Source: Principles of Statistics I, Texas A&M / Tamhane & Dunlop Ch. 3
Tags: observational study, experimental study, prospective study, retrospective study, sample survey, cross-sectional study, cohort study, case-control study, causation vs correlation, Tuskegee study, Declaration of Helsinki, research ethics, informed consent
Statistical studies fall into two broad categories: observational (the researcher watches) and experimental (the researcher intervenes). Within observational studies, the direction in time matters, giving us prospective, retrospective, and cross-sectional designs. Only experiments can establish causation; observational studies can only show correlation.
Observational study
A study that records information about subjects without applying any treatment. The researcher is a passive participant. The purpose is to describe a group or situation.
Experimental study
A study that records information about subjects while applying treatments and controlling study conditions. The researcher actively participates. The purpose is to determine whether a treatment causes a change in a response.
Sample survey
An observational study that provides information about a population based on a sample drawn at a specific point in time. Examples include political polls, market surveys, and customer satisfaction questionnaires.
Prospective study (longitudinal, forward-looking study)
An observational study that begins with a sample in the present and follows it forward in time to record the occurrence of specific outcomes.
Retrospective study (backward-looking study)
An observational study that begins with a sample in the present and collects information about outcomes that have already occurred in the past.
Cross-sectional study
An observational study involving data collected at a single point in time, often used to assess prevalence of conditions or to explore associations between variables.
Cohort study
A comparative study that follows a group of subjects forward in time to observe differences between those who develop a condition and those who do not. Typically prospective.
Case-control study
A comparative study that identifies two groups (one with a condition, one without) and gathers information about their past exposure to risk factors. Typically retrospective.
Correlation vs causation
Observational studies can demonstrate that two variables are related (correlation) but cannot prove that one causes the other. Only well-designed experiments can establish causation.
Confounding variable (confounder, lurking variable)
A variable not accounted for in the study that may influence both the treatment and the response, making it difficult to isolate the true effect.
The basic goal of most studies is to use a subset of a population to make a statement about the whole population.
In an observational study, the researcher simply records what is happening without intervening. Examples: political polls, traffic accident studies, epidemiological studies, the Challenger data.
In an experimental study, the researcher applies treatments and controls conditions to some degree. Examples: clinical trials, laboratory studies, agricultural field trials, greenhouse experiments, pilot plants in industry.
The critical distinction: observational studies can only show correlation. For instance, vegetarians have lower rates of heart disease, but this could be because they also smoke less and exercise more. An observational study cannot untangle these factors.
Experiments, because they involve controlled manipulation and randomisation, can establish cause-and-effect relationships.
Sample survey: snapshot at one point in time. The Bureau of Labor Statistics' Current Population Survey (CPS) is a classic example, sampling 60,000 households monthly to calculate unemployment rates (UR1 through UR6, each using progressively broader definitions of unemployment).
Prospective study: starts now, follows subjects forward. Example: tracking academic success of Head Start vs non-Head Start children; recording weight gain over one year in subjects who quit smoking.
Retrospective study: starts now, looks backward. Example: investigating whether colon cancer incidence is related to diet by collecting dietary histories from people with and without colon cancer.
Cross-sectional study: data collected at a single time point to assess prevalence. Example: randomly selecting 5,000 oral contraceptive users and 10,000 non-users and recording occurrence of myocardial infarction among women aged 40–44.
Retrospective studies are generally cheaper and faster to complete.
Retrospective studies suffer from recall errors (e.g. "What did you eat during the past three days?") and have no control over variables that may affect outcomes, such as genetics, occupation, or environment.
Prospective studies allow subjects to keep careful real-time records (diet diaries, daily weigh-ins) and can instruct subjects to avoid activities that might bias the study.
Prospective studies still cannot completely control confounding variables because they remain observational. Restricting the study to matched subgroups (similar occupation, ethnicity, location) can partially mitigate this.
Both prospective and retrospective studies are often comparative. The two main comparative designs are cohort studies and case-control studies.
A cohort study follows subjects forward in time, observing who develops a disease and who does not. This is prospective.
A case-control study starts with two groups (disease present, disease absent) and looks backward at risk factors. This is retrospective.
Case-control studies have a practical advantage for rare diseases: a random sample from the general population would contain very few individuals with the disease, making analysis difficult. By deliberately selecting subjects with the disease, the study ensures enough cases for meaningful comparison.
The Tuskegee Study of Untreated Syphilis (1932–1972) is a landmark example of unethical research. The U.S. Public Health Service enrolled 600 Black men (399 with syphilis, 201 without) without informed consent. Participants were told they were being treated for "bad blood" but never received proper treatment, even after penicillin became the standard cure in 1947. The study was originally projected to last six months but ran for 40 years.
An Associated Press story in 1972 prompted public outcry. An advisory panel concluded the study was "ethically unjustified" and ordered it stopped.
The Declaration of Helsinki (1964, amended multiple times through 2013) is the World Medical Association's statement of ethical principles for medical research involving human subjects. Core principles include: the health of the patient is the first consideration; informed consent is required; research must be based on sound scientific methodology; and benefits must outweigh risks.
⚠️ Identifying study type is a very common exam question. The key question: did the researcher apply a treatment (experimental) or just observe (observational)?
⚠️ The direction in time distinguishes prospective (forward) from retrospective (backward). Cross-sectional is a single snapshot.
⚠️ "Observational studies show correlation, not causation" is a foundational principle you will be tested on.
⚠️ Know the practical trade-offs: retrospective studies are cheaper and faster but suffer from recall bias and confounding; prospective studies are more controlled but costlier and slower.
⚠️ Case-control studies are retrospective. Cohort studies are prospective. Exams frequently ask you to identify which is which.
⚠️ The Tuskegee study is the go-to example for why informed consent and ethical oversight exist.
Q: A researcher records the blood pressure of 200 adults without prescribing any medication. Is this an observational or experimental study?
A: Observational. No treatment is applied; the researcher is passively recording information.
Q: A study follows 1,000 smokers and 1,000 non-smokers over 20 years to compare lung cancer rates. What type of study is this?
A: A prospective cohort study. Subjects are identified now and followed forward in time.
Q: Researchers identify 500 patients with Type 2 diabetes and 500 without, then compare their dietary histories from the past decade. What type of study is this?
A: A retrospective case-control study. Two groups are identified by outcome and their past exposures are compared.
Q: Can an observational study prove that eating red meat causes heart disease? Why or why not?
A: No. Observational studies can only show a correlation between red meat consumption and heart disease. Confounding variables (exercise habits, smoking, genetics) cannot be fully controlled, so causation cannot be established.
Q: Why are case-control studies preferred over random sampling when studying rare diseases?
A: A random sample from the population would contain very few individuals with the disease, making it difficult to draw meaningful conclusions. A case-control design deliberately includes enough disease cases for comparison.
Q: What was the primary ethical violation in the Tuskegee study?
A: The participants were not given informed consent. They were misled about the purpose and nature of the study, and they were denied effective treatment even after penicillin became available.
observational study, experimental study, prospective study, retrospective study, cohort study, case-control study, cross-sectional study, sample survey, correlation vs causation, confounding variable, lurking variable, informed consent, Tuskegee syphilis study, Declaration of Helsinki, research ethics, clinical trial, study design, types of statistical studies, STAT 301, Principles of Statistics I