Perceived warmth and competence predict callback rates in meta-analyzed North American labor market experiments

Carina I. Hausladen*, Marcos Gallo*, Ming Hsu, Adrianna C. Jenkins, Vaida Ona, Colin F. Camerer

* contributed equally

Correspondence Studies

• 169 studies analysed
• various grounds of discrimination:

How to systematize this literature?

Correspondence Studies

Social Perception

• Decades of social psychology literature have investigated the semantic dimensions that perceivers use to predict the character and intentions of other people.
• Some attributes are of greater importance for effectively coordinating social behavior than others and thus serve as fundamental dimensions of social perception.
• According to the Stereotype Content Model (Fiske et al. 2002), the most relevant criteria are the
  • persons’ intentions
  • and their ability to carry out their plans

Warmth

Competence

• How to systematize this literature?  

• Is there an easier way to measure and/or predict labour market discrimination in humans?

Lakisha

Lakisha

Warm

Competent

Prolific
Participant

☎

Callback

Data

Studies

CATEGORY

Hiring

Manager

☎

Callback

Gender

Race

Hiring

Manager

☎

Callback

Gender

Race

Hiring

Manager

☎

Callback

95% CI

Gender

Race

Hiring

Manager

☎

Callback

95% CI

Hiring

Manager

☎

Callback

Prolific
Participant

• 787 raters in total
• 85.9 per name

Prolific
Participant

Warm

Prolific
Participant

Warm

Competent

Intraclass Correlation Coefficient

• Single Rating

• Fixed Set of Raters 

• Consistency of Rating

  • How much do the ratings for the same name vary across different raters?

Warm

Warm

Gender

Race

Prolific
Participant

Prolific
Participant

Warm

Competent

Gender

Race

Prolific
Participant

Warm

Competent

How correlated are warmth and competence?

Prolific
Participant

Warm

Competent

PC1
explains 79.3%

of the variance.

PC2

explains 20.7%

of the variance.

Principal Component Analysis

PC1

Prolific
Participant

PC1
 

PC1

Callback

Effect Size \(\rho\)

\(\hat{\rho}\)

0

0.33

1

\(\hat{\rho}\)

Predictive Power

• \[ \text{callback}_i = \beta_0 + \beta_1 \text{PC1}_i + \beta_2 \text{PC2}_i + \epsilon_i \]
• trained a linear model on all names except one
• predict the callback for the left-out name

competence

warmth

median
58.2

median
61.8

competence

warmth

median
58.2

median
61.8

Alternative Specification: Meta-Regression

Meta-Regression

   Sincerely,

   Lakisha Washington

 Lakisha

 Washington

   Hello, I am active in an organisation as

   Treasurer of the Gay and Lesbian Alliance, and I am a

    member of the Jewish Student Alliance.

   Experience
   2017–2020 Front Desk Manager

   Education
   2010 B. Sc. in Public Relations

   Community Service
   2008–2010 Coordinator: Parent-Teacher-Association

   Hobbies
   Sailling, Polo, Classical Music

Names

Categories

Gender

Race

Sexual orientation

Religion

Employment gap

Age

Parenthood

SES

Disability

Nationality

8

4

1

1

2

2

2

2

2

TOTAL

21

Studies

CATEGORY

Prolific
Participant

• 200 raters in total
• 99.11 per category

Warm

ICC (3,1)

Intraclass Correlation Coefficient

Prolific
Participant

Warm

Competent

PC1
explains 80.7%

of the variance.

PC2

explains 19.3%

of the variance.

Principal Component Analysis

PC1

Meta-Regression

Meta-Regression

Conclusion

• PC1 explains 80% of the variance.
   
• Names: Warmth and competence perception predicted callback.
   
• Categories: The effects of social perception on callback rates are ambiguous.

Implications

Theories of Discrimination

• Statistical Discrimination
  • Employers use observable characteristics as proxies for unobservable traits.
  • Process is inefficient if characteristics are weakly correlated with job performance but strongly correlated with social perceptions.
• Taste-Based Discrimination
  • ​Personal prejudices/tastes held by employers
  • Observable characteristics elicit subjective judgements of warmth and competence.
• Institutional Discrimination
  • National context shapes the behaviour of social actors.
  • We can not add to this theory as we only focus on the North American labour market and, therefore, have no comparison.

Marcos

Gallo

Ming 

Hsu

Adrianna C.

Jenkins

Vaida

Ona

Colin F. 

Camerer

carinah@ethz.ch

slides.com/carinah

github.com/carinahausladen

Appendix

\(\tau^2\)

• \(\tau^2 = 0.08\)

• 95% CI [0.03–0.66]

\(I^2\)

• \(I^2 = 0.83\)
• 95% CI [0.69–0.91]

Prediction Interval

Prediction Interval:  [-0.40; 0.80]

Exercise

1. Assign Ratings:

   $$\text{Warmth}=45,\text{Competence}=10$$

2. PCA Loadings for PC1 and PC2:

   1.  

      PC1 Loadings=(−0.70 ​−0.70​)
   2. PC2 Loadings=(−0.70 0.70​)

       

      PC1 Loadings=(−0.7071−0.7071)
      PC2 Loadings=(−0.70710.7071)

3. Calculate PC1 Score:

   PC1X Æ A-12=(45×−0.7071)+(10×−0.7071)=−38.89PC1X Æ A-12​=(45×−0.70)+(10×−0.70)=−38.89

4. Calculate PC2 Score:

   PC2X Æ A-12=(45×−0.7071)+(10×0.7071)=−24.75PC2X Æ A-12​=(45×−0.70)+(10×0.70)=−24.75

• Formula Used:

  θ^X Æ A-12=θ+β1×PC1X Æ A-12+β2×PC2X Æ A-12+ϵX Æ A-12+ζX Æ A-12θ^X Æ A-12​=θ+β1​×PC1X Æ A-12​+β2​×PC2X Æ A-12​+ϵX Æ A-12​+ζX Æ A-12​

• Given Values:

  θ=−1.97,β1=1,β2=0.56θ=−1.97,β1​=1,β2​=0.56PC1X Æ A-12=−38.89,PC2X Æ A-12=−24.75PC1X Æ A-12​=−38.89,PC2X Æ A-12​=−24.75

• Calculation:

  θ^X Æ A-12=−1.97+(1×−38.89)+(0.56×−24.75)θ^X Æ A-12​=−1.97+(1×−38.89)+(0.56×−24.75)θ^X Æ A-12=−1.97−38.89−13.86=−54.72θ^X Æ A-12​=−1.97−38.89−13.86=−54.72

• Interpretation:

  $$\text{The predicted callback rate for "X Æ A-12" is -54.72\%, indicating a very low likeli}$$

Theoretical Models

Statistical discrimination (Arrow, 1998)

Unfair treatment of ethnic minorities can result from rational actions executed by profit-maximizing actors who are confronted with the uncertainties accompanying selection decisions.

Taste-based discrimination (Becker, 2010)

Discriminatory behavior is the result of people’s unfavorable attitudes toward ethnic minorities.

\[\tau^2\]

\[ \tau^2 \] is a measure of the variance of true effect sizes across studies. \[ \tau^2 \] =0.08 suggests that there is variability in the effect sizes across the studies that cannot be attributed to sampling error alone. This variability could be due to differences in study designs, populations, interventions, or other factors.

Confidence Interval (CI): The confidence interval provides a range in which we are fairly confident that the true value of \[\tau^2\]  lies. In our case, the 95% CI ranges from 0.03 to 0.66. This wide range indicates considerable uncertainty about the precise value of the variance. The lower bound (0.03) suggests that there is at least some heterogeneity, while the upper bound (0.66) indicates that the heterogeneity could be quite substantial.

Significance of Heterogeneity: The fact that the confidence interval does not include zero suggests that the heterogeneity is statistically significant. This means that the variance of true effect sizes is likely greater than zero, indicating that the effect sizes are not consistent across all studies.

Implications for Meta-Analysis: Significant heterogeneity, as indicated by our results, means that caution should be exercised in interpreting the overall effect size obtained from the meta-analysis. It suggests that the included studies are not estimating the same underlying effect size and that there may be subgroup differences or moderating variables that need to be explored.

Cochran’s Q:  $$ Q=\sum_{k=1}^{K} w_k  (\hat{\theta}_k - \hat{\theta})^2$$

 inverse of the study’s variance

competence

warmth

median
58.2

median
61.8

white

black

foreign

Lakisha Jones

Laurie Anderson

11–15%