PSYC2009 Study Guide - Final Guide: Stratified Sampling, Simple Random Sample, Autosuggestion
Experimental Design and Statistical Inference
• An experiment controls the influence of at least one variable, whereas a nonexperimental
study does not.
• Experimental control amounts to being able to hold a variable constant or systematically
manipulate it so that it is not influenced by anything other than the experiment.
• The motivation behind experimental control is the ability to make causal inferences, i.e.
outcome is due to only one variable.
• Extraneous variables - other possible causal variables whose influence is confounded with the
variable we're interested in.
• Matching - finding pairs of participants who are the same age, gender and have the same
severity of heart attack, and assigning one to anew and the other to an old treatment.
• Limitation - we usually cannot match on more than a few variables at a time and we often
don't know which variables may be confounded with the ones whose effects we want to
study.
• Randomised assignment - participants randomly assigned to treatment groups, resulting in
groups that differ on all variables by chance alone unless the new and old treatments really
differ.
• Limitation - it does not ensure that the groups will be identical - merely that they will differ by
chance only.
• Statistical methods enable us to distinguish whether the new treatment group's health is likely
to be really better than the old treatment group or whether an apparent difference could be
due to "luck of the draw".
• Why is this important? - we can kid ourselves into thinking a new treatment is better when the
effect is due to an extraneous variable.
Technical Terms in Experimental Designs
• Experimental conditions - correspond to states or values of the independent variables, eg:
new and old treatment.
• Randomised assignment of units - involves using a randomising device to allocate units to
conditions.
• Blinding - the experimenter and/or participant do not know which condition they have been
assigned to.
o Blinding the participant - ensures against auto-suggestion from the participant
regarding which condition they are placed in, eg: the placebo effect.
o Blinding the experimenter - ensures against unconsciously influencing the participant
or other aspects.
• DV - the outcome variable that the experimenter wishes to influence.
• IV - influences the dependent variable, regardless of whether it is under experimental control
or not.
• Experimental variable - an IV that is controlled by the experimenter. Some IVs are inherently
uncontrollable, eg: gender and age.
• Between-subjects design - expose each unit to only one condition.
• Randomised assignment pertains to this kind of design.
• Within-subjects design or repeated-measures - expose units to all conditions, eg: the study of
the effect of alcohol on driving performance. If we have two conditions, alcohol vs no alcohol
consumed, then we might compare the same people in both conditions.
Why would we use a within-subjects design rather than using a between-subjects design?
• Within-subjects designs have less random variation than between-subjects designs for the
same reason that a stratified random sample has less variation than a simple random sample -
a within-subjects design is a perfect matching strategy.
• Within-subjects designs require far fewer people than between-subjects.
find more resources at oneclass.com
find more resources at oneclass.com
