Six common misperceptions about teamwork

1

MGT 18: MANAGING DIVERSE TEAMS Table of Contents for Assigned Readings PROFESSOR: Mary A. McKay SUMMER I AND II 2016 All bolded items are in the reader. Others can be found via links embedded here and via TED (see Content folders by WEEK).

CLASS 1: THE BUSINESS CASE FOR DIVERSITY

1. Page, Scott E., “Making the Difference: Applying a Logic of Diversity.” Academy of Management

Perspectives (2007, November).

2. Banaji, M. R., Bazerman, M. H., & Chugh, D. (2003, December). “How (Un) Ethical Are You?” Harvard

Business Publishing Product #R0312D-PDF-ENG (skim for CLASS 1 but read thoroughly before CLASS 2)

3. Goldsmith, M. (2010, June 16). “Learn to Embrace the Tension of Diversity.”

http://blogs.hbr.org/goldsmith/2010/06/learn_to_embrace_the_tension_o.html

CLASS 2: SOCIAL IDENTITY THEORY: UNDERSTANDING INDIVIDUAL BEHAVIOR IN GROUPS AND TEAMS

4. Davidson, M. N. (2002, August). “Primer on Social Identity: Understanding Group Membership.”

Harvard Business Publishing Product #: UV0644-PDF-ENG

5. Sucher, S. J. (2007, November). “Differences at Work: The Individual Experience.” Harvard Business

Publishing Product # 608068-PDF-ENG

6. Sucher, S. J. (2007, November). “Social Identity Profile.” Harvard Business Publishing Product # 608091-

PDF-ENG

7. Ely, R. J., Vargas, I. (2004, December). “Managing a Public Image: Kevin Knight.” Harvard Business

Publishing Product # 405053-PDF-ENG

8. Polzer, J. T., Elfenbein, H. A. (2003, February). “Identity Issues in Teams.” Harvard Business School

Product # 403095-PDF-ENG

1

17

25

29

35

37

41

2

CLASS 3: AN INTRODUCTION TO GROUPS AND TEAMS

9. Katzenbach, Jon R., Smith, Douglas K. (2005, July). “The Discipline of Teams.” Harvard Business

Publishing Product # R0507P-PDF-ENG

10. Hackman, J. (2011, June 7). “Six Common Misperceptions About Team Work.”

http://blogs.hbr.org/cs/2011/06/six_common_misperceptions_abou.html

11. Coutu, D., & Beschloss, M. (2009, May). “Why Teams Don't Work.” Harvard Business Publishing Product

# R0905H-PDF-ENG

12. Huckman, R. S. and Staats, B. R. (2013, December). “The Hidden Benefits of Keeping Teams Intact.”

Harvard Business Publishing Product # F1312A-PDF-ENG

CLASS 4: UNDERSTAND BEFORE YOU ARE UNDERSTOOD: ESSENTIAL SKILLS FOR TEAM MEMBERSHIP

13. Edmondson, A. C. & Roloff, K. S. (2009, September). “Leveraging Diversity Through Psychological

Safety.” Harvard Business Publishing Product # ROT093-PDF-ENG.

14. Davidson, M. N. (2001). “Listening.” Darden Business Publishing Product # UVA-OB-0736.

15. Rosh, L. and Offermann, L. (2013, October). “Be Yourself, But Carefully.” Harvard Business Publishing

Product # R1310J-PDF-ENG

16. Connor, Jeffrey C. “It Wasn’t About Race. Or Was It?” Harvard Business Publishing # R00502-PDF-ENG.

CLASS 5: INTELLIGENCES: EMOTIONAL, SOCIAL AND CULTURAL

17. Goleman, Daniel (2004, January). “What Makes a Leader?” Harvard Business Publishing Product #

R0401H-PDF-ENG

18. Ross, Judith A. (2004, December). “Make Your Good Team Great.” Harvard Business Publishing Product

# U0812B-PDF-ENG

19. Goleman, D. & Boyatzis, R. (2008, September). “Social Intelligence and the Biology of Leadership.”

Harvard Business Publishing Product # R0809E-PDF-ENG

20. Earley, P. C. & Mosakowski, E. (2004, October). “Cultural Intelligence.” Harvard Business Publishing

Product # R0410J-PDF-ENG

51

61

69

73

79

87

93

101

113

115

123

3

CLASS 6: MIDTERM AND VIRTUAL/REMOTE TEAMS

21. Siebdrat, F., Hoegl, M., Ernst, H. (2009, July). “How to Manage Virtual Teams.” Harvard Business

Publishing Product # SMR322-PDF-ENG (CLASS 6 reading is covered on the final exam, not the midterm.)

CLASS 7: LEADING 21ST CENTURY TEAMS

22. Cardona, P. & Miller, Paddy. (2004, July). “Leadership in Work Teams.” Harvard Business Publishing

Product # IES087-PDF-ENG

23. Sitkin, S. B. & Hackman, J.R. “Developing Team Leadership: An Interview With Coach Mike Krzyzewski.”

Academy of Management Learning & Education, 2011, Vol. 10, No. 3, 494–501.

24. Useem, Michael. (2001, October). “Leadership Lessons of Mount Everest.” Harvard Business Publishing

Product # R0109B-PDF-ENG

25. Gallo, A. (2010, June 9). “Get Your Team to Stop Fighting and Start Working.”

http://blogs.hbr.org/hmu/2010/06/get-your-team-to-stop-fighting.html

26. Ellington-Booth, B. & Cates, K. L., “Growing Managers: Moving From Team Member to Team Leader.”

Harvard Business Publishing Product # KEL629-PDF-ENG.

CLASS 8: CULTURAL COMPETENCE

27. Corkindale, G. (2007, June 14). “Navigating Cultures.”

http://blogs.hbr.org/corkindale/2007/06/navigating_cultures.html

28. Brett, J. Behfar, K., Kern, M.C. (2006, November). “Managing Multicultural Teams.” Harvard Business

Publishing Product # R0611-PDF-ENG

29. Meyer, Erin (2014, May). “Navigating the Cultural Minefield.” Harvard Business Publishing Product #

R1405K-PDF-ENG

30. Meyer, Erin (2014, February). “How to Say This is Crap in Different Cultures.”

https://hbr.org/2014/02/how-to-say-this-is-crap-in-different-cultures/

31. Meyer, Erin. (2014, July). “Multicultural Teamwork: Accommodate Multiple Perspectives.”

http://knowledge.insead.edu/blog/insead-blog/multicultural-teamwork-accommodate-multiple-

perspectives-3489

131

137

157

165

173

185

193

4

CLASS 9: THE FUTURE OF TEAMS

32. Pentland, A. (2012, April). “The New Science of Building Great Teams.” Harvard Business School Product

# R1204C-PDF-ENG

33. Edmondson, A. (2012, April). “Teamwork on the Fly.” Harvard Business Publishing Product # R1204D-

PDF-ENG

34. Duhigg, C. “What Google Learned From Its Quest to Build the Perfect Team” (February 25, 2016).

http://www.nytimes.com/2016/02/28/magazine/what-google-learned-from-its-quest-to-build-the-

perfect-team.html?_r=0

CLASS 10: BECOMING A GLOBAL TEAM LEADER

35. Groysberg, B. and Connolly, K. (September 2013). “Great Leaders Who Make the Mix Work.” Harvard

Business Publishing Product # R1309D-PDF-ENG

36. Klau, M. “Twenty-first Century Leadership: It’s All About Values” (May 27, 2010).

http://blogs.hbr.org/imagining-the-future-of-leadership/2010/05/whose-values-the-gandhihitler.html

*Permission to reprint all selections granted to University Readers by the publishers for this individual course reader.

Please don’t photocopy – to do so would be a violation of copyright law.

199

211

221

E X C H A N G E

Making the Difference: Applying a Logic of Diversity by Scott E. Page

Executive Overview Each year, corporations spend billions of dollars on diversity training, education, and outreach. In this article, I explain why these efforts make good business sense and why organizations with diverse employees often perform best. I do this by describing a logic of diversity that relies on simple frameworks. Within these frameworks, I demonstrate how collections of individuals with diverse tools can outperform collections of high “ability” individuals at problem solving and predictive tasks. In problem solving, these benefits come not through portfolio effects but from superadditivity: Combinations of tools can be more powerful than the tools themselves. In predictive tasks, diversity in predictive models reduces collective error. It’s a mathe- matical fact that diversity matters just as much as highly accurate models when making collective predictions. This logic of diversity provides a foundation on which to construct practices that leverage differences to improve performance.

Along the moving sidewalks inside Paris’Charles de Gaulle airport, you cannot helpbut notice a sequence of HSBC advertise- ments meant to show diverse perspectives. One shows two identical pictures of a half-full glass of water. Across one glass, the caption reads moitié vide, under the other moitié plein. A second adver- tisement shows two identical pictures of an apple with a bite taken out. Défendu scrolls across one apple and recommandé across the other. These ads encourage us to think of HSBC as a firm that sees a problem from more than one perspective—and they also provide a welcome diversion from the inefficiencies of the airport. This multiple per- spective taking allows HSBC to add value, or so we are intended to believe.

The HSBC ads reflect a broader trend. Each

year, corporations spend billions related to pro- moting positive messages about diversity both in- ternally and externally. Why profit-seeking busi- nesses commit so many resources to constructing diverse workforces and creating welcoming orga- nizational cultures stems from two trends. First, businesses have become more global and hence more ethnically diverse. Firms sell to diverse con- sumers and hire from a diverse pool of candidates. The world, as has been said, is now flat, and consequently, organizations must cope with diver- sity. Second, the practice of work has become more team focused. The fixed hierarchy has given way to the evolving matrix (Mannix & Neale, 2006). In the past, welders positioned two stations apart on an assembly line need not get along. They need not validate one another’s worldview. The same cannot be said of a team of scriptwriters or oncologists, who must learn to understand the language and actions of one another.

This article is adapted from Scott E. Page’s book The Difference: How the Power of Diversity Creates Better Groups, Firms, Schools, and Societies, published in 2007 by Princeton University Press.

Scott E. Page (spage@umich.edu) is Professor of Complex Systems, Political Science, and Economics at the University of Michigan-Ann Arbor. He is also an external faculty member of the Santa Fe Institute.

6 NovemberAcademy of Management Perspectives

Copyright by the Academy of Management; all rights reserved. Contents may not be copied, e-mailed, posted to a listserv, or otherwise transmitted without the copyright holder’s express written permission. Users may print, download, or e-mail articles for individual use only.

1 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

This coincident emergence of diverse work- forces and team-based work makes leveraging di- versity a central concern of most organizations. A first question to ask is whether it’s a good thing from a business perspective. Does it hurt or help the bottom line? A substantial empirical literature addresses the question of whether diversity im- proves team performance (Williams & O’Reilly, 1998). A brief summary of that literature reveals that the answer depends on several factors. Par- ticularly important is what people believe (Ely & Thomas, 2001). If people do not believe in the value of diversity, then when part of a diverse team they’re not as likely to produce good out- comes. That expectations shape behavior and that behavior shapes outcomes should not come as a big surprise. How though to change expectations? How does an organization get its employees to believe that diversity leads to better outcomes? Taking out advertisements or printing up human resources documents with elaborate graphics and catchy tag lines won’t make it so. Managers and employees need, to quote Springsteen, “a reason to believe.”

Simple, clean logic can provide that reason. In this brief article, I derive links between cognitive differences among team members and better col- lective outcomes at specific tasks: problem solving and prediction. I build those links using conceptual frameworks that borrow from psychology, com- puter science, and economics. These links not only provide a foundation for understanding when, how, and if diversity produces benefits— the reason to believe—they also point toward specific policies and practices that can leverage the power of diversity.

The bottom line: Diversity can improve the bottom line. It may even matter as much as abil- ity.

DiversePerspectivesandHeuristics

I begin by formalizing the loose notion of a per-spective. No end of brochures and advertisementssing the praises of diverse perspectives, but what are they? Here, I define a perspective to be a representation of the set of the possible: the set of the semiconductor designs, welfare policies, or fall leather coats. Two people possess diverse perspec-

tives if they mentally represent the “set of the possible” differently. For example, one person might organize a collection of books by their au- thors’ last names; another person might organize those same books by color and size. One professor might arrange students’ names by class rank; an- other professor might order those same students alphabetically.

How a person represents the set of the possible determines “what is next to what.” For example, The Catcher in the Rye may seem rather discon- nected from Mao’s Little Red Book, but they are adjacent in a perspective that organizes books by color and size. Perspectives matter because “what is next to what” determines how a person locates new solutions. The linkage between perspectives and locating solutions can be clarified with an example. Suppose you are making butternut squash soup. You’ve pureed the sautéed onion and added the cream and baked squash, but the result tastes bland. Arrayed before you is an enormous spice rack. You’re thinking that perhaps you’ll add cumin. You sniff the cumin. It smells fine, but next to it, you see curry. So you smell the curry and decide it will be wonderful. You only try the curry because it sits adjacent to the cumin. Had the spices been arranged differently, say by color, you might have added cinnamon instead. What is next to what—in this case curry is next to cumin—determines where you look.

This same logic extends to almost any problem- solving situation: Two people with different perspec- tives test different potential improvements and increase the probability of an innovation.

Diverse perspectives may be the cause of most breakthroughs, but this does not mean that all diverse perspectives prove helpful. Someone who sees a problem from a different perspective will notice different candidate solutions. But those can- didate solutions need not improve the status quo. Diverse perspectives prove most valuable if they embed information relevant to the problem being solved. For example, in trying to increase fuel efficiency, a perspective that focuses on the weight of parts will likely yield good ideas. A perspective that considers their color probably won’t. Therefore, while organizations should en- courage bringing diverse perspectives to a prob-

2007 7Page

2 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

lem, they must also have some method for iden- tifying useful perspectives.

Perspectives describe how people see a prob- lem, but they do not fully capture the act of problem solving. When solving problems, people also use heuristics. Heuristics are methods or tools to find solutions. In my description of searching for a spice to add to the soup, I’ve assumed that you looked at adjacent jars. This is an example of a heuristic. Heuristics take many forms and vary in their sophistication from simple rules of thumb to complicated algorithms. To give a flavor for how heuristics operate, I describe here a famous simple heuristic known as do the opposite. In a classic episode of the television show Seinfeld, Jerry’s bumbling friend George Costanza comes to the realization that every decision he has made in his life has been the wrong one. This realization re- sults in an epiphany: He should do the opposite. He should do the reverse of whatever he thinks is best. If the rules in his head tell him to be kind, he should be rude. If they tell him to arrive early, he should show up late. If they tell him to dress casually, he should dress formally. The irony, of course, is that doing the opposite of what he thinks is right is the only “right” thing George has ever done, and by the end of the show he achieves personal and professional success. Diverse heuris- tics, like diverse perspectives, improve problem solving, but they do so in a different way.

Whereas perspectives change “what is next to what,” heuristics change how a person searches for solutions. Imagine two engineers working for a manufacturing company trying to improve the speed of an assembly line. The first engineer’s heuristic might be to try to break down individual tasks into smaller tasks. The second engineer’s heuristic may be to switch the order of the tasks. The two heuristics differ, and because they differ, they identify different candidate solutions, in- creasing the probability of a breakthrough.

This brief description of diverse perspectives and heuristics and how they operate reveals only part of the power of diversity. What I’ve shown is that by seeing problems differently (diverse per- spectives) and by looking for solutions in different ways (diverse heuristics), teams, groups, and orga- nizations can locate more potential innovations. I

now show that these individual improvements can be combined, creating superadditivity. Superaddi- tivity exists when the total exceeds the sum of the parts, when 1!1 " 3.

The idea that 1!1 " 3 may seem counterin- tuitive. Yet, when we add heuristics, either the two heuristics are the same (i.e., each points to the same solution, and therefore 1!1 " 1) or the two heuristics differ (in which case 1!1 " 3). Why three? Let’s do the math. Let’s go back to our assembly line problem. The first heuristic might advocate dividing a task that consists of six spot welds into two tasks. The second heuristic might advocate gluing on a piece of trim prior to the welding. The third heuristic comes from doing both—dividing the task and switching the order. Thus, any time you have two heuristics, you can create a third by combining the two heuristics. A similar logic shows that 1!1!1 " 7. Far from being a meaningless buzzword, superadditivity can be real, but only if people bring diverse perspec- tives and solutions to a problem.

The logic that diversity creates superadditive benefits differs from the standard portfolio analogy for diversity. According to the portfolio analogy, a firm wants diversity so as to be able to respond to diverse situations just as a stock investor wants a diverse portfolio of stocks. Just as a diverse port- folio guarantees a good payoff regardless of the state of the world, a diverse set of employees ensures that someone exists within the firm to handle any situation that arises. The portfolio analogy, though accurate in some cases, breaks down when applied to team-based problem solv- ing. There’s no give and take between stocks in a portfolio. One stock doesn’t say to another stock, “I never thought of the problem that way.” Nor can stocks build on solutions thought of by existing stocks. That just doesn’t make any sense.

I do not mean to imply that diversity does not provide insurance as suggested by the portfolio analogy. It does. However, the value of insurance against risk should not obscure the potentially larger superadditive benefits that accrue from hav- ing employees with diverse perspectives and heu- ristics.

Before moving on to more theoretical results, I

8 NovemberAcademy of Management Perspectives

3 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

want to inject a brief comment about identity diversity. In the framework that I have described, diverse perspectives and heuristics underpin diver- sity’s benefits. These more cognitively based no- tions of differences are distinct from identity- based distinctions such as race, gender, age, ethnicity, and so on. Though conceptually dis- tinct, cognitive and identity diversity often corre- late empirically. This correlation arises because perspectives and heuristics that people apply to problems do not come from thin air. They are the product of training, practice, and life experiences. How we see the world is informed and influenced by our values, our identities, and our cultures. People often reason by analogy. Each person’s unique set of life experiences provides the engine for these analogies. Diverse identities, therefore, often translate into diverse perspectives and heu- ristics.

ProblemSolving:Diversity TrumpsAbility

I have just outlined the basic logic for how di-verse perspectives and heuristics can improveproblem solving. I now want to push this logic a little further and touch on some formal results. First, I want to describe some experiments that I ran while an assistant professor at Caltech. For fun, I constructed a computer model of diverse problem solvers confronting a difficult problem. In my model, I represented diversity as differences in the ways problem solvers encoded the problem and searched for solutions, i.e., diverse perspectives and heuristics. I then stumbled upon a counterin- tuitive finding: Diverse groups of problem solv- ers—groups of people with diverse perspectives and heuristics—consistently outperformed groups composed of the best individual performers. So, if I formed two groups— one random (and therefore diverse) and one consisting of the best individual performers—the first group almost always did better. In other words, diversity trumped ability.

This counterintuitive finding led me to try to identify sufficient conditions for this to be true. What assumptions did I have to make for diverse groups, on average, to outperform groups of the best individuals? That turned out to be a rather

difficult task. So, following the logic of my own model, I enlisted the help of someone else, Lu Hong, a person with a different set of perspectives and heuristics than my own, to help me identify those conditions. Together, we found a set of conditions that, if they hold, imply that diversity trumps ability.

To show what these conditions are and why they matter, I will describe a simple model. Sup- pose that I begin with an initial pool of problem solvers from which I draw a random (e.g., diverse) team and a team of the best individual problem solvers. Each of these teams will have some mod- erate number of people, whereas the initial pool of people could be quite large. It could consist of everyone who works for a firm or every faculty member at a university. I then compare the col- lective performance of the team of the best prob- lem solvers against the collective performance of the randomly selected problem solvers.

Before I go too far, I want to remind you of the goal. Keep in mind that the diversity-trumps-abil- ity result won’t always hold. It holds given certain conditions. If, for example, the teams have only a single member, the team of the best problem solv- ers will consist of the best individual, and the team of random problem solvers will consist of a random person. Therefore, the first team will out- perform the second. Of course, in this case ability doesn’t trump diversity because the second team isn’t diverse. It has only one person. Thus, having the teams have more than one person will be a condition for the result to hold.

The question Lu and I asked was, what other conditions are needed? If those conditions are unrealistic, then we should not expect diversity to trump ability in practice. If those conditions seem mild, then perhaps we should. That’s one reason that we “do the math,” so that we can see when logic holds and when it doesn’t. Doing the math has other benefits as well, not the least of which is that we better understand how diversity produces benefits, which better enables us to leverage it in practice.

The first condition we identified relates to the difficulty of the problem. Easy problems don’t require diverse approaches.

2007 9Page

4 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

Condition 1: The Problem Is Difficult: No individual problem solver always locates the best solution.1

Without this condition, diversity cannot trump ability. If any individual problem solver always finds the best solution, then the collection of the best problem solvers (which by definition con- tains the best problem solver) always locates the best solution. For example, if we need to find the answer to a standard engineering problem, we can just ask an engineer who can give us the correct answer. We have no need to put together an interdisciplinary team. For harder problems, like designing an aircraft engine, we need a team. And that team needs diverse thinkers.

Condition 2: The Calculus Condition: The local optima of every problem solver can be written down in a list. In other words, all problem solvers are smart.

The second condition concerns the ability of the problem solvers. All of the possible problem solv- ers must have some ability to solve the problem. We cannot set loose a bunch of anthropologists and economists in the physics lab and hope they produce cold fusion. To formalize the idea that the problem solvers must have relevant cognitive tools, Lu and I assumed that the problem solvers got stuck in only a reasonable number of places. In the language of mathematics, such points are called local optima. We decided to call this restric- tion the Calculus Condition. We did this because people who know calculus can take derivatives, and therefore have a reasonable number of local optima. Here’s why. Think of a problem as creat- ing a mathematical function in which high values are good solutions. The derivative equals the slope of that function, which like the slope of a moun- tain is either positive (uphill), negative (down- hill), or zero (on a peak or a plateau). On a peak the derivative equals zero; the slope goes neither up nor down. Calculus enables a person to find points with derivatives equal to zero. Therefore, people who know calculus can find peaks. Econ- omists don’t know calculus when it comes to phys-

ics, but they probably do know calculus when asked about tax policy.

Condition 3: The Diversity Condition: Any solution other than the global optimum is not a local optimum for some non-zero percentage of problem solvers.

The third condition requires that for any proposed solution other than the global optimum, some problem solver can find an improvement on that solution. In formal terms, this means that the intersection of the problem solvers’ local optima contains only the global optimum. We call this the Diversity Condition, as it assumes diversity among the problem solvers. This condition does not say that given any solution some problem solver can immediately jump to the global opti- mum. That assumption would be much stronger and would rarely be the case. The assumption says, instead, that some problem exists who can find an improvement. That improvement need not be large. It need only be an improvement.

Condition4:Reasonably SizedTeamsDrawn from Lots of Potential Problem Solvers: The initial population of problem solvers must be large, and the teams of problem solvers working together must consist of more than a handful of problem solvers.

The final condition requires that the initial pool of problem solvers must be reasonably large and that the set of problem solvers who form the teams must not be too small. The logic behind this condition becomes clear in extreme cases. If the initial set consists of only 15 problem solvers, then the best ten should outperform a random ten. With so few problem solvers, the best ten cannot help but be diverse and therefore have different local optima. At the same time, the teams that work together must be large enough that the ran- dom collection can be sufficiently diverse. Think of it this way: We need to be selecting people from a big pool, and we need to be constructing teams that are big enough for diversity to come into play.

These four conditions—(a) the problem has to be hard, (b) the people have to be smart, (c) the people have to be diverse, and (d) the teams have to be reasonably big and chosen from a large pool—prove sufficient for diversity to trump abil-

1 If the best problem solver finds the optimal solution 99.9% of the time, the collection of randomly selected problem solvers will not outper- form the group of the best.

10 NovemberAcademy of Management Perspectives

5 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

ity. They’re not the only conditions under which the result holds, but if they’re satisfied, diversity will almost always trump ability.2

TheDiversity TrumpsAbility Theorem: Given conditions 1 to 4, a randomly selected collection of problem solvers almost always outperforms a collection of the best individual problem solvers.

This theorem is no mere metaphor, cute empirical anecdote, or small-sample empirical effect that may or may not be true with more trials. It’s a logical truth like the Pythagorean Theorem (Hong & Page, 2004). The reason diversity trumps ability is not deep: The best problem solv- ers likely have similar perspectives and heuristics. The random problem solvers bring diverse ways of thinking. Therefore, the best problem solvers all get stuck in the same places. The random problem solvers don’t.

The Diversity Trumps Ability Theorem implies that hiring people of high individual abilities may be less important than hiring people with diverse skills if those employees will work as part of a team. The logic of the theorem does not imply the irrelevance of ability. People need not remove those “my child is an honor student at Neil Arm- strong Junior High” bumper stickers from their minivans, nor should universities randomly allo- cate admission slots. Ability still matters, but so does diversity. And, as the theorem shows, once an ability threshold has been met, diversity mat- ters more than ability.

These comparisons between diversity and abil- ity require some care. Comparing ability to diver- sity is not unlike comparing a shiny apple to a fruit basket. Ability is a property of an individual—a nice shiny apple. Diversity is a property of a col- lection of people—a basket with many kinds of fruit. Rather than think of them as opposing con- cerns, we should see diversity and ability as com- plementary: The better the individual fruits, the better the fruit basket.

Problem solving is a central task for many or- ganizations, but it is not the only task for which diversity improves performance. Diversity also

improves collective predictions. As I now show, the so called “wisdom of crowds” comes not from having just smart people in the crowd, but from having smart people with diverse predictive models.

PredictiveModelsand theWisdomofCrowds

When a company decides which product tolaunch, when venture capitalists decidewhere to invest, and when stock analysts decide when to buy or sell, they’re making predic- tions. In putting together teams of people to make predictions, we might think that we want smart people, i.e., accurate individual predictors. And that’s true. But it’s also true that we want diverse predictors. We want people who differ in how they make predictions. Diversity should not be a second-order concern—multicolored sprinkles on the cake of ability; it merits equal billing. As in problem solving, diversity matters just as much as ability, and ideally, an organization or team would have an abundance of both.

To show the value of diversity in prediction, I need to define formally what I mean by a predictive model. Predictive models rely on interpretations. Interpretations are the mappings we make from the real world into categories. Categories, in turn, are conceptual boxes, or placeholders. For exam- ple, if I see a restaurant called Del Churro, I place it in the category Mexican restaurants. That may be true or it may not (most likely it is). I then predict that I’ll enjoy eating there because I like Mexican food.

These interpretations underpin statistical pre- dictive models as well. When stock analysts run regressions they restrict what dimensions, or at- tributes, they consider. In doing so, they create boxes. They use these boxes to construct a predic- tive model. In the best-selling book Blink, Mal- colm Gladwell describes several instances in which experts’ predictive models use very simple interpretations (2005). Gladwell loads his book with examples, including the story of an expert who instantly recognized a multimillion-dollar sculpture as fake even though scientific analysis had found otherwise, and one of an expert who can predict (using lots of analysis) whether a mar- ried couple will stay together just by looking at a2 In mathematics, the phrase almost always means with probability one.

2007 11Page

6 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

few dimensions of their relationship. Gladwell’s work shows the value of simple heuristics, an idea that receives a more formal treatment in the work of Gerd Gigerenzer and Peter Todd (1999).

As my example of the Mexican restaurant sug- gests, we need not think of predictive models as applying to just important events like stock mar- ket price changes or the causes of a disease. We apply predictive models almost every time we think. And our predictive models rely on inter- pretations. A popular predictive model for when a television show has reached its peak relies on categorizing episodes based on specific events. “Jumping the shark” (a reference to Fonzie jump- ing over a shark tank on his motorcycle, which signified the long decline of Happy Days, can take many forms. It could be a wedding that resolves building tension, or a death. It could be the ap- pearance of a special guest star. Nancy Reagan showing up on the television program Diff’rent Strokes was clearly jumping the shark.3

In making these predictions, be they about television shows or IPOs, people rely on predictive models that in turn rely on interpretations. Note that predictive models and interpretations differ from heuristics and perspectives. An interpreta- tion categorizes part of the world. It’s the mapping

of that big gray cloud into the category “rain cloud.” A predictive model tells us what we think will happen: “It looks like rain.” Predictive models are thoughts. Heuristics are courses of action. A heuristic tells us what to do: “It’s raining—let’s run for cover,” or what not to do: “We get just as wet by running, so let’s walk.” A perspective is not an action. It is a representation of the world. Each person possesses all of these: perspectives, heuris- tics, interpretations, and predictive models. And each of us differs in the particular collection of these tools that we hold inside our heads.

TheDiversityPrediction Theorem

Having constructed a model of how people make predictions, I can now turn to analyzing the role that diversity plays in the ability of a team, group, or crowd to make a prediction. I am going to consider some real-world data to show the impor- tance of diversity. If we are going to look at some data, we might as well look at something impor- tant. So let’s look at NFL draft predictions. (The actual reasons for considering this example are that the predictors have a stake in being correct and sufficient variance exists in the predictions to make the case interesting.)

The table below shows predictions for the top dozen picks in the 2005 NFL draft from seven3 See www.jumptheshark.com.

Table1 Experts’ Predictionsof2005NFLDraft

Player\Expert Wright Adler Fanball SNews Zimm Prisco Judge Crowd Smith 1 1 1 1 1 1 1 1.0 Brown 2 2 4 2 2 5 2 2.7 Edwards 3 3 2 7 3 2 3 3.3 Benson 4 4 13 4 8 4 8 5.9 Williams 8 5 5 5 4 13 4 6.4 Jones 16 9 6 8 6 6 9 8.1 Williamson 13 14 12 12 13 7 7 9.7 Rolle 6 6 8 10 9 8 6 7.9 Rogers 9 8 9 9 16 9 9 9.9 Williams 7 7 7 6 7 12 12 8.0 Ware 11 15 14 24 11 11 13 13.9 Merriman 12 11 3 11 12 10 11 10.1 Sq Error 158 89 210 235 112 82 75 34.4

12 NovemberAcademy of Management Perspectives

7 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

prognosticators.4 The players are listed in the or- der that they were selected. Each predictor pro- vides a ranking of the draftees. The names in the columns identify the predictors. These experts’ predictions came from detailed analyses. They don’t call them draft experts for nothing. These people—er, men—devote long days and nights to evaluating team needs, player skills, and a host of other factors.

The last row of the table shows the total squared error for each predictor. I calculated this number by summing the squared errors of the prediction for each player. To calculate that num- ber, I take the actual draft position of each player, subtract the predicted position, and square the difference. So, when the Sporting News (SNews) predicted Braylon Edwards to go seventh, their error on that pick was (3 – 7)2, which equals 16. If we look at the errors across the predictors, we see that they differ in their accuracy. The best has an error of 75. The worst predictor has an error of 235. The average of the individual errors equals 137.3. Comparing the accuracy of the individual predictors to the accuracy of the crowd reveals that the crowd is more accurate than any of its members.5 That won’t always be true, but the crowd will always be more accurate than its aver- age member. The wisdom of crowds, therefore, does exist, but the brilliance of crowds does only every once in a while.

By itself, this example doesn’t prove that diver- sity is valuable. It just shows that in this one instance a diverse group of predictors was more accurate than any member of the group. To show why diversity deserves the credit for the crowd’s success, I need to introduce one more statistical term, which I call the prediction diversity. Predic- tion diversity is nothing more than the variance of the experts’ predictions. A little math shows that in the NFL example, the prediction diversity equals 102.9. Notice the relationship between the crowd’s error (34.4), the average individual error (137.3), and the prediction diversity (102.9): Col-

lective error equals average error minus prediction diversity. This equality is not an artifact of our example. It is always true. I call this the Diversity Prediction Theorem.

TheDiversityPrediction Theorem: Collective Error " Average Individual Error # Prediction Diversity

Let’s think for a moment what this theorem means. It says that prediction diversity matters just as much as individual prediction accuracy when putting together a crowd of predictors. Equations such as this move us from some loose intuition that diverse points of view might be useful to an explicit characterization of how use- ful. Diversity isn’t just something of marginal value. Diversity matters just as much as individual ability. That is not a feel-good statement. It’s a mathematical fact.

Putting the Logic toWork

These two theorems—the Diversity TrumpsAbility Theorem and the Diversity PredictionTheorem—provide a foundation for claims that diversity provides benefits. That is, as they say, a good thing. But feeling good is not enough. Organizations can use this logic as more than a justification for policies that seek out diverse em- ployees. Organizations can leverage this logic to be more innovative and productive. In what fol- lows, I describe some direction for how this might be done.

Lesson#1:MoveBeyond thePortfolioAnalogy andPromote Interactions

As I mention above, many arguments for diversity lean on the portfolio analogy. For the same reason that a financial adviser advocates building a di- verse portfolio of stocks, a firm should have di- verse employees. The portfolio analogy sees diver- sity as a form of insurance. And it’s true that diversity often performs that function. However, the portfolio analogy misses a key part of the logic: the “superadditivity” of diverse tools. People have perspectives, heuristics, interpretations, and pre- dictive models.

When a collection of people work together to

4 The analysts are Scott Wright from NFL Countdown, James Adler from About.com, the Fanball Staff at Fanball.com, the Sporting News, Paul Zimmerman from Sports Illustrated, and Pete Prisco and Clark Judge from CBS Sportsline.

5 The crowd, in this case, is the sum of all of the analysts.

2007 13Page

8 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

solve a problem, and one person makes an im- provement, the others can often improve on this new solution even further. Problem solving is not the realization of a state but a process of innova- tion in which improvements build on improve- ments. This superadditivity can be found in many real-world examples. For example, attendees at the 1904 St. Louis World’s Fair could choose from a wide array of food choices: ice cream, cookies, cakes, waffles, and so on. One hot day during the fair, an ice cream vendor ran out of cups. A Syrian waffle vendor in the booth next door named Ernest Hami improvised by rolling up waffles to make cones. The rest, as they say, is history.6 The parts of the portfolio—the waffles and the ice cream—combined to create something new, and better: the ice cream cone. The key here is that the waffles and the ice cream interacted, and through that interaction produced a superadditive benefit. Diverse teams of people can produce similar gains, but they need to interact in order to do so.

Lesson#2: ContainMultitudes

The logic I have presented implies that rather than having a single perspective, interpretation, heuristic, or predictive model, people and organi- zations should have many. We must become Whitmanesque and contain multitudes. The ad- vantages of containing multitudes should be clear. Diverse perspectives and heuristics improve prob- lem solving. Diverse interpretations and predic- tive models lead to more accurate predictions. Crowds are not wise, but crowds of models are.

One way to maintain this diversity is to mimic evolution. From evolution, we know that diversity together with crude selective pressures can solve hard problems. In evolution, genetic mutation maintains diversity. Those mutations that in- crease fitness survive; those that do not fall by the wayside. The same effects occur within groups of people. Good attempts survive. Bad ones don’t. Experimentation can lead to a better “best” indi- vidual performer. More important, it can result in better collective performance. Increasing diversity

improves collective performance at prediction and problem solving.

Consider the following thought experiment. Suppose that we have to predict the amount of leather produced by a cow. This requires knowing the surface area of a cow. Even the complicated surfaces from calculus class are far more regular than your average cow. Fortunately, a book by John Harte (1985) on modeling offers a solution to this problem. We can imagine a spherical cow. I am going to ignore how we’d milk this spherical cow. Calculating the surface area of the spherical cow requires high school level math. That number won’t be correct. It’s an estimate, a prediction.

Someone else might decide to construct a dif- ferent predictive model. She might imagine cows shaped like boxes. She might even tape together a few Gateway computer boxes until she reached cow-like proportions. Someone else might imag- ine elliptical cows. Either of these two other mod- els may prove more accurate than the spherical cow model. If so, that’s great. That doesn’t mean that we should toss out the spherical cow model. The greatest benefit may well come from having multiple models that can be averaged into a crowd. The crowd of cow models may well be better than the best.

The amount of experimenting that makes sense depends upon the circumstances. Clearly, the lower the costs of experimenting and the more important the problem, the more we should ex- periment. We should also err on the side of more searching when a problem is connected to other problems. If we can understand how proteins fold, we can make headway on lots of other problems. Cognitive tools flow freely between domains. And by combing tools, we can find even larger break- throughs (Axelrod & Cohen, 2000).

Lesson#3: LookOutside: ConsultDissenters

When an organization confronts problems, it may lock in on a particular perspective. In an organi- zation, common perspectives facilitate communi- cation and the development of more advanced heuristics, but they also create common local op- tima. Thus, if one person gets stuck and if every- one sees the problem the same way, then everyone is stuck. Now, it could be that an organization’s

6 Unbeknownst to Hami, Italo Marchiony, a recent Italian immigrant to New York, had patented the ice cream cone in 1903.

14 NovemberAcademy of Management Perspectives

9 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

shared perspective makes every problem easy so that solutions are always optimal. Experience sug- gests otherwise. The only organizations that al- ways make optimal decisions lie nestled within the pages of introductory economics books. Given that organizations make mistakes and sometimes do so systematically, every so often they bring in people from the outside. These outside consult- ants are not necessarily smarter than the people who already work there. They’re just different. They bring with them different perspectives and heuristics that can improve outcomes.

An example from within the academy provides empirical grounding. Most universities organize themselves into departments. This hinders diver- sity for obvious reasons. These departments largely monitor themselves, so deans and provosts look for signs of external validation to see how these departments perform, such as the placement of graduate students, the number of publications by faculty in top journals, and the frequency of at- tempts by other schools to hire away faculty. These signals indicate if a department performs well, but they provide almost no clue for how a department could perform better. For this reason, universities periodically invite committees com- posed of scholars from other schools who work in the same discipline or a closely related one to provide suggestions for how the department might improve. How do they do this? They gather infor- mation about the current state of the department and advocate certain changes. Are these visitors more able than the people in the department? Probably not. But they are different. And they leverage those differences to make improvements.

These visiting committees can be thought of as a type of consultant. In fact, they are consultants (they just do not get paid as much as real consult- ants). And, clearly, this same line of thinking explains a benefit of consultants: They’re able to provide diversity to help departments or compa- nies improve. Sure, some companies trot out highly paid consultants in fancy suits to add cred- ibility to decisions that directors have already made—“Look, McKinsey agrees with me!” And yes, some consulting companies perform services that firms do not have the capacity or ability to do themselves, but many consultants do consult. And

when doing so they make improvements. (No, really.) Otherwise, there would not be so many consulting companies, and consultants wouldn’t be paid so much money. But the fact that these consultants add value does not mean that they are giants of the earth, smarter and more capable than others. A freshly minted MBA need not know more about dog food than Purina or more about manufacturing processes than General Motors or Toyota to add value. She need only be moderately capable and different. In difference lies value.

The careful reader will notice the subversive nature of this logic. I might have described visit- ing committees and consultants as experts, but I did not. Instead, I’ve described them as people who think differently. Visiting committees and consultants challenge the status quo. They are what Cass Sunstein (2003) might call “dissenters.” In politics dissenters identify new policy dimen- sions, and they force us to abandon our existing predictive models. Dissent is useful. Without it societies would falter. Organizational consult- ants—whether academic, nonprofit, or for prof- it—are dissenters too, paid dissenters.

Lesson#4: CreatePredictionMarkets

Given the potential wisdom of crowds, an organi- zation might benefit from creating internal infor- mation markets to make predictions. Information markets have substantial appeal to businesses and organizations. They can be highly accurate and low cost. Currently, most large companies and organizations hire people to construct models to predict future demands, sales, or, in the case of political parties, votes. Without these predictions long-range planning becomes difficult, if not im- possible. By creating an internal prediction mar- ket, an organization can leverage the wisdom of its own crowd. This prediction market could supple- ment or even replace the experts’ predictions. Some companies, such as Hewlett Packard and Google, have already done this. Chen and Plott (2002), for example, report that Hewlett Packard used managers to predict printer sales. The man- agers’ predictions proved to be as good as, and in some cases better than, the experts’.

Consider an auto company that wants to pre- dict what types of cars will sell best in the coming

2007 15Page

10 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

five years—a prediction that auto manufacturers address regularly, and one that they often get wrong. They could set up an internal market that includes all of the company’s engineers. This probably would not work. Their engineers proba- bly do not have much knowledge about consumer trends—they’re engineers! They may as well ask them to predict the Oscars. These information markets require that participants have reasonable models. For this reason, if these same engineers were asked to predict which of two vehicle designs would prove more durable, the information mar- ket would perform well. For this task, the engi- neers possess diverse and reasonable models (they understand different parts of the vehicle). Owing to that diversity, they can collectively predict well.

Lesson#5: FocusonRelevantDiversity

For organizations, what counts is relevant diver- sity, and again how much they should weigh di- versity relative to individual performance depends on the context. A firm that is hiring people for a job for which they have a well-defined ability measure may not reap many benefits from diver- sity. This would be true for a company hiring people to paint houses or to deliver messages by bicycle.

In contrast, consider a firm hiring people to design web pages. These potential hires would have to work together either directly or indirectly. In this case, the firm would want to consider diversity as much as ability. The firm should look for able people with diverse training, experiences, and identities. Unfortunately, human resources professionals can’t just look at someone and see her perspectives or heuristics.

The opaqueness of cognitive differences ex- plains why firms interview, administer tests, ask for recommendations, and sample previous work. They’re trying to make inferences about the tools applicants possess. Someone with a computer sci- ence degree probably knows more programming heuristics than someone with a degree in biology. And someone who has worked for five years sell- ing cars probably brings finer and more interesting interpretations of consumer types than someone who has been confined to a cubicle writing man-

uals for DVD players. But what of the undergrad riding the skateboard with all of those tattoos and piercings? Many people in corporate America would think, “He looks different from us.” Does that mean that they should hire him? The answer depends. If the kid on the skateboard knows the equivalent of calculus for the problem—if the job at hand is, say, designing bowling shirts or tennis shoes—they might want to think about doing so. But if the firm invests in derivatives and the skateboarder stopped taking math classes in fifth grade, then the firm would do better to look elsewhere.

Enlightened employers seek out diversity. Ow- ing to their success, Google can hire almost any- one they like. Google could just hire the top students from the engineering schools like MIT, Caltech, Stanford, Illinois, Michigan, Georgia Tech, and Cal-Berkeley. These people would all be smart, but they might be trained similarly. They also might have had similar college experi- ences. And they might be far from representative in the identity groups to which they belong.

Given that Google organizes itself in work teams that solve problems, success depends on both ability and diversity. That’s why Google doesn’t pursue a strategy of hiring only the people with the best grades from the best schools. In their own description of “who we’re looking for,” Google’s first criterion is diversity—“people with broad knowledge and expertise in many different areas of computer science and mathematics”—as is their last: “people with diverse interests and skills.” People who think alike get stuck. So Google samples widely. They look for diversity in training, experience, and identity. Computer sci- ence graduates from Santa Clara work alongside former math professors. But Google is also aware of the Calculus Condition. They seek diverse peo- ple with knowledge in mathematics and computer science. They’re not seeking poets. That said, if a good mathematical epidemiologist showed up at their door, they’d hire her.

Many identity attributes correlate with or influence how we think. Leveraging diversity re- quires more than greater racial and gender bal- ance. Forgetting this can result in lost opportuni- ties. The United States Army has substantial

16 NovemberAcademy of Management Perspectives

11 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

identity diversity at every rank. But because of the hierarchical nature of the military, they do not have much age diversity within a rank, so the people making the same kinds of decisions and giving advice to the same people are likely all about the same age. This reduces perspective and predictive model diversity. Some of the strongest evidence in all of the empirical diversity literature relates to demographic diversity. Those who arrive at the same time think the same way (Pfeffer, 1982). Therefore, maintaining age diversity can be crucial to success.

Firms might also test applicants for cognitive diversity relative to one another and to their cur- rent employees. Testing for diversity isn’t as hard as it sounds. One consulting company asks job applicants to predict the annual sales for a stan- dard household product, something like rubber bands, peanut butter, lug nuts, or size C batteries. This company wants to identify applicants who understand that total demand equals the sum of individual demands (recall the Calculus Condi- tion in the Diversity Trumps Ability theorem). They also want to identify people who think dif- ferently. They achieve both goals by learning if and how the applicants segment the market of consumers. Among those applicants who get the question about the C batteries, those who parse households in interesting ways, such as households with male children, have a good chance of getting hired. Those applicants who divide the country into regions probably do not. And yet the com- pany wouldn’t want all people who identified young boys as big users of batteries. They’d want some people who identified other market seg- ments, like campers. Asking silly questions doesn’t just get silly answers, it reveals diversity in think- ing. That’s why Google asks prospective employ- ees how many golf balls fit in a school bus.

Lesson#6: The SamuelPaulBowieCaveat

We can go too far in pursuing skill difference. In our pursuit of diversity, we must keep in mind the need to balance diversity with ability. We need only recall the 1984 NBA draft, in which the Portland Trail Blazers picked Samuel Paul Bowie, a seven-foot center from Kentucky, over a forward from North Carolina named Michael Jordan, per-

haps the greatest basketball player of all time. Reasons for the Bowie pick vary. Some claim that Jordan’s talents had been obscured by North Caro- lina’s team-oriented style of play.

I’m willing to cut the Blazers some slack. Port- land’s error could have resulted from having the wrong predictive model. Portland executives had reason to believe in the value of a good center. They had won a title just a few years earlier with an injury-prone Bill Walton at the pivot. Further supporting their case, only one team from 1959 to 1984 had won an NBA title without an all-star center. Add to this the fact that Portland already had an excellent tandem at small forward and shooting guard—Clyde Drexler and Jim Paxson— and the Bowie decision looks reasonable. But with the benefit of hindsight, choosing Bowie, an ex- ample of choosing diversity over ability, looks silly. If Michael Jordan’s available, draft him. Sometimes ability trumps diversity.

Lesson#7:Avoid Lumpingby Identityand Stereotyping

Employers often use identity as a crude proxy for cognitive diversity. And it’s true that the types of cognitive diversity that I’ve discussed correlate with identity. Even so, organizations probably can do better than to rely on coarse identity classifi- cations to categorize people. People are multifac- eted and multi-tooled. We all have different ex- periences and training as well as different identities. Those experiential and training differ- ences also translate into diverse toolboxes.

Mapping people into identity groups often over-lumps. Placing a recent immigrant from Nairobi, Kenya; a grandson of a sharecropper from the Mississippi Delta; and the daughter of a den- tist from Barrington, Illinois, into the same cate- gory—African Americans—obscures differences, as does placing the granddaughter of a miner from Copper Harbor, Michigan; a son of Gloria Vanderbilt (that would be Anderson Cooper); and a recently married former au pair from Lithua- nia into the box labeled “non-Hispanic white.” Similarly, having an Asian American box that lumps together people whose ancestors came from Singapore, Malaysia, China, Japan, and Korea bunches together diverse cultural identities. Each

2007 17Page

12 This course pack is for use in Professor McKay’s MGT 18 Summer 2016 course. Further reproduction is prohibited.

of these lumps of people, if unpacked, would prove cognitively diverse.

This lumping also ignores combinations of identities. A group consisting of five French men, three Korean men, two Kenyan women, and a woman from Singapore contains a man and Ken- yans but it does not contain Kenyan men and therefore may not be able to look at the problem in the same way that a Kenyan man might. And again, there is no single way in which a Kenyan man would look at a problem.

This insight also can be used to temper our enthusiasm for pipelines used to recruit minorities. These programs nurture potential employees or students from underrepresented groups. They may improve numbers, but they can limit the amount of cognitive diversity that a firm gets. By hiring only African American engineers who graduated from Berkeley and attended the same summer internship program, a company like Cisco sacri- fices cognitive diversity on the altar of identity accounting. Their employees look different, but they may not think differently. Thus, the use of pipelines probably has a negative effect on the benefits of diversity. It probably reduces the per- formance of identity-diverse firms. The greater identity diversity gained through the pipeline could be more than offset by their lack of experi- ential, demographic, or training diversity. Far bet- ter that Cisco forms a consortium of companies to create multiple pipelines to obtain what might be called within-lump diversity. Or even better, per- haps society might be structured in such a way that those pipelines are not needed.

Lumping people by identity group creates ste- reotypes and stigmatization. Many people think men are smarter than women, that people who grew up on farms work harder, and that Italians can cook better than the English. We describe people as typical Europeans or as fraternity boys. These stereotypes are predictive models. They place people in categories and make predictions based on those categorizations. If informed by lots of experience, these predictive models may be more accurate than not, provided we’ve lumped correctly. It is probably empirically true that on average, Italians probably are better cooks than English people, and frat boys do eat a lot of food

(and drink a lot of beer). But some do not. No evidence suggests that men are on average smarter than women.

Stereotypes, therefore, are to be avoided. In addition to being crude predictive models, they create three other problems. First, because stereo- types are predictive models about people, and not about physical phenomena, they can influence behavior and become self-reinforcing (Jackson & Fryer, 2002). People may evaluate women as less effective than men at task performance, even if by objective standards the women perform as well. This might happen if enough people carry around a crude predictive model that says that men are better workers than women. This can reduce in- centives for women to work hard, and thus the stereotypes become self-fulfilling. Any stereo- type—that Asians do better in math, that Indians are better spellers, that British people are wittier, or that African Americans are more creative— can induce self-fulfilling behavior. If we make stereotypical inferences about people who belong to an identity group, we reduce their incentives to accumulate tools outside these stereotypes. We limit opportunity. To use Glenn Loury’s phrasing (2000; 2002), stereotypic predictive models stig- matize.