Transcript Experiments
Experiments and Quasi-Experiments Overview • Correlation v. Causation • Three Criteria for causation • How to craft manipulations Correlation v. Causation Depressed Mood Cause? Impaired Sleep Depressed Mood Cause? Impaired Sleep Depressed Mood Impaired Sleep Family Conflict Correlation v. Causation • Finding: Women who have a baby after age 40 are more likely to live page 100. • Finding: The greater the quantity of ice cream sold, the greater the number of murders. • Finding: The greater the number of Churches, the greater the amount of crime. • Finding: The more a person weighs, the larger his/her vocabulary. Three Criteria for Causation: (1) random assignment of Ss (2) to two or more conditions (3) which differ in terms of (only) IVs (1) Random Assignment • What -- every subject has an equal chance of being assigned to different conditions • Why -- prevent systematic and non-treatment differences among subjects in different conditions • How -- Same as Random sampling: • IDEALLY - Identify every member of the Sample, assign them a number (of each condition), and then use random number generator to pick the number you need • IN REALITY – Quota sampling is typically used, so randomly generate numbers for each condition, and then when subject walks through the door, assign them to that condition. (2) Two or more conditions HANDOUT #1 Define Topic of Interest #2 Create the manipulation A. What is a manipulation B. How many conditions and/or manipulations – 1 IV 1) Two Levels 2) Three+ Levels C. How many conditions and/or manipulations – 2+ IV 1) 2 x 2 2) 2+ x 2+ 3) 2+ x 2+ x 2+ (3) Which differ in terms of (only) IVs HANDOUT #3 Evaluate the Manipulation A. Prevent Confounds B. Use Manipulation Check C. Get Feedback D. Pilot Test Quasi-Experiments • Contains aspects of both experiments and non-experiments because deficient in at least one of the three aspects of experimental designs • (1) Hybrid = Adding non-experimental factor that you can’t randomly assign (e.g., demographics, personality traits, etc). Pros/Cons - Can’t prove causation because no random assignment - Can add many non-experimental factors to any/all experiments (e.g., ask questions at end of study about demographics, personality traits, etc. Quasi-Experiments • (2) Matched-pairs = matching pairs of subjects on key variables and assigning each pair to separate condition Pros/Cons - Can’t prove causation because no random assignment - More power because reduced error variance • (3) Within-subjects = measuring/manipulating same subjects at two or more times. Pros/Cons - Can’t prove causation because same subjects in each condition - More power because compares each subject to themselves (so less error variance) and more power because more subjects - Must control for order effects by counter-balancing or Latinsquares Quasi-Experiments • (4) Mixed-designs = containing both between-subjects and within-subjects designs Pros/Cons - Can’t prove causation for within-subjects designs • (5) Single-n design = Manipulating single person in an AB (ABABABAB, etc design) Pros/Cons - Can’t prove causation because not random assignment to two or more groups Quasi-Experiments When do I choose which design? (1) Choose experiments (2) If practical issues prevent you from conducting experiment, then those same practical issues will dictates which quasiexperimental design you use.