1 00:00:00,000 --> 00:00:02,465 [SQUEAKING] 2 00:00:02,465 --> 00:00:04,437 [RUSTLING] 3 00:00:04,437 --> 00:00:07,395 [CLICKING] 4 00:00:09,583 --> 00:00:10,750 FRANK SCHILLBACH: All right. 5 00:00:10,750 --> 00:00:11,792 I'm going to get started. 6 00:00:11,792 --> 00:00:15,250 Welcome to lecture 16 of 14.13. 7 00:00:15,250 --> 00:00:17,680 We talked a lot about utility from beliefs 8 00:00:17,680 --> 00:00:20,950 and how, in particular, anticipatory utility, 9 00:00:20,950 --> 00:00:23,650 utility about thinking about the future and what 10 00:00:23,650 --> 00:00:27,940 might happen in the future, may affect people's utility. 11 00:00:27,940 --> 00:00:30,040 And then how that in turn might affect 12 00:00:30,040 --> 00:00:33,370 how people, A, choose when to consume 13 00:00:33,370 --> 00:00:36,370 or when to engage in certain activities. 14 00:00:36,370 --> 00:00:39,490 Because anticipatory utility provides some motivation 15 00:00:39,490 --> 00:00:41,980 to push things forward, at least a little bit, 16 00:00:41,980 --> 00:00:44,930 so people can look forward to positive events. 17 00:00:44,930 --> 00:00:47,890 Second, we talked about information acquisition 18 00:00:47,890 --> 00:00:52,030 about the idea that people, if they had derived utility 19 00:00:52,030 --> 00:00:56,500 from beliefs, that might affect how much information they 20 00:00:56,500 --> 00:00:58,240 might want to acquire. 21 00:00:58,240 --> 00:00:59,950 In particular, we talked about the idea 22 00:00:59,950 --> 00:01:03,100 that when you think about potentially negative 23 00:01:03,100 --> 00:01:06,855 information that you might receive-- 24 00:01:06,855 --> 00:01:08,980 in particular, we talked about Huntington's disease 25 00:01:08,980 --> 00:01:12,790 where people had a negative health information-- 26 00:01:12,790 --> 00:01:16,510 that they might not want that information with the motivation 27 00:01:16,510 --> 00:01:18,970 that they would like to make themselves 28 00:01:18,970 --> 00:01:20,990 feel better about themselves. 29 00:01:20,990 --> 00:01:23,350 They might think that the healthier than they 30 00:01:23,350 --> 00:01:25,480 actually are, they might look forward 31 00:01:25,480 --> 00:01:28,150 to a healthy life at least for two more years. 32 00:01:28,150 --> 00:01:32,140 And that might depress their willingness 33 00:01:32,140 --> 00:01:35,350 to gather or receive information. 34 00:01:35,350 --> 00:01:40,000 Now we talked a little bit about then a model 35 00:01:40,000 --> 00:01:41,710 and how to think about this. 36 00:01:41,710 --> 00:01:45,760 First, we talked about a model where people did not 37 00:01:45,760 --> 00:01:48,550 have the choice whether they could manipulate 38 00:01:48,550 --> 00:01:49,900 the information themselves. 39 00:01:49,900 --> 00:01:52,660 So we just talked about when somebody 40 00:01:52,660 --> 00:01:55,690 wants to seek or reject information, 41 00:01:55,690 --> 00:01:57,793 but we have the constraint. 42 00:01:57,793 --> 00:01:58,960 We had imposed a constraint. 43 00:01:58,960 --> 00:02:00,400 Let me go back for a second. 44 00:02:00,400 --> 00:02:02,590 We had a constraint imposed where we said, well, 45 00:02:02,590 --> 00:02:05,332 the person needs to be-- 46 00:02:05,332 --> 00:02:08,770 have-- hold correct beliefs conditional on the information 47 00:02:08,770 --> 00:02:11,085 that they have been exposed to so far. 48 00:02:11,085 --> 00:02:13,210 So essentially, that was just an idea of, like, OK, 49 00:02:13,210 --> 00:02:17,410 here's some information that the person could gather. 50 00:02:17,410 --> 00:02:19,003 But the constraint was this number one 51 00:02:19,003 --> 00:02:20,920 that I'm showing you here is that beliefs need 52 00:02:20,920 --> 00:02:24,370 to be correct, as in like the person is a [INAUDIBLE] 53 00:02:24,370 --> 00:02:28,210 conditional on the information that they have received so far. 54 00:02:28,210 --> 00:02:30,895 And then we said, well, if you think about this, 55 00:02:30,895 --> 00:02:35,860 if this is just about wanting to gather information and be 56 00:02:35,860 --> 00:02:39,130 considered so far the case where the person could not actually 57 00:02:39,130 --> 00:02:41,200 affect future outcomes, so that's 58 00:02:41,200 --> 00:02:43,760 the case of Huntington's disease where there's no cure. 59 00:02:43,760 --> 00:02:45,320 There's nothing you can do about it. 60 00:02:45,320 --> 00:02:46,600 The question was just like, would you 61 00:02:46,600 --> 00:02:48,475 like to receive some information that's going 62 00:02:48,475 --> 00:02:50,800 to happen in the future or not? 63 00:02:50,800 --> 00:02:54,460 Then we-- and I'm not going to go to this in detail very much. 64 00:02:54,460 --> 00:02:58,125 We then looked at the conditions that 65 00:02:58,125 --> 00:03:01,240 are involved for when does the person want to know about this. 66 00:03:01,240 --> 00:03:02,800 And the condition was essentially 67 00:03:02,800 --> 00:03:06,190 about this function of f of p, where f of p 68 00:03:06,190 --> 00:03:09,940 was essentially the utility derived from beliefs. 69 00:03:09,940 --> 00:03:13,110 And the condition was either if f is concave or f is convex. 70 00:03:13,110 --> 00:03:16,180 If f was concave, the person is information-averse. 71 00:03:16,180 --> 00:03:19,570 If f was convex, the person is information-loving 72 00:03:19,570 --> 00:03:22,540 and would like to find out about those things. 73 00:03:22,540 --> 00:03:24,430 Now then we said, well, consider now also 74 00:03:24,430 --> 00:03:26,440 the possibility that people might 75 00:03:26,440 --> 00:03:28,720 be able to manipulate their beliefs. 76 00:03:28,720 --> 00:03:32,050 So you might say, what are the reasons? 77 00:03:32,050 --> 00:03:34,713 Why might you ever want to hold correct beliefs? 78 00:03:34,713 --> 00:03:36,130 And in the above framework, if you 79 00:03:36,130 --> 00:03:37,750 could choose what your p is, there's 80 00:03:37,750 --> 00:03:41,800 really no reason whatsoever to not choose p equals 1. 81 00:03:41,800 --> 00:03:44,200 Remember, p here was the probability 82 00:03:44,200 --> 00:03:48,070 of your health being good, of being HD, Huntington's Disease 83 00:03:48,070 --> 00:03:50,270 negative, of not having the disease. 84 00:03:50,270 --> 00:03:52,150 So if there's no negative consequences 85 00:03:52,150 --> 00:03:54,340 of what's going to happen in the future, 86 00:03:54,340 --> 00:03:55,990 surely you have all the incentives 87 00:03:55,990 --> 00:03:58,510 in the world to deceive yourself and make yourself think 88 00:03:58,510 --> 00:04:00,770 that you are healthy, right? 89 00:04:00,770 --> 00:04:02,927 And so now then the person was like, well-- 90 00:04:02,927 --> 00:04:04,510 and this is what happens [INAUDIBLE].. 91 00:04:04,510 --> 00:04:05,710 The expression that's here. 92 00:04:05,710 --> 00:04:09,490 There's essentially f of 1, and that's larger than f of p 93 00:04:09,490 --> 00:04:11,050 for any p that are there. 94 00:04:11,050 --> 00:04:12,270 And then if the future-- 95 00:04:12,270 --> 00:04:14,170 this is the term here-- what is actually 96 00:04:14,170 --> 00:04:16,029 going to happen in the future, if that's 97 00:04:16,029 --> 00:04:22,630 independent of your beliefs and the sense of that's 98 00:04:22,630 --> 00:04:26,740 going to happen anyway, you might as well 99 00:04:26,740 --> 00:04:30,970 make yourself believe that p is large. 100 00:04:30,970 --> 00:04:31,480 P is 1. 101 00:04:31,480 --> 00:04:32,710 You're healthy. 102 00:04:32,710 --> 00:04:36,350 Now then what we left things at was the question of, 103 00:04:36,350 --> 00:04:39,310 why might you not want to choose f of one anyway? 104 00:04:39,310 --> 00:04:44,630 Why might you not want to say p equals 1? 105 00:04:44,630 --> 00:04:45,880 And that's a question to you. 106 00:04:56,090 --> 00:04:58,910 AUDIENCE: If your belief of what's going to happen 107 00:04:58,910 --> 00:05:03,200 might affect the future outcome in a negative way, you-- 108 00:05:07,710 --> 00:05:09,150 I guess if there were a treatment 109 00:05:09,150 --> 00:05:11,830 available for Huntington's disease, 110 00:05:11,830 --> 00:05:14,070 you would rather know whether you have it. 111 00:05:14,070 --> 00:05:15,070 FRANK SCHILLBACH: Right. 112 00:05:15,070 --> 00:05:18,180 So Huntington's is not the greatest of all examples 113 00:05:18,180 --> 00:05:23,130 that I chose to start with, because there's no cure. 114 00:05:23,130 --> 00:05:24,060 But you're right. 115 00:05:24,060 --> 00:05:25,950 For example, in particular, for diseases 116 00:05:25,950 --> 00:05:28,320 such as HIV or the like, surely it 117 00:05:28,320 --> 00:05:29,580 would be very helpful to know. 118 00:05:29,580 --> 00:05:32,880 Because then you can engage in proper treatment that might 119 00:05:32,880 --> 00:05:34,597 actually help you get better. 120 00:05:34,597 --> 00:05:36,180 Presumably, you only get the treatment 121 00:05:36,180 --> 00:05:38,880 if you actually know what the disease is like. 122 00:05:38,880 --> 00:05:40,643 In the specific Huntington's disease, 123 00:05:40,643 --> 00:05:42,060 we also talked about a few things. 124 00:05:42,060 --> 00:05:45,000 We talked about other actions that you might be able to take. 125 00:05:45,000 --> 00:05:52,680 For example, if you are able to save for retirement, 126 00:05:52,680 --> 00:05:55,300 if you are able to go to travels before you get sick. 127 00:05:55,300 --> 00:05:57,720 Or there was questions about whether people 128 00:05:57,720 --> 00:06:01,770 want to have children, questions about like what partners 129 00:06:01,770 --> 00:06:03,760 they want to be with and so on and so forth. 130 00:06:03,760 --> 00:06:06,330 So if there's a bunch of other economic choices, 131 00:06:06,330 --> 00:06:10,080 or other choices in life that depend on whether the person is 132 00:06:10,080 --> 00:06:12,930 positive or negative, it seems like that person 133 00:06:12,930 --> 00:06:14,070 should want to know. 134 00:06:14,070 --> 00:06:16,500 And then deluding yourself might get in the way 135 00:06:16,500 --> 00:06:18,960 of making optimal decisions. 136 00:06:18,960 --> 00:06:21,360 But broadly speaking, just to summarize, 137 00:06:21,360 --> 00:06:24,000 if there is no action item, as in like if there's 138 00:06:24,000 --> 00:06:27,370 some information about the future that where 139 00:06:27,370 --> 00:06:31,142 really, even if you knew the information, 140 00:06:31,142 --> 00:06:32,850 there's nothing you could do about what's 141 00:06:32,850 --> 00:06:35,220 going to happen in the future, it's 142 00:06:35,220 --> 00:06:38,160 not obvious why you not want to just delude yourself and think 143 00:06:38,160 --> 00:06:39,690 like things are rosier. 144 00:06:39,690 --> 00:06:43,020 The world is looking better than it actually is. 145 00:06:43,020 --> 00:06:45,060 Because it might just make you happier, at least 146 00:06:45,060 --> 00:06:47,640 in the moment, even if things in the future 147 00:06:47,640 --> 00:06:51,080 might turn out to be bad. 148 00:06:51,080 --> 00:06:54,030 And so this is what we already just discussed. 149 00:06:54,030 --> 00:06:57,850 So incorrect beliefs can lead to mistaken decisions. 150 00:06:57,850 --> 00:07:01,570 Well, that's correct, but sort of overly positive beliefs 151 00:07:01,570 --> 00:07:03,640 are an economically important indication 152 00:07:03,640 --> 00:07:05,200 of utility from beliefs. 153 00:07:05,200 --> 00:07:08,440 That is to say on the one hand, you might-- 154 00:07:08,440 --> 00:07:09,940 and this is what we just discussed-- 155 00:07:09,940 --> 00:07:12,065 you want to believe that you're healthy if it makes 156 00:07:12,065 --> 00:07:15,790 you feel better about yourself or the currently or the future. 157 00:07:15,790 --> 00:07:19,870 So you might want to convince yourself that that's the case. 158 00:07:19,870 --> 00:07:21,580 But this is what we just said. 159 00:07:21,580 --> 00:07:23,767 Overoptimism distorts decision making 160 00:07:23,767 --> 00:07:25,600 in some ways-- for example, health behavior, 161 00:07:25,600 --> 00:07:27,290 whether you want to seek treatment, 162 00:07:27,290 --> 00:07:28,870 but also whether you want to adjust 163 00:07:28,870 --> 00:07:31,040 to potentially bad events and some other ways 164 00:07:31,040 --> 00:07:34,540 in economic choices. 165 00:07:34,540 --> 00:07:37,000 So you might not want to-- so you might not 166 00:07:37,000 --> 00:07:44,410 want to be overoptimistic because of those distortions. 167 00:07:44,410 --> 00:07:46,330 And then the optimal expectations 168 00:07:46,330 --> 00:07:48,530 will then trade off these two things. 169 00:07:48,530 --> 00:07:51,190 On the one hand, you're healthy and maybe happier right now, 170 00:07:51,190 --> 00:07:52,480 thinking you're healthy. 171 00:07:52,480 --> 00:07:55,300 On the other hand-- or thinking that the future will be bright 172 00:07:55,300 --> 00:07:56,090 anyway. 173 00:07:56,090 --> 00:07:58,090 On the other hand, it might make-- 174 00:07:58,090 --> 00:08:00,620 change your choices in some bad way. 175 00:08:00,620 --> 00:08:02,990 Now, there was another reason that people mentioned, 176 00:08:02,990 --> 00:08:06,470 which was potential for disappointment, right? 177 00:08:06,470 --> 00:08:08,110 If you think the future is always 178 00:08:08,110 --> 00:08:12,310 going to be great, even if you can't effect it in any way 179 00:08:12,310 --> 00:08:14,570 well, at some point reality is going to set in, 180 00:08:14,570 --> 00:08:16,573 and then you might get really disappointed. 181 00:08:16,573 --> 00:08:18,490 We haven't really talked about this very much. 182 00:08:18,490 --> 00:08:19,948 That's a way in which you can think 183 00:08:19,948 --> 00:08:23,390 about perhaps people potentially choosing their reference point. 184 00:08:23,390 --> 00:08:26,420 So if you think people have referenced dependent utility, 185 00:08:26,420 --> 00:08:29,110 and if you're really overoptimistic 186 00:08:29,110 --> 00:08:32,110 very much in a way that allows you to potentially choose 187 00:08:32,110 --> 00:08:35,960 your expectations and your reference point. 188 00:08:35,960 --> 00:08:40,299 And then you might not want to be overly optimistic, 189 00:08:40,299 --> 00:08:42,220 because your reference point might be then 190 00:08:42,220 --> 00:08:44,440 too high and any outcomes that you're 191 00:08:44,440 --> 00:08:49,000 gonna eventually receive you might sort of view worse 192 00:08:49,000 --> 00:08:51,530 because you expect and believe things to happen. 193 00:08:51,530 --> 00:08:55,780 So to the extent that being true, or being overly 194 00:08:55,780 --> 00:08:58,780 optimistic leads to disappointment, 195 00:08:58,780 --> 00:09:02,770 you might not want to engage in overoptimism, even if there's 196 00:09:02,770 --> 00:09:05,470 nothing you can do about the outcomes, 197 00:09:05,470 --> 00:09:07,690 because of that potential disappointment. 198 00:09:07,690 --> 00:09:10,432 But in general for decision making 199 00:09:10,432 --> 00:09:11,890 with anticipatory utility, at least 200 00:09:11,890 --> 00:09:15,400 some overoptimism leads to higher utility than realism. 201 00:09:15,400 --> 00:09:19,000 Because essentially it makes you feel better in the moment. 202 00:09:19,000 --> 00:09:20,860 But there's this trade off potentially 203 00:09:20,860 --> 00:09:25,270 with how good you feel in the moment versus how disappointed 204 00:09:25,270 --> 00:09:30,880 you might be in the future in case you're too overoptimistic. 205 00:09:30,880 --> 00:09:32,230 Any questions about this? 206 00:09:36,590 --> 00:09:37,850 Oh, sorry. 207 00:09:41,160 --> 00:09:46,620 OK, so then let me show you some other overoptimistic beliefs 208 00:09:46,620 --> 00:09:51,630 in addition to the evidence [INAUDIBLE] showed you. 209 00:09:51,630 --> 00:09:54,030 So there's a classic study by Weinstein 210 00:09:54,030 --> 00:09:55,770 [INAUDIBLE] that asks students to make 211 00:09:55,770 --> 00:09:57,930 judgments of their students' chances 212 00:09:57,930 --> 00:09:59,220 for a number of outcomes. 213 00:09:59,220 --> 00:10:02,610 It's similar to the question that I had asked you 214 00:10:02,610 --> 00:10:04,208 at the beginning of class. 215 00:10:04,208 --> 00:10:06,000 There are two measures that Weinstein uses. 216 00:10:06,000 --> 00:10:08,160 One is what's called comparative judgment. 217 00:10:08,160 --> 00:10:11,040 This is-- excuse me. 218 00:10:13,970 --> 00:10:15,620 This is how much more or less likely 219 00:10:15,620 --> 00:10:18,020 the average student thinks the event will 220 00:10:18,020 --> 00:10:20,732 happen to them relative to the average student. 221 00:10:20,732 --> 00:10:22,190 And the second one is what's called 222 00:10:22,190 --> 00:10:25,333 the optimistic/pessimistic ratio, the number of students 223 00:10:25,333 --> 00:10:26,750 who think their chances are better 224 00:10:26,750 --> 00:10:29,690 than the average classmates divided by the number who 225 00:10:29,690 --> 00:10:31,170 think their chances are worse. 226 00:10:31,170 --> 00:10:33,735 So these are both sort of measures of overoptimism. 227 00:10:33,735 --> 00:10:35,360 They measure slightly different things, 228 00:10:35,360 --> 00:10:38,610 but they're broadly-- they're fairly highly correlated. 229 00:10:38,610 --> 00:10:41,270 So what does Weinstein find? 230 00:10:41,270 --> 00:10:44,112 Clear evidence of overoptimism when 231 00:10:44,112 --> 00:10:46,070 you look at different things about stuff that's 232 00:10:46,070 --> 00:10:49,010 going to happen in your life, owning a house, salary 233 00:10:49,010 --> 00:10:51,570 of larger than $10,000 a year-- 234 00:10:51,570 --> 00:10:52,910 I think this is 1980. 235 00:10:52,910 --> 00:10:54,510 That's been a while ago-- 236 00:10:54,510 --> 00:10:57,470 traveling to Europe, living past age 80 and on. 237 00:10:57,470 --> 00:11:00,590 The comparative judgment is always fairly high. 238 00:11:00,590 --> 00:11:03,630 So that number would be 0 if people were, on average, 239 00:11:03,630 --> 00:11:04,130 realistic. 240 00:11:04,130 --> 00:11:06,510 Remember, comparative judgment is how much more or less likely 241 00:11:06,510 --> 00:11:08,300 the average student thinks the event will 242 00:11:08,300 --> 00:11:11,810 happen to them relative to the average actual average student. 243 00:11:11,810 --> 00:11:16,070 And the optimistic/pessimistic ratio, 244 00:11:16,070 --> 00:11:21,240 that should be one if people were not overoptimistic. 245 00:11:21,240 --> 00:11:23,295 It's also clearly a large N1. 246 00:11:23,295 --> 00:11:24,920 Remember, that's the number of students 247 00:11:24,920 --> 00:11:27,650 who think the chances are better than the average classmates 248 00:11:27,650 --> 00:11:30,920 divided by the number who think their chances are worse. 249 00:11:30,920 --> 00:11:32,510 But it works for positive events, 250 00:11:32,510 --> 00:11:34,410 but also works for negative events. 251 00:11:34,410 --> 00:11:37,490 So when you ask people about drinking problems, suicide, 252 00:11:37,490 --> 00:11:39,950 divorce, heart attacks, all sorts of bad things, 253 00:11:39,950 --> 00:11:42,080 people think they're now, of course, 254 00:11:42,080 --> 00:11:48,470 less likely to happen to them, these kinds of events. 255 00:11:48,470 --> 00:11:50,660 And then they end up optimistic/pessimistic ratio, 256 00:11:50,660 --> 00:11:51,380 that's positive. 257 00:11:51,380 --> 00:11:55,040 That's just because it's like the optimist flips, so people 258 00:11:55,040 --> 00:11:57,650 are way more optimistic, or-- 259 00:11:57,650 --> 00:12:00,110 about-- they think they're not going 260 00:12:00,110 --> 00:12:03,380 to have drinking problems, get divorced, get heart attacks, 261 00:12:03,380 --> 00:12:03,900 and so on. 262 00:12:03,900 --> 00:12:07,730 So people tend to be very optimistic about their future 263 00:12:07,730 --> 00:12:11,240 lives compared with when you asked 264 00:12:11,240 --> 00:12:14,230 them to compare it to others. 265 00:12:14,230 --> 00:12:16,477 Now there's other examples of that. 266 00:12:16,477 --> 00:12:18,060 Couples believe there's a small chance 267 00:12:18,060 --> 00:12:20,960 that their marriage will end. 268 00:12:20,960 --> 00:12:22,710 Small business owners think their business 269 00:12:22,710 --> 00:12:26,370 is far more likely to succeed than their typical similar 270 00:12:26,370 --> 00:12:27,210 business. 271 00:12:27,210 --> 00:12:29,485 Smokers understand the health risks of smoking 272 00:12:29,485 --> 00:12:31,860 but don't believe this risk applies specifically to them. 273 00:12:31,860 --> 00:12:34,350 There's a long list of kind of these kinds of behaviors. 274 00:12:34,350 --> 00:12:37,950 People tend to have very-- hold rosy beliefs 275 00:12:37,950 --> 00:12:42,630 about their future, even in the presence of a sort 276 00:12:42,630 --> 00:12:43,900 of objective information. 277 00:12:43,900 --> 00:12:45,600 So even if you give people information 278 00:12:45,600 --> 00:12:47,400 about the health risks of smoking, 279 00:12:47,400 --> 00:12:48,850 they understand the health risks. 280 00:12:48,850 --> 00:12:50,308 But then they're like, well, that's 281 00:12:50,308 --> 00:12:51,390 not going to apply to me. 282 00:12:51,390 --> 00:12:54,960 Of course, there's no good reason to actually dismiss. 283 00:12:54,960 --> 00:12:57,090 It should apply to anybody. 284 00:12:57,090 --> 00:13:01,560 So people just want some things to be true presumably 285 00:13:01,560 --> 00:13:05,710 because it makes them happier in some ways. 286 00:13:05,710 --> 00:13:08,250 So then beyond future prospects, people also 287 00:13:08,250 --> 00:13:11,310 tend to have overly positive views about their abilities 288 00:13:11,310 --> 00:13:12,460 and traits. 289 00:13:12,460 --> 00:13:15,540 So, for example, 99% of drivers think they're 290 00:13:15,540 --> 00:13:17,280 better than the average driver. 291 00:13:17,280 --> 00:13:20,167 94% of professors at the University of Nebraska 292 00:13:20,167 --> 00:13:22,500 think they're better teachers than the average Professor 293 00:13:22,500 --> 00:13:23,640 at the University. 294 00:13:23,640 --> 00:13:26,880 I don't know what this number looks like at MIT, 295 00:13:26,880 --> 00:13:31,520 but probably it's larger than 50%, as well. 296 00:13:31,520 --> 00:13:34,500 So [INAUDIBLE],, you can find this kind of evidence 297 00:13:34,500 --> 00:13:35,910 from a range of domains. 298 00:13:35,910 --> 00:13:38,430 In some cases, there's some other potential explanation 299 00:13:38,430 --> 00:13:40,180 for this, but broadly speaking there's 300 00:13:40,180 --> 00:13:41,790 a pretty robust evidence that people 301 00:13:41,790 --> 00:13:45,630 are overconfident, overly positive about, 302 00:13:45,630 --> 00:13:48,360 A, what's going to happen to them in the future, and B, 303 00:13:48,360 --> 00:13:53,280 about thinking about their own skills and abilities. 304 00:13:53,280 --> 00:13:56,610 Now if you look at MIT students, however, in 305 00:13:56,610 --> 00:14:01,200 contrast when you ask about students different questions, 306 00:14:01,200 --> 00:14:03,390 and this is from a recent survey from a few years 307 00:14:03,390 --> 00:14:07,350 ago that MIT asks students. 308 00:14:07,350 --> 00:14:08,910 One question here is academically, 309 00:14:08,910 --> 00:14:11,300 I would consider myself above average at MIT. 310 00:14:13,830 --> 00:14:17,280 People say about 31% of students say they're above average. 311 00:14:17,280 --> 00:14:20,160 In particular, female students tend 312 00:14:20,160 --> 00:14:23,205 to think they're below average. 313 00:14:23,205 --> 00:14:25,330 People might have biased beliefs about the average. 314 00:14:25,330 --> 00:14:27,870 So what you think is an average MIT student might actually 315 00:14:27,870 --> 00:14:31,830 be like the fifth or whatever, some of the highest percentile, 316 00:14:31,830 --> 00:14:34,320 because essentially you're exposed to all the success 317 00:14:34,320 --> 00:14:37,830 and math Olympics and whatever, wins 318 00:14:37,830 --> 00:14:40,470 that people receive, which is just not what 319 00:14:40,470 --> 00:14:42,210 the average student is like. 320 00:14:42,210 --> 00:14:44,520 There's, of course, also very severe selection 321 00:14:44,520 --> 00:14:47,730 in terms of there's so many smart and brilliant people 322 00:14:47,730 --> 00:14:52,080 at MIT that, in a way, people's perception is quite biased. 323 00:14:52,080 --> 00:14:55,380 The questions here were asked about the average at MIT, 324 00:14:55,380 --> 00:14:59,490 but even there I think people might just 325 00:14:59,490 --> 00:15:01,860 be biased in terms of thinking what's the average. 326 00:15:01,860 --> 00:15:04,230 There might also be some worries about disappointment 327 00:15:04,230 --> 00:15:06,438 in the sense-- and this is what I was saying before-- 328 00:15:06,438 --> 00:15:09,600 when you think about being overconfident, 329 00:15:09,600 --> 00:15:10,810 maybe in some ways-- 330 00:15:10,810 --> 00:15:14,190 so one issue with, or one reason why you might not 331 00:15:14,190 --> 00:15:16,305 end up being overconfident for a long time 332 00:15:16,305 --> 00:15:18,150 is if you receive feedback. 333 00:15:18,150 --> 00:15:20,970 And at MIT, people receive lots of exams 334 00:15:20,970 --> 00:15:24,660 and so on where you can learn or interact with others, 335 00:15:24,660 --> 00:15:28,050 but you can notice how people are doing overall. 336 00:15:28,050 --> 00:15:30,450 And if you're worried about being disappointed, 337 00:15:30,450 --> 00:15:33,060 or you have been disappointed several times already, 338 00:15:33,060 --> 00:15:38,010 then thinking you are the smartest person in class 339 00:15:38,010 --> 00:15:40,470 might not be a good idea because you might get disappointed 340 00:15:40,470 --> 00:15:41,490 again. 341 00:15:41,490 --> 00:15:44,610 And then we'll talk a little bit more 342 00:15:44,610 --> 00:15:47,790 about gender in a different lecture later. 343 00:15:47,790 --> 00:15:50,632 That's a very common theme-- 344 00:15:50,632 --> 00:15:55,748 one second-- is that female MIT students, or in general women, 345 00:15:55,748 --> 00:15:57,540 but in particular also female MIT students, 346 00:15:57,540 --> 00:16:02,790 are particularly under-confident. 347 00:16:02,790 --> 00:16:05,640 We don't exactly know why that is, 348 00:16:05,640 --> 00:16:08,760 but it's a very robust finding, not just at MIT. 349 00:16:08,760 --> 00:16:11,550 In some ways, then, there's still a question-- 350 00:16:11,550 --> 00:16:14,150 so it's a very nice observation that in a way, 351 00:16:14,150 --> 00:16:18,100 beliefs can also be a form of self-motivation. 352 00:16:18,100 --> 00:16:22,050 So if you are really interested in academic success 353 00:16:22,050 --> 00:16:26,620 and being really good at exams and so on and so forth, 354 00:16:26,620 --> 00:16:28,950 if you are under-confident, that can 355 00:16:28,950 --> 00:16:30,450 be a motivator in saying that you're 356 00:16:30,450 --> 00:16:31,680 really worried about failing. 357 00:16:31,680 --> 00:16:33,810 You're worried about doing badly and so on. 358 00:16:33,810 --> 00:16:35,910 Then you work extremely hard, and then you 359 00:16:35,910 --> 00:16:36,870 do better in exams. 360 00:16:36,870 --> 00:16:39,610 You surprise yourself positively. 361 00:16:39,610 --> 00:16:43,650 That can be an important motivator. 362 00:16:43,650 --> 00:16:45,630 Of course, that could also go the other way. 363 00:16:45,630 --> 00:16:46,963 It could be really discouraging. 364 00:16:46,963 --> 00:16:49,643 If you think you're really terrible at everything, 365 00:16:49,643 --> 00:16:50,310 then you might-- 366 00:16:50,310 --> 00:16:53,760 at the end of the day, you might not study at all anymore, 367 00:16:53,760 --> 00:16:54,850 because what's the point? 368 00:16:54,850 --> 00:16:57,880 So there's a bit of a trade-off here as well. 369 00:16:57,880 --> 00:17:04,170 And it could be that if you think that people are-- when 370 00:17:04,170 --> 00:17:06,310 they're under-confident, that leads to motivation 371 00:17:06,310 --> 00:17:09,660 and it makes them better at school. 372 00:17:09,660 --> 00:17:11,940 Plus, they're going to be less disappointed 373 00:17:11,940 --> 00:17:13,800 and get positive surprises. 374 00:17:13,800 --> 00:17:15,079 That could be an explanation. 375 00:17:15,079 --> 00:17:16,829 I think there's a bit of a question again. 376 00:17:16,829 --> 00:17:20,880 Why is that the case at MIT and not in the overall population? 377 00:17:20,880 --> 00:17:23,250 But it might have to do with frequency 378 00:17:23,250 --> 00:17:26,532 of feedback, and school, and so on and so forth. 379 00:17:26,532 --> 00:17:28,740 And maybe also just with social norms or other things 380 00:17:28,740 --> 00:17:32,730 that maybe are somewhat trickier to explain. 381 00:17:32,730 --> 00:17:34,960 I was teaching Harvard undergrads as a grad student. 382 00:17:34,960 --> 00:17:38,700 I'm not so sure that Harvard students are under-confident, 383 00:17:38,700 --> 00:17:40,720 to be honest. 384 00:17:40,720 --> 00:17:43,350 I haven't seen the survey evidence on that, 385 00:17:43,350 --> 00:17:46,980 but my sense is that MIT students 386 00:17:46,980 --> 00:17:50,400 are more under-confident compared to Harvard students. 387 00:17:50,400 --> 00:17:51,450 Let's leave it at that. 388 00:17:54,245 --> 00:17:55,620 But I think it's true if you look 389 00:17:55,620 --> 00:17:58,500 at Caltech and other type of schools 390 00:17:58,500 --> 00:18:00,420 that are very similar to MIT. 391 00:18:00,420 --> 00:18:03,390 I think-- my guess is you'll find similar results 392 00:18:03,390 --> 00:18:04,740 in those kinds of surveys. 393 00:18:07,970 --> 00:18:08,630 OK. 394 00:18:08,630 --> 00:18:10,940 So the next thing we're talking about 395 00:18:10,940 --> 00:18:13,610 is very quickly is what's called ego utility. 396 00:18:13,610 --> 00:18:16,490 So partly I want to talk about anticipatory utility, which 397 00:18:16,490 --> 00:18:19,070 was the idea that people want to look forward 398 00:18:19,070 --> 00:18:21,560 to good things in life. 399 00:18:21,560 --> 00:18:26,900 And therefore, they're-- have inflated beliefs about 400 00:18:26,900 --> 00:18:28,590 what's going to happen in the future. 401 00:18:28,590 --> 00:18:34,870 A very related aspect is what's called ego utility, which is-- 402 00:18:34,870 --> 00:18:37,610 so just to recap, inflated beliefs 403 00:18:37,610 --> 00:18:39,770 might be explained by anticipatory utility 404 00:18:39,770 --> 00:18:44,180 essentially to say higher ability means better 405 00:18:44,180 --> 00:18:46,790 future prospects, and people want to convince themselves 406 00:18:46,790 --> 00:18:48,080 that they have high ability. 407 00:18:48,080 --> 00:18:50,210 And therefore, if you think you're really smart, 408 00:18:50,210 --> 00:18:52,970 that means good things are going to happen in the future. 409 00:18:52,970 --> 00:18:56,600 But it might also just be, like, right now, 410 00:18:56,600 --> 00:18:58,460 you're just feeling better about yourself, 411 00:18:58,460 --> 00:19:03,770 and therefore you have inflated beliefs about yourself. 412 00:19:03,770 --> 00:19:06,283 So that's often tricky to distinguish empirically 413 00:19:06,283 --> 00:19:07,700 in the sense of like, I might just 414 00:19:07,700 --> 00:19:12,200 feel like I'm good looking, smart, and so on and so forth, 415 00:19:12,200 --> 00:19:14,570 because it makes me feel good right now. 416 00:19:14,570 --> 00:19:16,520 Or I might want to hold those beliefs, 417 00:19:16,520 --> 00:19:20,540 because that means good things is going 418 00:19:20,540 --> 00:19:22,550 to happen in the future, and I will have 419 00:19:22,550 --> 00:19:25,290 a bright future going forward. 420 00:19:25,290 --> 00:19:29,300 So these things are slightly tricky to disentangle, 421 00:19:29,300 --> 00:19:32,630 and both of them probably play an important role 422 00:19:32,630 --> 00:19:34,940 at the end of the day. 423 00:19:34,940 --> 00:19:39,610 So now one thing I'm going to talk about a little bit more 424 00:19:39,610 --> 00:19:42,650 also later in the semester is about mental health. 425 00:19:42,650 --> 00:19:46,030 And so there is this literature that argues 426 00:19:46,030 --> 00:19:48,610 that positive illusions are-- 427 00:19:48,610 --> 00:19:51,700 promote, in fact, psychological well-being. 428 00:19:51,700 --> 00:19:53,710 So that's the argument that is here, 429 00:19:53,710 --> 00:19:58,120 that overconfident is, in fact, vital and important 430 00:19:58,120 --> 00:20:00,130 for maintaining mental health. 431 00:20:00,130 --> 00:20:05,470 So overconfident and that makes people happier. 432 00:20:05,470 --> 00:20:07,510 People are better able to care for others 433 00:20:07,510 --> 00:20:10,270 if they think they, themselves, are doing better. 434 00:20:10,270 --> 00:20:13,780 They're better at doing creative and productive work. 435 00:20:13,780 --> 00:20:16,540 That's also better over confident helps people 436 00:20:16,540 --> 00:20:19,090 manage negative feedback. 437 00:20:19,090 --> 00:20:22,510 So overall, all of those things are [INAUDIBLE] 438 00:20:22,510 --> 00:20:25,270 so this literature argues that overconfidence actually 439 00:20:25,270 --> 00:20:27,730 helps people with their everyday lives. 440 00:20:27,730 --> 00:20:29,410 In particular, helps people maintain 441 00:20:29,410 --> 00:20:32,360 a positive-- a good mental health. 442 00:20:32,360 --> 00:20:37,450 And there's this term called depressive realism, which 443 00:20:37,450 --> 00:20:40,990 is to say that realistic expectation might actually be 444 00:20:40,990 --> 00:20:42,280 detrimental to mental health. 445 00:20:42,280 --> 00:20:44,620 So that the world is just difficult in various ways 446 00:20:44,620 --> 00:20:49,880 and bad things are happening in many lives, 447 00:20:49,880 --> 00:20:54,503 and if people are not overly optimistic about what's 448 00:20:54,503 --> 00:20:56,170 going to happen in the future and so on, 449 00:20:56,170 --> 00:20:59,260 that might be actually bad for their mental health. 450 00:20:59,260 --> 00:21:02,410 We'll return to this issue on happiness 451 00:21:02,410 --> 00:21:04,110 and mental health in a later lecture, 452 00:21:04,110 --> 00:21:05,720 but I wanted to flag as these things. 453 00:21:05,720 --> 00:21:08,810 They're very much linked to each other. 454 00:21:08,810 --> 00:21:17,350 And another reason perhaps why overoptimism might be, in fact, 455 00:21:17,350 --> 00:21:22,390 good or not detrimental is because not only makes people 456 00:21:22,390 --> 00:21:24,790 happier, but it might also help people 457 00:21:24,790 --> 00:21:26,400 protect their mental health. 458 00:21:29,560 --> 00:21:30,060 OK. 459 00:21:30,060 --> 00:21:33,713 So then just very briefly, and this is more like-- 460 00:21:33,713 --> 00:21:35,630 there's some research in some of those things. 461 00:21:35,630 --> 00:21:37,640 Some of this is more anecdotal. 462 00:21:37,640 --> 00:21:39,080 What are some factors that affects 463 00:21:39,080 --> 00:21:41,550 the extent of positive biases? 464 00:21:41,550 --> 00:21:44,180 So people tend to have greater biases, in fact, with respect 465 00:21:44,180 --> 00:21:46,908 to prospects and traits that are personally important to them. 466 00:21:46,908 --> 00:21:48,450 On the one hand, you might say, well, 467 00:21:48,450 --> 00:21:51,020 people should also be better informed about those things. 468 00:21:51,020 --> 00:21:54,200 But to the extent that it affects their utility, 469 00:21:54,200 --> 00:21:55,550 they might be more biased. 470 00:21:55,550 --> 00:21:57,090 If it's stuff that's really important to you, 471 00:21:57,090 --> 00:21:59,423 you might be more biased because that's precisely what's 472 00:21:59,423 --> 00:22:01,700 going to make you happier. 473 00:22:01,700 --> 00:22:03,980 Available or imminent objective information 474 00:22:03,980 --> 00:22:06,980 tends to decrease biases as in like-- 475 00:22:06,980 --> 00:22:09,680 and this is what I was talking about with MIT students. 476 00:22:09,680 --> 00:22:11,930 When you think about academic performance, 477 00:22:11,930 --> 00:22:17,190 you get lots and lots of feedback every single semester. 478 00:22:17,190 --> 00:22:20,990 So in a way, it's more difficult to be overly optimistic, 479 00:22:20,990 --> 00:22:24,440 or to maintain an overly optimistic picture 480 00:22:24,440 --> 00:22:29,300 of your academic self if you've got all these objective signals 481 00:22:29,300 --> 00:22:30,587 over and over again. 482 00:22:30,587 --> 00:22:32,670 So one prediction here would be that, for example, 483 00:22:32,670 --> 00:22:34,970 if you looked at first, second, third, and fourth year 484 00:22:34,970 --> 00:22:40,550 students, that perhaps some of the under-confidence 485 00:22:40,550 --> 00:22:42,950 only comes later in the semester. 486 00:22:42,950 --> 00:22:47,370 Of course, there's other reasons why that might not be the case. 487 00:22:47,370 --> 00:22:49,710 Then if feedback about the prospect or any prospect 488 00:22:49,710 --> 00:22:51,710 is more ambiguous and subject to interpretation, 489 00:22:51,710 --> 00:22:53,362 biases tend to be greater. 490 00:22:53,362 --> 00:22:55,070 So in particular, people are particularly 491 00:22:55,070 --> 00:22:56,660 biased in situations where people 492 00:22:56,660 --> 00:22:59,240 get ambiguous feedback where you can 493 00:22:59,240 --> 00:23:00,680 interpret things either way. 494 00:23:00,680 --> 00:23:04,880 So either you were really smart and did something really great, 495 00:23:04,880 --> 00:23:06,200 or you're just lucky. 496 00:23:06,200 --> 00:23:09,050 And people tend to, in many things in life, 497 00:23:09,050 --> 00:23:11,240 people tend to interpret sort of lucky things 498 00:23:11,240 --> 00:23:14,390 that happened in their lives, tend to think 499 00:23:14,390 --> 00:23:16,840 that those are due to skills. 500 00:23:16,840 --> 00:23:20,330 And there's this very nice work in a book by Robert Frank-- 501 00:23:20,330 --> 00:23:23,570 he's at Cornell University-- who sort of argues that people tend 502 00:23:23,570 --> 00:23:27,170 to interpret a lot of things that happen in their lives that 503 00:23:27,170 --> 00:23:28,760 really are just mostly luck. 504 00:23:28,760 --> 00:23:32,600 They tend to think that it's due to their amazing abilities 505 00:23:32,600 --> 00:23:34,017 and so on. 506 00:23:34,017 --> 00:23:35,600 And that's the reason for that, often, 507 00:23:35,600 --> 00:23:38,030 is that a lot of this feedback, or a lot of this information 508 00:23:38,030 --> 00:23:38,572 is ambiguous. 509 00:23:38,572 --> 00:23:40,190 You can interpret it either way. 510 00:23:40,190 --> 00:23:42,950 And people like to think that good things are because they 511 00:23:42,950 --> 00:23:45,710 did really great, and bad things are because they just 512 00:23:45,710 --> 00:23:47,422 happen to be unlucky. 513 00:23:47,422 --> 00:23:49,130 If people-- and that's quite interesting. 514 00:23:49,130 --> 00:23:51,500 If people feel like they have control over the outcomes, 515 00:23:51,500 --> 00:23:55,460 biases tend to be greater. 516 00:23:55,460 --> 00:23:58,520 Expertise sometimes increases biases, 517 00:23:58,520 --> 00:24:01,040 but not for experts who get very good feedback. 518 00:24:01,040 --> 00:24:04,640 So meteorologists or the like who get lots of feedback 519 00:24:04,640 --> 00:24:08,325 all the time, they actually tend to not be overconfident, 520 00:24:08,325 --> 00:24:10,700 because, again, if you get so much feedback all the time, 521 00:24:10,700 --> 00:24:14,500 it's hard to sort of maintain your overoptimism. 522 00:24:14,500 --> 00:24:15,130 OK. 523 00:24:15,130 --> 00:24:17,080 So now one question you might have is, 524 00:24:17,080 --> 00:24:20,410 well, I showed you a bunch of data information 525 00:24:20,410 --> 00:24:23,860 on biased beliefs but not a lot on actual action. 526 00:24:23,860 --> 00:24:26,497 Of course, the testing behavior, and the Huntington's 527 00:24:26,497 --> 00:24:28,330 disease, and so on [INAUDIBLE] some actions. 528 00:24:28,330 --> 00:24:30,790 But a lot of the stuff on the Weinstein and other studies 529 00:24:30,790 --> 00:24:32,740 were just about self-appointed beliefs. 530 00:24:32,740 --> 00:24:34,660 And one question you might have is, well, 531 00:24:34,660 --> 00:24:35,800 this is about self reports. 532 00:24:35,800 --> 00:24:38,860 What about your preference on choices that people make? 533 00:24:38,860 --> 00:24:42,147 So there's some evidence that's from lab experiments, 534 00:24:42,147 --> 00:24:44,230 and there's some other evidence that I'll show you 535 00:24:44,230 --> 00:24:46,090 that is about actual choices. 536 00:24:46,090 --> 00:24:48,880 So one very nice experiment is the paper by Eil and Rao 537 00:24:48,880 --> 00:24:50,690 from 2010. 538 00:24:50,690 --> 00:24:53,650 The way this works is people are given feedback 539 00:24:53,650 --> 00:24:55,540 about people doing IQ tests, and they're 540 00:24:55,540 --> 00:24:58,240 rated according to their physical attractiveness 541 00:24:58,240 --> 00:25:01,570 by others in the study. 542 00:25:01,570 --> 00:25:04,030 And they get feedback about this IQ test 543 00:25:04,030 --> 00:25:06,100 score and their physical attractiveness. 544 00:25:06,100 --> 00:25:08,308 These two things are chosen presumably 545 00:25:08,308 --> 00:25:09,850 because people care a lot about them. 546 00:25:09,850 --> 00:25:11,500 People care about how smart they are. 547 00:25:11,500 --> 00:25:14,320 People also care about how good looking they are. 548 00:25:14,320 --> 00:25:16,060 And then so in the study, the authors 549 00:25:16,060 --> 00:25:19,540 elicit people's prior beliefs, the beliefs at the beginning 550 00:25:19,540 --> 00:25:23,270 before they receive information, about a rank between one 551 00:25:23,270 --> 00:25:23,770 and 10. 552 00:25:23,770 --> 00:25:27,040 So people are put in groups of 10 and then asked about, 553 00:25:27,040 --> 00:25:28,600 in a group of 10 people, how do you 554 00:25:28,600 --> 00:25:31,900 think you rank compared to others? 555 00:25:31,900 --> 00:25:35,090 And then people get some information, 556 00:25:35,090 --> 00:25:37,720 and they get, in particular, bilateral comparisons 557 00:25:37,720 --> 00:25:39,670 with another participant in the group. 558 00:25:39,670 --> 00:25:43,150 And then the authors elicit their posterior, so beliefs 559 00:25:43,150 --> 00:25:45,550 after receiving this information, and also 560 00:25:45,550 --> 00:25:48,070 the willingness to pay for true ranks. 561 00:25:48,070 --> 00:25:51,550 And broadly speaking, what the authors find 562 00:25:51,550 --> 00:25:54,520 is there's asymmetric processing of objective information 563 00:25:54,520 --> 00:25:55,720 about the self. 564 00:25:55,720 --> 00:25:58,250 In particular, when people receive positive news, 565 00:25:58,250 --> 00:26:00,040 so when somebody says you've been compared 566 00:26:00,040 --> 00:26:03,970 to another person in the study and you're better looking, 567 00:26:03,970 --> 00:26:07,600 or you have a better IQ score than this other person, 568 00:26:07,600 --> 00:26:09,730 people tend to be roughly Bayesian, 569 00:26:09,730 --> 00:26:12,550 as in they tend to be pretty good at actually updating 570 00:26:12,550 --> 00:26:14,140 based on that information. 571 00:26:14,140 --> 00:26:18,100 In contrast, when people receive unfavorable news, 572 00:26:18,100 --> 00:26:20,740 they tend to essentially discount the signals, 573 00:26:20,740 --> 00:26:23,050 and they tend to essentially not really 574 00:26:23,050 --> 00:26:24,300 react to that information. 575 00:26:24,300 --> 00:26:26,050 So that's very much consistent with people 576 00:26:26,050 --> 00:26:28,330 who are very happy to receive positive feedback 577 00:26:28,330 --> 00:26:30,190 and actually good at integrating it, 578 00:26:30,190 --> 00:26:33,250 as in they're good at doing the math 579 00:26:33,250 --> 00:26:35,620 as a Bayesian would want them to. 580 00:26:35,620 --> 00:26:37,870 But once they get negative feedback, they essentially 581 00:26:37,870 --> 00:26:40,690 mostly just discount that information, presumably 582 00:26:40,690 --> 00:26:44,890 and very much consistently with motivated beliefs. 583 00:26:44,890 --> 00:26:48,387 People tend to want to have positive images of themselves. 584 00:26:48,387 --> 00:26:50,470 There's also some evidence of people's willingness 585 00:26:50,470 --> 00:26:53,440 to pay for information if they think the information will 586 00:26:53,440 --> 00:26:55,990 be good, and not willing to pay for information 587 00:26:55,990 --> 00:26:57,820 if they think the information will be bad, 588 00:26:57,820 --> 00:26:59,590 so if they're low in the ranks. 589 00:26:59,590 --> 00:27:03,790 Again, people want to hear good things about themselves, 590 00:27:03,790 --> 00:27:07,930 and they learn more or they're willing to learn more when 591 00:27:07,930 --> 00:27:09,400 they receive positive news. 592 00:27:09,400 --> 00:27:11,830 They're also willingness-- willing to pay more 593 00:27:11,830 --> 00:27:14,950 for information that they think will be positive. 594 00:27:14,950 --> 00:27:17,560 There's some other evidence that's similar to that. 595 00:27:17,560 --> 00:27:20,230 There's a very nice-- or a similar paper by Mobius et al 596 00:27:20,230 --> 00:27:22,550 that's very similar to Eil and Rao. 597 00:27:22,550 --> 00:27:27,100 There's some very nice work by Florian Zimmermann 598 00:27:27,100 --> 00:27:28,750 in motivated memory. 599 00:27:28,750 --> 00:27:31,210 So that paper argues that it's not just 600 00:27:31,210 --> 00:27:33,040 about when people update immediately. 601 00:27:33,040 --> 00:27:35,080 So when people are given positive or negative 602 00:27:35,080 --> 00:27:37,580 information, they update differently. 603 00:27:37,580 --> 00:27:41,800 But what Florian Zimmermann shows is that in his work 604 00:27:41,800 --> 00:27:44,420 that when people are given this information, in fact, 605 00:27:44,420 --> 00:27:46,450 in his experiment there is no asymmetric 606 00:27:46,450 --> 00:27:47,560 updating in the short run. 607 00:27:47,560 --> 00:27:49,810 So when people are given this information about-- this 608 00:27:49,810 --> 00:27:53,710 is about how they did in a test in Germany, 609 00:27:53,710 --> 00:27:58,300 in the short run people seem to actually be not overly 610 00:27:58,300 --> 00:28:00,070 optimistic in their updating. 611 00:28:00,070 --> 00:28:03,130 But then a month later when the author goes back to people 612 00:28:03,130 --> 00:28:05,350 and asks them about the information that they have 613 00:28:05,350 --> 00:28:07,000 received and their beliefs, people 614 00:28:07,000 --> 00:28:09,640 tend to remember positive signals more 615 00:28:09,640 --> 00:28:10,960 than negative signals. 616 00:28:10,960 --> 00:28:13,780 And that suggests that memory can 617 00:28:13,780 --> 00:28:16,780 play an important role in the formation of motivated beliefs. 618 00:28:16,780 --> 00:28:18,760 When stuff happens, good stuff, people 619 00:28:18,760 --> 00:28:22,510 remember more, and bad things, not so much. 620 00:28:22,510 --> 00:28:25,180 Now those are all lab type experiments 621 00:28:25,180 --> 00:28:28,100 where people are incentivized to give correct answers. 622 00:28:28,100 --> 00:28:31,570 But in a way, it's somewhat contrived. 623 00:28:31,570 --> 00:28:34,210 There's also quite a bit of evidence in the real world 624 00:28:34,210 --> 00:28:37,600 setting, in particular, when it comes to finance or trading 625 00:28:37,600 --> 00:28:38,560 behavior. 626 00:28:38,560 --> 00:28:42,010 For example, there's Barber and Odean that shows that-- 627 00:28:42,010 --> 00:28:44,650 looks at overconfident small scale investors. 628 00:28:44,650 --> 00:28:47,862 Men, for example, tend to trade a lot more than women. 629 00:28:47,862 --> 00:28:49,570 We also know from other settings that men 630 00:28:49,570 --> 00:28:53,470 tend to be more overconfident than women, and those men, 631 00:28:53,470 --> 00:28:57,580 then, also lose more money or make less money plausibly 632 00:28:57,580 --> 00:28:59,140 due to overconfidence. 633 00:28:59,140 --> 00:29:01,390 With this type of evidence, it's often a little tricky 634 00:29:01,390 --> 00:29:04,210 because it's hard to say, is it to due to gender? 635 00:29:04,210 --> 00:29:06,460 Is it due to overconfidence and maybe other things 636 00:29:06,460 --> 00:29:08,710 that are correlated with this? 637 00:29:08,710 --> 00:29:10,325 But there's quite a bit of evidence 638 00:29:10,325 --> 00:29:13,960 on that kind of behavior that essentially overconfidence 639 00:29:13,960 --> 00:29:19,090 leads to worse decision here in this case trading decisions. 640 00:29:19,090 --> 00:29:22,960 There's also a very nice paper by Ulrike Malmendier and Tate 641 00:29:22,960 --> 00:29:24,550 on managerial hubris. 642 00:29:24,550 --> 00:29:26,200 They have a clever way of identifying 643 00:29:26,200 --> 00:29:30,010 overconfident managers, which essentially is managers that 644 00:29:30,010 --> 00:29:32,470 are overconfident-- they are holding stock options, 645 00:29:32,470 --> 00:29:33,800 if they own stocks. 646 00:29:33,800 --> 00:29:36,740 So you're not supposed to do that, 647 00:29:36,740 --> 00:29:39,910 and you will most likely do that only if you think your company 648 00:29:39,910 --> 00:29:43,060 is going to do really great compared to other companies. 649 00:29:43,060 --> 00:29:45,640 Instead, you exercise-- you should exercise your stock 650 00:29:45,640 --> 00:29:49,540 options and diversify and not to be overly invested 651 00:29:49,540 --> 00:29:50,630 in your own firm. 652 00:29:50,630 --> 00:29:53,980 But if you think that you're really great 653 00:29:53,980 --> 00:29:57,580 or your firm is really great, you [INAUDIBLE].. 654 00:29:57,580 --> 00:30:00,100 And precisely those overconfident managers 655 00:30:00,100 --> 00:30:02,170 engage their business in more mergers, 656 00:30:02,170 --> 00:30:05,900 and those mergers, Malmendier and Tate show, in fact, 657 00:30:05,900 --> 00:30:07,450 are not good. 658 00:30:07,450 --> 00:30:09,790 People tend to do too many mergers that 659 00:30:09,790 --> 00:30:12,040 essentially destroy value, presumably 660 00:30:12,040 --> 00:30:13,430 due to overconfidence. 661 00:30:13,430 --> 00:30:17,783 So there's that kind of evidence as well. 662 00:30:17,783 --> 00:30:19,450 So overall, we think there's quite a bit 663 00:30:19,450 --> 00:30:21,520 of evidence of people systematically holding 664 00:30:21,520 --> 00:30:23,320 overoptimistic beliefs. 665 00:30:23,320 --> 00:30:27,730 And those overoptimistic beliefs are affecting people's choices 666 00:30:27,730 --> 00:30:29,860 in a way that makes them worse off. 667 00:30:29,860 --> 00:30:33,310 In some ways, in particular, in finance choices 668 00:30:33,310 --> 00:30:38,170 where people tend to lose or destroy value. 669 00:30:38,170 --> 00:30:38,840 OK. 670 00:30:38,840 --> 00:30:42,640 Any questions on this before I move 671 00:30:42,640 --> 00:30:45,460 on to heuristics and biases? 672 00:30:51,480 --> 00:30:52,140 OK. 673 00:30:52,140 --> 00:30:59,412 So then we talked about three of our four issues here already. 674 00:30:59,412 --> 00:31:01,620 The last thing that we haven't talked about very much 675 00:31:01,620 --> 00:31:03,810 is to say Bayesian learning is just really hard, 676 00:31:03,810 --> 00:31:05,850 and people might just not be good 677 00:31:05,850 --> 00:31:08,520 at it, not for any motivated reasons 678 00:31:08,520 --> 00:31:11,790 or because of any utility reasons [INAUDIBLE] 679 00:31:11,790 --> 00:31:13,735 anticipatory [INAUDIBLE] utility. 680 00:31:13,735 --> 00:31:15,360 But it just might be really hard to do, 681 00:31:15,360 --> 00:31:18,060 and people essentially use heuristics 682 00:31:18,060 --> 00:31:22,160 because it's too hard to compute these things on their own. 683 00:31:22,160 --> 00:31:25,750 Now we talked about this already quite a bit. 684 00:31:25,750 --> 00:31:28,630 Lots of economic choices are made under uncertainty. 685 00:31:28,630 --> 00:31:32,600 Almost anything in life involves some uncertainty, 686 00:31:32,600 --> 00:31:35,260 so we need to know something about the likelihood 687 00:31:35,260 --> 00:31:36,310 of relevant events. 688 00:31:36,310 --> 00:31:40,850 And if you decide which topic in the course 689 00:31:40,850 --> 00:31:42,490 to focus on when you study for an exam, 690 00:31:42,490 --> 00:31:44,620 and you only have like one night to do so. 691 00:31:44,620 --> 00:31:48,220 If a basketball coach decides whether to leave tired players 692 00:31:48,220 --> 00:31:50,710 in the game, you kind make some probabilistic judgment 693 00:31:50,710 --> 00:31:53,230 about which player is going to be best. 694 00:31:53,230 --> 00:31:56,650 Many medical, managerial, educational, career decisions, 695 00:31:56,650 --> 00:31:58,390 essentially all those kinds of decisions 696 00:31:58,390 --> 00:32:02,183 are involving probabilities based on probability 697 00:32:02,183 --> 00:32:04,600 of good things happening, some probabilities of bad things 698 00:32:04,600 --> 00:32:06,610 happen, and you need to make some estimates 699 00:32:06,610 --> 00:32:08,480 about those probabilities. 700 00:32:08,480 --> 00:32:14,600 Now how do make individuals probabilistic 701 00:32:14,600 --> 00:32:16,740 judgement of this kind? 702 00:32:16,740 --> 00:32:18,720 So far when we talked about risk preferences 703 00:32:18,720 --> 00:32:21,360 earlier in the semester, we essentially looked at, 704 00:32:21,360 --> 00:32:24,168 how do people make these choices for given probabilities, right? 705 00:32:24,168 --> 00:32:25,710 That was essentially prospect theory. 706 00:32:25,710 --> 00:32:28,560 That was about Kahneman and Tversky, 1979, 707 00:32:28,560 --> 00:32:30,990 which was how do people think about risk? 708 00:32:30,990 --> 00:32:33,270 And what we always assumed was that the probability 709 00:32:33,270 --> 00:32:35,398 distribution was given. 710 00:32:35,398 --> 00:32:37,440 There's a 50% chance of something good happening. 711 00:32:37,440 --> 00:32:40,200 There's a 50% chance of something bad happening, 712 00:32:40,200 --> 00:32:43,500 and how do people make those kinds of choices? 713 00:32:43,500 --> 00:32:46,170 What does their utility function look like? 714 00:32:46,170 --> 00:32:48,180 Now we're going to talk about, how do people 715 00:32:48,180 --> 00:32:49,472 learn about such probabilities? 716 00:32:49,472 --> 00:32:52,110 How, in the first place, do they update 717 00:32:52,110 --> 00:32:54,060 what probabilities are like? 718 00:32:54,060 --> 00:32:57,240 Which is, of course, a very important input 719 00:32:57,240 --> 00:32:59,360 in their decision making. 720 00:32:59,360 --> 00:33:03,540 Now, this is based on pioneering work by Kahneman and Tversky, 721 00:33:03,540 --> 00:33:05,650 even earlier than the '79 paper. 722 00:33:05,650 --> 00:33:08,920 This is, in particular, a paper on 1974, 723 00:33:08,920 --> 00:33:14,220 which is in the reading list and on the course website. 724 00:33:14,220 --> 00:33:15,940 Now first, I want to start with-- 725 00:33:15,940 --> 00:33:17,820 and this is important in behavioral economics 726 00:33:17,820 --> 00:33:21,000 to acknowledge and try to emphasize 727 00:33:21,000 --> 00:33:24,900 people are pretty rational and pretty good in various ways, 728 00:33:24,900 --> 00:33:28,080 in the sense of people get lots of things right. 729 00:33:28,080 --> 00:33:30,030 In particular, people are pretty good in terms 730 00:33:30,030 --> 00:33:35,100 of broad directions of people's probability judgments. 731 00:33:35,100 --> 00:33:37,290 Here's a very simple example. 732 00:33:37,290 --> 00:33:40,450 Imagine you're deciding whether to see the new James Cameron 733 00:33:40,450 --> 00:33:40,950 movie. 734 00:33:40,950 --> 00:33:42,450 This could be a James Cameron movie. 735 00:33:42,450 --> 00:33:45,270 It could be a Korean drama or anything else. 736 00:33:45,270 --> 00:33:47,612 People think about they want to watch something new. 737 00:33:47,612 --> 00:33:49,070 Well, how are you going to do that? 738 00:33:49,070 --> 00:33:50,880 Well, you can read some online reviews. 739 00:33:50,880 --> 00:33:53,340 You hear some opinions from your friends. 740 00:33:53,340 --> 00:33:55,590 You look at some Rotten Tomatoes or other ratings, 741 00:33:55,590 --> 00:33:58,830 and then you try to make those kinds of choices. 742 00:33:58,830 --> 00:34:01,080 And so you start-- maybe you probably would, like, OK, 743 00:34:01,080 --> 00:34:04,230 have I seen a James Cameron movie before, 744 00:34:04,230 --> 00:34:07,530 or I said Korean drama, whatever you're interested in. 745 00:34:07,530 --> 00:34:09,850 have I seen something like this before? 746 00:34:09,850 --> 00:34:11,940 How much did I like the previous one? 747 00:34:11,940 --> 00:34:13,846 That could be your prior that you start with. 748 00:34:13,846 --> 00:34:15,929 And then you get these other pieces of information 749 00:34:15,929 --> 00:34:18,872 of online reviews, various opinions from friends, 750 00:34:18,872 --> 00:34:21,330 or a Rotten Tomato ratings, and how do you use those, then? 751 00:34:21,330 --> 00:34:23,056 What do you do with those? 752 00:34:23,056 --> 00:34:24,639 So essentially what you're going to do 753 00:34:24,639 --> 00:34:26,920 is you take some-- you got some prior, which 754 00:34:26,920 --> 00:34:30,340 is about James Cameron movies or whatever to start with. 755 00:34:30,340 --> 00:34:33,100 And that's like that-- 756 00:34:33,100 --> 00:34:35,465 that's what you use to start with in the first place. 757 00:34:35,465 --> 00:34:37,840 And then you get essentially these pieces of information. 758 00:34:37,840 --> 00:34:40,870 And if they're positive pieces of information 759 00:34:40,870 --> 00:34:43,840 about those specific movies, then you're going to update up. 760 00:34:43,840 --> 00:34:46,330 And if they're bad, if your friends don't like it, 761 00:34:46,330 --> 00:34:49,048 as you said, they update that. 762 00:34:49,048 --> 00:34:51,340 Notice that it could also be that you have a friend who 763 00:34:51,340 --> 00:34:54,790 has terrible taste in movies, and if that friend 764 00:34:54,790 --> 00:34:58,460 does not like a movie, that you update the other way and so on. 765 00:34:58,460 --> 00:35:02,200 So it doesn't need to be that your friend's movie ratings are 766 00:35:02,200 --> 00:35:04,537 positively correlated with yours. 767 00:35:04,537 --> 00:35:05,870 That could still be informative. 768 00:35:05,870 --> 00:35:08,245 If somebody has terrible movie taste and likes something, 769 00:35:08,245 --> 00:35:11,200 that could be actually good news for the particular movie. 770 00:35:11,200 --> 00:35:13,222 But exactly-- this is what I have here. 771 00:35:13,222 --> 00:35:14,680 Essentially what you're going to do 772 00:35:14,680 --> 00:35:18,940 is a version of this would be you start with your base 773 00:35:18,940 --> 00:35:20,530 rate or your prior. 774 00:35:20,530 --> 00:35:25,090 And then essentially you use the various pieces of information 775 00:35:25,090 --> 00:35:27,045 and then adjust your probability up or down. 776 00:35:27,045 --> 00:35:29,170 Suppose you have to sort of [INAUDIBLE] willingness 777 00:35:29,170 --> 00:35:31,087 to pay or your probability of seeing the movie 778 00:35:31,087 --> 00:35:33,760 or doing other things, or your probability of liking 779 00:35:33,760 --> 00:35:37,270 the movie, you adjust it up or down based 780 00:35:37,270 --> 00:35:39,640 on the different pieces of information that we get. 781 00:35:39,640 --> 00:35:42,050 And so generally people are pretty-- 782 00:35:42,050 --> 00:35:45,250 and this is a pretty reasonable and mathematical, 783 00:35:45,250 --> 00:35:47,540 mathematically well-founded procedure. 784 00:35:47,540 --> 00:35:50,800 This is actually pretty close to what a Bayesian would do. 785 00:35:50,800 --> 00:35:52,965 So people are actually pretty good at this. 786 00:35:52,965 --> 00:35:55,570 They're pretty-- they understand the basics of forming 787 00:35:55,570 --> 00:35:56,830 likelihood estimates. 788 00:35:56,830 --> 00:35:59,410 They know the direction in which features 789 00:35:59,410 --> 00:36:01,540 a baseline likelihood and available information 790 00:36:01,540 --> 00:36:02,680 should affect estimates. 791 00:36:02,680 --> 00:36:06,770 So overall, people are pretty good. 792 00:36:06,770 --> 00:36:11,260 Now what's much harder to do is not just a direction, but also 793 00:36:11,260 --> 00:36:13,480 the extent to which you should update. 794 00:36:13,480 --> 00:36:18,340 That is to say, people know which direction to move, 795 00:36:18,340 --> 00:36:21,940 as in like if you and I like similar movies 796 00:36:21,940 --> 00:36:23,650 and I tell you this movie was great, 797 00:36:23,650 --> 00:36:25,990 it's pretty clear that you should move upwards. 798 00:36:25,990 --> 00:36:27,670 But what's much harder to understand 799 00:36:27,670 --> 00:36:29,500 is by how much should you move upwards 800 00:36:29,500 --> 00:36:32,080 in your estimate of how good the movie or the probability 801 00:36:32,080 --> 00:36:35,260 that you like the movie is. 802 00:36:35,260 --> 00:36:37,300 And that's true for many things in life. 803 00:36:37,300 --> 00:36:39,910 People are pretty good at the direction in terms 804 00:36:39,910 --> 00:36:43,000 of where to move up or down. 805 00:36:43,000 --> 00:36:46,630 What they're much worse at is how far 806 00:36:46,630 --> 00:36:48,800 to adjust their estimates. 807 00:36:48,800 --> 00:36:51,700 And that's where they should be using Bayes' Rule, 808 00:36:51,700 --> 00:36:53,980 but that's extremely cognitively demanding. 809 00:36:53,980 --> 00:36:59,630 And so in many situations people tend to not get that right. 810 00:36:59,630 --> 00:37:03,620 Now, just to sort of remind you, and I think in recitation 811 00:37:03,620 --> 00:37:05,580 you discussed this as well, what is Bayes' Rule 812 00:37:05,580 --> 00:37:07,360 or how does Bayes' Rule work? 813 00:37:07,360 --> 00:37:09,540 Let me go very quickly through that 814 00:37:09,540 --> 00:37:12,930 and then focus on that next evidence. 815 00:37:12,930 --> 00:37:13,910 Here's one example. 816 00:37:13,910 --> 00:37:17,760 Suppose you have a coin that you start off thinking as a-- 817 00:37:17,760 --> 00:37:20,060 is fair with probability of 2/3. 818 00:37:20,060 --> 00:37:22,880 That means the coin is biased towards heads 819 00:37:22,880 --> 00:37:25,010 or the coin is heads and tails. 820 00:37:25,010 --> 00:37:28,430 It's biased towards heads with probability 1/3, in which case 821 00:37:28,430 --> 00:37:31,700 heads comes up 75% of the time. 822 00:37:31,700 --> 00:37:35,840 Now if you flip the coin and it comes up H, heads, 823 00:37:35,840 --> 00:37:39,650 what's the probability of that, that it's fair? 824 00:37:39,650 --> 00:37:45,380 Clearly, it should be 2/3, because you've got some signal, 825 00:37:45,380 --> 00:37:46,940 but by how much should you adjust 826 00:37:46,940 --> 00:37:48,140 is actually very hard to do. 827 00:37:48,140 --> 00:37:49,848 So you can think about this in your head, 828 00:37:49,848 --> 00:37:52,460 and you realize it's actually kind of hard to do, 829 00:37:52,460 --> 00:37:54,138 unless you can actually write it down. 830 00:37:54,138 --> 00:37:56,180 So suppose you had to make a very quick decision. 831 00:37:56,180 --> 00:37:57,890 In particular, it's actually very hard 832 00:37:57,890 --> 00:37:59,735 to do this quickly on your own. 833 00:37:59,735 --> 00:38:01,610 You can do this particularly with one signal, 834 00:38:01,610 --> 00:38:03,250 but suppose you get several signals 835 00:38:03,250 --> 00:38:06,097 and it gets much harder and harder over time. 836 00:38:06,097 --> 00:38:07,180 Now how would you do this? 837 00:38:07,180 --> 00:38:08,750 You can sort of-- in this particular case, 838 00:38:08,750 --> 00:38:10,345 you can do a graphical illustration. 839 00:38:10,345 --> 00:38:11,720 Again, I'll do this very quickly, 840 00:38:11,720 --> 00:38:13,933 but you can read about it overall. 841 00:38:13,933 --> 00:38:15,850 You can think about, there's sort of two types 842 00:38:15,850 --> 00:38:19,840 of coins or like urns. 843 00:38:19,840 --> 00:38:22,690 There's a fair coin that's represented 844 00:38:22,690 --> 00:38:26,080 by an urn with equal number of heads and tail balls. 845 00:38:26,080 --> 00:38:27,670 And there's an unfair coin, which 846 00:38:27,670 --> 00:38:31,930 is represented by an urn in which 75% of the balls are H. 847 00:38:31,930 --> 00:38:34,060 So you don't know whether the coin is fair, 848 00:38:34,060 --> 00:38:36,170 so you don't know which urn you're drawing from. 849 00:38:36,170 --> 00:38:37,660 And one way to think about this is 850 00:38:37,660 --> 00:38:39,460 imagine you're drawing from an urn that 851 00:38:39,460 --> 00:38:41,380 contains both of those urns. 852 00:38:41,380 --> 00:38:43,780 Of course, you need to then also take into account 853 00:38:43,780 --> 00:38:47,390 the probability of the fair ball being 2/3. 854 00:38:47,390 --> 00:38:51,460 So the fair urn has twice as many balls as the unfair one. 855 00:38:51,460 --> 00:38:53,680 So you can sort of represent it in that way. 856 00:38:53,680 --> 00:38:56,070 And then suppose you draw a ball randomly 857 00:38:56,070 --> 00:38:57,520 and it's H, what's the probability 858 00:38:57,520 --> 00:38:59,230 that it came from the fair urn? 859 00:38:59,230 --> 00:39:03,430 Then again, if you've taken probability classes and so on, 860 00:39:03,430 --> 00:39:07,540 that should be relatively simple for you to calculate. 861 00:39:07,540 --> 00:39:10,340 But here in the graphical way, you can look this very simple. 862 00:39:10,340 --> 00:39:12,910 There's essentially seven red balls. 863 00:39:12,910 --> 00:39:16,480 There's four in the fair coin and three in the unfair coin. 864 00:39:16,480 --> 00:39:21,580 So the probability that it came from the fair coin is 4/7. 865 00:39:21,580 --> 00:39:24,400 Similarly, if you draw a T, then the probability 866 00:39:24,400 --> 00:39:28,000 it came from the fair urn is 4/5, right? 867 00:39:28,000 --> 00:39:29,878 And that's, in some sense-- 868 00:39:29,878 --> 00:39:32,420 you can write down the math, and it's pretty clear and so on. 869 00:39:32,420 --> 00:39:36,200 But in a way, it's hard to do this in your head in the sense 870 00:39:36,200 --> 00:39:38,500 that you know it should be lower than 2/3, 871 00:39:38,500 --> 00:39:40,600 but how much lower and how much should you adjust 872 00:39:40,600 --> 00:39:41,800 is actually kind of hard. 873 00:39:41,800 --> 00:39:45,700 And here the answer happens to be 4/7. 874 00:39:45,700 --> 00:39:49,820 And so here's sort of the formal way of doing this. 875 00:39:49,820 --> 00:39:51,070 I'm not going to go into this. 876 00:39:51,070 --> 00:39:54,490 But trust me, the answer is 4/7. 877 00:39:54,490 --> 00:39:57,465 Now, even this simple example of Bayes' rule, 878 00:39:57,465 --> 00:39:58,840 maybe it was very simple for you. 879 00:39:58,840 --> 00:40:01,150 But it's somewhat difficult to follow 880 00:40:01,150 --> 00:40:04,090 in the sense of would you have gotten the first sentence right 881 00:40:04,090 --> 00:40:06,040 immediately if I gave you 10 seconds? 882 00:40:06,040 --> 00:40:07,090 Maybe, maybe not. 883 00:40:07,090 --> 00:40:11,810 But probably quite a few of you would have gotten it wrong. 884 00:40:11,810 --> 00:40:15,250 Now, in addition, then, if you try 885 00:40:15,250 --> 00:40:18,370 to apply these kinds of rules or the base rule in new situations 886 00:40:18,370 --> 00:40:20,590 with multiple pieces of different information, 887 00:40:20,590 --> 00:40:22,540 it's far more difficult. Suppose I 888 00:40:22,540 --> 00:40:25,660 told you to draw like 17 times and get those signals 889 00:40:25,660 --> 00:40:29,050 and without replacement and with replacement and so on. 890 00:40:29,050 --> 00:40:31,810 Things would get way, way more difficult. 891 00:40:31,810 --> 00:40:34,510 And the urn example is, in fact, a much easier one. 892 00:40:34,510 --> 00:40:37,960 Because there are at least I'm telling you exactly what 893 00:40:37,960 --> 00:40:39,430 the probabilities are. 894 00:40:39,430 --> 00:40:41,020 But in many real-world situations, 895 00:40:41,020 --> 00:40:42,130 you don't even know that. 896 00:40:42,130 --> 00:40:44,150 A friend telling you the movie is good or bad, 897 00:40:44,150 --> 00:40:46,900 it's not clear how to read that signal, 898 00:40:46,900 --> 00:40:50,290 how do you think about that signal as a whole. 899 00:40:50,290 --> 00:40:52,390 So but most people don't know the precise rule 900 00:40:52,390 --> 00:40:53,380 of Bayes' rule. 901 00:40:53,380 --> 00:40:55,680 I guess, at MIT, most students do. 902 00:40:55,680 --> 00:40:58,040 But in the general population, people don't. 903 00:40:58,040 --> 00:41:00,040 And even if they do, they can't or don't want 904 00:41:00,040 --> 00:41:03,140 to think so hard in many cases. 905 00:41:03,140 --> 00:41:06,130 So what people do instead, they use intuitive shortcuts 906 00:41:06,130 --> 00:41:10,140 to make judgments of likelihoods. 907 00:41:10,140 --> 00:41:12,080 Now, that's a good thing. 908 00:41:12,080 --> 00:41:14,590 So we want people to use these shortcuts because otherwise 909 00:41:14,590 --> 00:41:16,720 they will just freeze and not be able to make any choices. 910 00:41:16,720 --> 00:41:18,553 And so, in some sense, it's good that people 911 00:41:18,553 --> 00:41:22,470 are sort of simplifying problems and they make some choices 912 00:41:22,470 --> 00:41:24,040 quickly. 913 00:41:24,040 --> 00:41:25,680 So in some sense, it's good. 914 00:41:25,680 --> 00:41:29,810 But it also leads to systematic mistakes. 915 00:41:29,810 --> 00:41:31,860 Because in some sense, you can use shortcuts. 916 00:41:31,860 --> 00:41:34,950 And they help you to get things approximately right. 917 00:41:34,950 --> 00:41:38,430 But now, if we can understand their shortcuts, 918 00:41:38,430 --> 00:41:40,260 then we can also understand, potentially, 919 00:41:40,260 --> 00:41:42,555 their systematic mistakes. 920 00:41:45,730 --> 00:41:47,730 So what people tend to do is they 921 00:41:47,730 --> 00:41:52,680 focus on one or a small set of the situations 922 00:41:52,680 --> 00:41:54,120 at hand that seems most relevant. 923 00:41:54,120 --> 00:41:56,470 They focus on that to make their decision. 924 00:41:56,470 --> 00:41:59,760 And often, then, they systematically 925 00:41:59,760 --> 00:42:04,020 neglect other more complicated other issues 926 00:42:04,020 --> 00:42:07,200 of the situation, which makes the likelihood 927 00:42:07,200 --> 00:42:09,450 of their estimates typically incorrect. 928 00:42:09,450 --> 00:42:11,910 And now the question is, can we sort of understand 929 00:42:11,910 --> 00:42:13,350 what people focus on? 930 00:42:13,350 --> 00:42:17,250 And can we understand what causes these systematic biases 931 00:42:17,250 --> 00:42:19,740 to sort of make clear predictions on how 932 00:42:19,740 --> 00:42:21,135 people do things wrongly. 933 00:42:21,135 --> 00:42:23,010 And then, perhaps, if you want to help people 934 00:42:23,010 --> 00:42:26,100 make better choices, you can provide them 935 00:42:26,100 --> 00:42:29,900 with valid information as well. 936 00:42:29,900 --> 00:42:34,240 Now, one particularly interesting issue 937 00:42:34,240 --> 00:42:37,460 is sequences of random outcomes. 938 00:42:37,460 --> 00:42:39,640 So as I just showed you, if you sort of drop balls 939 00:42:39,640 --> 00:42:42,370 from an urn, that's a very relevant situation, 940 00:42:42,370 --> 00:42:45,400 not because people in the real world in real-world situations, 941 00:42:45,400 --> 00:42:49,690 drop balls from urns, but rather because, in many situations, 942 00:42:49,690 --> 00:42:52,510 you do actually get repeated signals over time 943 00:42:52,510 --> 00:42:54,715 that you should use to update. 944 00:42:54,715 --> 00:42:56,590 So an investor might observe past performance 945 00:42:56,590 --> 00:42:59,710 of mutual funds before deciding which one to invest in. 946 00:42:59,710 --> 00:43:01,900 A patient of a doctor might observe the outcome 947 00:43:01,900 --> 00:43:03,580 of prior surgeries before deciding 948 00:43:03,580 --> 00:43:07,450 whether to undertake the surgery or which doctor to choose. 949 00:43:07,450 --> 00:43:10,090 A coach can observe the recent performance of a basketball 950 00:43:10,090 --> 00:43:13,060 player before deciding whether to put the player in the game. 951 00:43:13,060 --> 00:43:15,070 There's lots of kinds of situations where people 952 00:43:15,070 --> 00:43:17,710 get repeated signals over time and then 953 00:43:17,710 --> 00:43:21,130 have to try to infer likelihoods of probabilities 954 00:43:21,130 --> 00:43:23,940 of certain events to happen. 955 00:43:23,940 --> 00:43:25,690 So that's a very common thing in the world 956 00:43:25,690 --> 00:43:28,810 that we should try and understand. 957 00:43:28,810 --> 00:43:32,210 Now, one systematic pattern that we see in the world 958 00:43:32,210 --> 00:43:34,480 is what's called the gambler's fallacy. 959 00:43:34,480 --> 00:43:37,720 That's the false belief that, in a sequence of independent draws 960 00:43:37,720 --> 00:43:39,520 from a distribution, an outcome that 961 00:43:39,520 --> 00:43:43,180 hasn't occurred for a while is more likely to come up 962 00:43:43,180 --> 00:43:44,420 on the next block. 963 00:43:44,420 --> 00:43:46,795 So the important part and what's doing a lot of work here 964 00:43:46,795 --> 00:43:48,700 is independent draws. 965 00:43:48,700 --> 00:43:51,080 Suppose there's a distribution of outcomes 966 00:43:51,080 --> 00:43:53,740 and there's independent draws, which 967 00:43:53,740 --> 00:43:57,250 means that, essentially, what happened in the last two, 968 00:43:57,250 --> 00:43:59,350 three, or four draws should not have 969 00:43:59,350 --> 00:44:02,200 any effect on what's going to happen in the next draw. 970 00:44:02,200 --> 00:44:05,260 So if you play roulette or any sort of poker or the like, 971 00:44:05,260 --> 00:44:09,850 if you get read several times in a row, that 972 00:44:09,850 --> 00:44:14,200 has exactly no impact on how likely it is that red or black 973 00:44:14,200 --> 00:44:17,500 is coming up in the next draw. 974 00:44:17,500 --> 00:44:19,900 But people tend to have sort of this almost like folk 975 00:44:19,900 --> 00:44:23,200 knowledge of, well, if red came up a few times, 976 00:44:23,200 --> 00:44:25,270 now black is due, and the other way around. 977 00:44:30,240 --> 00:44:32,330 And that's true in many different situations. 978 00:44:32,330 --> 00:44:35,100 But the important part here is that there's 979 00:44:35,100 --> 00:44:36,900 a sequence of independent draws. 980 00:44:36,900 --> 00:44:43,200 So as to say, these are situations 981 00:44:43,200 --> 00:44:46,020 where the probability distribution is actually null. 982 00:44:46,020 --> 00:44:49,350 That is to say, when you play roulette or you go to a casino, 983 00:44:49,350 --> 00:44:52,440 it's not like you need to learn about what's 984 00:44:52,440 --> 00:44:55,695 the probability that the roulette is fair or not. 985 00:44:55,695 --> 00:44:58,620 You know that there's some monitoring and so on in those 986 00:44:58,620 --> 00:44:59,670 situations. 987 00:44:59,670 --> 00:45:02,640 You know, essentially, that the chance of getting red or black 988 00:45:02,640 --> 00:45:04,140 is the same. 989 00:45:04,140 --> 00:45:07,050 And yet people tend to make these kinds of updates. 990 00:45:07,050 --> 00:45:08,800 There's a nice paper by Gold and Hester, 991 00:45:08,800 --> 00:45:12,630 an old paper that shows this quite nicely, 992 00:45:12,630 --> 00:45:15,870 in which subjects are told the coin with a black 993 00:45:15,870 --> 00:45:18,810 and a red side would be flipped 25 times. 994 00:45:18,810 --> 00:45:21,570 And the experiment is slightly sneaky here, 995 00:45:21,570 --> 00:45:23,730 because essentially actually they 996 00:45:23,730 --> 00:45:26,250 reported people a predetermined sequence of events. 997 00:45:26,250 --> 00:45:27,940 So there's some deception involved here, 998 00:45:27,940 --> 00:45:32,110 which is a little bit trickier but it's mostly fine. 999 00:45:32,110 --> 00:45:34,380 So what they see is essentially the subject 1000 00:45:34,380 --> 00:45:37,650 sees 17 mixed coin flips. 1001 00:45:37,650 --> 00:45:39,870 So they see a sequence of coin flips. 1002 00:45:39,870 --> 00:45:41,650 It's red and black and so on. 1003 00:45:41,650 --> 00:45:45,180 And then, at the end of it, you see essentially 1004 00:45:45,180 --> 00:45:47,520 one black and four reds. 1005 00:45:47,520 --> 00:45:49,980 And so, again, this is supposed to be 1006 00:45:49,980 --> 00:45:53,190 like independent draws, which to make this case, 1007 00:45:53,190 --> 00:45:58,140 in this experiment, they sort of rigged the last five draws, 1008 00:45:58,140 --> 00:45:59,760 which is not quite ideal. 1009 00:45:59,760 --> 00:46:02,652 But let's just go with this for a bit. 1010 00:46:02,652 --> 00:46:04,860 And so what essentially people see in this experiment 1011 00:46:04,860 --> 00:46:14,670 is mostly evidence of red and black 1012 00:46:14,670 --> 00:46:16,770 where roughly red and black is even. 1013 00:46:16,770 --> 00:46:18,765 If anything-- so the 17 are mixed-- 1014 00:46:18,765 --> 00:46:20,130 these are actually draws-- 1015 00:46:20,130 --> 00:46:23,940 if anything, people saw more red than they saw black. 1016 00:46:23,940 --> 00:46:26,460 Then, on the 23rd flip, after seeing 1017 00:46:26,460 --> 00:46:30,540 these 22 flips to start with, the participants 1018 00:46:30,540 --> 00:46:33,300 were given a choice between 70 points 1019 00:46:33,300 --> 00:46:37,368 for sure or 100 points if the next flip was their color. 1020 00:46:37,368 --> 00:46:38,910 And points are essentially stuff they 1021 00:46:38,910 --> 00:46:42,060 can get money for eventually. 1022 00:46:42,060 --> 00:46:45,090 And so now it's randomly chosen whether half of the subjects 1023 00:46:45,090 --> 00:46:48,270 color was red and half of them was black. 1024 00:46:48,270 --> 00:46:52,170 And so now the propensity to take the 70 points reveals 1025 00:46:52,170 --> 00:46:55,290 the beliefs about the odds that the next flip would 1026 00:46:55,290 --> 00:46:56,842 be their color. 1027 00:46:56,842 --> 00:47:00,060 So if you think the probability is 1028 00:47:00,060 --> 00:47:04,110 50%, which is kind of like, given the information that you 1029 00:47:04,110 --> 00:47:08,610 got, you should probably roughly think it's 50%, 1030 00:47:08,610 --> 00:47:10,450 and if you're risk-averse in particular, 1031 00:47:10,450 --> 00:47:13,650 you should truly choose the 70 points for sure. 1032 00:47:13,650 --> 00:47:17,520 Because an expectation, if you get p times 100, if it's 50%, 1033 00:47:17,520 --> 00:47:20,290 you get, essentially, 50. 1034 00:47:20,290 --> 00:47:23,310 So you're only got to choose the "100 points if the next flip is 1035 00:47:23,310 --> 00:47:29,520 your color" if you really think your color is likely to occur. 1036 00:47:29,520 --> 00:47:31,980 And in particular if your probability of that occurring 1037 00:47:31,980 --> 00:47:35,850 is like 70% or higher, notice that that even 1038 00:47:35,850 --> 00:47:38,070 assumes risk neutrality. 1039 00:47:38,070 --> 00:47:39,840 So if people are risk-neutral, if they 1040 00:47:39,840 --> 00:47:43,450 think the probability is 70%, they should be different. 1041 00:47:43,450 --> 00:47:45,120 And again, if they're risk neutral 1042 00:47:45,120 --> 00:47:49,180 and they choose number 2 here, their color, well, that's 1043 00:47:49,180 --> 00:47:52,100 only the case if they think the probability of the color 1044 00:47:52,100 --> 00:47:53,410 occurring is higher than 70%. 1045 00:47:56,070 --> 00:47:58,290 And so given the evidence that they saw, 1046 00:47:58,290 --> 00:48:00,660 it's going to be rarely the case that-- 1047 00:48:00,660 --> 00:48:03,960 or people should really, given the evidence that's there, 1048 00:48:03,960 --> 00:48:06,837 the best thing they can do is essentially-- 1049 00:48:06,837 --> 00:48:08,670 or the thing is set up such that they really 1050 00:48:08,670 --> 00:48:11,500 should choose number one. 1051 00:48:11,500 --> 00:48:14,970 But of course, the idea is here, if people 1052 00:48:14,970 --> 00:48:18,330 think that it was four times red, so now black is due, 1053 00:48:18,330 --> 00:48:22,350 the prediction would be that people would be more likely 1054 00:48:22,350 --> 00:48:28,320 or will choose item number 2 if their color is black. 1055 00:48:28,320 --> 00:48:35,150 And so 24 of 29 red subjects chose 1056 00:48:35,150 --> 00:48:37,970 to take the sure thing, which essentially is option 1 here 1057 00:48:37,970 --> 00:48:39,110 that I showed you. 1058 00:48:39,110 --> 00:48:42,170 And eight of 30 of the black subjects did. 1059 00:48:42,170 --> 00:48:46,200 So these are essentially the people who choose their color. 1060 00:48:46,200 --> 00:48:48,200 So if people have black as their color, 1061 00:48:48,200 --> 00:48:50,120 they essentially choose this option 1062 00:48:50,120 --> 00:48:52,880 because they really think now black is due 1063 00:48:52,880 --> 00:48:57,650 because there's been essentially four reds in a row. 1064 00:48:57,650 --> 00:49:00,170 Now, again, if you're told-- and that's 1065 00:49:00,170 --> 00:49:02,120 a little bit in play in this experiment-- 1066 00:49:02,120 --> 00:49:05,900 about these being independent events, which for coin flips, 1067 00:49:05,900 --> 00:49:07,880 really they should be independent events, 1068 00:49:07,880 --> 00:49:09,320 that doesn't make any sense. 1069 00:49:09,320 --> 00:49:12,248 You should, if anything, think red 1070 00:49:12,248 --> 00:49:13,790 is more likely because the coin might 1071 00:49:13,790 --> 00:49:17,270 be sort more likely to be red. 1072 00:49:17,270 --> 00:49:20,000 But sort of thinking that black is more likely 1073 00:49:20,000 --> 00:49:23,170 surely is not a good idea. 1074 00:49:23,170 --> 00:49:25,898 And it might essentially make you lose money. 1075 00:49:25,898 --> 00:49:27,440 Now, there's some interesting variant 1076 00:49:27,440 --> 00:49:32,030 of this for which they set, for some variants, 1077 00:49:32,030 --> 00:49:37,890 the 23rd coin flip was delayed by some time, 24 minutes. 1078 00:49:37,890 --> 00:49:40,110 And what they then find is, in fact, 1079 00:49:40,110 --> 00:49:43,760 weaker evidence of what's called the gambler's fallacy, which 1080 00:49:43,760 --> 00:49:48,560 seems to say that once you let the coin rest for a while, 1081 00:49:48,560 --> 00:49:50,510 people seem to think that letting the coin 1082 00:49:50,510 --> 00:49:56,210 rest for a while, which sort of makes it more likely, I guess, 1083 00:49:56,210 --> 00:49:59,780 that the streak continues, or less likely that essentially 1084 00:49:59,780 --> 00:50:01,560 black now is the case. 1085 00:50:01,560 --> 00:50:04,820 So people seem to these fairly-- 1086 00:50:04,820 --> 00:50:07,010 and I think that's fair to say-- irrational beliefs 1087 00:50:07,010 --> 00:50:10,670 about or biased beliefs about what's going to happen next. 1088 00:50:10,670 --> 00:50:15,920 And that seems to be easily swayed by even relatively 1089 00:50:15,920 --> 00:50:16,830 small things. 1090 00:50:16,830 --> 00:50:19,520 It's like the coin needs to revert. 1091 00:50:19,520 --> 00:50:23,440 But if you wait for 24 minutes, not so much anymore. 1092 00:50:23,440 --> 00:50:25,940 There's quite a bit of work on the gambler's fallacy overall 1093 00:50:25,940 --> 00:50:26,780 in various settings. 1094 00:50:26,780 --> 00:50:29,652 That's just one setting to illustrate this. 1095 00:50:29,652 --> 00:50:31,610 There's quite a bit of other work in that area. 1096 00:50:31,610 --> 00:50:34,505 And it's a fairly robust finding that people 1097 00:50:34,505 --> 00:50:37,160 have found in the literature. 1098 00:50:37,160 --> 00:50:40,010 Now, a second pattern that people find 1099 00:50:40,010 --> 00:50:42,590 is what's called the hot-hand fallacy. 1100 00:50:42,590 --> 00:50:45,720 That's the idea that, in particular basketball fans 1101 00:50:45,720 --> 00:50:49,100 or other sort of types of sports fans-- 1102 00:50:49,100 --> 00:50:52,040 fans, players, coaches-- believe that there 1103 00:50:52,040 --> 00:50:54,830 is systematic day-to-day operations in players' 1104 00:50:54,830 --> 00:50:55,760 shooting performance. 1105 00:50:55,760 --> 00:50:57,427 And that's for basketball, but it's also 1106 00:50:57,427 --> 00:50:59,400 true for other sports. 1107 00:50:59,400 --> 00:51:04,100 So the idea is that the performance of a player 1108 00:51:04,100 --> 00:51:06,500 may sometimes be predictably better than expected 1109 00:51:06,500 --> 00:51:09,240 on the basis of the player's overall record. 1110 00:51:09,240 --> 00:51:11,690 And so what people would say is, well, the player 1111 00:51:11,690 --> 00:51:14,540 is on fire today or is a streak shooter and so on. 1112 00:51:14,540 --> 00:51:17,690 And so the idea is that "on fire" today 1113 00:51:17,690 --> 00:51:23,880 means that he or she is more likely to hit his or her shots 1114 00:51:23,880 --> 00:51:25,160 than on other days. 1115 00:51:25,160 --> 00:51:27,290 So that's to say prediction is that made 1116 00:51:27,290 --> 00:51:29,060 shots should cluster together. 1117 00:51:29,060 --> 00:51:31,250 Like on one day you happen to be really good, maybe, 1118 00:51:31,250 --> 00:51:32,000 in the first half. 1119 00:51:32,000 --> 00:51:37,430 The second half, and conditional of having made a few shots, 1120 00:51:37,430 --> 00:51:41,247 the next shot should be more likely than sort 1121 00:51:41,247 --> 00:51:43,820 of the unconditional probability. 1122 00:51:43,820 --> 00:51:46,280 There's lots of work on this issue, 1123 00:51:46,280 --> 00:51:49,610 starting by Gilovich and Vallone and Tversky in 1985. 1124 00:51:49,610 --> 00:51:56,110 People have gone back and forth between saying in fact 1125 00:51:56,110 --> 00:51:58,870 there is sort of such a thing as a hot hand or not. 1126 00:51:58,870 --> 00:52:00,340 The initial claims were essentially 1127 00:52:00,340 --> 00:52:01,360 there is no such thing. 1128 00:52:01,360 --> 00:52:04,480 People believe that there's a hot hand going on 1129 00:52:04,480 --> 00:52:09,010 and players are really running hot, but in fact, they're not. 1130 00:52:09,010 --> 00:52:10,660 Some other evidence later showed maybe 1131 00:52:10,660 --> 00:52:11,952 there is actually such a thing. 1132 00:52:11,952 --> 00:52:13,700 And it was contradicted again. 1133 00:52:13,700 --> 00:52:14,890 So it's kind of complicated. 1134 00:52:14,890 --> 00:52:17,710 But overall it seems to be that people tend to be-- 1135 00:52:17,710 --> 00:52:19,900 players, fans, and so on-- tend to be 1136 00:52:19,900 --> 00:52:25,270 quite overoptimistic about the hot-handed streaks happening 1137 00:52:25,270 --> 00:52:28,200 in reality. 1138 00:52:28,200 --> 00:52:31,810 Now, one question is that that seems kind of odd in some ways. 1139 00:52:31,810 --> 00:52:36,720 And so, on the one hand, the hot-hand fallacy 1140 00:52:36,720 --> 00:52:40,840 and the gambler's fallacy seem to be opposites of each other. 1141 00:52:40,840 --> 00:52:42,900 That is to say, the gambler's fallacy 1142 00:52:42,900 --> 00:52:44,430 is the belief that the next outcome 1143 00:52:44,430 --> 00:52:47,820 is likely to be different from the previous ones. 1144 00:52:47,820 --> 00:52:50,100 If I show you, if you're gambling, 1145 00:52:50,100 --> 00:52:51,690 you've got like four reds, now you say 1146 00:52:51,690 --> 00:52:52,857 it's going to be black next. 1147 00:52:52,857 --> 00:52:54,510 So if you have three heads, then you 1148 00:52:54,510 --> 00:52:57,220 think that your coin is going to show some tails, that's 1149 00:52:57,220 --> 00:53:01,140 essentially saying you saw a bunch of outcomes of one kind, 1150 00:53:01,140 --> 00:53:03,340 the next one will likely be different. 1151 00:53:03,340 --> 00:53:05,940 On the other hand, then hot-hand fallacy is saying, well, 1152 00:53:05,940 --> 00:53:07,560 it's a belief that the next outcome 1153 00:53:07,560 --> 00:53:10,395 is likely to be similar to the previous ones. 1154 00:53:10,395 --> 00:53:13,797 Now, what's going on here, you can think of both of these 1155 00:53:13,797 --> 00:53:15,630 as a consequence of what's called the belief 1156 00:53:15,630 --> 00:53:17,335 in the law of small numbers. 1157 00:53:17,335 --> 00:53:18,240 Now, what is that? 1158 00:53:18,240 --> 00:53:20,490 You should all know the law of large numbers, which 1159 00:53:20,490 --> 00:53:23,322 is, in large samples of independent draws 1160 00:53:23,322 --> 00:53:24,780 from a distribution, the proportion 1161 00:53:24,780 --> 00:53:27,540 of different outcomes closely reflects the underlying 1162 00:53:27,540 --> 00:53:28,200 probability. 1163 00:53:28,200 --> 00:53:31,453 So essentially once you have sufficiently many draws 1164 00:53:31,453 --> 00:53:33,120 from some distribution, the distribution 1165 00:53:33,120 --> 00:53:38,220 will converge to what you're drawing from. 1166 00:53:38,220 --> 00:53:41,370 Now, what's the belief in the law of small numbers? 1167 00:53:41,370 --> 00:53:43,860 Well, it's the belief that, in small samples, 1168 00:53:43,860 --> 00:53:45,360 the proportion of different outcomes 1169 00:53:45,360 --> 00:53:47,930 should reflect the underlying probabilities. 1170 00:53:47,930 --> 00:53:49,680 So usually, the law of large numbers, it's 1171 00:53:49,680 --> 00:53:51,750 called the law of large numbers because you 1172 00:53:51,750 --> 00:53:53,550 need the large number of draws. 1173 00:53:53,550 --> 00:53:55,110 That's why it's called that way. 1174 00:53:55,110 --> 00:53:58,140 But people seem to think that the law of large numbers 1175 00:53:58,140 --> 00:54:01,650 also applies to small samples. 1176 00:54:01,650 --> 00:54:06,720 And so how does it explain our puzzle? 1177 00:54:06,720 --> 00:54:09,990 Well, then there's a question of does the person know 1178 00:54:09,990 --> 00:54:12,720 the underlying distribution? 1179 00:54:12,720 --> 00:54:14,970 So if you think you know the underlying distribution-- 1180 00:54:14,970 --> 00:54:18,930 for example, if you think for sure that the coin is fair 1181 00:54:18,930 --> 00:54:23,290 or the roulette table is not cheating you, you might say, 1182 00:54:23,290 --> 00:54:26,970 well, roughly, I should see as many reds 1183 00:54:26,970 --> 00:54:30,390 as I should see blacks or I should see as many heads 1184 00:54:30,390 --> 00:54:33,750 as I should see tails. 1185 00:54:33,750 --> 00:54:37,050 And so now, if I have a sample of like four draws, 1186 00:54:37,050 --> 00:54:39,090 if I already have like four heads, 1187 00:54:39,090 --> 00:54:41,220 then I might sort of think, in that small sample, 1188 00:54:41,220 --> 00:54:45,720 I need to see, essentially, a roughly equal fraction 1189 00:54:45,720 --> 00:54:47,070 of heads and tails. 1190 00:54:47,070 --> 00:54:49,110 And therefore, if I have seen four reds, 1191 00:54:49,110 --> 00:54:51,210 the next one is due to be black. 1192 00:54:51,210 --> 00:54:54,030 And of course that's not true because essentially 1193 00:54:54,030 --> 00:54:58,567 the large number does not apply to small samples 1194 00:54:58,567 --> 00:55:00,400 because it's called the law of large numbers 1195 00:55:00,400 --> 00:55:02,460 and not the law of small numbers. 1196 00:55:02,460 --> 00:55:05,160 But mistakenly, people seem to think the law 1197 00:55:05,160 --> 00:55:08,830 of small numbers applies. 1198 00:55:08,830 --> 00:55:13,180 In contrast, if the person does not know the distribution, 1199 00:55:13,180 --> 00:55:14,860 the belief in the law of small numbers 1200 00:55:14,860 --> 00:55:17,573 can lead to the hot-hand fallacy. 1201 00:55:17,573 --> 00:55:19,240 That's to say, if you're trying to learn 1202 00:55:19,240 --> 00:55:24,760 about a distribution of shots or the underlying probability, 1203 00:55:24,760 --> 00:55:28,090 if you see some good events in a row, if you see three heads 1204 00:55:28,090 --> 00:55:31,630 or tails and so on, if you see a player making three 1205 00:55:31,630 --> 00:55:35,410 shots in a row, you might sort of try to infer, oh, today 1206 00:55:35,410 --> 00:55:36,970 is a good day or a bad day. 1207 00:55:36,970 --> 00:55:38,758 And people tend to over-infer. 1208 00:55:38,758 --> 00:55:41,050 They tend to think they can learn more from these three 1209 00:55:41,050 --> 00:55:42,610 shots than they actually do. 1210 00:55:42,610 --> 00:55:47,890 Really, to be able to estimate whether a player is 1211 00:55:47,890 --> 00:55:51,700 running hot on a given day, you need quite a few draws. 1212 00:55:51,700 --> 00:55:53,980 But people think, if a player makes three shots, 1213 00:55:53,980 --> 00:55:57,130 that already means that his probability of making 1214 00:55:57,130 --> 00:56:00,230 good shots on that day are way higher than it actually is. 1215 00:56:00,230 --> 00:56:03,970 So essentially people mistakenly seem to over-infer, 1216 00:56:03,970 --> 00:56:07,090 from a small number of observations, 1217 00:56:07,090 --> 00:56:09,760 what the underlying probability distribution is. 1218 00:56:09,760 --> 00:56:13,060 And in that sense, the belief in the law of small numbers 1219 00:56:13,060 --> 00:56:16,400 can lead to the hot-hand fallacy as well. 1220 00:56:16,400 --> 00:56:21,250 So these are essentially two very different conclusions 1221 00:56:21,250 --> 00:56:24,970 that seem to come from the same underlying issue. 1222 00:56:24,970 --> 00:56:27,920 Then the last pattern I'm going to show you-- 1223 00:56:27,920 --> 00:56:29,920 by the way, there's more patterns such as those, 1224 00:56:29,920 --> 00:56:32,560 but I'm sort of just showing you three of them to give you 1225 00:56:32,560 --> 00:56:34,050 a sense of what's going on. 1226 00:56:34,050 --> 00:56:36,340 Remember the example that I showed you 1227 00:56:36,340 --> 00:56:41,380 in the first lecture, which was a question about base rates, 1228 00:56:41,380 --> 00:56:45,130 base-rate neglect, which is the question about 1 in 100 people 1229 00:56:45,130 --> 00:56:45,850 have HIV. 1230 00:56:45,850 --> 00:56:49,570 If we have a test that is 99% accurate, if a person tests 1231 00:56:49,570 --> 00:56:52,347 positive, what's the probability of having the disease? 1232 00:56:52,347 --> 00:56:53,680 And so I showed you this before. 1233 00:56:53,680 --> 00:56:55,190 The true answer is 50%. 1234 00:56:55,190 --> 00:56:59,230 But quite a few if you answered 99%. 1235 00:56:59,230 --> 00:57:04,120 And the plausible explanation is that people probably 1236 00:57:04,120 --> 00:57:07,600 forgot to take into account how few people in the population 1237 00:57:07,600 --> 00:57:10,470 have the disease to start with. 1238 00:57:10,470 --> 00:57:13,250 And so this is what's called the base-rate neglect. 1239 00:57:13,250 --> 00:57:18,140 When people are given some new information, people tend to-- 1240 00:57:18,140 --> 00:57:22,670 in this case, I guess, a test being 99% accurate, 1241 00:57:22,670 --> 00:57:25,220 people tend to focus on that piece of information. 1242 00:57:25,220 --> 00:57:28,620 They tend to neglect the underlying base rate, which, 1243 00:57:28,620 --> 00:57:31,160 in this case, is like one in 100 people 1244 00:57:31,160 --> 00:57:34,550 tend to have the disease to start with. 1245 00:57:34,550 --> 00:57:38,360 And again, that's a natural consequence 1246 00:57:38,360 --> 00:57:41,090 of focusing too much on one central aspect 1247 00:57:41,090 --> 00:57:45,080 offhand, which is the 99% accuracy of the test. 1248 00:57:45,080 --> 00:57:47,783 In a way, that, again, can be useful. 1249 00:57:47,783 --> 00:57:49,700 But it's sort of systematically getting people 1250 00:57:49,700 --> 00:57:51,110 to make wrong choices. 1251 00:57:51,110 --> 00:57:53,930 And there's quite a bit of evidence 1252 00:57:53,930 --> 00:57:55,650 showing this in the literature. 1253 00:57:55,650 --> 00:57:57,560 And here's, again, what you should be doing 1254 00:57:57,560 --> 00:57:59,690 and here's the probability of this. 1255 00:57:59,690 --> 00:58:02,620 Trust me that this really is 50%. 1256 00:58:02,620 --> 00:58:07,560 Now, the last piece here, in terms 1257 00:58:07,560 --> 00:58:12,450 of biases or systematic biases, what I showed you so far was, 1258 00:58:12,450 --> 00:58:17,650 well, people have a hard time making right choices even 1259 00:58:17,650 --> 00:58:20,810 in simple informational environments. 1260 00:58:20,810 --> 00:58:24,400 Now, two-dimensional learning is even more complicated. 1261 00:58:24,400 --> 00:58:27,910 And what I mean by two-dimensional is to say, 1262 00:58:27,910 --> 00:58:31,780 if you try to learn about news or the accuracy of news, 1263 00:58:31,780 --> 00:58:34,900 if you watch news online or anywhere else, 1264 00:58:34,900 --> 00:58:37,900 you don't need to only learn about somebody gives you 1265 00:58:37,900 --> 00:58:40,060 some information and you try to update based 1266 00:58:40,060 --> 00:58:42,340 on that information, but rather you 1267 00:58:42,340 --> 00:58:43,960 also need to learn at the same time 1268 00:58:43,960 --> 00:58:47,230 about the accuracy of the piece of information 1269 00:58:47,230 --> 00:58:48,430 that's given to you. 1270 00:58:48,430 --> 00:58:51,580 That is to say, if you watch something online, any article, 1271 00:58:51,580 --> 00:58:53,860 not only you need to sort of understand 1272 00:58:53,860 --> 00:58:55,360 what to learn from that information, 1273 00:58:55,360 --> 00:58:57,280 that piece of information, but you also 1274 00:58:57,280 --> 00:58:59,380 need to understand the underlying source 1275 00:58:59,380 --> 00:58:59,880 Of it. 1276 00:58:59,880 --> 00:59:01,547 And there's quite a bit of work recently 1277 00:59:01,547 --> 00:59:07,230 on fake news and the question on, can people in fact 1278 00:59:07,230 --> 00:59:08,940 detect fake news? 1279 00:59:08,940 --> 00:59:11,770 Does it have to do with partisanship in particular? 1280 00:59:11,770 --> 00:59:14,790 So you might think that Republicans or Democrats one 1281 00:59:14,790 --> 00:59:17,530 certainly news to be true. 1282 00:59:17,530 --> 00:59:19,620 So you might sort of think that Republicans 1283 00:59:19,620 --> 00:59:23,010 are more likely to want to believe views that are biased 1284 00:59:23,010 --> 00:59:24,900 towards Republicans, Democrats are 1285 00:59:24,900 --> 00:59:26,700 more likely to believe news that are 1286 00:59:26,700 --> 00:59:29,400 more likely to be Democrats. 1287 00:59:29,400 --> 00:59:35,310 And therefore, they might be more likely to trust fake news 1288 00:59:35,310 --> 00:59:36,880 or fall for fake news. 1289 00:59:36,880 --> 00:59:39,330 David Rand, in fact, seems like it has actually-- 1290 00:59:39,330 --> 00:59:41,910 and at least in their work seems to have-- 1291 00:59:41,910 --> 00:59:47,910 by the way, he's at Sloan, in marketing, in the marketing 1292 00:59:47,910 --> 00:59:49,170 group-- 1293 00:59:49,170 --> 00:59:52,080 he seems to find that it's less about partisanship but rather 1294 00:59:52,080 --> 00:59:54,178 about people being not paying attention 1295 00:59:54,178 --> 00:59:56,220 and being somewhat lazy and making their choices. 1296 00:59:56,220 --> 00:59:59,220 And once you draw their attention to, 1297 00:59:59,220 --> 01:00:01,140 here's an article that might be fake, 1298 01:00:01,140 --> 01:00:04,920 people might seem to be actually quite good at learning 1299 01:00:04,920 --> 01:00:08,450 about the accuracy of it. 1300 01:00:08,450 --> 01:00:10,458 Now let me just summarize for a bit 1301 01:00:10,458 --> 01:00:12,250 and then see whether there's any questions. 1302 01:00:12,250 --> 01:00:15,320 So what I've shown you is essentially using Bayes' rule 1303 01:00:15,320 --> 01:00:17,540 is difficult, or I gave you some sense of it. 1304 01:00:17,540 --> 01:00:19,880 I showed you three systematic deviations 1305 01:00:19,880 --> 01:00:22,010 from Bayesian updating, which is the gambler's 1306 01:00:22,010 --> 01:00:25,430 fallacy, the hot-hand fallacy, and base rate neglect. 1307 01:00:25,430 --> 01:00:27,620 These are all well known deviations from Bayesian 1308 01:00:27,620 --> 01:00:28,760 learning. 1309 01:00:28,760 --> 01:00:31,640 Now, 1 and 2, I have argued, can be 1310 01:00:31,640 --> 01:00:36,010 explained by the belief in the law of small numbers. 1311 01:00:36,010 --> 01:00:39,190 And there are a really important real-world implications 1312 01:00:39,190 --> 01:00:41,320 of those types of outcomes. 1313 01:00:41,320 --> 01:00:44,542 For example, one very nice recent paper-- 1314 01:00:44,542 --> 01:00:46,000 Maddie talked about it a little bit 1315 01:00:46,000 --> 01:00:47,760 in recitation, but only very briefly-- 1316 01:00:47,760 --> 01:00:51,970 is this is paper by Kelly Shue and co-authors about decision 1317 01:00:51,970 --> 01:00:54,778 making by judges, umpires, and loan officers, 1318 01:00:54,778 --> 01:00:57,070 where essentially, if you're a judge or a loan officer, 1319 01:00:57,070 --> 01:00:59,590 you see a bunch of applications over time. 1320 01:00:59,590 --> 01:01:06,280 And then judges seem to essentially engage 1321 01:01:06,280 --> 01:01:08,770 in what's the gambler's fallacy. 1322 01:01:08,770 --> 01:01:12,910 That's to say, if they see three applicants in a row that 1323 01:01:12,910 --> 01:01:15,470 happen to be very good or very bad, 1324 01:01:15,470 --> 01:01:19,120 that tends to affect the following applicant. 1325 01:01:19,120 --> 01:01:25,630 That is to say, judges, when they decide about parole 1326 01:01:25,630 --> 01:01:28,720 and so on, if they have given parole several times in a row, 1327 01:01:28,720 --> 01:01:30,760 they might think the next person who 1328 01:01:30,760 --> 01:01:35,090 is in line, if they have granted parole several times in a row, 1329 01:01:35,090 --> 01:01:37,750 the next person then would be less likely to get parole, 1330 01:01:37,750 --> 01:01:40,930 even though of course these are independent events. 1331 01:01:40,930 --> 01:01:43,330 It just happens to be that one person happens 1332 01:01:43,330 --> 01:01:47,570 to be after or before a certain applicant in a random way. 1333 01:01:47,570 --> 01:01:49,450 So there's really important decisions 1334 01:01:49,450 --> 01:01:53,440 that the gambler's fallacy might affect. 1335 01:01:53,440 --> 01:01:55,933 And more recently, when you think about base-rate neglect, 1336 01:01:55,933 --> 01:01:57,600 in some sense, the example that I showed 1337 01:01:57,600 --> 01:02:00,920 you seems to be a fairly academic example in some ways. 1338 01:02:00,920 --> 01:02:02,620 But when you think about-- 1339 01:02:02,620 --> 01:02:05,800 and I don't want to talk too much about COVID-19 since you 1340 01:02:05,800 --> 01:02:08,260 see that in your life too much already anyway-- 1341 01:02:08,260 --> 01:02:12,730 but if you think about tests for COVID-19, 1342 01:02:12,730 --> 01:02:14,985 there is now talk about antibody tests. 1343 01:02:14,985 --> 01:02:18,130 And particularly as tests that seem to be highly informative. 1344 01:02:18,130 --> 01:02:20,720 In some sense, their sensitivity and specificity, 1345 01:02:20,720 --> 01:02:23,410 meaning the type 2 errors are pretty lower. 1346 01:02:23,410 --> 01:02:26,960 Their sensitivity and specificity is very high. 1347 01:02:26,960 --> 01:02:30,350 These tests tend to be highly accurate. 1348 01:02:30,350 --> 01:02:33,220 But in fact, how much you can learn from those 1349 01:02:33,220 --> 01:02:36,160 tests is actually very limited if the overall fraction 1350 01:02:36,160 --> 01:02:38,800 of people who are infected is not that high. 1351 01:02:38,800 --> 01:02:41,290 And that's exactly the example that I just showed you. 1352 01:02:41,290 --> 01:02:43,000 If the fraction of people who are 1353 01:02:43,000 --> 01:02:46,240 positive in the overall population is relatively low, 1354 01:02:46,240 --> 01:02:47,410 you are doing a test. 1355 01:02:47,410 --> 01:02:51,580 And essentially, receiving a negative signal actually 1356 01:02:51,580 --> 01:02:56,050 is not giving you a lot of information overall. 1357 01:02:56,050 --> 01:02:59,110 And people tend to miss that part precisely 1358 01:02:59,110 --> 01:03:02,130 because of base-rate neglect. 1359 01:03:02,130 --> 01:03:03,380 Let me stop here for a second. 1360 01:03:03,380 --> 01:03:04,650 That was a little bit fast. 1361 01:03:04,650 --> 01:03:07,478 So I want to see if there's any comments or questions. 1362 01:03:22,930 --> 01:03:25,300 And then, only in the last 10 minutes, 1363 01:03:25,300 --> 01:03:28,150 talk a little bit about heuristics and biases. 1364 01:03:28,150 --> 01:03:30,040 And this is what Kahneman and Tversky really 1365 01:03:30,040 --> 01:03:33,100 are very well known for. 1366 01:03:33,100 --> 01:03:36,640 So the way to think about, then, biases and probability 1367 01:03:36,640 --> 01:03:38,350 judgments is what we already discussed. 1368 01:03:38,350 --> 01:03:39,940 The starting point is that applying 1369 01:03:39,940 --> 01:03:41,620 the laws of probability and statistics 1370 01:03:41,620 --> 01:03:43,450 is often impossibly hard. 1371 01:03:43,450 --> 01:03:45,670 So people use their quick and intuitive judgments 1372 01:03:45,670 --> 01:03:47,350 to make likelihood estimates. 1373 01:03:47,350 --> 01:03:50,230 And so there's a seminal work by Kahneman and Tversky 1374 01:03:50,230 --> 01:03:51,950 and lots of subsequent work that tries 1375 01:03:51,950 --> 01:03:56,560 to think about these biases. 1376 01:03:56,560 --> 01:03:58,660 If you're interested in learning more about this, 1377 01:03:58,660 --> 01:04:03,295 Thinking Fast and Slow by Danny Kahneman 1378 01:04:03,295 --> 01:04:06,860 is really a terrific book about thinking about this. 1379 01:04:06,860 --> 01:04:08,060 Now, what's a heuristic? 1380 01:04:08,060 --> 01:04:11,020 It's an informal algorithm that generates an approximate answer 1381 01:04:11,020 --> 01:04:12,910 to a problem, quickly. 1382 01:04:12,910 --> 01:04:15,520 And therefore, because it's informal, 1383 01:04:15,520 --> 01:04:18,340 it's kind of hard to model. 1384 01:04:18,340 --> 01:04:22,460 Well, the previous things that I showed you before-- 1385 01:04:22,460 --> 01:04:24,110 let me just go back for a second-- 1386 01:04:24,110 --> 01:04:26,090 the gambler's fallacy, hot-hand fallacy, 1387 01:04:26,090 --> 01:04:27,530 and base-rate neglect-- 1388 01:04:27,530 --> 01:04:36,020 here, you can write down models that capture these phenomena 1389 01:04:36,020 --> 01:04:37,515 pretty well. 1390 01:04:37,515 --> 01:04:39,140 For example, for the base-rate neglect, 1391 01:04:39,140 --> 01:04:42,770 you could just have a parameter like how much weight people put 1392 01:04:42,770 --> 01:04:45,740 in the base rate, how much weight 1393 01:04:45,740 --> 01:04:48,110 do people put in the new information that they get. 1394 01:04:48,110 --> 01:04:50,000 And then that parameter will essentially-- 1395 01:04:50,000 --> 01:04:51,470 you can estimate that parameter and you 1396 01:04:51,470 --> 01:04:52,850 can write down a model that's not 1397 01:04:52,850 --> 01:04:54,920 Bayesian but close to Bayesian. 1398 01:04:54,920 --> 01:04:57,830 Similarly, for the hot-hand fallacy and gambler's, you 1399 01:04:57,830 --> 01:05:00,050 can model that people essentially 1400 01:05:00,050 --> 01:05:01,970 apply the law of small numbers. 1401 01:05:01,970 --> 01:05:05,550 And again, that's a model you can write down and estimate. 1402 01:05:05,550 --> 01:05:08,320 In contrast, some of the stuff that I'm showing you next 1403 01:05:08,320 --> 01:05:11,490 is much more difficult to model because in that sense, 1404 01:05:11,490 --> 01:05:15,090 some basic laws of probability do 1405 01:05:15,090 --> 01:05:16,650 just not apply anymore in the sense 1406 01:05:16,650 --> 01:05:18,233 that people don't respect them anymore 1407 01:05:18,233 --> 01:05:20,190 when they make their belief updating. 1408 01:05:20,190 --> 01:05:22,810 Let me show you, in a second, what I mean by that. 1409 01:05:22,810 --> 01:05:25,020 But again, heuristics have a good and bad side. 1410 01:05:25,020 --> 01:05:28,368 So they speed up and make possible cognition. 1411 01:05:28,368 --> 01:05:30,660 So they help you make decisions in your everyday lives. 1412 01:05:30,660 --> 01:05:32,520 And without heuristics, it would be 1413 01:05:32,520 --> 01:05:35,680 really hard to make any decisions overall. 1414 01:05:35,680 --> 01:05:37,710 But because they're shortcuts, they occasionally 1415 01:05:37,710 --> 01:05:40,630 produce incorrect answers or biases. 1416 01:05:40,630 --> 01:05:42,870 So these are essentially unintended side effects 1417 01:05:42,870 --> 01:05:44,880 of adaptive processes. 1418 01:05:44,880 --> 01:05:48,930 And that makes it very useful to study heuristics and biases 1419 01:05:48,930 --> 01:05:50,260 together. 1420 01:05:50,260 --> 01:05:52,230 In the same way as you would study vision 1421 01:05:52,230 --> 01:05:55,350 and optical illusion together, studying them 1422 01:05:55,350 --> 01:05:59,530 jointly is very helpful because they're sort of the product 1423 01:05:59,530 --> 01:06:01,200 or the result of the same thing. 1424 01:06:01,200 --> 01:06:01,920 Things are hard. 1425 01:06:01,920 --> 01:06:03,330 Therefore, they use heuristics. 1426 01:06:03,330 --> 01:06:05,460 And because they use heuristics, that 1427 01:06:05,460 --> 01:06:09,470 leads to systematic biases. 1428 01:06:09,470 --> 01:06:12,230 OK, so now I'm going to show a few of those heuristics just 1429 01:06:12,230 --> 01:06:14,450 to give you a sense of what these are. 1430 01:06:14,450 --> 01:06:15,950 One of the most well-known ones is 1431 01:06:15,950 --> 01:06:18,140 what's called the representativeness heuristic, 1432 01:06:18,140 --> 01:06:19,880 which is about Linda. 1433 01:06:19,880 --> 01:06:24,000 Linda is 31 years old, single, outspoken, and very bright. 1434 01:06:24,000 --> 01:06:25,257 She majored in philosophy. 1435 01:06:25,257 --> 01:06:26,840 As a student, she was deeply concerned 1436 01:06:26,840 --> 01:06:29,120 with issues of discrimination and social justice 1437 01:06:29,120 --> 01:06:33,650 and has also participated in anti-nuclear demonstrations. 1438 01:06:33,650 --> 01:06:37,010 The question becomes, then, rank the following statements 1439 01:06:37,010 --> 01:06:39,440 from most to least probable. 1440 01:06:39,440 --> 01:06:41,900 And so here's a statement. 1441 01:06:41,900 --> 01:06:45,540 You can read about them, all sorts of different things. 1442 01:06:45,540 --> 01:06:48,510 In particular, what Kahneman and Tversky are focusing on 1443 01:06:48,510 --> 01:06:51,200 is items 6 and 8. 1444 01:06:51,200 --> 01:06:56,090 And item 6 is "Linda is a bank teller" 1445 01:06:56,090 --> 01:06:58,400 and item 8 is "Linda is a bank teller 1446 01:06:58,400 --> 01:07:02,150 and is active in the feminist movement." 1447 01:07:02,150 --> 01:07:04,880 Now, one thing that should be clear 1448 01:07:04,880 --> 01:07:07,670 is that 6 is more likely to an 8, 1449 01:07:07,670 --> 01:07:10,790 because essentially 8 is sort of conditioning on two statements, 1450 01:07:10,790 --> 01:07:13,280 while 6 is only conditioning on one. 1451 01:07:13,280 --> 01:07:18,050 So if Linda is a number 8, of course it's 1452 01:07:18,050 --> 01:07:21,020 implied that, if she's a bank teller and something else, 1453 01:07:21,020 --> 01:07:22,990 of course then she also is a bank teller. 1454 01:07:22,990 --> 01:07:27,440 So should always be the case that 6 is more likely than 8. 1455 01:07:27,440 --> 01:07:30,980 But what people tend to do quite consistently 1456 01:07:30,980 --> 01:07:34,760 is that, in fact, people rank it more likely that Linda 1457 01:07:34,760 --> 01:07:36,420 is both a bank teller and a feminist 1458 01:07:36,420 --> 01:07:37,790 than that she's a bank teller. 1459 01:07:37,790 --> 01:07:39,590 And there's quite a bit of work on this. 1460 01:07:39,590 --> 01:07:41,987 And showed this in various ways. 1461 01:07:41,987 --> 01:07:43,820 But essentially it's what's it the violation 1462 01:07:43,820 --> 01:07:45,445 of the conjunction law, which is a very 1463 01:07:45,445 --> 01:07:47,330 basic law of probability. 1464 01:07:47,330 --> 01:07:49,887 And precisely because it's such a basic law of probability, 1465 01:07:49,887 --> 01:07:52,220 it's hard to actually model this because essentially you 1466 01:07:52,220 --> 01:07:54,590 can't use simple things that should 1467 01:07:54,590 --> 01:07:57,900 be true in probability theory. 1468 01:07:57,900 --> 01:07:59,760 There's some potential concerns about this, 1469 01:07:59,760 --> 01:08:01,133 people misunderstanding this. 1470 01:08:01,133 --> 01:08:03,300 But there's a subsequent experiment that essentially 1471 01:08:03,300 --> 01:08:05,760 shows that that's not the case. 1472 01:08:05,760 --> 01:08:08,430 Now, what's going on here is what Kahneman and Tversky 1473 01:08:08,430 --> 01:08:11,340 called the representativeness heuristic, which people 1474 01:08:11,340 --> 01:08:15,090 essentially use similarity or representativeness as a proxy 1475 01:08:15,090 --> 01:08:17,040 for probabilistic thinking. 1476 01:08:17,040 --> 01:08:19,770 So based on the available information, 1477 01:08:19,770 --> 01:08:21,430 people form a mental image of what 1478 01:08:21,430 --> 01:08:29,100 Linda might be like and then ask about how likely it is that-- 1479 01:08:29,100 --> 01:08:30,600 for example, she's a schoolteacher-- 1480 01:08:30,600 --> 01:08:33,300 they might ask them how similar is my picture of Linda 1481 01:08:33,300 --> 01:08:35,319 to that of the school teacher? 1482 01:08:35,319 --> 01:08:37,649 And so that turns this similarity judgment 1483 01:08:37,649 --> 01:08:40,810 into a probability judgment. 1484 01:08:40,810 --> 01:08:43,560 So here, of course, the example is very much 1485 01:08:43,560 --> 01:08:46,319 rigged in the sense of they were asking things 1486 01:08:46,319 --> 01:08:48,600 about discrimination, social justice, 1487 01:08:48,600 --> 01:08:50,819 and anti-nuclear demonstration and so on, which 1488 01:08:50,819 --> 01:08:53,790 made people think, well, that's a person that's likely to be 1489 01:08:53,790 --> 01:08:55,529 part of the feminist movement. 1490 01:08:55,529 --> 01:08:57,990 And then people focus on that when making these choices. 1491 01:08:57,990 --> 01:09:00,720 And they forget the fact that, essentially, 6 1492 01:09:00,720 --> 01:09:05,292 must be, by definition, more likely than 8. 1493 01:09:05,292 --> 01:09:06,959 So that's a common thing that people do. 1494 01:09:06,959 --> 01:09:09,390 So there's quite a bit of evidence of that. 1495 01:09:12,982 --> 01:09:14,899 And again, that's a very reasonable heuristic. 1496 01:09:14,899 --> 01:09:17,689 And it probably works in many cases. 1497 01:09:17,689 --> 01:09:22,220 But it also leads to very predictably bad choices. 1498 01:09:22,220 --> 01:09:24,740 And it's a poor predictor of true probability 1499 01:09:24,740 --> 01:09:27,770 in several situations. 1500 01:09:27,770 --> 01:09:30,529 Similarly, what's called the availability heuristic, 1501 01:09:30,529 --> 01:09:33,560 people assess the probability of an event 1502 01:09:33,560 --> 01:09:35,720 by ease of which instances or occurrences can 1503 01:09:35,720 --> 01:09:37,279 be brought to mind. 1504 01:09:37,279 --> 01:09:39,720 So when you ask people, for example, 1505 01:09:39,720 --> 01:09:44,569 about are there more suicides or homicides in the US each year, 1506 01:09:44,569 --> 01:09:48,770 people will think about suicides and homicides 1507 01:09:48,770 --> 01:09:49,970 that they can recall. 1508 01:09:49,970 --> 01:09:52,430 They might think about what they watched on TV and so on. 1509 01:09:52,430 --> 01:09:54,020 And they judge the frequency of each 1510 01:09:54,020 --> 01:09:56,450 based on how many instances they can recall. 1511 01:09:56,450 --> 01:09:59,600 And people will be much more likely to recall homicides 1512 01:09:59,600 --> 01:10:01,730 because they are much more salient in the world. 1513 01:10:01,730 --> 01:10:03,800 And that leads people to think that murders 1514 01:10:03,800 --> 01:10:06,500 are way more common, which in fact is not the case. 1515 01:10:06,500 --> 01:10:09,710 Again, it's a very sensible heuristic that people use. 1516 01:10:09,710 --> 01:10:11,960 More often than not, it's easier to recall things that 1517 01:10:11,960 --> 01:10:14,670 are more common or probable. 1518 01:10:14,670 --> 01:10:16,760 So that's reasonable overall. 1519 01:10:16,760 --> 01:10:18,560 But again, there are, predictably, 1520 01:10:18,560 --> 01:10:20,420 things that are then more salient, 1521 01:10:20,420 --> 01:10:22,790 that get more attention, people tend 1522 01:10:22,790 --> 01:10:27,020 to think they're much more likely than they in fact are. 1523 01:10:27,020 --> 01:10:30,530 Let me skip the familiarity and anchoring and adjustment 1524 01:10:30,530 --> 01:10:34,580 and just sort of like conclude. 1525 01:10:34,580 --> 01:10:38,270 So what have you learned about beliefs overall? 1526 01:10:38,270 --> 01:10:40,010 So we studied several reasons why 1527 01:10:40,010 --> 01:10:42,200 people might miss information and fail to learn 1528 01:10:42,200 --> 01:10:43,220 we talk about attention. 1529 01:10:43,220 --> 01:10:44,887 Attention is limited, and therefore they 1530 01:10:44,887 --> 01:10:45,920 might miss things. 1531 01:10:45,920 --> 01:10:48,410 You talked about, why might they miss important things? 1532 01:10:48,410 --> 01:10:52,110 Well, because they might have wrong theories of the world. 1533 01:10:52,110 --> 01:10:54,440 Then we discussed people might derive utility 1534 01:10:54,440 --> 01:10:56,690 from wrong beliefs and therefore might actually not 1535 01:10:56,690 --> 01:10:59,420 want to learn, or might be systematically engaged 1536 01:10:59,420 --> 01:11:01,102 in trying to deceive themselves. 1537 01:11:01,102 --> 01:11:02,810 And then we talked about people who might 1538 01:11:02,810 --> 01:11:05,990 be bad at Bayesian learning. 1539 01:11:05,990 --> 01:11:08,357 Now, what's important here, I want 1540 01:11:08,357 --> 01:11:10,940 you to take away it's important to understand these underlying 1541 01:11:10,940 --> 01:11:12,770 reasons why people are misinformed 1542 01:11:12,770 --> 01:11:15,200 because that might lead to vastly different policy 1543 01:11:15,200 --> 01:11:16,340 implications. 1544 01:11:16,340 --> 01:11:19,160 For example, if you were to make information salient, 1545 01:11:19,160 --> 01:11:23,640 make sense, if you think people miss 1546 01:11:23,640 --> 01:11:26,670 information and important things in the world-- 1547 01:11:26,670 --> 01:11:29,600 so if you, for example, think that people just might not 1548 01:11:29,600 --> 01:11:32,210 be aware of the fact that certain foods have more 1549 01:11:32,210 --> 01:11:35,690 calories than others, then making that information very 1550 01:11:35,690 --> 01:11:37,820 salient to people when they purchase things, 1551 01:11:37,820 --> 01:11:39,512 drawing their attention to that stuff, 1552 01:11:39,512 --> 01:11:40,970 could be really important and could 1553 01:11:40,970 --> 01:11:44,970 be really powerful in helping people make better choices. 1554 01:11:44,970 --> 01:11:47,700 Or if they have wrong theories of the world, 1555 01:11:47,700 --> 01:11:49,590 they think calories are not that important 1556 01:11:49,590 --> 01:11:52,143 or if they miss the fact that smoking causes cancer, 1557 01:11:52,143 --> 01:11:54,810 well, then we should really sort of draw their attention to them 1558 01:11:54,810 --> 01:11:56,070 and help them understand it. 1559 01:11:56,070 --> 01:11:57,750 And that's what a lot of labeling 1560 01:11:57,750 --> 01:12:01,680 and a lot of making things salient often is about. 1561 01:12:01,680 --> 01:12:04,980 Now, but, if people don't want to learn, 1562 01:12:04,980 --> 01:12:06,810 if people actually have motivated beliefs 1563 01:12:06,810 --> 01:12:11,700 in certain ways, and in fact they know that, in some ways, 1564 01:12:11,700 --> 01:12:16,770 deep down inside, they know the relevant information already-- 1565 01:12:16,770 --> 01:12:19,290 smokers, for example, might know exactly 1566 01:12:19,290 --> 01:12:22,350 that smoking is bad for the health. 1567 01:12:22,350 --> 01:12:24,630 So providing them with information about smoking 1568 01:12:24,630 --> 01:12:28,560 will not actually make them learn. 1569 01:12:28,560 --> 01:12:31,650 Or like if you wanted to give somebody information about how 1570 01:12:31,650 --> 01:12:33,600 to best take care of your health when it comes 1571 01:12:33,600 --> 01:12:36,780 to Huntington's, well that's not helpful if the person 1572 01:12:36,780 --> 01:12:39,900 doesn't actually want to believe that they have Huntington's. 1573 01:12:39,900 --> 01:12:42,030 They're never going to read what you send them 1574 01:12:42,030 --> 01:12:44,430 if they want to maintain their positive image 1575 01:12:44,430 --> 01:12:48,620 about their positive beliefs about their health. 1576 01:12:48,620 --> 01:12:52,910 And in fact, when people derive actually from beliefs, 1577 01:12:52,910 --> 01:12:56,210 correcting those beliefs can make them worse off. 1578 01:12:56,210 --> 01:12:58,670 If somebody wants to believe that for the next-- 1579 01:12:58,670 --> 01:13:00,390 they know the probability of having 1580 01:13:00,390 --> 01:13:02,330 Huntington's is reasonably high, but they 1581 01:13:02,330 --> 01:13:05,210 want to maintain the beliefs that they're healthy 1582 01:13:05,210 --> 01:13:08,030 and they want to be happy for the next 10 or 20 years, 1583 01:13:08,030 --> 01:13:10,550 and then maybe, at some point, the disease will break out. 1584 01:13:10,550 --> 01:13:14,870 But their choice is they want to think that they're healthy 1585 01:13:14,870 --> 01:13:18,300 and they do not want to adjust their behavior. 1586 01:13:18,300 --> 01:13:21,950 So who are we then to tell them otherwise that's their choice? 1587 01:13:21,950 --> 01:13:24,530 If it makes them less happy, we might actually 1588 01:13:24,530 --> 01:13:25,885 make them worse off. 1589 01:13:25,885 --> 01:13:27,260 So in some ways, there's at least 1590 01:13:27,260 --> 01:13:29,870 some argument for respecting people's choices 1591 01:13:29,870 --> 01:13:33,470 if they want to be deluded, if they want to delude themselves, 1592 01:13:33,470 --> 01:13:35,540 and the consequences are not severe in the sense 1593 01:13:35,540 --> 01:13:37,670 of these situations where they can't really 1594 01:13:37,670 --> 01:13:41,300 do much about that, then in fact leaving people uninformed 1595 01:13:41,300 --> 01:13:44,740 might be the right thing to do. 1596 01:13:44,740 --> 01:13:47,960 And then understanding systematic biases 1597 01:13:47,960 --> 01:13:51,110 based on kind of help improve decisions. 1598 01:13:51,110 --> 01:13:53,150 That's, in some sense, a lot less controversial 1599 01:13:53,150 --> 01:13:54,942 if you think that people are systematically 1600 01:13:54,942 --> 01:13:57,200 making wrong choices because they learned wrongly 1601 01:13:57,200 --> 01:13:59,735 and they just have wrong information in their heads, 1602 01:13:59,735 --> 01:14:01,860 sort of providing them with the correct information 1603 01:14:01,860 --> 01:14:06,390 seems like unambiguously something [INAUDIBLE].. 1604 01:14:06,390 --> 01:14:10,372 That's all I have on beliefs for you. 1605 01:14:10,372 --> 01:14:12,330 Next time, we're going to talk about projection 1606 01:14:12,330 --> 01:14:13,500 and attribution bias. 1607 01:14:13,500 --> 01:14:15,720 But I'm also happy to answer any questions 1608 01:14:15,720 --> 01:14:18,420 in the next few minutes that you have on this lecture 1609 01:14:18,420 --> 01:14:22,220 or on the summary that I just showed you