| C["Simple Random<br/>Systematic<br/>Stratified<br/>Cluster"]
    B -->|"Non-Probability"| D["Convenience<br/>Purposive<br/>Snowball<br/>Quota"]
    C --> E["Ensures Generalizability<br/>Higher Cost"]
    D --> F["Faster<br/>Biased<br/>Lower Cost"]
    E --> G["Analyze Statistically"]
    F --> H["Qualitative Insights"]
    G --> I["Report Findings"]
    H --> I