hi
Observational studies-
Retrospective Study:
In a retrospective study, researchers identify subjects or cases that have already experienced the outcome or event of interest.
They then look back at historical data to gather information about the factors or exposures that might have led to that outcome.
Example: A study examining medical records of patients who had a disease to identify risk factors.
Prospective Study:
In a prospective study, researchers identify subjects before the outcome or event occurs.
They then collect data over time as events happen, tracking the subjects to see if they develop the outcome of interest.
Example: Follow a group of healthy people over time to see who develops a certain disease and record data on potential risk factors as they go.
Issues with observational studies
High Chance of Lurking Variables
Cause and effect relationships are difficult to establish( can't show causation)
How can we show causation? Experiment
Experiment- A study intended to establish a cause and effect with random assignment of subjects to treatments
Researcher intentionally manipulates the factors of treatments
Subjects are assigned treatments at random
The response variable (Y) is observed for different groups and is compared
Levels: Specific Values that are chosen for a given factor, ex: If I wanted to see how a half dose of fertilizer worked in comparison to a full dose, that would be an additional level of the chosen factor.
Treatment: The combination of specific levels from all the factors that a subject receives
For example, if you’re testing different types of exercise and diet, one treatment might be “30 minutes of running + high-protein diet,” while another might be “30 minutes of yoga + low-carb diet.”
Key elements of experimental design:
Control
Randomization
Replication
Blocking
Control: Outside of the factor being tested all other sources of variation are kept as similar as possible for all groups
Control = keeping things the same across all groups.
Control group = a group that doesn’t get the treatment, used for comparison.
Randomization: Treatments should be assigned randomly to create groups that are equivalent (reduces confounding)
Replication: Treatments should be applied to the largest number of subjects possible
Blocking: Group similar subjects when appropriate to reduce variability within groups (NOT required in experimental design)
Example: In a diet study, you could block participants by age, so the diet’s effects aren’t confounded by age differences.
We’ll know if treatments made a difference if the differences are considered statistically significant, bigger than what might be expected from randomization alone
Differences Between Sample Surveys and Experiments
Sample Surveys:
Purpose: Collect information from a sample to make inferences about a larger population.
Randomization in Surveys: Randomly select people or subjects to get a representative snapshot of the population.
Example: Polling a random sample of people to estimate public opinion on an issue.
Experiments:
Purpose: Test if one factor causes a change in another (cause-and-effect).
Randomization in Experiments: Randomly assign treatments to subjects, not just pick random people. This helps ensure any effect seen is due to the treatment, not other factors.
Example: Giving one group a new medication and another group a placebo to see if the medication has an effect.
Key Difference: Surveys focus on gathering data from a sample to generalize about a population. Experiments aim to test specific treatments or interventions to determine cause-and-effect relationships.
Control Group
Definition: A control group is a group that doesn’t receive the treatment, serving as a baseline to compare the effects of the treatment.
Example: In a drug study, the control group might receive a placebo, allowing researchers to see if the drug makes a real difference.
Blinding
Blinding: Used to prevent bias by keeping certain information hidden.
Single-Blind: Only participants don’t know which treatment they’re getting.
Double-Blind: Both participants and experimenters don’t know who’s receiving which treatment.
Example: In a taste test, participants don’t know which drink is which (single-blind), or neither participants nor researchers know (double-blind) until results are analyzed.
Placebo
Placebo: A fake treatment that resembles the real treatment but has no effect, used to see if the actual treatment has a real impact.
Example: A sugar pill that looks like the drug being tested.
Completely Randomized Design
Definition: All subjects have an equal chance of receiving any treatment, with no grouping by any factors.
Example: Testing a new fertilizer by randomly assigning plants to either a new or standard fertilizer group.
Randomized Block Design
Definition: Subjects are grouped into blocks based on similar characteristics, and treatments are randomly assigned within each block.
Example: Testing a medication by blocking participants by age group to ensure effects aren’t influenced by age differences.
Matched Pairs Design
Definition: Pairs of subjects are matched based on similar traits, and each pair receives different treatments for comparison.
Example: Pairing people by age and weight in a diet study, with one getting a new diet and the other a regular diet.
Confounding Factors
Definition: An outside variable that affects both the treatment and outcome, making it hard to determine the treatment's real effect.
Example: In a study on exercise and weight loss, diet could be a confounding factor since it also impacts weight.
Lurking vs. Confounding Variables
Lurking Variable: Creates an association that appears causal, but it isn’t.
Example: Shoe size and reading ability in kids are linked by age, not by one causing the other.
Confounding Variable: Interferes with the relationship between treatment and outcome.
Example: Diet in a study on exercise and weight.
Statistical Significance
Definition: Differences are "statistically significant" if they are unlikely to have occurred by chance, suggesting a real effect.
Example: If a drug group shows much greater improvement than the placebo group, the effect might be statistically significant.
The Importance of Sample Size
Larger Sample Sizes: Help ensure results are reliable and not just due to chance.
Example: Testing a vaccine on thousands of people rather than just a few to get dependable results.
Experimental Design Vocabulary
Stratify (for Sampling): Group similar subjects in surveys.
Block (for Experiments): Group similar subjects in experiments to reduce variability.
Example: Blocking by gender in an experiment on medication effects
Matched Pairs
Definition: Subjects are paired based on similarities, with each pair receiving different treatments to directly compare outcomes.
Example: In a sleep study, pairing participants with similar habits and giving one a new sleep aid and the other a placebo.