COMM 88 Lecture Notes - Lecture 15: Observational Error, Random Assignment, Time Series
Comm 88 Lecture 15
May 24, 2018
Quasi-experiments
•Time series design
•Multiple time series design
•Y1 Y2 Y3 Y4 X1 Y5 Y6 Y7 Y8 (group 1)
•Y1 Y2 Y3 Y4 X2 Y5 Y6 Y7 Y8 (group 2)
•Variation: give comparison group treatment and keep measuring
•So far, all of our experiments have been “between subjects” designs
Within-subjects design
•Every subject is in every condition
•Example: pilots’ reaction time to warning lights
•Possible between subjects design, with 18 pilots
•R X1 (red) Y (reaction time) group 1
•R X2 (green) Y (reaction time) group 2
•R X3 (yellow) Y (reaction time) group 3
•Problems - small N’s: much random error
•As a within-subject design: each pilot reacts to each light
•1st) red light → reaction, 2nd) green light → reaction, 3rd) yellow light → reaction
•Now N = 18 in each condition (↑power, ↓random error)
•Problem: carry-over/order effects: practice effects, fatigue effects , first treatment/contrast
effects
•Solution: counterbalance orders
•Randomly assign order of treatments to each subject
Laboratory vs. Field Experiments
•Laboratory experiments - bring subjects into highly controlled setting
•Field experiments - manipulate IV(s) in the “real world”
•Ex: littering studies - parking garage littering in clean vs. littered areas
•Ex: random sample and random assignment of movie “She Said No”, watched movie or not
•More natural setting/behavior → higher external validity
•Less reactivity (or none!)
•But harder to maintain experimental control
