sp 2015 cp probability & statistics observational studies vs. experiments chapter 11
DESCRIPTION
Observational Studies Simply observing regular occurrences Researchers do not assign choices Can be RETROSPECTIVE or PROSPECTIVETRANSCRIPT
SP 2015CP PROBABILITY & STATISTICS
Observational Studiesvs.
Experiments
Chapter 11
Observational Study vs. Experiment
Observational StudyObserve only, no manipulation of factors is used
ExperimentFactors are manipulated to create treatments and randomly assign subjects to the different treatments
Observational Studies
Simply observing regular occurrences
Researchers do not assign choices
Can be RETROSPECTIVE or PROSPECTIVE
Observational Studies
Useful for:Discovering trends and potential relationshipsUsed in public health and marketing
Observational StudiesDO NOT demonstrate cause-and-effect relationships
Retrospective Study
Not based on random samplesTypically used to estimate:
Differences between groups Associations between variables
Typically focus on small segment of entire population
Retrospective Study - STEPS
Steps:Select SubjectsDetermine their previous
conditions/behaviors Exam historic information
Pull records from data bases/sources (prior grades, classes, etc...)
Ask subjects questions in order to gather knowledge of past events
Retrospective Study - EXAMPLE
Example:Identify people with a diseaseLook at their history, heritage to
determine things which may be related to their condition.
Retrospective Study - WARNING
Data collected through asking subjects to recall past events tends to have errors!!!!!!
What did you eat last week, exactly?
Retrospective Study – WARNING 2
Lurking Variables
Prospective Study
Observe subjects over time
Identifying subjects in advanceCollecting data as events unfold
No treatments/conditions are intentionally controlled
Example
Experiments
Manipulate factors to create treatments
Can show cause-and-effect relationships
Requires random assignment of subjects to treatments
Subjects are also referred to as Participants/ Experimental Units
Compares results from different treatment groups
Experiments - BASICS
Requires a random assignment of subjects to treatments
Study the relationship between two or more variables
Must have at least one of each:Explanatory variable (Factor) to
manipulate&
Response variable to measure
Experimenter - ROLE
Actively and deliberately manipulates factors to control the details of the possible treatments
RANDOMLY assigns the subjects to the treatments
Observes the response variableCompares responses for different
groups of subjects who have been treated differently.
Experiments - WHO
Humans are commonly called subjects or participants
Other individuals (rats, days, petri dishes of bacteria) are commonly referred to by the more generic term experimental units.
Experiments - WHAT
Factor – Variable whose levels are manipulated by the experimenter
Levels - Specific values chosen for a factorex. In a sleep study, we might assign participants to sleep for 4, 6, or 8 hours
Treatment – Process applied to a group of subjects. Treatments are the different levels of the factor.
Example - Continued
Treatments: Typical portions of 2 dog foods (1 from original company, 1 we know is
safe)
Response: Veterinarian’s assessment of test animals health
Experimental Design - 3 Principles
ControlRandomizeReplicate
Experimental Design - 3 Principles - CONTROL
We control sources of variation other than the factors we are testing by making conditions as similar as possible for all treatment groups.
For human subjects, we try to treat them alike.
Controlling sources of variation makes it easier to detect any differences caused by the treatments.
Control - Methods
Control treatment - baseline measurement
Control group - experimental units to whom the control treatment is applied
Placebo, Blinding
Blinding
People are so good at picking up subtle cues about treatments that it’s important to keep anyone who could affect the outcome or the measurement of the response from knowing which subjects have been assigned to which treatments.
2 groups: People who can influence the results (interviewer,
subjects, test administrator) People who evaluate the results (researcher,
doctors, judges)
Blinding
Single Blinding – when every individual in 1 of the groups is blinded
Double Blinding – when every individual in both groups is blinded
Recall: 2 groups People who can influence the results People who evaluate the results
Placebo & Placebo Effect
Placebo - A “fake” treatment that looks just like the treatments being tested.
Placebos are the best way to blind subjects from knowing whether they are receiving the treatment or not.
Placebo effect - when subjects treated with a placebo improve.
It’s not unusual for 20% or more of subjects given a placebo to report reduction in pain, improved movement, or greater alertness, or even to demonstrate improved health or performance.
Placebo controls are so effective that you shoulduse them as an essential tool for blinding whenever possible.
Blocking
We use blocking when there are pre-existing differences between groups of experimental units.
Randomization is introduced when we randomly assign treatments within each block.
Experimental Design - 3 Principles - RANDOMIZE
Allows us to equalize the effects of unknown or uncontrollable sources of variation.
Note: Randomization cannot eliminate the effects of these sources, but it should spread them out fairly equally across the treatment levels
Experimental Design - 3 Principles - Replicate
2 Types of Replication
Type 1 - Within the experiment Apply each treatment to several subjects
Type 2 – Additional Experiment Replication of an entire experiment with different subjects
Example - Continued
In the experiment comparing animal food, how could you implement the principles of
control, randomization, and replication?•Control the portion sizes•Reduce possible variability from other factors:• (Standardize other aspects of their environments—housing the dogs
in similar pens and ensuring that each got the same amount of water, exercise, play, and sleep time.)
• Restrict the experiment to a single breed and age • Assign dogs to the two feed treatments randomly.
•To try and equalize traits, pre-existing conditions, and other unknown influences•Replicate by assigning more than one dog to each treatment to allow for variability among individual dogs.•If time and funding, possibly replicate the entire experiment using, for example, a different breed of dog.
Diagram of Experimental Procedure
Your Turn! – CW1
On the following slide, you will be asked to design an experiment. Follow these steps:
State what you want to know (in general)Specify the response variableSpecify the factors and treatments/levelsSpecify the Experimental UnitsDesign the experiment (Control, Replicate, Random)Create a diagram to show the processHow will you display resultsReflect – are the observations made, significant?
Your Turn! – CW1