PSYCH339 Lecture Notes - Lecture 10: Coefficient Of Determination, Psycinfo, Personnel Selection
67 views3 pages
10 Aug 2016
School
Department
Course
Professor
Lecture 10
Cross validation: degree to which derived from one sample are useful for predicting
performance in another sample
We can use a technique called multiple regressions to figure out the best way to
combine those two predictors
Essentially we want to add the tests together
Predicted performance = test 1+ test 2
Predicted grade school = GPA + GRE
Regression tells us how much to weight to give each test
Predicted performance = 17.54 + test 1 * 0.85 + test 2 * 1.04
Predicted performance is also called composite score
But here’s the problem
We need to cross validate
Apply those weights to another sample
Typically Rd > Rc
D = derivation (original sample)
C = cross validation sample
In my example Rc = 0.68
This decrease n validity is called shrinkage
Typical recommendation? Just add them together
Unit weighting
Predicted = test 1 + test 2
Easier to explain to clients anyway
Meta analysis
Also called validity generalization in context of selection
Average of many validity studies
oWeighted by sample size
find more resources at oneclass.com
find more resources at oneclass.com