The Diversity Bonus

0:06 hi 0:07 my name is scott paige and i'm a 0:08 professor at the raw school of business 0:10 at the university of michigan 0:11 what i want to talk to you today about 0:12 is how is it when is it 0:15 that diverse groups outperform 0:16 homogeneous groups 0:18 and the logic that i'm going to present 0:20 is something that i call diversity 0:21 bonuses 0:22 it explains why is it that diverse 0:24 groups gives you that extra 5 0:26 10 over homogeneous groups 0:29 and the idea is going to be this is the 0:31 diversity ends up being really 0:33 beneficial 0:33 when the task before you the problem at 0:35 hand is really complex 0:37 if it's easy you don't need a team and 0:39 you don't need a diverse team 0:41 it's when the problems are hard so what 0:43 do i mean by complex 0:44 by complex i mean something that's 0:46 difficult to explain or 0:48 engineer or evolve or predict but to 0:50 think of the global economy 0:52 that's complex if you think of middle 0:53 school that's complex you think of the 0:55 brain 0:56 that's complex if i think about 0:58 something like you know cutting a 2x4 to 1:00 a specific length 1:01 that's not complex and i can just have 1:03 one competent person do that job 1:05 so if i look at trying to predict or 1:07 forecast but in the economy 1:08 some things are going to be complex like 1:10 oil prices 1:12 and other things like oil production are 1:14 going to be less complex 1:15 and so where we need to put teams 1:16 together where we get these bonuses and 1:18 diversity 1:19 are on the more complex tasks now to 1:21 walk you through the logic 1:23 it's not going to be some sort of like 1:25 fits all and now we sprinkle on some 1:27 diversity 1:27 and we do a lot better we actually have 1:29 to have a mechanistic understanding of 1:31 what is the nature of the task 1:33 what is the type of diversity and how 1:35 does the type of diversity make us 1:36 better at that task 1:38 so let's start with prediction or 1:40 forecasting so now we've got it is we've 1:41 got to figure out should we hire this 1:43 person 1:43 what's going to happen in this market 1:45 what's possibly going to happen in this 1:46 region of the world 1:48 and what we want to do is we want to see 1:49 what are the advantages of putting a 1:50 diverse group of people 1:51 together in a room over just having a 1:53 whole bunch of like homogeneous people 1:55 or maybe just one really smart person 1:58 so let me start with some data google 2:00 which gets about four million job 2:02 applicants a year 2:03 has to figure out a way to sort of sort 2:04 through those so what they do at first 2:06 pass is use a computer algorithm so what 2:08 you see in this graph at the 50 percent 2:09 line 2:10 is sort of if you just took someone from 2:12 the algorithm half would be above 2:13 average half would be below average 2:15 if they just have one person interview 2:18 they get a bonus up to 76 so they're 26 2:21 percent better than the algorithm 2:23 now the point isn't that the person is 2:24 smarter than the algorithm the point is 2:26 that the person is different from the 2:27 algorithm 2:28 now they had a second interviewer it 2:30 goes from 76 to 81 2:32 a third from 81 to 84 a fourth from 84 2:35 to 86. 2:36 so what you get is by adding 2:38 interviewers adding this diversity 2:40 you end up having a 36 chance of hiring 2:43 someone better than average 2:44 whereas if you had one person you might 2:46 be 26 better than average 2:48 now the point here is when they're 2:49 adding more interviewers they're not 2:51 adding on people that are smarter 2:53 they're just adding on people that are 2:54 different each one of us when we 2:56 interview someone we see that person 2:57 through a different lens 2:59 and what you see is the crowd is more 3:01 accurate 3:02 than an individual now this phenomenon 3:04 of groups being better than individuals 3:06 is sometimes called the wisdom of crowds 3:08 that comes from a book by jim sirwiki 3:09 where he talks about this west of 3:11 england fat stock and poultry exhibition 3:13 held in the 19th century where 787 3:16 people guessed the weight of a steer 3:17 average guess was 197 pounds 3:20 actual weight of this steer was 198 3:22 pounds so this is a great anecdote 3:24 they're within a pound but we want to 3:26 move from anecdote to science 3:28 because i understand when do we need the 3:30 team when don't we 3:31 so what does the science say what the 3:33 science says is we can write down a 3:35 theorem a formal mathematical identity 3:37 that says the following 3:38 the crowd's error equals the average 3:40 error of the people in it 3:42 minus their diversity what does that 3:44 mean so the crowds there that trimming 3:46 standoff sees the crowd's prediction and 3:48 that greek 3:48 term theta that little you know circle 3:50 with a line there 3:51 that's the true value so the crowd's 3:54 error you might think 3:55 is the average of the individual errors 3:57 like your stocks in a portfolio but it 3:59 turns out it's not 4:00 the crowd is more accurate than the 4:02 average person and the amount 4:03 by which it's more accurate is the 4:05 diversity of their predictions because 4:07 if some predictions are high and some 4:08 predictions are low 4:09 you end up the crowd being more accurate 4:11 than the average person in it 4:13 now this isn't just some sort of like 4:14 feel-good idea this is a mathematical 4:17 fact 4:18 so if we look at the data from galton 4:19 sphere the crowd remember i said was up 4:22 by a pound well we 4:23 square the errors here because it's 4:24 statistics we do squared errors they're 4:26 actually up by 1.4 pounds 4:28 so when we square it we get two the 4:30 average air was over 70 pounds so when 4:32 we square that we get something a little 4:33 over 5 000. 4:35 the reason the crowd was correct is the 4:38 crowd was also diverse 4:39 so what you get is the crowd air is 4:41 really small because you're pretty smart 4:42 people 4:43 but those people happen to be diverse 4:45 this is the key 4:46 and this is also why complexity is so 4:48 important why do we need diversity on 4:50 complex problems 4:52 so if the pump is complex nobody's going 4:54 to be right right the average air is 4:55 going to be pretty big 4:57 we want the crowds here to be small so 4:59 the only way we can make the crowds air 5:01 small the average air is big 5:02 is by making the crowd diverse now this 5:06 happens and this plays out 5:08 everywhere right it's a mathematical 5:09 fact so let me show you some data from 5:10 economic forecasts 5:11 this is worked by professors at duke 5:13 university what they have is 28 000 5:16 forecasts by professional economist 5:18 six economic indicators the crowd is 21 5:22 better than the average economist now 5:23 it's not 28 000 economists on each 5:25 prediction so typically it's like 40 or 5:26 50 economists making predictions 5:28 the crowd was 21 better than the average 5:30 economist 5:31 now even more impressive than that if 5:33 you look at this red line 5:35 the red line starts at one that's the 5:36 year for an average economist 5:38 if you just add in a second economist 5:40 you do nine percent better 5:42 even more amazing than that take the 5:44 purple line that purple is the very best 5:47 economy so the best economist in this 5:48 data set 5:49 is 10 better than average economist when 5:52 you average her 5:53 and a lot of state it wasn't her was 5:54 abby cohen from goldman sachs 5:56 when you average her with the second 5:58 best economist 5:59 who by definition is not 10 better you 6:02 get 17 6:03 better that's what i mean by a bonus ten 6:06 percent plus nine percent gives you 6:07 seventeen percent 6:09 the reason why is because they're 6:10 diverse so this is the key logic 6:14 when you think about why you have 6:16 diversity for a portfolio 6:18 you do so because you want to minimize 6:19 risk right so if one stock gets eight 6:21 percent one gets 12 and one loses eight 6:23 right on average you get about five 6:25 percent but with predictors and this is 6:27 the data from the economist 6:28 the best is nine percent better the next 6:30 best nine percent better the next best 6:31 nine percent better the next best eight 6:32 percent better when you average them 6:34 22 percent better bonus so whatever 6:38 logic you have whatever reason you have 6:40 for diverse stocks 6:41 which is risk mitigation you know which 6:43 diversity you have in your stock 6:44 portfolio 6:45 you should have more diversity within 6:47 your employees because employees don't 6:48 just mitigate risk 6:50 they do that also but they give you a 6:52 bonus and that bonus in this case is 6:54 enormous right it's 6:55 the value add from the bonus is larger 6:57 than the value out from having the best 6:58 predictor in the market 7:00 all right that's prediction let's look 7:03 at problem solving the logic here's 7:05 entirely different but again we're going 7:06 to see a bonus 7:07 so let's think about how problem solving 7:08 works you've got some sort of difficult 7:10 problem you've got a bunch of you know 7:11 problem solvers that set loose on it 7:13 like they represent it in some way 7:14 they think about how they're going to 7:15 find better solutions i find an 7:17 improvement that lindy builds on my 7:18 improvement 7:19 what you can show here lu hong and i 7:21 showed this in a mathematical paper in 7:22 2001 7:24 that i think a whole bunch of really 7:25 competent problem solvers these are 7:26 represented as kind of mathematical 7:28 algorithms 7:29 i'm actually better off taking a random 7:31 group of competent ones 7:33 than taking a collection of the very 7:34 best and the reason why 7:36 is the collections very best aren't 7:37 diverse so when one of them gets stuck 7:39 at a point 7:39 since they're all representing the 7:41 problem in similar ways they all get 7:43 stuck at that same 7:44 point when i have a diverse group of 7:45 problem solvers they get stuck at 7:47 different points 7:48 now this was a somewhat controversial 7:49 paper so there's a bunch of follow-ons 7:52 and everybody who's done a follow-on 7:54 has found pretty much the exact same 7:55 thing and here's one by john kleinberg 7:57 and michael ragu from cornell 7:59 they said okay let's think of problem 8:00 solvers as people who sort of like come 8:02 in with a set of ideas 8:04 right and then what happens in the room 8:06 we combine those ideas an interesting 8:07 way we build up those ideas 8:09 what they show is that if you do 8:11 something interesting in the group as 8:12 long as it's like it's kind of like 8:13 adding up 8:14 and taking an average if you do 8:15 something interesting with these 8:16 solutions 8:17 then there's no test that you could 8:20 apply to individuals 8:21 such that the best team consists of 8:24 the people who scored highest on that 8:26 test so in other words if you're putting 8:27 together a team 8:29 you can't put together a team thinking 8:30 in terms of individual performance you 8:32 want to think of that team in terms of 8:33 diversity 8:35 now these are theoretical results what 8:36 about empirically well this was so 8:38 sort of surprising to people that people 8:40 then tested this in a lot of cases and 8:42 one of the most amazing tests is 8:44 using iq tests so this is raven's 8:45 progressive iq test right here which is 8:47 the next pattern in the sequence 8:50 so this is a wonderful work where people 8:51 wanted to see would an iq test 8:54 work to put together a team to take an 8:57 iq test 8:58 and the answer is no if you want to put 9:01 together a team to take an iq test 9:02 you actually don't want the people with 9:04 the highest iq now why is that 9:06 but what you want is you want people who 9:08 have high contextual iq 9:10 relative to the other people in the 9:11 group what is contextual like 9:13 contextual iq is this suppose that like 9:15 i don't score that well in an iq test 9:17 but i happen to get the questions right 9:20 that that really smart people 9:22 get wrong so i get questions right that 9:24 like other people are sort of confused 9:25 on 9:26 then you want me in the group now yeah i 9:28 miss a lot of easy questions but that 9:30 doesn't matter because the rest of the 9:31 groups get it right 9:32 so you want sort of people whose correct 9:35 answers are correlated differently 9:37 right who get different correct answers 9:39 than the p than most of the people 9:41 so if what you're trying to do is take 9:43 an iq test as a team you actually don't 9:45 want the people who score highest 9:46 individually 9:47 you want people who think differently 9:50 all right third type of task creative 9:52 tasks suppose you're trying to come with 9:54 just creative solutions just generate 9:56 ideas 9:57 what about diversity here well let me 9:59 give an example this is something that 10:00 sometimes knows the brick test or the 10:01 alternative uses test we have to think 10:03 of 10:03 how many ideas can i think of you know 10:06 using a brick for 10:08 so i did a version of this with a 10:09 corporate client where i said okay 10:10 suppose you've got 10 000 plastic straws 10:12 sitting in the basement 10:13 what can you do with it right how many 10:15 ideas can you think of 10:17 this is known as the alternative uses 10:18 test the individual ability of a person 10:21 is how many ideas they can think of on 10:22 their own 10:23 the ability of the group is how many is 10:24 the group can think of so i did this 10:26 with the corporate client here's the 10:28 results so one person person a the 10:30 highest performing person thought of 31 10:32 ideas the lowest performing person 10:34 person j thought of five ideas you could 10:36 say okay a is the most creative b is the 10:38 second most creative and so on 10:39 but when i combine these people i get 10:41 103 10:43 ideas because what i can do is i can 10:44 sort of take the union of all their 10:45 ideas if there's a duplicate copy i 10:47 throw it out 10:48 and what you see is the group is far far 10:50 more creative 10:51 than any one person in the room and in 10:53 this case if a person be in person c 10:56 even though they have the same 10:57 individual diversity person c 10:59 actually thought of a lot more unique 11:01 ideas so relative to the group for since 11:03 he had more value than person b 11:05 even though on their own they were 11:07 equally valuable 11:09 so what we see in all these cases is the 11:12 sum 11:12 is better than the parts and it happens 11:14 in the form of bonuses 11:16 but let's think about this carefully 11:17 what is the form of diversity in 11:19 prediction it's different 11:20 categorizations of world different 11:22 models 11:22 and problem solving it's different 11:24 representations of the problems 11:26 different tools different analogies 11:27 different ways of kind of climbing the 11:28 hill thinking of little improvements 11:30 and at creative tasks it's different 11:32 sets of possibilities it's just what can 11:34 you think of 11:35 and then what is the logic in prediction 11:38 the logic is by averaging these diverse 11:40 models 11:40 you kind of eradicate biases right some 11:43 of the estimates that are too high and 11:44 some that are too low those kind of 11:45 cancel out 11:46 in problem solving it's this notion of 11:48 iteratively improving 11:49 so i'm stuck somewhere somebody else 11:51 thinks of something new 11:53 stuart kaufman calls this the adjacent 11:55 possible 11:56 i'm stuck i can't think of anything 11:57 better lindy can think of something new 11:59 her adjacent possible is different than 12:01 mine 12:02 making the group smarter on a creative 12:04 task 12:05 it's about the union of ideas right so 12:07 it's my ideas 12:08 lindy's ideas but somebody else's ideas 12:10 and the union of those sets 12:12 is more than any one person knows in 12:14 each case 12:15 diversity bonus in each case a different 12:18 logic 12:20 okay what about identity diversity 12:21 because this is a course on 12:23 diversity equity and inclusion what 12:25 about identity 12:26 because i've been talking about 12:27 different ways that people think 12:30 so you can think about identity 12:31 diversity who we are 12:33 affects how we think and how we think 12:36 leads to better outcomes but we can 12:38 unpack that a little bit more because 12:39 what do we mean by identity diversity we 12:41 mean things like gender race age sexual 12:43 orientation ethnicity social class where 12:45 we grew up religion 12:47 those things map into the information we 12:50 have the knowledge the heuristics we 12:51 develop how we represent the world the 12:53 causal models in our head our analogies 12:55 the books we read 12:56 the places we visit what's inside our 12:59 head 13:00 is this kind of messy bundle of who we 13:02 are so if you want 13:04 cognitive diversity you have to have 13:07 identity diversity and the mechanisms 13:09 which that play out are actually 13:10 four-fold so first as we just mentioned 13:12 identity diversity correlates with 13:13 cognitive diversity so if you want a 13:14 diverse room 13:15 in terms of how that people think you 13:17 need diversity and how people look 13:19 second people think differently and 13:21 harder 13:22 when they're around identity diverse 13:23 people so this has worked from catherine 13:25 phillips 13:26 if i'm in a room with people just like 13:27 myself i kind of take it easy 13:29 when i'm in a room with people who are 13:30 different from me i just think harder i 13:32 generate new ideas 13:34 third identity diversity influences what 13:37 people are experts in 13:38 so it's not just how we filter the world 13:41 in terms of through our identities 13:42 but also what really interests people is 13:45 strongly a function of who they are 13:47 and so our expertise is going to 13:48 correlate with what we know 13:51 and then last when you're drawing from a 13:53 diverse pool yes it's larger but it's 13:55 also 13:56 more diverse so one reason you want even 13:58 if there were no differences 14:00 in people based on identity diversity by 14:03 hiring from an identity diverse pool 14:05 yes you're getting more talent but 14:06 you're also getting more diverse talent 14:09 all right sum it up diversity 14:13 improves performance when the task is 14:15 hard when it's easy 14:17 someone can just handle it it improves 14:19 performance when the task is hard 14:21 and it's cognitive and identity 14:23 diversity that really drive this 14:25 because who we are affects how we think 14:30 last when you think about what's going 14:31 to give you sustainable competitive 14:32 advantage 14:33 it's going to be your performance on 14:35 these complex tasks 14:37 thank you very much 14:45 you