Lecture 11 Flashcards
Binomial effect size (validit coeffcient)
How well a test predicts or measures the outcome it’s supposed to asses.
Sensitivity
How well the test diagnoses people who have a condition (reality =yes) as having a condition.
-True positive rate
Specificty
how well the test diagnoses people who don’t have a condition (reality=no) as not having a condition.
- True negative rate
Construct bias
If the test measures the construct differently for different groups
ex: culture-language, age, gender,test situation
Prediction bias
when test scores predict outcomes differently for different groups. (ex: predicts success for one not for other)
Happiness level example
Item: I feel awesome. Objectively americans and eastern europeans will feel the same but test score for americans will be higher.
Construct bias- reponse process
1-The relation between the latent variable and an item is affected by external variable (gender)
2-Works differently for different groups (unfair)
Differential Item functioning
occurs when people from different groups (e.g., male vs. female) with the same ability or trait level have different probabilities of answering a test item correctly or similarly. (ex:For example, a math problem that requires knowledge of sports might disadvantage people unfamiliar with sports, even if they are equally good at math.)
No construcy bias if?
1- Item doesn’t show differential item functioning.
2-The regression line is the same for males and females (if there is only one regression line.
yes construct bias if?
1- Item shows differential item functioning.
2-The regression line differs for different groups either a different intercept or slope.
Methods to detect construct bias
1-Group differences in factor loading
2-Differential Item functioning
3-Discrimination Index
4-Mantel-Haenzel test
