Sampling Techniques, Observational Studies, and Experimental Design
Principles of Simple Random Sampling
Definition of Simple Random Sample (SRS):
- In a Simple Random Sample (), each individual of the population has an equal chance of being selected, regardless of the selection of other individuals.
- An is inherently unbiased because it prevents systematic favoritism toward any subset of the population.
Evaluation of Sampling Methods ( Students from a Class of ):
- Alphabetic Roster Assignment with a Random Number Generator:
- Procedure: Assign each student on the class roster a number from to in alphabetical order. Utilize a random number generator to select two distinct numbers between and . The corresponding students constitute the sample.
- Status: This constitutes a valid Simple Random Sample ().
- Justification: Every individual in the population possesses an equal probability of being chosen, and the selection is entirely objective.
- Alphabetic Roster Assignment with Human Selection:
- Procedure: Assign each student on the class roster a number from to in alphabetical order. Ask a student to "think of two numbers between and ." The students whose numbers are called form the sample.
- Status: This does not constitute a Simple Random Sample ().
- Bias Present: Human selection bias. Humans cannot generate truly random numbers; psychological preferences, subconscious number biases, and patterns prevent equal selection probability. Therefore, an individual cannot manually choose in an .
- Physical Lottery Draw (Names in a Hat):
- Procedure: Write every student's name on a slip of paper, place all slips into a hat, mix thoroughly, and draw two names blindly.
- Status: This constitutes a valid Simple Random Sample ().
- Justification: Thorough mixing provides a physical lottery mechanism where each name has an identical likelihood of being picked.
- Volunteer / Hand-Raising Method:
- Procedure: Solicit volunteers from the classroom and pick the first two students who raise their hands.
- Status: This does not constitute a Simple Random Sample ().
- Bias Present: Voluntary response and selection bias. The most eager, outgoing, or attentive students sitting at the front are disproportionately selected, while shy or disengaged students have a zero probability of selection.
Survey Design, Wording, and Ordering Biases
Question Wording Bias in Sensitive Research:
- Survey A (High Stigma / Judgmental Framework):
- Title and Header: "Alcoholism and drug addiction survey"
- Query: "How many alcoholic drinks do you consume in a typical week?"
- Response Format: Categorical options consisting of:
- None
- One or two
- Three to Five
- More than five
- Psychological and Measurement Impact: The judgmental title induces shame and social stigma. Respondents systematically gravitate toward responses perceived as "safe" or "socially acceptable" rather than providing an honest report, biasing results downward.
- Survey B (Neutral / Low Stigma Framework):
- Title and Header: "Alcohol and recreational drug use survey"
- Query: "How many drinks containing alcohol do you consume in a typical day or week?"
- Response Format: Quantitative numerical entry:
- "About _____ drinks per day/week (circle one)"
- Psychological and Measurement Impact: Neutral language eliminates shame, allowing respondents to answer honestly without fear of being labeled. Recording numerical data as a continuous quantitative variable preserves exact counts rather than forcing responses into broad categorical bins.
Question Ordering and Priming Bias ("Are You Happy?"):
- Survey A Order:
- Question : "How happy are you with your life in general?" (Scale of to )
- Question : "How many dates did you have last month?"
- Observed Effect: Responses to general life happiness and dating frequency remain largely unrelated, producing negligible correlation.
- Survey B Order:
- Question : "How many dates did you have last month?" (Quantitative explanatory variable)
- Question : "How happy are you with your life in general?" (Scale of to ; Quantitative response variable)
- Observed Effect: Researchers find a significant positive association where individuals reporting more dates also report higher life satisfaction. Asking about dating frequency first primes the respondent to focus on romantic success when evaluating general happiness, anchoring their overall emotional evaluation to that specific domain.
Association Versus Causation
Fundamental Distinctions:
- Association: Two variables are associated if values of one variable tend to be related to the values of the other. The variables are simply connected or correlated without any implied directional influence.
- Causal Association: Two variables are causally associated if changing the value of one variable directly influences or changes the value of the other variable (a direct cause-and-effect relationship).
Linguistic Markers and Phrasing Conventions:
- Causal Phrasing:
- " leads to " (e.g., smoking leads to cancer)
- " results in " (e.g., smoking results in cancer)
- " reduces/increases " (e.g., smoking increases cancer risk)
- Associational Phrasing:
- "When we see , we also see "
- "People who do tend to also do "
- Explicitly avoids language suggesting that causes or directly influences .
Classification and Transformation of Analytical Statements:
- Statement 1: "Children who drink skim milk tend to have more obesity problems than children who drink whole milk."
- Classification: Association (signaled by the phrase "tend to").
- Causal Transformation: "Drinking skim milk leads to obesity in children."
- Statement 2: "Texting while driving increases your chance of having a car crash."
- Classification: Causal association (signaled by the active verb "increases").
- Associational Transformation: "People who text while driving tend to have more crashes."
- Statement 3: "Taking your Statistics homework seriously will increase your test grades."
- Classification: Causal association (signaled by the phrase "will increase").
- Statement 4: "Students who take their Statistics homework more seriously get better grades in Statistics."
- Classification: Association (signaled by the phrase "get better grades" indicating observed co-occurrence without direct proof of causation).
Confounding Variables in Observational Data
Definition and Mechanics of Confounding:
- A confounding variable is an external variable that simultaneously influences both the explanatory variable and the response variable.
- A confounding variable provides an alternative, non-causal explanation for an observed statistical association between the explanatory and response variables.
Case Studies in Confounding:
- Green Clothing and Skin Cancer:
- Observed Link: Wearing green clothing has been linked to an increased incidence of skin cancer.
- Explanatory Variable: Wearing green clothing (Categorical: Yes/No).
- Response Variable: Skin cancer diagnosis (Categorical: Yes/No).
- Confounding Variable: Irish heritage / ethnicity (Categorical: Yes/No).
- Mechanistic Explanation: People of Irish descent frequently wear green clothing due to cultural traditions and simultaneously possess fair skin complexions that genetically heighten vulnerability to skin cancer. The association between green clothing and skin cancer is entirely non-causal.
- Firefighters and Property Damage:
- Observed Link: An increased number of firefighters at a scene is linked to greater total property damage.
- Explanatory Variable: Number of firefighters (.
- Response Variable: Dollar amount of property damage.
- Confounding Variable: Size and severity of the fire.
- Mechanistic Explanation: Larger, more severe fires necessitate sending more firefighters and independently cause greater property destruction. Firefighters do not cause the increased damage.
- Designer Clothing and Longevity:
- Observed Link: People who wear designer clothing have a longer life expectancy.
- Explanatory Variable: Wearing designer clothing.
- Response Variable: Life expectancy (longevity).
- Confounding Variable: Personal wealth / socioeconomic status.
- Mechanistic Explanation: Wealthy individuals have the disposable income required to purchase luxury designer clothes and simultaneously have access to superior healthcare, better nutrition, and healthier living environments, all of which directly prolong lifespan.
Observational Studies Versus Controlled Experiments
Core Operational Definitions:
- Observational Study: A study design in which the researcher does not actively manipulate or control any variable, but merely observes, records, and measures variables as they naturally occur.
- Experiment: A study design in which the researcher actively manipulates and controls the explanatory variable by applying specific treatments to experimental units.
Comparative Study Classifications:
- Nut Consumption and Cholesterol Levels:
- Study Design 1: Participants are divided into two assigned groups where one group is provided a daily portion of nuts to consume and the other is instructed to follow a strict nut-free diet. Classification: Experiment (the researcher actively dictates the explanatory variable).
- Study Design 2: Participants maintain their personal diets freely and complete self-reported food logs noting their nut intake. Classification: Observational Study (the researcher records natural behavior without intervention).
- Hypertension and Diastolic Blood Pressure:
- Design 1: Researchers access hospital medical records of patients who independently chose different medications for hypertension and measure their diastolic blood pressure. Classification: Observational Study (treatments were selected electively by patients and clinicians, not assigned by the researcher).
- Design 2: Researchers assign hypertensive patients randomly to two distinct treatment regimens and record their resulting diastolic blood pressure. Classification: Experiment (treatments are controlled and imposed by the researchers).
Randomization in Sampling and Experimental Design
Experimental Removal of Confounding:
- In an observational framework connecting designer clothing to life expectancy, wealth acts as an uncontrolled confounder.
- Hypothetical Experiment: Researchers select subjects who do not currently wear designer clothing and randomly assign them into two cohorts:
- Treatment Group: Awarded an annual shopping spree to obtain designer clothing free of charge and instructed to wear it consistently.
- Control Group: Receives no clothing stipend and continues wearing non-designer attire.
- Assumption: Subjects retain and wear the apparel rather than selling or trading it.
- Inference Outcome: If an association between designer clothing and increased life expectancy persists, it supports a causal link. Random assignment evenly distributes wealthy and non-wealthy subjects across both the treatment and control groups, neutralizing wealth as an influential factor.
- Principle: By neutralizing the systematic influence of confounding variables, an experiment allows researchers to support conclusions of causal association. An observational study cannot eliminate confounding and therefore cannot establish causality.
Structural Topologies of Research Designs:
- Observational Study with a Random Sample:
- Path:
- Locus of Randomization: Occurs between the target population and the sample selection.
- Primary Function: Enables generalization of observed sample statistics to the broader population.
- Randomized Comparative Experiment:
- Path:
- Locus of Randomization: Occurs between the volunteer participant pool and assignment to treatment groups.
- Primary Function: Balances confounding factors across groups to permit valid causal inference.
The Two-Dimensional Study Evaluation Matrix:
- Dimension 1: Random Selection from the Population:
- Question: Was the sample randomly selected from the population?
- Outcome Yes: Findings can be generalized from the sample to the broader population (indicative of a well-executed observational study or generalized experimental sample).
- Outcome No: Findings cannot be generalized beyond the sample group due to sampling bias.
- Dimension 2: Random Assignment of Explanatory Variable:
- Question: Was the explanatory variable randomly assigned to participants?
- Outcome Yes: Researchers can draw definitive conclusions regarding causality because confounding variables are controlled (indicative of an experiment).
- Outcome No: Researchers cannot establish causality due to potential confounding variables influencing the relationship (indicative of an observational study).