•Most (if not all) classic studies in Behavioural Economics began in a lab
•In a lab, you can exert better control on an experiment. This means:
–We can create “worlds” (conditions) in which certain combinations of factors occur
–We can rule out the effects of factors that we are not interested in studying
1)To document behaviouralphenomena
2)To develop a theory to explain them
3)To study effect sizes
4)To reconcile and test across theories with conflicting predictions
5)To test for the efficacy ofbehaviour-change interventions
1)Control and Treatment conditions
–Control: Typically the absence of the cause
–Treatment: In choice architecture, the “intervention” you are trying to test
–A condition refers to the collection of specific tasks and stimuli that the experimenter presents to a participant.
2)Treatments may take the form of many factors, each factor could have multiple levels
–Price framing is a factor which has two levels: aggregate and pennies-a-day
3)Effect or Outcome: Attitudes, preference strengths or choice.
1)A factor refers to any theoretical variable that you would like to vary and hence test for the effects of
–For instance: Amount prepaid for a basketball ticket
2)Levels: Refers to the different categories or strengths at which that variable is shown to participants
–For instance: Two levels (Low = $20, High = $80)
–Or, four levels (Free, $30, $60, $90)
•The aspect of a process or task that the experimenter changes across experimental groups is a manipulation.
•For example: The information contained in two questions is identical, however the fonts might be different
Background Variables and Randomization
–A set of variables that are “given” and not manipulated
–Results must be interpreted in the context of the background
2)Randomization: Allocating participants to conditions randomly minimizes selection bias and allocation bias (confounding).
–Example: Biases in field studies of credit card versus cash / checks
–Research Assistants: Typically "blind" to details of experiment design in order to minimize any biases on their part
•The use of multiple methods with a view to double (or triple) check the results.
–Multiple sets of background variables
–Data analysis techniques
•A before and after (intervention) design
–Ideally there is also a control condition where there is no intervention
•A simple two condition design
–Control and Treatment
•Fully crossed designs
–If you have two factors each with two levels, a fully crossed design will have all combinations, so 2 x 2 = 4 conditions
–Similarly one could have 3 x 3 x 4 [Three factors, two of them with three levels and one with four]