A researcher applies a treatment to part of the population (treatment group) and does not apply the treatment to another part (control group).
The purpose is to observe responses after applying the treatment and compare outcomes between the groups.
In the transcript, examples come from medical realities and popular TV portrayals:
Cancer trials often use a cancer drug vs a placebo to form a treatment group and a control group.
The placebo is a harmless treatment that looks like the real one; subjects in the control group usually do not know it’s a placebo.
Responses from both groups are recorded and compared to assess effectiveness.
Ethics/real-world caveat mentioned in the talk: there are sometimes tensions around access to trials and rule-breaking in helping patients get in (contextualized in the pop-media example).
Basic vocabulary summary:
Treatment group: receives the active treatment.
Control group: receives no active treatment (often a placebo).
Placebo: a fake treatment that appears identical to the real treatment.
Experimental group vs control group: all subjects in experiments are divided into these two kinds of groups.
Observation vs experiment distinction:
If you change something in the environment, you are conducting an experiment.
If you simply observe without changing the environment, you are making an observation.
Classic example in talk: watching how many times you yawn in class is an observation if you don’t alter anything; changing conditions (e.g., providing different stimuli) makes it an experiment.
Practical framing:
In experiments, randomization and control are used to isolate effects of the treatment.
In observations, natural conditions are recorded with no intentional manipulation.
Quick real-world relevance:
Designing studies (treatment vs control) is foundational for assessing cause-and-effect.
Placebos and blinding (participants not knowing they’re in control) help reduce bias in responses.
Key examples discussed in the talk
Vitamin D trial (experimental):
Setup: seven patients received 4000 units of vitamin D3; another group of 70 patients received a placebo.
Treatment: the vitamin D3 (specifically, 4000 IU vitamin D3).
Conclusion (from the speaker): this is an experiment; treatment identified as vitamin D3.
Opinion survey on US economy (observational):
Researchers called many adults and asked them to rate current US economic conditions.
No manipulation of environment or treatment; therefore, observation.
Elk counting by Pennsylvania Game Commission (observational):
Observational study; data collection without deliberate experimental manipulation.
Whale tagging/presence in natural environment (observational with a note on capture/release):
Whales are captured or tagged briefly, then released back into the environment; the mechanism is observational in the main sense, with an interaction with the environment.
Medical information transparency survey (observational):
1033 US adults; no treatment or manipulation; observed opinions.
Intensive program to reduce systolic blood pressure (experimental):
Treatment: intensive program to lower SBP to < or equal to 120 mm Hg.
Outcome: reduced risk of death.
Social media and teenagers’ brains (comparison of observation vs experiment):
Observation route: scan brains while teenagers view their own social media feeds (no manipulation).
Experimental route: show a set of photographs with a varying number of likes (altering the environment) to study brain responses.
Conclusion in talk: supplying photographs with varying likes constitutes an experiment; simply scanning brains while viewing their own feeds is an observation.
Sampling methods (six methods) — overview and quick definitions
There are two concepts that involve grouping of individuals:
Stratified sampling: split the population into groups (strata) and take a sample from each group (some from every group).
Cluster sampling: divide into subgroups (clusters) and sample entire clusters (100% from some clusters).
The other four methods deal with patterns or effort level in selecting individuals:
Systematic sampling: use a fixed pattern (e.g., every nth person) to select individuals; the “nth” pattern is explicit and repeatable.
Convenience sampling: select the easiest or most convenient respondents (lazy approach).
Voluntary sampling: individuals volunteer themselves to be part of the sample (self-selection); no one tells them to participate.
Simple random sampling: every member of the population has an equal chance of being selected; often described as a blind draw from a hat or a random selection process.
Summary mnemonic from the talk:
If it involves groups, think stratified or cluster.
If there is a clear repeating pattern, think systematic.
If it’s lazy or convenient, think convenience.
If participation is voluntary, think voluntary.
If it’s a true random draw, think simple random.
Quick examples from the talk that illustrate the six methods:
Stratified: Dividing students by majors and sampling from each major (some from every group).
Cluster: Selecting entire classes or groups (e.g., picking certain classes and surveying all students in those classes).
Systematic: Selecting every 10th person.
Convenience: Asking the 10 people closest to you in a class or outside a party.
Voluntary: People choosing to respond to a survey on their own, without being asked directly.
Simple random: Selecting participants via a random process such as random telephone numbers or a random draw.
Practice problem themes from the talk:
Classifying scenarios into observation vs experiment and identifying the treatment when it’s an experiment.
Examples included: limited manipulation vs no manipulation, and how the presence of manipulation changes the classification.
Bias in sampling and practical considerations
Potential sources of bias are discussed in the hurricane-grid sampling example:
If only certain grids with more damage are selected, the sample may not represent overall damage distribution (biased sampling).
If many grids are easy to reach and sampled preferentially, this convenience introduces bias.
Two biased scenarios discussed:
Sampling only grids with certain characteristics (e.g., higher damage) could over- or under-estimate true damage patterns.
Convenience sampling (selecting grids that are easy to reach) can bias results toward those areas rather than the population as a whole.
Non-bias/de-bias considerations to note from the talk: