Transcript Issues in Designing Controlled Experiments

Issues in Designing Controlled
Experiments
Experimental Units
• Items that to be used in experiment should be
representative of the population of interest.
– Insects/Animals/Plants of a particular
gender/species/maturity
– Patients with a given condition and age/size status
– Primary household shopper for families of a
particular size/socio-economic status
– Steel rods of a specified length/radius
Treatments/Conditions
• Should represent all possible levels that can
provide insight to research problem
– Compare 4 widely used varieties of fertilizer
– Compare positive and negative valence in risk
preference experiment.
– Compare a new test drug with currently available
standard (not ethical to include placebo)
– Compare 5 distinct heating temperatures in cooking
optimization study
Assignment of Units to Treatments
• When no external information is available or
units are homogeneous, assign them at random
to treatments using random mechanism.
COMPLETELY RANDOMIZED DESIGN
• When external information is available, units
can be lumped into homogeneous blocks and
randomized to treatments within blocks
RANDOMIZED BLOCK DESIGN
Other Issues
• Blinding: Whenever possible, subjects and researchers
making measurements should be unaware of which
treatment subject was exposed to.
• Crossover Designs: When subjects can be assigned to
all treatments in a study without affecting subsequent
measurements, subjects can act as blocks. They should
receive the various treatment conditions in random
order with sufficient time between exposures.
– Bicyclists endurance at multiple caffeine doses
– Bioavailabilty studies of various drug formulations in people
• Repeated Measures: Many studies measure the same
subjects under a single condition at multiple time points