Presentation is loading. Please wait.

Presentation is loading. Please wait.

Big Bad Racists, Subtle Prejudice and Minority Victims: An Agent Based Analysis of the Dynamics of Racial Inequality Quincy Thomas Stewart Department of.

Similar presentations


Presentation on theme: "Big Bad Racists, Subtle Prejudice and Minority Victims: An Agent Based Analysis of the Dynamics of Racial Inequality Quincy Thomas Stewart Department of."— Presentation transcript:

1 Big Bad Racists, Subtle Prejudice and Minority Victims: An Agent Based Analysis of the Dynamics of Racial Inequality Quincy Thomas Stewart Department of Sociology Indiana University CSDE, University of Washington February 2011

2 Introduction Research on race begins with a statistic. ▫“Black unemployment is twice that of whites.” ▫“Blacks and Latinos earn significantly less than whites.” ▫Point: Highlight difference in Quality of Life

3 Introduction Research on race begins with a statistic. Theory is then used to explain the emergence of this racial disparity. ▫“Employer discrimination leads to unemployment and lower wages.” ▫Point: Highlight the dynamic processes that lead to racial inequality.

4 Introduction Research on race begins with a statistic. Theory is then used to explain the emergence of this racial disparity. Challenge: Quantitative analyses of racial difference largely rely on static, cross-sectional data that do not capture the micro-level dynamics described by theory. ▫Disputes about support for theory. ▫Data mining for explanatory variables.

5 Introduction Solution: Adopt a broad methodologically heterogeneous research program to examine dynamic experience of race (i.e., group process). ▫Qualitative Research ▫Simulations ▫Social Experiments ▫Audits ▫Traditional Quantitative Analyses

6 Aim Use agent-based model (i.e., simulation) to shed light on the link between dynamic theories of micro-level behavior and the macro-phenomena of racial inequality. Question: Can a small group of discriminators create racial disparities via the emergence of subtle prejudice and minority victims? ▫Change expectations about racial groups and inspire boundely rational biased treatment.

7 Outline Review: Race and Status Construction Theories. Outline of Simulation: Nash Bargaining Game Results ▫Graphs of inequality ▫Multivariate analysis of simulation Discussion

8 Background—State of Race Racial inequality continues to be a characteristic of American society. Quandary: How can a system of inequality persist in a country that values equality, justice and meritocratic achievement?

9 Background—State of Race Racial inequality continues to be a characteristic of American society. Quandary: How can a system of inequality persist in a country that values equality, justice and meritocratic achievement? Persistence of inequality and quandary has led to an array of theories with related policy prescriptions.

10 Background—Theories of Race Discrimination–A very large number of persons with a “taste for discrimination” drive inequality. ▫Contact Theory ▫Support has waned for two reasons: 1.Overt racial attitudes have declined significantly. 2.Surge of scholars championing structural causes.

11 Background—Theories of Race Discrimination–A very large number of persons with a “taste for discrimination” drive inequality. Structural–Inequality results from widespread, systemic bias that inspires discriminatory treatment in a host of contexts. ▫Race is part of the social machinery that distributes private rewards and public privilege. ▫Racism is more than an attitude/belief.

12 Background—Theories of Race Discrimination–A very large number of persons with a “taste for discrimination” drive inequality. Structural–Inequality results from widespread, systemic bias that inspires discriminatory treatment in a host of contexts. Minority Victim–Product of rational minority disinvestments resulting from lack of opportunity. ▫Ogbu: Acting White

13 Background—Victim Theory Oppositional Culture –Black youth respond to low opportunities and marginalization by forming a social space where certain “dominant” behaviors are marginalized. ▫Similar to Feedback in Statistical Discrimination Stereotype Threat –Minorities underperform out of fear for confirming a negative group stereotype.

14 Background—Theories of Race Discrimination–A very large number of persons with a “taste for discrimination” drive inequality. Structural–Inequality results from widespread, systemic bias that inspires discriminatory treatment in a host of contexts. Minority Victim–Product of rational minority disinvestments resulting from lack of opportunity. ▫Ogbu: Acting White ▫Stereotype Threat

15 Background—Race Discrimination and Structural theory locate source of inequality outside of minority actors. ▫Neither theory indicates the extent to which persons or institutions must be saturated with racist ideology to create/maintain inequality

16 Background—Race Discrimination and Structural theory locate source of inequality outside of minority actors. Victim theories locate the source of inequality in the bounded rational responses of minorities. Does the creation and maintenance of racial inequality require an abundance of racist (institutional) actors? To what extent does the presence of minority victims influence inequality? Status Construction Theory

17 Background—Status Construction Focus: How nominal characteristics (i.e., race) acquire status value. ▫Centers on micro-level mechanisms that lead to race developing a reputation that is linked to behavior and attitudes in a specific context. ▫Strong support from social experiments.

18 Background—Status Construction Focus: How nominal characteristics (i.e., race) acquire status value. Idea: Relationship between race and an important resource is sufficient to create wider beliefs (i.e., prejudice) that buttress inequality and lead to unfair treatment.

19 Background—Status Construction Focus: How nominal characteristics (i.e., race) acquire status value. Idea: Relationship between race and an important resource is sufficient to create wider beliefs (i.e., prejudice) that buttress inequality and lead to unfair treatment. Encounters between persons of different characteristics (i.e., race) and resources drive the acceptance of the status belief—the stereotype.

20 Background—Double Difference Ridgeway (1991) writes: “Encounters among those that differ in both resources and the nominal characteristic…lead eventually to the creation of consensual beliefs about the characteristic’s status value.”

21 Support for Status Construction Several social experiments support the theory. Ridgeway et al. (1998) found that: “…[after] two doubly different encounters, both the [dominant actors] and, importantly, [subordinate actors] formed beliefs that people in the [dominant] category were higher in status” (347). People learn status beliefs in interactions and by witnessing interactions among doubly dissimilar persons (Ridgeway & Erickson 2000).

22 Race & Status Construction Structural and Discrimination theory suggest institutions and/or actors discriminate to shape racial inequality. Victim theory indicates that minority groups’ rational responses to lack of opportunity influences disparities. Status Construction theory highlights the micro-level mechanisms that create/maintain racial inequality.

23 Race & Status Construction Structural and Discrimination theory suggest institutions and/or actors discriminate to shape racial inequality. Victim theory indicates that minority groups’ rational responses to lack of opportunity influences disparities. Status Construction theory highlights the micro-level mechanisms that create/maintain racial inequality. Taken together, these theories suggest that institutional and/or individual discrimination leads to resource differences among groups of persons that may then be used by dominant and subordinate actors to augment inequality in future interactions. Maybe a small group of discriminators can create racial inequality through the emergence of subtle prejudice and minority victims…let’s simulate it to see!

24 Methods: Simulation Simulations of social phenomena have been used for decades. ▫Schelling’s (1971) work on residential segregation is perhaps the most popular simulation in social science. ▫More recently, simulations have been used to study disease processes, self enforcing norms and household demand.

25 Outline of Simulation Nash Bargaining Game (Axtell et al. 2001) 200 agents endowed with race, age, wealth, memory, and an initial bid ranging from 10 to 90. Agents randomly paired to bid on a good with maximum value of 100—blind bids. ▫If  of bids >100, agents both receive no earnings ▫If  of bids <100, agents each earn their bid. ▫Most optimal, rational bid is 50.

26 Simulation: Example Encounter Agent A bid: 60 Agent B bid: 30 Total Bid = 90 (bid<100). Agent A earns 60, and B earns 30. A: 60! B: 30! Yeah!!!!!

27 Outline of Simulation Nash Bargaining Game 200 agents endowed with race, age, wealth, memory, and an initial bid ranging from 10 to 90. Agents randomly paired to bid on good with maximum value of 100—blind bids. Agents use recent memory (e.g., 6 encounters) to inform future bids. ▫New bids = 100 minus average competitor bid.

28 Simulation: Example Encounter 1 Agent A: Bid 60 Agent B: Bid 30 Total Bid = 90. Agent A earns 60 and B earns 30. Memory: 2 nd Encounter with different agents Agent A: Bid 70 [100 minus 30 (avg. bid)]. Agent B: Bid 40 (100 minus 60). Each makes rational bid based on local info. A: 70! B: 40! C D

29 Outline of Simulation II Nash Bargaining Game In certain simulations, I insert a small number of dominant discriminators. ▫Always increase their bid when they encounter a subordinate agent—subordinates deserve less. In certain simulations, I allow agents to invoke race-specific memories in making bids. ▫Each agent’s bid based average bid received by agents of that race group in the past. ▫Boundedly rational bid based on past experience.

30 Outline of Simulation III The addition of discriminators and race-specific memories allow me to critically examine three mechanisms of inequality. 1.Discriminators—How do they contribute to disparities? 2.Subtle prejudice—How does it develop among dominants and add to disparities? 3.Minority victims—How might this rational response (i.e., low bid) add to inequality? 4.How might these parameters intersect and jointly contribute to inequality?

31 How does it work? Discriminators (i.e., big bad racists) increase bids to subordinates and lower their demands. Subtle Prejudice emerges among other dominants because subordinates begin to bid less than dominants—these dominants also make higher bids to subordinates. Minority Victims emerge in response to high bids from others—discriminators and agents with subtle prejudice—and make lower bids.

32 Classic Solution: Example The game can be written as a two-player normal form game G = {S 1, S 2 ; u 1, u 2 } where S is strategy and u is payout. If one assumes agents are using a uniform random mixed strategy making bids between 10 and 90, the Nash equilibrium will be a response of 50. S 1 = [10,90]S 2 = 50

33 Classic Solution: Example The game can be written as a two-player normal form game G = {S 1, S 2 ; u 1, u 2 } where S is strategy and u is payout. If subordinates assume dominants are using a uniform random mixed strategy making bids between 30 and 90, the Nash equilibrium will be a pure strategy response of 40. In turn, the best response by dominants is a pure strategy of 60. S 1 = [30,90]S 2 = 40S 1 = 60

34 Classic Solution: Example The game can be written as a two-player normal form game G = {S 1, S 2 ; u 1, u 2 } where S is strategy and u is payout. One may also solve this inversely for when dominants believe subordinates are playing deflated strategy. This example is insightful, but lacks depth (e.g., prejudice, population composition, etc).

35 Outline of Simulation: Summary I ran a total of 2,500 simulations of the bargaining game with varying parameters. ▫% dominant agents, % discriminators, magnitude of discrimination, agent life expectancy, agent memory length, % of wealth inherited, use of race-specific memory, % of encounters with discrimination, # of bidding encounters, cost of life, and level of random noise for agent memory and bids. Question: Can a small group of discriminators create racial disparities via the emergence of subtle prejudice and minority victims?

36 Results: Graphs Present several models with each building on the former. Average of 50 runs of simulation of 125 bidding encounters with varying parameters. Baseline: Do racial disparities in average bids and earnings emerge in simple model? Full: How do racial disparities in average bids and earnings emerge with discriminators, subtle prejudice and minority victims?

37 Results: Graphs Figure 1a: Average Bid by Racial Group—Baseline

38 Results: Graphs Figure 2a: Average Earnings by Racial Group—Baseline

39 Results: Graphs Figure 3a: Mean Racial Disparity in Bid & Earnings—Baseline

40 Results: Graphs Baseline: No racial difference.

41 Figure 3b: Mean Racial Disparity in Bid & Earnings Discrimination Model (10% increasing bid by 10 points) Results: Graphs *Significant Mean Bid Diff. of Mean Earnings Diff. of -1.47

42 Results: Graphs Baseline: No racial difference. Discrimination: A small number of discriminators leads to small but significant racial difference in mean bids and earnings.

43 Results: Graphs Figure 3c: Mean Racial Disparity in Bid & Earnings Discrimination, Dom. & Sub. Memory Model (Full)

44 Results: Graphs Baseline: No racial difference. Discrimination: A small number of discriminators leads to small but significant racial difference in mean bids and earnings. Full Model: Varying racial differences in bids and large discrepancies in earnings.

45 Results: Multivariate Analysis Regress “Group Earnings Disparity” on various simulation parameters. How do discrimination, subtle prejudice and minority victims each contribute to the emergence and maintenance racial inequality?

46 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

47 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

48 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

49 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

50 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

51 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

52 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

53 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

54 Table 3: Multivariate OLS Regression of Average Group Earnings a Disparities on Simulation Parameters—Games with Discrimination & Memory Model 1Model 2Model 3Model 4Model 5Model 6 Intercept *** *** *** *** *** Discrimination *** *** *** *** Dominant Memory *** *** *** *** Subordinate Memory *** *** *** *** # Rounds w/o Disc. N NN N Y Y # Rounds N NN N Y Y % Subord. N NN N Y Y % Subord. Squared N NN N Y Y % Wealth Inherited N NN N Y Y Initial Wealth Disp. N NN N Y Y Life Expectancy N NN NNY Memory Size N NN NNY Memory Noise N NN NNY Bid Noise N NN NNY Cost of Life N NN NNY r2r n2500 df Notes: * = p<0.05, **= p<0.01, ***= p<0.001.

55 Results: Multivariate Analysis Regress “Group Earnings Disparity” on various simulation parameters. How do discrimination, subtle prejudice and minority victims each contribute to the emergence and maintenance racial inequality? ▫Discrimination creates difference, but subtle prejudice and victims drive significant disparity. ▫Racial composition influences emergence—more minorities leads to smaller disparity.

56 Results: Multivariate Analysis It may be that the discriminators are fueling the inequality present in the system. Does removing discriminators change the inequality in the system?

57 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept Discrimination Dominant Memory Minority Memory # Rounds w/o Disc. # Rounds % Subord. % Subord. Squared % Wealth Inherited Initial Wealth Disparity Life Expectancy Memory Size Memory Noise Bid Noise Cost of Life r2r2 n df Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

58 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept Discrimination Dominant Memory Minority Memory # Rounds w/o Disc. # Rounds % Subord. % Subord. Squared % Wealth Inherited Initial Wealth Disparity Life Expectancy Memory Size Memory Noise Bid Noise Cost of Life r2r2 n df Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

59 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept Discrimination Dominant Memory Minority Memory # Rounds w/o Disc. # Rounds % Subord. % Subord. Squared % Wealth Inherited Initial Wealth Disparity Life Expectancy Memory Size Memory Noise Bid Noise Cost of Life r2r2 n df Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

60 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** Discrimination *** Dominant Memory *** Minority Memory *** # Rounds w/o Disc.Y # RoundsY % Subord.Y % Subord. SquaredY % Wealth Inherited Y Initial Wealth DisparityY Life ExpectancyY Memory SizeY Memory NoiseY Bid NoiseY Cost of LifeY r2r n463 df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

61 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** Discrimination *** Dominant Memory *** Minority Memory *** # Rounds w/o Disc.Y # RoundsY % Subord.Y % Subord. SquaredY % Wealth Inherited Y Initial Wealth DisparityY Life ExpectancyY Memory SizeY Memory NoiseY Bid NoiseY Cost of LifeY r2r n463 df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

62 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** Discrimination *** Dominant Memory *** Minority Memory *** # Rounds w/o Disc.Y # RoundsY % Subord.Y % Subord. SquaredY % Wealth Inherited Y Initial Wealth DisparityY Life ExpectancyY Memory SizeY Memory NoiseY Bid NoiseY Cost of LifeY r2r n463 df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

63 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** Discrimination *** Dominant Memory *** Minority Memory *** # Rounds w/o Disc.Y # RoundsY % Subord.Y % Subord. SquaredY % Wealth Inherited Y Initial Wealth DisparityY Life ExpectancyY Memory SizeY Memory NoiseY Bid NoiseY Cost of LifeY r2r n463 df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

64 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** Discrimination *** Dominant Memory *** *** Minority Memory *** *** # Rounds w/o Disc.YY # RoundsYY % Subord.YY % Subord. SquaredYY % Wealth Inherited YY Initial Wealth DisparityYY Life ExpectancyYY Memory SizeYY Memory NoiseYY Bid NoiseYY Cost of LifeYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

65 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** Discrimination *** Dominant Memory *** *** Minority Memory *** *** # Rounds w/o Disc.YY # RoundsYY % Subord.YY % Subord. SquaredYY % Wealth Inherited YY Initial Wealth DisparityYY Life ExpectancyYY Memory SizeYY Memory NoiseYY Bid NoiseYY Cost of LifeYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

66 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** Discrimination *** Dominant Memory *** *** Minority Memory *** *** # Rounds w/o Disc.YY # RoundsYY % Subord.YY % Subord. SquaredYY % Wealth Inherited YY Initial Wealth DisparityYY Life ExpectancyYY Memory SizeYY Memory NoiseYY Bid NoiseYY Cost of LifeYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

67 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** Discrimination *** Dominant Memory *** *** *** Minority Memory *** *** *** # Rounds w/o Disc.YYY # RoundsYYY % Subord.YYY % Subord. SquaredYYY % Wealth Inherited YYY Initial Wealth DisparityYYY Life ExpectancyYYY Memory SizeYYY Memory NoiseYYY Bid NoiseYYY Cost of LifeYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

68 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** Discrimination *** Dominant Memory *** *** *** Minority Memory *** *** *** # Rounds w/o Disc.YYY # RoundsYYY % Subord.YYY % Subord. SquaredYYY % Wealth Inherited YYY Initial Wealth DisparityYYY Life ExpectancyYYY Memory SizeYYY Memory NoiseYYY Bid NoiseYYY Cost of LifeYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

69 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** *** Discrimination *** Dominant Memory *** *** *** *** Minority Memory *** *** *** *** # Rounds w/o Disc.YYYY # RoundsYYYY % Subord.YYYY % Subord. SquaredYYYY % Wealth Inherited YYYY Initial Wealth DisparityYYYY Life ExpectancyYYYY Memory SizeYYYY Memory NoiseYYYY Bid NoiseYYYY Cost of LifeYYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

70 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** *** Discrimination *** Dominant Memory *** *** *** *** Minority Memory *** *** *** *** # Rounds w/o Disc.YYYY # RoundsYYYY % Subord.YYYY % Subord. SquaredYYYY % Wealth Inherited YYYY Initial Wealth DisparityYYYY Life ExpectancyYYYY Memory SizeYYYY Memory NoiseYYYY Bid NoiseYYYY Cost of LifeYYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

71 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** *** Discrimination *** Dominant Memory *** *** *** *** Minority Memory *** *** *** *** # Rounds w/o Disc.YYYY # RoundsYYYY % Subord.YYYY % Subord. SquaredYYYY % Wealth Inherited YYYY Initial Wealth DisparityYYYY Life ExpectancyYYYY Memory SizeYYYY Memory NoiseYYYY Bid NoiseYYYY Cost of LifeYYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

72 Table 5: Multivariate OLS Regressions of Recent Group Earnings a Disparities on Simulation Parameters—Stratified into 4 Groupings based on Bargaining Iterations since Discrimination Stopped Model A: LT b 10 Model B: 10-to-50 Model C: 50-to-100 Model D: GT b 100 Intercept *** *** *** *** Discrimination *** Dominant Memory *** *** *** *** Minority Memory *** *** *** *** # Rounds w/o Disc.YYYY # RoundsYYYY % Subord.YYYY % Subord. SquaredYYYY % Wealth Inherited YYYY Initial Wealth DisparityYYYY Life ExpectancyYYYY Memory SizeYYYY Memory NoiseYYYY Bid NoiseYYYY Cost of LifeYYYY r2r n df14 Notes: * = p<0.05, **= p<0.01, ***= p< Each of the four models (i.e., A through D) is mutually exclusive, and, as a result, highlights the relationsip between the variables under four distinct conditions pertaining to the time since discrimination in the game ceased to exist. a. Recent group earnings disparities refer to the average group difference in earnings over the final 10 encounters of the bargaining game. b. The terms LT and GT refer to “Less Than” and “Greater Than,” respectively.

73 Results: Multivariate Analysis It may be that the discriminators are fueling the inequality present in the system. Does removing discriminators change the inequality in the system? ▫Discrimination is only a significant determinant of inequality in early interactions. ▫Inequality persists in system as a result of subtle prejudice and minority victims. ▫Modest attenuation in contribution of these factors to inequality.

74 Discussion: Results Simulation results reveal that several factors can simultaneously (using bounded rationality) lead to racial inequality. ▫Discriminatory actors and institutions may set the table—put men on base. ▫Subtle prejudice (i.e., indifference) and minority victims may jointly drive in the runs. ▫Foundation for other simulations including:  Evolutionary Models (i.e., rationality of prejudice)  Incorporating Structure (i.e., variation in pot size)  Antiracists (i.e., innovators) and Sympathetic Ties (i.e., networks)

75 Discussion: Results Simulation results reveal that several factors can simultaneously (using bounded rationality) lead to racial inequality. It is only a simulation, but it reveals how dynamic process may work. ▫Example: A racist teacher/school system may give black students less attention/resources than whites. ▫Mechanism: White students/families demand more, and black students/families demand less of good. ▫Consequence: Larger disparity in achievement.

76 Discussion: Results Simulation results reveal that several factors can simultaneously (using bounded rationality) lead to racial inequality. It is only a simulation, but it reveals how dynamic process may work. ▫Example: Wages of Whiteness in a social milieu. ▫Mechanism: Whites demand more of public wage and blacks demand less of good. ▫Consequence: Larger disparity in public and psychological wage in other social spheres.

77 Conclusion It is very likely that several dynamic factors interact to shape the emergence and maintenance of racial inequality. Understanding how they interact is crucial step in determining the policy mechanisms that may undermine and eradicate inequality. Future researchers must (collaboratively) use an array of methods—simulation, ethnography, etc—to shed light on how behaviors (e.g., indifference) of actors and groups of actors shape racial inequality.

78 Big Bad Racists, Subtle Prejudice and Minority Victims: An Agent Based Analysis of the Dynamics of Racial Inequality Quincy Thomas Stewart Department of Sociology Indiana University CSDE, University of Washington February 2011

79 Supplemental Slides

80 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

81 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

82 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

83 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

84 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

85 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

86 Table 1: Descriptive Statistics for Simulation Parameters MeanMinimumMaximum Agent Life Exp % Discriminators Magnitude of Disc Disc. (Mag. x % Disc) % Subord. Agents Memory Size Memory Noise Bid Noise a Cost of Life % Wealth Inherited # Rounds # Rounds w/ Disc # Rounds w/o Disc Avg. Earning Disparity b Avg. Wealth Disparity b Recent Earning Disparity c Notes: The total number of simulations run (N) was 2,500. All of the parameters shown here–and others–randomly varied across the simulations based on a uniform probability distribution across a pre-specified range. a. Bid noise is measured as the absolute value of the change in bid per agent in the simulation. It is a product of the prevalence and magnitude of bid noise each run of the simulation. b. Average group disparities refer to the average group difference in earnings/wealth over the course of all encounters in a bargaining game. c. Recent earnings disparities refer to the average group difference in earnings over the final 10 encounters of a bargaining game.

87 Discussion: Results Simulation results reveal that several factors can simultaneously (and using bounded rationality) lead to racial inequality. It is only a simulation, but it reveals how dynamic process may work. Advantage: If theory plays out like we think, what is the aggregate result? ▫Gender Inequality: Hochschild’s Second Shift ▫What rational, local interactions lead to the emergence of women’s greater contribution to household duties?

88 Discussion: Results Simulation results reveal that several factors can simultaneously (and using bounded rationality) lead to racial inequality. It is only a simulation, but it reveals how dynamic process may work. Advantage: If theory plays out like we think, what is the aggregate result? ▫Bonilla-Silva: Racism without Racists ▫How might non-racist, rational behaviors (e.g., denial, indifference) contribute to the maintenance of inequality?


Download ppt "Big Bad Racists, Subtle Prejudice and Minority Victims: An Agent Based Analysis of the Dynamics of Racial Inequality Quincy Thomas Stewart Department of."

Similar presentations


Ads by Google