Transcript 08-25 lecture

Research Design
8/25
Overview
•
•
•
•
•
•
Components of scientific studies
Types of scientific studies
Inferring causation
Independent and dependent variables
Confounds, random assignment
Quasi-independent variables
Components of Scientific Studies
• Scientific study: Basic unit of empirical research
• Variables
– Anything that can take on multiple values
• Height, IQ, reaction time, extraversion, favorite color
– Measured in scientific studies
• Hypothesis
– Conjecture about how the world works
– Prediction about how variables relate
• Taller people are smarter
• This drug improves memory
• Blue is more popular than red
• Data (singular: datum)
– Results of measurements
– Values of variables
• IQ of Subject 4
• Reaction time of Subject 12 on Trial 23
Types of Scientific Studies
• Experiment
– Involves some sort of intervention or manipulation
– Researcher sets some variable(s) and assesses
effect on other variable(s)
• Vary number of items in a memory list
• Different drugs to different rats
– Allows inference of causation
• List length affects memory
• Drugs differentially affect lever pressing
• Non-experimental study
– Purely observational
– Measure naturally occurring variables and examine
relationships
• Row of classroom, exam grade
– Can't be sure about causation
Non-experimental Studies
• Measure variables without influencing
–
–
–
–
Row of classroom, exam grade
Time spent outside, depression
Bicycles currently owned, lifetime head injuries (6, 4)
Apples per week, colds per year
• Correlation
– Relationship between variables, in terms of what values co-occur
• More apples, fewer colds
• Smarter people tend to like the color red
– All that can be inferred from non-experimental studies
– Does not say what causes what
• Problems with inferring causation
– Reverse causation
– Third variable problem
– Self-selection
Reverse Causation
• Researcher expects X causes Y, but actually Y causes X
• Depression and time outdoors
– Might predict outdoors alleviates depression
– Might find such a correlation
– But, depression might reduce desire for activity
• XY or YX both mean X and Y co-occur
– If you only measure co-occurrence, can’t tell difference
• Solution: Intervention
– Manipulate X
– Any resulting effect on Y must be caused by X, not vice versa
Experiment Group
Time Outdoors
Depression
Depr(Outdoor Group) < Depr(Indoor Group)
Third-variable Problem
• X and Y might co-vary because they’re both caused by Z
• Apples and colds
– Overall health-consciousness could increase apples and reduce
colds
– People who eat more apples would also tend to get fewer colds
– But, no direct causal relationship
• Solution: Intervention (again)
– Manipulate X
– Shouldn’t affect Z
– Any effect on Y must be direct
Attitude
Apples
Experiment Group
Colds(Apple Group) < Colds(No-apple Group)
Colds
Self-selection
• Differences between groups of people can be due to who chooses
to be in which group
– Not necessarily consequence of group membership
• Math GREs by major
– Physics majors might do better than Psych
– Does physics make you better at math?
– Kids good at math more likely to choose Physics
• Height by sport
– Playing basketball makes you taller?
• Effects of alternative medicine
• Can view as reverse causation
– Being tall makes you better at basketball
• Can view as 3rd-variable problem
– Math aptitude affects both major choice and GRE
Experiments
• Independent variable (IV)
– Manipulated by researcher
– Drug/placebo, training time, priming
• Dependent variable (DV)
– Measured by researcher
– Pain tolerance, proficiency, reaction time
• Intervention assures causality
Attitude
X
Apples
Colds
Confounds and Control
• Importance of experimental control
– Only manipulate the IV
– Hold everything else constant
• Confound
– Variable that accidentally covaries with IV
– Subject expectations about drug effects
– Familiarity with experimental context
• Control means not having confounds
– Necessary for knowing effect is due to IV
Random Assignment
• Values of IV must be chosen at random for each subject
• Only way to assure causal relationship
• 3rd variable again
– Outright cheating
– Time of semester
Ability
Experiment
Group
Performance
Quasi-independent Variables
• Some variables can’t be manipulated, but can
be used to create groups
– Sex, age, birthplace
• Sometimes causal direction is obvious
– Height, men vs. women
– Hockey enjoyment, Canadians vs. Americans
• Allows non-experimental study to be treated like
an experiment
– Grouping variable is quasi-independent
– Can treat other variables like DVs