COMM 88 Lecture Notes - Lecture 13: Internal Validity, Selection Bias
Comm 88 Lecture 13
May 17, 2018
Factorial Designs
•“Interaction effect” between IVs
•The unique effect of the combination of IVs
•The effect of one IV depends on the levels of the other IV(s)
•Examples for a Music X Task interaction: music reduces learning when styling reading, but
enhances it when studying math
•To test for an interaction effect, graph the cell means
•There is an interaction effect if the lines are not parallel
•So although music lowered scores overall, the effect was really only when studying
reading. Music had no effect when studying math.
•A word about factors (IVs)…
•In one design, can have as IVs/factors:
•Manipulated variable(s)
•Ex: music exposure, study tasks
•Subject variable(s)
•Ex: gender, personality traits, TV use (hi/lo)
•Can only make causal conclusions about manipulated IVs (not subject variables)
•If no manipulated variables at all, then it’s not an experiment (it’s a survey with factorial-
type set-up)
•Factorial designs (if…)
•Note re the term “pretest”: also used to refer to “trial run” of some aspect of your study (pilot
test)
Pre-experiments and Threats to Internal Validity
•If NOT a TRUE experiment or if do experiment improperly, then →
•Alternative explanations for results become possible (i.e., threaten internal validity)
•“Pre-experiments” - may sound like experiments (IV manipulation), but no RA & many threats
to internal validity
•One-shot case study
•X1 Y (group 1)
•One-group pretest-posttest design
•Y1 X1 Y2 (group 1)
•Static group comparison
•X1 Y (group 1)
•X2 Y (group 2)
•Threats to internal validity
•Selection bias
•History effect
•Threats related to pre-testing (or measures over time):
•Maturation
•Testing/sensitization
•Statistical regression (to the mean) - phenomenon that happens by chance
•
