Greenwald: Statistically small effects of IAT can have societally large effects

Question 1

Why can meta-analyses have different results? Name 3 ways and give an example

Answer 1

Different methods:
- Different studies and effects input
- Different statistical decisions
- Missing moderators (correspondence, social sensitivity)

Meta analyses of IAT scores on racial bias predicting behavior

Question 2

What did Greenwald et al. find in his meta analysis on effects of IAT?

Answer 2

He found moderate predictive validity, echoing preceding years’ demonstrations of moderate predictive validity for self-report attitude measures

Question 3

What was the difference in research goal for Greenwald and Oswald?

Answer 3

Greenwald: examine predictive validity

Oswald: investigate correlations of IAT measures with measures indicative of discrimination, regardless whether predictive relations were expected

Question 4

What is the conclusion on Oswald’s findings? Why was it different than Greenwald’s?

Answer 4

Smaller effect of IAT on discrimination, primarily because of the difference in purpose of their meta-analysis

So Oswald included a substantial minority of effect sizes for which no correlations were expected. So IAT is poor predictor

Question 5

What was the biggest difference in methods between Greenwald and Oswald?

Answer 5

Different policies for including effect sizes

Question 6

On which 2 expectations were Greenwalds findings based?

Answer 6

• Measures of attitude toward a group should predict behavior favorable or unfavorable
• Measures of stereotype of group should predict stereotype-consistent behavior

Question 7

Describe Oswald’s inclusion policy

Answer 7

Included any study where the criterion measured some form of discrimination

Question 8

What are ICC’s? What was the difference between Oswald’s and Greenwald’s research?

Answer 8

IAT criterion correlations

Oswald included more ICC’s than Greenwald did

Question 9

Why did Oswald have a larger number of samples? Give 2 reasons

Answer 9

• They requested raw data to compute ICC’s (IAT criterion correlations) that authors hadn’t reported
• They subdivided samples into subsamples so different ICCs could be computed

Question 10

What was Oswald’s result in comparing mean ICC effect size between Oswald’s and Greenwald’s?

Answer 10

Two values weren’t significantly different

Question 11

According to the writers of this paper, which 2 steps did they take to arrive at their conclusion?

Answer 11

1. Oswald’s and Greenwald’s findings had very similar implications for the variance explained by IAT measures
2. Statistically small effects may have societally large effects

Question 12

When do statistically small effects have societally important effects? Give 2 conditions

Answer 12

• If they apply for many people
• If they apply repeatedly to the same person

Question 13

What is consequential validity?

Answer 13

This part of construct validity looks at whether the scores from a test really mean what they’re supposed to and how they affect decisions or actions. It considers if the test is actually useful and what could happen because of it

Especially in regard to sources of invalidity related to bias, fairness and distributive justice

Question 14

When may IAT measures and other predictors be used? Give 2 conditions

Answer 14

• Identify persons especially prone to committing discrimination
• Understand system-level discrimination

Question 15

What is the four-fifths rule?

Answer 15

It tests whether a protected class (identified by race, color etc.) has been treated in discriminatory fashion

Question 16

When can protected class members be treated by courts as indicating adverse impact/discrimination according to the four-fifths rule?

Answer 16

If it receives receives favorable outcomes less than 80% as often as a comparison class

Question 17

What is the criterion of the four-fifths rule? What does translation of this rule into a correlation coefficient require (2)?

Answer 17

80% (4/5)

Requires:
1. Base rate
2. Class proportion
If they are 50%, the four-fifths rule gives: R=-0,111

Question 18

What does it mean when base rate and class rate are 50%

Answer 18

Half of the population receives favorable treatment and the protected class is half of the population