Simple Random Sampling: Every member of the population has an equal chance of being selected, and every possible sample of size n has an equal chance of being selected.
- Assign a unique number to each member of the population (1 to 500)
- Use a random number generator or table to select 50 unique numbers
- Select the corresponding members for the sample
Number all 500 students from 1 to 500
Use a random number generator to select 50 unique numbers between 1 and 500
Select the 50 students corresponding to the randomly generated numbers
Ensure each student had an equal probability of selection (50/500 = 10%)
Assign numbers 1-500 to students, generate 50 random numbers, select corresponding students. Each student has 10% chance of selection.
• Unbiased: Each member has equal chance of selection
• Simple: Easy to implement with small populations
• Statistically valid: Allows for probability calculations
• May lack representativeness: Could miss important subgroups
• Requires complete list: Need access to full population
• Logistical challenges: Difficult for large, dispersed populations
Stratified Sampling: Population is divided into non-overlapping groups (strata), then samples are taken from each stratum proportional to its size in the population.
Female: 600/1000 = 60%, Male: 400/1000 = 40%
Female: 60% × 100 = 60 students, Male: 40% × 100 = 40 students
Select 60 females randomly from 600, and 40 males randomly from 400
Total sample: 60 females + 40 males = 100 students
Select 60 females using simple random sampling from 600, and 40 males using simple random sampling from 400.
Sample size for stratum = (Stratum size / Total population) × Total sample size
• Guaranteed representativeness: Ensures all subgroups are represented
• More precise estimates: Reduces sampling error
• Better comparisons: Allows analysis within subgroups
Cluster Sampling: Population is divided into clusters (usually geographic or organizational units), some clusters are randomly selected, and all members of selected clusters are included in the sample.
Each school represents one cluster (total of 50 schools)
Randomly select a number of schools (e.g., 10 schools) to ensure approximately 500 students
Survey all students in the selected 10 schools
If each school averages 50 students, 10 schools = 500 students
Randomly select 10 schools from 50, then survey all students in those schools (approximately 500 total).
| Method | Process | Advantages | Disadvantages |
|---|---|---|---|
| Simple Random | Individual selection | Unbiased, fair | Expensive, may miss subgroups |
| Stratified | Proportional subgroup sampling | Representative, precise | Requires detailed population info |
| Cluster | Group selection | Cheap, convenient | Less precise, potential bias |
Multi-stage Sampling: A complex sampling method that combines multiple sampling techniques in stages, often starting with cluster sampling followed by other methods within selected clusters.
Select 10 districts randomly from 20 districts
Identify teacher categories (e.g., subject areas) as strata within each selected district
Within each selected district, randomly select 50 teachers proportionally from different subject strata
Total: 10 districts × 50 teachers per district = 500 teachers
First randomly select 10 districts, then within each district identify subject strata and randomly select 50 teachers proportionally from each stratum.
• Cost-effective: Reduces travel and administrative costs
• Practical: Suitable for large, complex populations
• Representative: Maintains diversity while reducing complexity
• Increased complexity: More difficult to analyze statistically
• Potential bias: Each stage introduces potential sources of bias
• Sample size calculations: More complex due to multiple stages
Comparative Analysis: Evaluating different sampling methods based on criteria like representativeness, cost, feasibility, and statistical validity.
Population: 5000 employees across 5 departments, Sample size: 250 (5%)
Advantages: Unbiased, simple; Disadvantages: May not represent all departments equally
Departments as strata: 5% of each department (proportional allocation)
Select 1-2 departments as clusters (non-proportional representation)
Stratified sampling recommended for departmental representation
Recommend stratified sampling with departments as strata. Select 5% of employees from each department to ensure all departments are represented proportionally.
| Method | Representation | Cost | Complexity | Best For |
|---|---|---|---|---|
| Simple Random | May vary | Medium | Low | Homogeneous populations |
| Stratified | Guaranteed | High | Medium | Heterogeneous populations |
| Cluster | Limited | Low | Low | Large geographic areas |
Population: The entire group of individuals we want to study
Sample: A subset of the population selected for study
Parameter: A numerical characteristic of a population
Statistic: A numerical characteristic of a sample
- Define the population: Clearly specify who/what you want to study
- Choose the sampling method: Based on population characteristics and research goals
- Determine sample size: Balance between precision and practical constraints
- Implement the sampling: Follow the chosen method precisely
- Verify representativeness: Check that sample reflects population characteristics
• Sampling fraction: \( \frac{n}{N} \)
• Proportional allocation: \( n_h = N_h \times \frac{n}{N} \)
• Standard error depends on sampling method and sample size
Questions & Answers
Question: I'm confused about when to use stratified vs. cluster sampling. Both involve dividing the population into groups. Can you clarify the key difference?
Answer: Great question! The key difference lies in how sampling occurs within the groups:
- Stratified sampling: We divide the population into strata (groups), then take a sample FROM EACH STRATUM. The goal is to ensure representation from all important subgroups.
- Cluster sampling: We divide the population into clusters, then randomly select ENTIRE CLUSTERS and include ALL members of selected clusters. The goal is to reduce costs and increase efficiency.
Analogy: Think of a pizza. For stratified sampling, you'd cut slices from each topping area (representing different strata). For cluster sampling, you'd randomly select whole slices (clusters) and eat everything on those slices.
When to use: Use stratified when you want guaranteed representation from important subgroups. Use cluster when the population is naturally grouped and you want to reduce logistical costs.
Question: In stratified sampling, why do we maintain the same proportion in each stratum? Couldn't we take different proportions?
Answer: You raise an excellent point! There are actually TWO types of stratified sampling:
- Proportional allocation: Sample size from each stratum is proportional to its size in the population (the standard approach)
- Disproportional allocation: Different sampling rates for different strata, used when some strata are more important or variable
For example, if you're studying a rare disease, you might oversample the affected population to get enough cases for meaningful analysis, even though they're a small percentage of the total population.
However, for most basic applications (like your grade level), proportional allocation is the standard because it maintains the same representation as the population.
Formula for proportional allocation: \( n_h = N_h \times \frac{n}{N} \), where \( n_h \) is sample size for stratum h, \( N_h \) is population size of stratum h, n is total sample size, and N is total population size.
Question: How does sample size affect the accuracy of our results? Is bigger always better?
Answer: Generally, larger samples provide more accurate estimates, but there are important nuances:
- Law of diminishing returns: Accuracy increases with the square root of sample size, so doubling sample size only improves accuracy by about 40%
- Sampling method matters: A well-designed smaller sample can be more accurate than a poorly designed larger sample
- Cost considerations: At some point, the marginal gain in accuracy doesn't justify the increased cost
Mathematical relationship: The standard error of a sample mean is proportional to \( \frac{1}{\sqrt{n}} \), where n is the sample size.
Practical advice: Focus on proper sampling method and reducing bias rather than just increasing sample size. A representative sample of 1000 is usually much better than a biased sample of 10,000.