Can 10 people tell you what millions think?
"Study of 12 people proves new diet works!" "I surveyed my 5 friends - they all agree!" The problem? SAMPLE SIZE (too few) and REPRESENTATIVENESS (not diverse enough). Let's learn to spot weak samples!
SAMPLE = small group studied to learn about a LARGER population. Can't ask all Americans their opinion? Ask a sample! But for reliable results: (1) Sample must be LARGE enough, (2) Sample must be REPRESENTATIVE (match the larger group's diversity)!
TOO SMALL = unreliable! Flip a coin 3 times: might get 3 heads (100%!). Flip 1,000 times: approaches 50/50. LARGER samples reduce the impact of random chance and outliers. General rule: Hundreds minimum for population studies, thousands better!
REPRESENTATIVE SAMPLE = reflects the diversity of full population! Surveying 1,000 college students about ALL Americans? Not representative (too young, educated, etc). Need mix of: ages, locations, backgrounds, income levels. SIZE ≠ quality if sample is biased!
Be skeptical when: • Sample size under 100 (for population claims), • Only from one location/group, • Self-selected (online polls - only motivated people respond!), • No info about HOW sample was chosen. Always ask: "How many? Who exactly? How were they selected?"
Good samples must be both LARGE ENOUGH and REPRESENTATIVE of the population!
Sample size:
• Too small: Random chance dominates (unreliable)
• Large enough: Patterns emerge, chance evens out
• Rule of thumb: 100s minimum, 1000s+ ideal for population studies
Representativeness:
Sample should MIRROR the population in key characteristics:
• Age distribution
• Geographic spread
• Income levels
• Education levels
• Gender balance
• Ethnic diversity
Bad sampling examples:
• "I asked 10 people at the mall" (tiny + biased location)
• "1,000 Twitter users said..." (self-selected, not representative)
• "Study of Harvard students shows..." (not representative of all students)
Good sampling:
• Random selection from full population
• Stratified (ensuring diverse representation)
• Large enough for statistical reliability
• Transparent about methodology
Critical questions:
1. How many people?
2. Who were they exactly?
3. How were they chosen?
4. Do they represent the full population?
Remember: 1,000 biased people < 100 well-chosen people!