Experimental control

1
Experimental control

• Measures to reduce confound
• Single blind vs.double blind to reduce what
  bias?
• Pretest to control for pretest sensitization
• (e.g., threats to internal validity, e.g.,
  history, maturation).
• Random assignment

2
Random Assignment

• Why necessary?
• Example social support group
• How to assign randomly?
• Assign pairs of subjects
• Between subject treatment A vs. B assign equal
  n to each treatment.
• Stratified random assignment example
• Computer generated random numbers

3
Example Random assignment using computer
generated random numbers http//www.random.org/nfo
rm.html
Web Interface to theTrue Random NumbersMads
Haahr, July 1999 Fill out this form to generate
genuine random numbers. Generate ____random
integers (maximum 10,000). Smallest value
____(limit -1000,000,000). Largest value
____(limit 1000,000,000). Format in ____columns.
4
79 47 3 13 90 35 22 14 50 97 1
33 68 70 62 9 44 65 43 37 22
18 28 78 23 29 31 54 56 34 67 66
35 11 24 44 84 99 73 23 77 72
88 7 8 8 26 74 31 99 37 27 53
83 67 7 86 79 99 20
W/ duplicates
5
Table of Random Table Digits

• 35233 76520 13586 34673

37542 04805 64892 74296 24805
Odd treatment, even control

• 35233 76520 13586 34673

37542 04805 64892 74296 24805
S1 Treatment S2, 3 Control S4,5 Treatment
6
Fill out this form to generate randomized
sequences. Smallest value ___(limit
-1000,000,000). Largest value ___(limit
1000,000,000). The length of the sequence ___(the
largest minus the smallest value plus 1) can be
no greater than 10,000.
W/o duplicate
7
Fill out this form to generate randomized
sequences. Smallest value _1__(limit
-1000,000,000). Largest value _40__(limit
1000,000,000). The length of the sequence (the
largest minus the smallest value plus 1) can be
no greater than 10,000.
W/o duplicate
8
12 22 29 13 21 6 32 8 30 38 17 24 9 40 3 31 19 16
14 28 7
12 22 29 13 21 6 32 8 30 38 17 24 9 40 3
31 19 16 14 28 7 1
9
Experimental Designs

• Between-subjects (nested design)
• Example
• Within-subjects (crossed design, repeated
  measures)
• Different types of within factorial design
  Example
• Mixed model design
• Example

10
Within subjects design

• Order effect
• Counterbalancing
• Latin squares
• order of administration
• 1 2 3
• Sequence1 A B C
• Sequence2 A C B
• Sequence3 B A C
• Sequence4 B C A
• Sequence5 C A B
• Sequence6 C B A

11
Solomon four-group design
Cause Group I Group II Group III Group
IV Pretest yes no yes no Treatment yes
yes no no Sensitization yes no no
no Extraneous yes yes yes yes Effects
12
Solomon four-group design
Groups Random Obs Treat Obs Assignment Group
I R O X O Group II R X O Group
III R O O Group IV R O
13
Solomon four-group design
Groups Random Obs Treat Obs Assignment Group
I R O X O Group II R X O Group
III R O O Group IV R O
Insufficient controls