Ch. 7 - Experimental Designs
Tuesday, October 30, 20121:00 PM
PSYC2520: Introduction to Experimental Psychology
Beginning BehavioralResearch: A Conceptual Primer (7thEd. 2012) Rosnow& Rosenthal
Chapter 7: RandomizedExperiments and CausalInterference
What is the purpose of randomized experiments?
• Focused statistical tests ask precise questions of data and include:
○ All t tests
○ Only F tests with 1 degree of freedom in the numerator
○ Only chi-square tests with 1 degree of freedom
• Randomized designs
○ Participants are randomly assigned to the experimental groups or conditions
○ Safeguard against biased assignments of samples to the different groups
○ Permits the use of statistical analyses that require certain data characteristics
○ It distributes the characteristics of the sampling units over the experimental and control conditions in a way that
should not bias the outcome of the experiment
Although, it cannot guarantee equality in the characteristics of the sampling units that are assigned to
○ Randomized designs may be unethical in certain instances when a potentially life-saving drug is randomly assigned
to only some participants
How is randomassignment accomplished?
• Presorting booklets or questionnaires
• Blindly drawing names
• Flipping a coin
• Consulting a table of random numbers
What are between-subjectsdesigns?
• Between-subject design-- participants are exposed to one condition only
• Also called nested designs
• Two-group between-subject designs are analyzed using a t test for independent samples
What is the formativelogic of experimentalcontrol?
• Mill's methods-- the idea that when two independent groups are comparable in all respects except for some
intervention or manipulated variable (the experimental treatment) that is operating in one group but not the other, that
experimental treatment is implicated as the probable agent responsible for the observed differences between the two
groups on the dependent measures
○ Method of agreement-- "If X, then Y"
If we find two or more instances in which Y occurs, and if only X is present on each occasion, then X might be
at least a sufficient condition of Y
○ Method of difference-- "If not-X, then not-Y"
If the presumed effect (Y) does not occur when the presumed cause (X) is absent, then X is probably
suspected to be a necessary condition of Y
• In order to say that X causes Y, X must be both sufficient and necessary
What are within-subject designs?
• Within-subjects design-- subjects are exposed to each condition
• Repeated-measures design-- a within-subject design in which actions were measured after each condition
• Also called crossed designs
• Importance of controlling order of administration • Importance of controlling order of administration
• T test for non-independent samples is used
What are factorial designs?
• Two independent variables or more: two-ways, three-ways, etc.
○ Between subjects
○ Within subjects
What is meant by counterbalancing the conditions?
• Counterbalancing-- rotating the sequences to account for the problem of systematic differences between successive
treatments (or measurements)
• Latin square-- a specific statistical design that has counterbalancing built in; characterized by a square array of numbers
or letters (representing treatment conditions) where each letter appears once and only once in each row and in each
How do scientistslogically puzzleout efficient causality?
• By covariation, temporal precedence, and internal validity
What conditionspose a threat to internalvalidity?
• Quasi-experiments-- research using non-randomized designs
• Between & Within Subject Variables
○ Researcher expectancy
○ Participant expectancy
• Between Subject Variables
○ Participant selection
○ Loss of participants
• Within Subject Variables
○ Maturation and