Different Designs

• between subjects
• within subjects
• mixed designs
• single-subject

Error Variance

• The effect of extraneous variables
• Not including the independent variable
• Can't control for every variable
• Strange events do happen (even with randomization)!
• What is the analog in the stats we have been doing?

Juggling Aptitude
 

Control Group	2 drinks
25	13
24	19
18	22
29	18
19	23
mean=23	mean=19

 
 
 
 
 
 

Reducing Error Variance

• Controlled Experimental Setting
• Limits impact of extraneous variables
• Match subjects on certain criteria
• Equating Stimuli
• naming experiment for frequency effects
• use all words or some class?
• external validity?
• Increase the effectiveness of your variables
• Strong manipulation of independent variable
• more sensitive dependent variable
• better measurement techniques
• Random Assignment
• How will you do this in your experiment?
• Run a lot subjects, correct stats

Randomized between subjects (one factor, 2 levels) design

• Safest Design
• Random Assignment
• Least Power (large error variance)
• Of course, you can have more than 2 groups
• multiple control groups
• really just like the other groups (or levels)

Matched Groups Design

• Equate subjects on some criteria
• Randomly assign a member of each pair to a condition
• Advantage
• potentially reduces error variance
• Disadvantages
• time consuming
• may not equating on relevant factors
• Of course, you can do this with more than two groups

Within Subjects Design

• the ultimate matched design
• each subject see every level of the independent variable
• What would this mean for our naming/frequency experiment?
• Low error variance, lots of power!

Dangers of repeated measures?

• fatigue
• Carryover effects
• memory experiment
• rehearsal vs rehearsal/imagery
• practice effects
• may not follow instructions

Sources of Carryover

• Learning
• learn something in one treatment that helps you do better in another
• Fatigue
• Do worse in later treatments because you are worn out in later treatments
• Habituation
• Reduced sensitivity
• Sensitization
• Increased sensitization (e.g., previous foot shock and then noise)
• Contrast
• difficulty level
• rating scales (similarity)
• Adaptation/tolerance
• subject changes during experiment

Dealing with Carryover

Counterbalancing

• Naming experiment?
• Spread carryover equally over all treatments
• Complete counterbalancing
• k! where k is the number of conditions
• A,B,C
• A,C,B
• B,C,A
• B,A,C
• C,A,B
• C,B,A
• 10 stimuli
• 3,628,900 orders!
• Randomized Counterbalancing
• naming/frequency
• easy to do
• relies on chance
• most used method
• Partial Counterbalancing
• e.g.,every item in every position
• Latin Square
• Random Rotation Method
• Still carryover effects may not be symmetric
• memory experiment
• differential carryover effect
• May be so interesting that you study it in its own right (make it a variable)
• irreversible
• not naive anymore

Latin Squares

• Each condition appears at each ordinal position
• each condition precedes and succeeds each other condition at least once
• each condition does not precede and succeed each other condition at all ordinal positions

Constructing a Latin Square

• Assume you have N conditions (where N is even)
• 1) randomly order conditions
• 2) Row 1 of the square
• 1,2,N,N-1,4,N-2,5,N-3,....
• 3) the second row is generated by adding 1 to each entry in row 1
• N becomes 1
• 4) repeat step 3 to generate the rest of the rows up to row N
• 5) randomly assign subjects to rows

Random Rotation Method

• Another incomplete within subjects design
• Begin with a single random order
• for each subject rotate the random order one position to the right
• last condition becomes first
• 1 2 3 4 5
• 5 1 2 3 4
• 4 5 1 2 3
• 3 4 5 1 2
• 2 3 4 5 1

Other techniques

• Practice-highly trained subjects
• Stable
• No learning, sensitization, or habituation

Should you use a within-subjects design?

• Obviously it depends.
• How many subjects do you need?
• Carryover effects?