Data Collection Via Experiments

Slide Information

Remember the two strategies for data collection

Observational Studies (First lecture)

Experiments

Recruit a group of people and assign them a treatment

Can give evidence that treatment causes a response

we control lurking variables to do this

Lurking variables are variables in the study that you haven't controlled for, but may affect the outcome
lurking variable

Experiments try to control for as many lurking variables as possible

Structure of experiments exper

Experiment Vocab

Subject - Individual to measure

Response variable (y) - outcome of the experiment ie damage

Explanatory variable (x) - variable used to explain the response ( num fire fighters )

Treatment - the condition or conditions applied to a subject or individual in an experiment ( 5 fire fighters)

Control - a treatment with supposedly zero effects

Double-blind study - An experiment where the individual and researcher doesn't know which treatment is applied.

Confounding - a situation where a lurking variable, in addition to the explanatory variable is affecting the response

Principles of Valid experiments

Control/Comparison - Control lurking variable by including comparison treatments, using homogeneous subjects; used to measure placebo effect

Randomization - neutralize effects of lurking variables by randomly assigning subjects to treatments

Replication - assign more than one subject to each treatment group

double blinding

Main classes of good experiments

Randomized controlled experiment

Randomized block experiment - matched pairs as a special case

Randomly split all subjects into treatment groups

Separate rats by size first then randomize within each size

There is a problem with this solved by randomized block experiment

Used when individuals are similar

use when individuals are similar within block , but very different from block to block

removed confounding of lurking variables with response variable

yields more precise estimates of chance variation which makes detection of statistical significance easier

reduces chance variation by removing variation associated with the lurking (blocking) variable

Matched Pairs?

Explanatory variable: 2 level factor?

Block: 2 subjects who are very similar (twins)

Randomly assign 1 subject within each block to treatment

Example

Company boot example boots

Placebo effect

Diagnostic Effect

Problem: Diagnosis of subjects biased by preconceived notions about effectiveness of treatment

Solution: blind the diagnoser

Problems

Placebo

Diagnostic effect

Lack of realism

Hawthorne Effect

People in an experiment behave differently from how they would normally behave

Ethical hidden observation

Non compliance

Belmont Report - Ethical guidlines

Respect for persons with given informed consent

Beneficence - do no harm

Justice - can't exploit subjects

Equal distributions of burdens and benefits of study

experiments reviewed by an institutional review board (IRB) to ensure they follow ethical guidlines

Key Terms to understand

Causation

Lurking Variables

Subject

Response Variable

Explanatory Variable /factor

Treatment

Control

Double - blind

Placebo effect

Diagnostic Bias

Data ethics

Randomized Controlled Experiment

Randomized Block Experiment

Matched Pairs Experiment

Reading 2.2

Using a coin flip to decide which treatment a patient receives is called a "randomized experiment"

Principles of experimental design

Controlling

Researches assign treatments to cases, and try to control any differences between groups.

So, like if people take a pill by sipping water, or by drinking a whole glass of water. The researcher tells everyone to take the pill in the same way

Randomization

Randomize patients into treatment groups to account for variable that cannot be controlled.

So, like, some patients might be more susceptible to diseased because of diet.

This would be a confounding variable

That means it is associated with both the explanatory and response variables. Randomization of patients in treatment and control groups helps to even things out

Replication

You have to replicate. Size of replication is also important

There is a replication crisis

Pseudoreplication is when individual observations under different treatments are heavily dependent on each other

Blocking

click to edit