Week 7: (Quasi-) Experimental Design Flashcards
experimental study does allow for causal claims by using three ingredients
- manipulation of X, and consequently observing Y
- random assignment of subject to levels of X, preventing pre-treatment difference in DV and confounders
- experimental control over variables of non-interest such as Z
levels of IV
- level I: experimental condition: some treatment
- level II: control condition: no treatment / dummy treatment
random assignment methods: blocking
e.g., we suspect gender has an influence. Take block of equal size of gender and randomly allocate
matched random assignment
- Pair subjects on expected lurking variable in sets
- Members of set randomly allocated to different level of IV: matched on lurking var
- Special case of blocking where we have 2 treatments and we can pair subjects
between-subjects
every subject experience only 1 condition
within-subjects
every subject experience every condition
pros and cons of within-subjects
+ no confounding subject variance
+ larger ‘statistical power’
+ less participants needed
- order/carry-over effects. solution: counterbalancing
internal validity
is variation in DV not caused by other variation than IV
Total variance = ?
total variance = treatment variance (maximize) + confound variance (eliminate) + error variance (minimize)
treatment variance
variation in DV due to variation in IV between conditions (lang lab vs. control)
confound variance
variation in DV due to systematic co-variation in lurking variable Z (IQ)
error variance
non-systematic differences between study objects (subjects, items) within a condition
threats to internal validity: selection
e.g., more talented pupils select themselves for Lab condition
threats to internal validity: attrition
selective drop-out of participants
threats to internal validity: testing effect
taking vocabulary test before training influences performance after treatment
