1 00:00:00,640 --> 00:00:02,980 The following content is provided under a Creative 2 00:00:02,980 --> 00:00:04,370 Commons license. 3 00:00:04,370 --> 00:00:06,580 Your support will help MIT Open CourseWare 4 00:00:06,580 --> 00:00:10,670 continue to offer high quality educational resources for free. 5 00:00:10,670 --> 00:00:13,240 To make a donation or to view additional materials 6 00:00:13,240 --> 00:00:17,200 from hundreds of MIT courses, visit MIT Open CourseWare 7 00:00:17,200 --> 00:00:18,086 at ocw.mit.edu. 8 00:00:21,807 --> 00:00:24,140 PROFESSOR: So today we talk about ridership forecasting. 9 00:00:24,140 --> 00:00:26,890 Last time we talked about cost modeling. 10 00:00:26,890 --> 00:00:30,320 So we were looking at models to estimate 11 00:00:30,320 --> 00:00:31,760 costs of [INAUDIBLE] changes. 12 00:00:31,760 --> 00:00:34,200 Today we look at how ridership responds, 13 00:00:34,200 --> 00:00:38,360 which is the other big question the agencies, [INAUDIBLE] 14 00:00:38,360 --> 00:00:41,670 agencies have when a network will suffer changes. 15 00:00:41,670 --> 00:00:44,240 So we'll talk about route ridership prediction 16 00:00:44,240 --> 00:00:47,840 needs and issues, alternative approaches 17 00:00:47,840 --> 00:00:51,170 to ridership prediction and forecasting. 18 00:00:51,170 --> 00:00:54,860 We'll look at some examples from several agencies, 19 00:00:54,860 --> 00:00:59,930 notably the Toronto Transit commission and [INAUDIBLE].. 20 00:00:59,930 --> 00:01:03,650 We'll look at more advanced sort of DIS based models, 21 00:01:03,650 --> 00:01:06,440 simultaneous equations models, and show 22 00:01:06,440 --> 00:01:08,810 some examples of software packages 23 00:01:08,810 --> 00:01:10,760 that people use for this. 24 00:01:10,760 --> 00:01:15,200 So one note before we see is that all of this lecture 25 00:01:15,200 --> 00:01:19,100 focuses on short run root level prediction methods. 26 00:01:19,100 --> 00:01:25,160 We're not talking about regional level effects of, say, 27 00:01:25,160 --> 00:01:28,490 the land use influence on ridership, or the total amount. 28 00:01:28,490 --> 00:01:31,230 We're talking about specifically transit ridership. 29 00:01:31,230 --> 00:01:32,870 And it's a little more short run. 30 00:01:32,870 --> 00:01:39,030 So you'll see how that plays a role as we move forward. 31 00:01:39,030 --> 00:01:40,460 OK, so what are the roles? 32 00:01:40,460 --> 00:01:44,420 Why do we need ridership models? 33 00:01:44,420 --> 00:01:47,020 Of course, predicting ridership is tied to predicting revenue, 34 00:01:47,020 --> 00:01:47,520 right? 35 00:01:47,520 --> 00:01:51,860 So when there is any change an agency is interested in, 36 00:01:51,860 --> 00:01:54,350 will they lose ridership or gain ridership, 37 00:01:54,350 --> 00:01:58,100 and what impact will that have on their fair bidding? 38 00:01:58,100 --> 00:02:02,030 So first role is predicting ridership revenues 39 00:02:02,030 --> 00:02:03,110 or total fare changes. 40 00:02:03,110 --> 00:02:05,540 This is the most typical example. 41 00:02:05,540 --> 00:02:08,900 All agencies go through periodic fare changes 42 00:02:08,900 --> 00:02:12,090 unless they are completely free. 43 00:02:12,090 --> 00:02:15,020 So usually you want system-wide predictions. 44 00:02:15,020 --> 00:02:18,990 And the approaches are fairly specific calculations. 45 00:02:18,990 --> 00:02:22,010 So you might estimate some risk elasticity 46 00:02:22,010 --> 00:02:26,630 of the response, average response of ridership 47 00:02:26,630 --> 00:02:27,920 to fare changes. 48 00:02:27,920 --> 00:02:29,860 You could use a time theory [INAUDIBLE] model. 49 00:02:29,860 --> 00:02:31,818 That's a better approach if you want to control 50 00:02:31,818 --> 00:02:34,580 for external factors. 51 00:02:34,580 --> 00:02:38,690 And the best methods use a two-stage market segment model, 52 00:02:38,690 --> 00:02:41,114 recognizing that different segments of the population 53 00:02:41,114 --> 00:02:42,530 might have different elasticities, 54 00:02:42,530 --> 00:02:45,865 different responses to a fair change. 55 00:02:45,865 --> 00:02:47,990 The second role is predicting ridership and revenue 56 00:02:47,990 --> 00:02:50,990 for general agency planning and budget [INAUDIBLE],, 57 00:02:50,990 --> 00:02:56,030 so not necessarily fair changes. 58 00:02:56,030 --> 00:02:59,690 Here, again, we need a system-wide prediction. 59 00:02:59,690 --> 00:03:06,062 And the key elements are trend projection and another-- 60 00:03:06,062 --> 00:03:08,270 this is another place where a time series [INAUDIBLE] 61 00:03:08,270 --> 00:03:10,370 model plays a role. 62 00:03:10,370 --> 00:03:12,260 Here we're not necessarily looking 63 00:03:12,260 --> 00:03:15,800 at a response to something, not anything 64 00:03:15,800 --> 00:03:18,290 that the agency has control over, for example. 65 00:03:18,290 --> 00:03:21,110 But you might have the impacts of gas prices 66 00:03:21,110 --> 00:03:24,260 or other external factors, population growth, 67 00:03:24,260 --> 00:03:29,300 things like that, so trends, seasonality, things like that. 68 00:03:29,300 --> 00:03:31,490 Then there's predicting ridership revenue 69 00:03:31,490 --> 00:03:33,310 as a result of service changes. 70 00:03:33,310 --> 00:03:35,460 This is the other very typical example. 71 00:03:35,460 --> 00:03:39,200 So here we are interested in more detailed root level 72 00:03:39,200 --> 00:03:41,990 prediction as opposed to system wide prediction. 73 00:03:41,990 --> 00:03:45,940 And therefore, we need to take into account 74 00:03:45,940 --> 00:03:48,360 the higher resolution set of variables, 75 00:03:48,360 --> 00:03:51,870 like the periods of operation, the headway or frequency 76 00:03:51,870 --> 00:03:54,960 of service, route configuration, stop spacing, 77 00:03:54,960 --> 00:03:58,521 and service type because of demand response 78 00:03:58,521 --> 00:03:59,520 to all of these factors. 79 00:04:02,480 --> 00:04:05,500 And which factors affect transit ridership? 80 00:04:05,500 --> 00:04:06,920 We mentioned a few of them. 81 00:04:06,920 --> 00:04:08,510 Broadly speaking, we can divide them 82 00:04:08,510 --> 00:04:11,810 into two categories-- exogenous factors that are not 83 00:04:11,810 --> 00:04:15,590 under the control of the agency, and endogenous factors which 84 00:04:15,590 --> 00:04:17,690 the agency has some degree of control over. 85 00:04:17,690 --> 00:04:22,820 So exogenous factors include auto ownership and availability 86 00:04:22,820 --> 00:04:24,230 and operating costs. 87 00:04:24,230 --> 00:04:27,470 The cheaper it is to own and maintain and operate 88 00:04:27,470 --> 00:04:30,020 your own private automobile, the less likely 89 00:04:30,020 --> 00:04:32,240 you are to take transit. 90 00:04:32,240 --> 00:04:35,930 Fuel prices and availability, same thing. 91 00:04:35,930 --> 00:04:39,770 Demographics-- it's age, gender, other income. 92 00:04:39,770 --> 00:04:45,770 All these things affect ridership and revenues. 93 00:04:45,770 --> 00:04:47,180 And then the activity system. 94 00:04:47,180 --> 00:04:52,910 So tractors, job centers, population density, 95 00:04:52,910 --> 00:04:56,640 things like that also are affecting ridership. 96 00:04:56,640 --> 00:05:00,380 So to the degree that there's an exogenous change in the system, 97 00:05:00,380 --> 00:05:04,220 population, employment distributions, 98 00:05:04,220 --> 00:05:07,730 these things can influence ridership, demand 99 00:05:07,730 --> 00:05:10,370 for public transportation. 100 00:05:10,370 --> 00:05:12,260 So usually these things are assumed 101 00:05:12,260 --> 00:05:13,770 to be fixed in the short run. 102 00:05:13,770 --> 00:05:18,920 And we said that we would focus on short run predictions. 103 00:05:18,920 --> 00:05:22,634 Often these things are not considered in models. 104 00:05:22,634 --> 00:05:23,300 Some of them do. 105 00:05:23,300 --> 00:05:24,770 We'll look at some exceptions. 106 00:05:24,770 --> 00:05:27,380 And then there's endogenous factors. 107 00:05:27,380 --> 00:05:31,052 And these are the ones that an agency can control, so fare, 108 00:05:31,052 --> 00:05:33,260 the head of way which relates to how long people have 109 00:05:33,260 --> 00:05:36,020 to wait for service, route structure which relates 110 00:05:36,020 --> 00:05:39,590 to the walking time, access and egress times, and the ride 111 00:05:39,590 --> 00:05:40,820 time. 112 00:05:40,820 --> 00:05:43,830 And then there is crowding and reliability. 113 00:05:43,830 --> 00:05:46,340 So obviously if we change frequency, 114 00:05:46,340 --> 00:05:47,900 we might change crowding. 115 00:05:47,900 --> 00:05:50,390 And also reliability through cycle 116 00:05:50,390 --> 00:05:52,010 time and available recovery time, 117 00:05:52,010 --> 00:05:54,120 as we've seen in previous lectures. 118 00:05:54,120 --> 00:06:01,370 So crowding is interesting because as crowding increases, 119 00:06:01,370 --> 00:06:03,530 that makes transit less comfortable, right? 120 00:06:03,530 --> 00:06:08,480 So that has an impact to, say, decrease the demand. 121 00:06:08,480 --> 00:06:10,190 And then if demand is to decrease, 122 00:06:10,190 --> 00:06:12,710 then that would really reduce crowding. 123 00:06:12,710 --> 00:06:18,310 So there is some sort of loop there, a causal loop which 124 00:06:18,310 --> 00:06:22,290 reaches some equilibrium, at least, 125 00:06:22,290 --> 00:06:24,560 when we look at snapshots. 126 00:06:24,560 --> 00:06:28,730 Crowding and reliability are difficult to measure, 127 00:06:28,730 --> 00:06:32,020 especially reliability is difficult to measure 128 00:06:32,020 --> 00:06:32,800 without ADL. 129 00:06:32,800 --> 00:06:35,350 And crowding is difficult to measure without ADC. 130 00:06:35,350 --> 00:06:38,950 So usually these things are not accounted 131 00:06:38,950 --> 00:06:42,130 for in ridership prediction models if you look at practice. 132 00:06:42,130 --> 00:06:43,600 These things are often excluded. 133 00:06:43,600 --> 00:06:48,190 They are important, though, so they do effect demand. 134 00:06:48,190 --> 00:06:52,240 Reliability has a big impact on demand, 135 00:06:52,240 --> 00:06:55,740 according to revealed preference studies. 136 00:06:55,740 --> 00:06:57,222 OK, so-- 137 00:06:57,222 --> 00:06:57,930 AUDIENCE: Can I-- 138 00:06:57,930 --> 00:06:58,596 PROFESSOR: Yeah. 139 00:06:58,596 --> 00:06:59,540 [INTERPOSING VOICES] 140 00:06:59,540 --> 00:07:04,910 AUDIENCE: In the exogenous factors of auto availability, 141 00:07:04,910 --> 00:07:08,500 now in this world we're living where [INAUDIBLE],, 142 00:07:08,500 --> 00:07:10,150 does that fall under auto availability, 143 00:07:10,150 --> 00:07:11,607 even though it's not a function-- 144 00:07:11,607 --> 00:07:13,690 PROFESSOR: But that is an exogenous factor, right? 145 00:07:13,690 --> 00:07:15,330 However you want to classify it. 146 00:07:15,330 --> 00:07:18,040 So the question is, transportation network 147 00:07:18,040 --> 00:07:23,140 companies like Uber and Lyft and Fasten, et cetera, 148 00:07:23,140 --> 00:07:27,430 all of these companies provide not necessarily auto ownership. 149 00:07:27,430 --> 00:07:29,770 But they do affect-- 150 00:07:29,770 --> 00:07:32,440 they provide a different mode of transportation 151 00:07:32,440 --> 00:07:34,690 that competes in some cases and other cases 152 00:07:34,690 --> 00:07:36,880 might complement public transportation. 153 00:07:36,880 --> 00:07:38,920 So that's an exogenous factor. 154 00:07:38,920 --> 00:07:40,540 It's not necessarily under the control 155 00:07:40,540 --> 00:07:42,500 of the public transportation agency. 156 00:07:42,500 --> 00:07:44,020 And therefore it's not-- 157 00:07:44,020 --> 00:07:47,860 I haven't seen any demand model that captures it, 158 00:07:47,860 --> 00:07:51,280 although there is some research here at MIT that 159 00:07:51,280 --> 00:07:54,441 is trying to look at that because that's an emerging 160 00:07:54,441 --> 00:07:54,940 trend. 161 00:07:54,940 --> 00:08:00,880 And it's important to see if we can capture those effects. 162 00:08:00,880 --> 00:08:03,940 AUDIENCE: In certain places [INAUDIBLE].. 163 00:08:03,940 --> 00:08:06,040 PROFESSOR: We don't know, right? 164 00:08:06,040 --> 00:08:07,830 We're still investigating. 165 00:08:07,830 --> 00:08:09,397 AUDIENCE: But you said we consider 166 00:08:09,397 --> 00:08:10,480 those factors to be fixed. 167 00:08:10,480 --> 00:08:12,790 But actually, TNC is an example of something 168 00:08:12,790 --> 00:08:15,490 which is rather dynamic in terms of pricing. 169 00:08:15,490 --> 00:08:17,140 PROFESSOR: So I didn't say that we want 170 00:08:17,140 --> 00:08:18,520 to consider them to be fixed. 171 00:08:18,520 --> 00:08:22,180 I'm saying that if we look at what happens in practice, what 172 00:08:22,180 --> 00:08:25,240 the models that agencies have implemented and used 173 00:08:25,240 --> 00:08:30,550 to study, to predict ridership and revenue, those models, 174 00:08:30,550 --> 00:08:32,890 if you were to go survey the industry, 175 00:08:32,890 --> 00:08:36,039 and even some models in academia tend to assume 176 00:08:36,039 --> 00:08:37,669 these things to be fixed. 177 00:08:37,669 --> 00:08:40,510 And part of that has to do with the fact 178 00:08:40,510 --> 00:08:46,490 that the predictions of interest are shorter on predictions. 179 00:08:46,490 --> 00:08:51,370 For example, I'm going to raise fares by $0.25. 180 00:08:51,370 --> 00:08:54,370 So should I include the effect of TNCs? 181 00:08:54,370 --> 00:08:55,120 Yes or no? 182 00:08:55,120 --> 00:08:55,679 Maybe. 183 00:08:55,679 --> 00:08:56,220 I don't know. 184 00:08:56,220 --> 00:08:58,030 But it could be a factor, right? 185 00:08:58,030 --> 00:09:02,661 So good question. 186 00:09:02,661 --> 00:09:03,160 OK. 187 00:09:03,160 --> 00:09:08,450 So the traditional approach is reactive. 188 00:09:08,450 --> 00:09:10,990 So if there is an exogenous change, 189 00:09:10,990 --> 00:09:13,740 the reaction is to monitor that change 190 00:09:13,740 --> 00:09:17,730 and maybe respond to it in service planning process. 191 00:09:17,730 --> 00:09:21,250 An endogenous change, such as a fare increase, 192 00:09:21,250 --> 00:09:23,360 it means that we're sort of modifying the system 193 00:09:23,360 --> 00:09:25,480 and if we see a ridership change, 194 00:09:25,480 --> 00:09:28,570 we might then react to the ridership change. 195 00:09:28,570 --> 00:09:32,650 So it's always catching up to what we see. 196 00:09:32,650 --> 00:09:37,420 So no attempt to anticipate impacts prior to the changes. 197 00:09:37,420 --> 00:09:40,420 Current practice, typically little attention 198 00:09:40,420 --> 00:09:44,410 is paid to the problem in many public transportation agencies, 199 00:09:44,410 --> 00:09:47,560 unless it's fare increases or a major capital project, right? 200 00:09:47,560 --> 00:09:52,120 So we just set an example of TNCs being a different thing. 201 00:09:52,120 --> 00:09:54,850 We haven't really seen serious attention 202 00:09:54,850 --> 00:09:58,780 until very recently of what impact that 203 00:09:58,780 --> 00:10:02,740 would have on our ridership. 204 00:10:02,740 --> 00:10:07,510 Typical or traditional planning models are ineffective. 205 00:10:07,510 --> 00:10:09,280 Generally, they're not lethal enough 206 00:10:09,280 --> 00:10:13,100 to apply them for a short run prediction. 207 00:10:13,100 --> 00:10:16,550 So we're thinking here about four-step modeling. 208 00:10:16,550 --> 00:10:17,960 It's expensive to run. 209 00:10:17,960 --> 00:10:19,480 It takes time. 210 00:10:19,480 --> 00:10:22,120 So for short run predictions, they 211 00:10:22,120 --> 00:10:24,760 tend not to have the high [INAUDIBLE] 212 00:10:24,760 --> 00:10:28,030 solution that you need to decide something about a specific bus 213 00:10:28,030 --> 00:10:29,150 route, for example. 214 00:10:29,150 --> 00:10:31,540 So these are typically not applied 215 00:10:31,540 --> 00:10:35,170 to specific case studies. 216 00:10:35,170 --> 00:10:37,250 Yes, question? 217 00:10:37,250 --> 00:10:38,311 AUDIENCE: What is TNC? 218 00:10:38,311 --> 00:10:40,060 PROFESSOR: Transportation Network Company, 219 00:10:40,060 --> 00:10:48,400 so Uber, Lyft, Via, Fasten, all these providers. 220 00:10:48,400 --> 00:10:51,970 So ad hoc judgmental methods dominate. 221 00:10:51,970 --> 00:10:57,760 So now these methods are not the most accurate, obviously. 222 00:10:57,760 --> 00:11:00,990 But they are useful, especially if there's a small change 223 00:11:00,990 --> 00:11:03,400 and you have many proposals on the table, 224 00:11:03,400 --> 00:11:07,150 you have people experienced with the area or the network. 225 00:11:07,150 --> 00:11:10,330 They might have a good sense of at least the direction 226 00:11:10,330 --> 00:11:16,730 and the overall level of change that one might expect. 227 00:11:16,730 --> 00:11:19,610 But they tend to be quite inaccurate 228 00:11:19,610 --> 00:11:24,230 for disaggregate predictions, long term predictions, or very 229 00:11:24,230 --> 00:11:26,580 large changes in the network. 230 00:11:26,580 --> 00:11:29,090 So if you're doing something small, maybe OK. 231 00:11:29,090 --> 00:11:31,520 If you're doing something significant, not so much. 232 00:11:36,480 --> 00:11:41,610 OK, so we have four approaches here to modeling, right, is it? 233 00:11:41,610 --> 00:11:45,990 Ordered from in order of increasing sophistication, 234 00:11:45,990 --> 00:11:48,150 starting with professional judgment, 235 00:11:48,150 --> 00:11:50,370 then non-committal survey techniques, 236 00:11:50,370 --> 00:11:52,040 then cross-sectional data models, 237 00:11:52,040 --> 00:11:53,580 then time series data models. 238 00:11:53,580 --> 00:11:56,220 And let's talk a little bit about each one. 239 00:11:56,220 --> 00:11:59,460 Professional judgment, so this is 240 00:11:59,460 --> 00:12:03,390 when you bring in as an agency either experts 241 00:12:03,390 --> 00:12:05,910 within the agency or consultants whose 242 00:12:05,910 --> 00:12:08,440 job is to predict ridership. 243 00:12:08,440 --> 00:12:11,850 They have a lot of experience doing so. 244 00:12:11,850 --> 00:12:14,850 They might have local knowledge. 245 00:12:14,850 --> 00:12:17,910 And of course, the problem is that there 246 00:12:17,910 --> 00:12:20,730 haven't been any scientific studies that indicate 247 00:12:20,730 --> 00:12:22,490 the accuracy of these methods. 248 00:12:22,490 --> 00:12:26,560 They tend not to be reproducible every time you call someone in. 249 00:12:26,560 --> 00:12:29,070 That's sort of a one shot deal, right? 250 00:12:29,070 --> 00:12:33,960 So you're not really able to replicate those results easily 251 00:12:33,960 --> 00:12:37,290 or understand what led to the prediction. 252 00:12:37,290 --> 00:12:41,040 This reflects lack of faith in formal models in some cases, 253 00:12:41,040 --> 00:12:43,620 lack of data or technical expertise 254 00:12:43,620 --> 00:12:45,270 to support the development from all, 255 00:12:45,270 --> 00:12:48,270 or a lack of time or budget to do so. 256 00:12:48,270 --> 00:12:51,090 And maybe most importantly relative on importance 257 00:12:51,090 --> 00:12:52,920 of the topic of ridership prediction 258 00:12:52,920 --> 00:12:56,070 in general to an agency is compared 259 00:12:56,070 --> 00:13:00,640 to the impact of changes or impacts on existing ridership. 260 00:13:00,640 --> 00:13:04,230 So agencies tend to worry more about the reaction 261 00:13:04,230 --> 00:13:06,240 of the public that currently rides 262 00:13:06,240 --> 00:13:09,030 their service than the general increase 263 00:13:09,030 --> 00:13:10,110 or decrease in ridership. 264 00:13:10,110 --> 00:13:12,780 Although, they do they are concerned about that. 265 00:13:12,780 --> 00:13:16,750 They worry more about the last four. 266 00:13:16,750 --> 00:13:20,700 So any questions on professional judgment? 267 00:13:20,700 --> 00:13:23,850 We all know what this is, right? 268 00:13:23,850 --> 00:13:24,780 OK. 269 00:13:24,780 --> 00:13:27,600 Survey-based methods, the next level of sophistication. 270 00:13:27,600 --> 00:13:34,170 So here we go out and ask people, 271 00:13:34,170 --> 00:13:38,430 would you ride a bus route if it were 272 00:13:38,430 --> 00:13:41,040 extended to Kendall Square. 273 00:13:41,040 --> 00:13:46,300 Or if we provide a new train line, 274 00:13:46,300 --> 00:13:49,090 would you do this instead of driving your own car? 275 00:13:49,090 --> 00:13:49,810 Things like this. 276 00:13:49,810 --> 00:13:50,934 And how much would you pay? 277 00:13:50,934 --> 00:13:52,870 So you can survey people and ask them, 278 00:13:52,870 --> 00:13:54,280 would you take this or not? 279 00:13:57,810 --> 00:14:01,094 You lay out a proposal and you ask people for their reaction. 280 00:14:01,094 --> 00:14:03,010 This isn't called non-committal because people 281 00:14:03,010 --> 00:14:05,470 are saying that they would do it or not do it. 282 00:14:05,470 --> 00:14:07,230 But they are not committed to doing so. 283 00:14:07,230 --> 00:14:10,030 And there's no promise that they will do so. 284 00:14:10,030 --> 00:14:12,400 So the stated preference might be 285 00:14:12,400 --> 00:14:14,380 quite different from the revealed preference 286 00:14:14,380 --> 00:14:18,760 if one were to actually follow up with what these people 287 00:14:18,760 --> 00:14:19,730 actually, the-- 288 00:14:19,730 --> 00:14:20,580 question? 289 00:14:20,580 --> 00:14:24,974 AUDIENCE: Is there a sense of accuracy [INAUDIBLE]?? 290 00:14:24,974 --> 00:14:26,140 PROFESSOR: It varies widely. 291 00:14:26,140 --> 00:14:28,840 And there are many papers and lots 292 00:14:28,840 --> 00:14:30,430 of studies that have looked at this. 293 00:14:30,430 --> 00:14:31,030 So-- 294 00:14:31,030 --> 00:14:32,860 AUDIENCE: [INAUDIBLE] can you just give us a sense? 295 00:14:32,860 --> 00:14:34,193 PROFESSOR: So, yeah, right here. 296 00:14:34,193 --> 00:14:37,690 So typically we have a survey. 297 00:14:37,690 --> 00:14:41,620 You might ask, say, 500 people or 300 people 298 00:14:41,620 --> 00:14:43,150 whether they would do this or not. 299 00:14:43,150 --> 00:14:46,885 A typical example is a roadside survey where you stop cars 300 00:14:46,885 --> 00:14:48,760 and you ask them, would you have taken a park 301 00:14:48,760 --> 00:14:53,587 and ride using the metro system or an express bus line? 302 00:14:53,587 --> 00:14:55,420 And they would say yes or no and off you go. 303 00:14:55,420 --> 00:14:57,070 So noncommittal, right? 304 00:14:57,070 --> 00:15:00,370 There's no-- there's no necessity to actually do 305 00:15:00,370 --> 00:15:01,480 that if you said yes. 306 00:15:01,480 --> 00:15:04,810 So the problem is that there's this noncommittal bias. 307 00:15:04,810 --> 00:15:11,620 People tend to say yes more than they would actually do so. 308 00:15:11,620 --> 00:15:13,900 This has to do with sort of psychological factors. 309 00:15:13,900 --> 00:15:19,490 Sometimes the desire somewhat subconscious to be friendly-- 310 00:15:19,490 --> 00:15:21,740 there is a perception that the person doing the survey 311 00:15:21,740 --> 00:15:24,110 is interested in [INAUDIBLE] result. 312 00:15:24,110 --> 00:15:25,880 And therefore, as human beings, we 313 00:15:25,880 --> 00:15:29,880 try to be friendly or try to be positive, right? 314 00:15:29,880 --> 00:15:31,130 This is psychological. 315 00:15:31,130 --> 00:15:32,180 It's subconscious. 316 00:15:32,180 --> 00:15:33,200 It affects results. 317 00:15:33,200 --> 00:15:35,660 So what happens-- what ends up happening 318 00:15:35,660 --> 00:15:37,610 is that there's an adjustment factor 319 00:15:37,610 --> 00:15:39,390 for the non-committal bias. 320 00:15:39,390 --> 00:15:42,390 And this is where professional judgment comes in. 321 00:15:42,390 --> 00:15:44,140 So you do have some survey technique. 322 00:15:44,140 --> 00:15:46,720 But at the end, you need this fudge factor. 323 00:15:46,720 --> 00:15:51,200 You multiply it by anywhere between 5% and 50%. 324 00:15:51,200 --> 00:15:54,230 And what that range-- what that specific number is 325 00:15:54,230 --> 00:15:57,950 depends on the specifics of a problem, who you survey, 326 00:15:57,950 --> 00:16:01,760 how many people you survey, the experience with the person 327 00:16:01,760 --> 00:16:03,650 conducting the study. 328 00:16:03,650 --> 00:16:05,480 Any questions on non-committal surveys? 329 00:16:05,480 --> 00:16:06,938 AUDIENCE: Yeah, maybe it could also 330 00:16:06,938 --> 00:16:10,600 be that I would say yes because I personally wouldn't use it, 331 00:16:10,600 --> 00:16:12,725 but I have a kid who might use it. 332 00:16:12,725 --> 00:16:14,450 And so I know someone who would use it. 333 00:16:14,450 --> 00:16:16,970 So I say, yeah, yeah, I use it. 334 00:16:16,970 --> 00:16:17,990 And I might even be-- 335 00:16:17,990 --> 00:16:18,823 [INTERPOSING VOICES] 336 00:16:18,823 --> 00:16:20,330 --blatantly lying about myself. 337 00:16:20,330 --> 00:16:22,991 But I say, oh, well, if they provide this service, 338 00:16:22,991 --> 00:16:23,990 then my kid will use it. 339 00:16:23,990 --> 00:16:25,304 And so I want to say yes. 340 00:16:25,304 --> 00:16:26,720 PROFESSOR: I'll repeat the comment 341 00:16:26,720 --> 00:16:30,560 for the benefit of people listening on the web later on. 342 00:16:30,560 --> 00:16:35,090 So the comment is that another reason for a noncommittal bias 343 00:16:35,090 --> 00:16:39,530 is that you might know people who would write it. 344 00:16:39,530 --> 00:16:42,620 Or another instance of this is, maybe you ask, 345 00:16:42,620 --> 00:16:44,540 would you today have done this? 346 00:16:44,540 --> 00:16:46,010 And maybe today you wouldn't have, 347 00:16:46,010 --> 00:16:47,960 but you think, typically I would, right? 348 00:16:47,960 --> 00:16:51,459 So then you want to be helpful and give what is most typical. 349 00:16:51,459 --> 00:16:53,750 Problem with that is that if everybody's is doing that, 350 00:16:53,750 --> 00:16:55,250 you have a bias, right? 351 00:16:55,250 --> 00:16:58,500 Because you are doing, hopefully, a random sample. 352 00:16:58,500 --> 00:17:02,295 So you could have surveyed my spouse or my kid, 353 00:17:02,295 --> 00:17:02,920 but you didn't. 354 00:17:02,920 --> 00:17:04,380 You picked someone at random. 355 00:17:04,380 --> 00:17:09,800 If everyone you pick is trying to answer for someone else 356 00:17:09,800 --> 00:17:12,349 or answer for what I would have done on a different day 357 00:17:12,349 --> 00:17:15,609 and that all leads to a positive bias, then in aggregate, 358 00:17:15,609 --> 00:17:16,920 you could have a problem. 359 00:17:16,920 --> 00:17:21,380 So it doesn't help if people are dishonest and trying 360 00:17:21,380 --> 00:17:23,690 to be helpful, right? 361 00:17:23,690 --> 00:17:24,829 Any questions-- yeah? 362 00:17:24,829 --> 00:17:25,454 AUDIENCE: Yeah. 363 00:17:25,454 --> 00:17:28,190 Is there any way to randomize the direction 364 00:17:28,190 --> 00:17:29,210 of the question and-- 365 00:17:29,210 --> 00:17:29,500 [INTERPOSING VOICES] 366 00:17:29,500 --> 00:17:30,710 PROFESSOR: Yes, but that's-- 367 00:17:30,710 --> 00:17:31,930 so that's the next method. 368 00:17:31,930 --> 00:17:34,440 So we'll talk about that right now. 369 00:17:34,440 --> 00:17:38,030 So this is not recommended because this bias can 370 00:17:38,030 --> 00:17:41,254 be significant and it's hard to quantify. 371 00:17:41,254 --> 00:17:45,920 Now there's something called stated preference analysis 372 00:17:45,920 --> 00:17:49,520 or conjoint analysis. 373 00:17:49,520 --> 00:17:52,280 And is where we apply statistical techniques to try 374 00:17:52,280 --> 00:17:53,730 to correct for that bias. 375 00:17:53,730 --> 00:17:59,240 So this originated in mathematical psychology. 376 00:17:59,240 --> 00:18:05,780 And what we do is that it's a rigorous, detailed sign 377 00:18:05,780 --> 00:18:07,017 of the experiment. 378 00:18:07,017 --> 00:18:09,350 And usually what you do is that instead of asking people 379 00:18:09,350 --> 00:18:12,350 if they would take this or not, you say, 380 00:18:12,350 --> 00:18:14,420 would you rather take this option or this option? 381 00:18:14,420 --> 00:18:15,560 And you put some bundles. 382 00:18:15,560 --> 00:18:18,410 And you ask them to trade them off multiple times. 383 00:18:18,410 --> 00:18:20,810 You say, well, what if it cost one more dollar? 384 00:18:20,810 --> 00:18:23,370 What if-- and you have multiple levels. 385 00:18:23,370 --> 00:18:26,390 So people start contradicting themselves sometimes 386 00:18:26,390 --> 00:18:28,310 in these trade-offs. 387 00:18:28,310 --> 00:18:32,360 And you can detect the biases using those contradictions, 388 00:18:32,360 --> 00:18:37,100 or at least you can rank the relative importance 389 00:18:37,100 --> 00:18:39,200 of different factors to that person 390 00:18:39,200 --> 00:18:43,130 and then apply those factors to correct for the bias. 391 00:18:43,130 --> 00:18:48,240 So this is, again, this area's evolving. 392 00:18:48,240 --> 00:18:52,460 It is being applied already by high end consulting firms 393 00:18:52,460 --> 00:18:54,490 and also by researchers. 394 00:18:54,490 --> 00:18:58,945 So if you find research papers on the [INAUDIBLE] 395 00:18:58,945 --> 00:19:00,320 stated preference surveys, you'll 396 00:19:00,320 --> 00:19:03,870 find these kinds of techniques being applied. 397 00:19:03,870 --> 00:19:08,180 This is very useful for new services or new service areas. 398 00:19:08,180 --> 00:19:10,840 Why is that? 399 00:19:10,840 --> 00:19:13,510 Why would this technique be especially useful for that? 400 00:19:13,510 --> 00:19:14,260 Yeah, [INAUDIBLE]? 401 00:19:14,260 --> 00:19:18,430 AUDIENCE: Because there is an existing status 402 00:19:18,430 --> 00:19:22,760 quo that they're implicitly comparing against. 403 00:19:22,760 --> 00:19:25,150 PROFESSOR: If there isn't any service to start with, 404 00:19:25,150 --> 00:19:28,670 you're talking about a new line or a new area of service 405 00:19:28,670 --> 00:19:31,760 where transit isn't really present or not 406 00:19:31,760 --> 00:19:33,640 present to the degree that you're proposing, 407 00:19:33,640 --> 00:19:37,480 then there's no good way of looking at the existing amount 408 00:19:37,480 --> 00:19:39,010 and extrapolate from there. 409 00:19:39,010 --> 00:19:41,980 So then you have to ask people. 410 00:19:41,980 --> 00:19:47,500 And this is a method to correct for [INAUDIBLE] bias. 411 00:19:47,500 --> 00:19:51,800 One more comment-- the MTA useless in the 1980s. 412 00:19:51,800 --> 00:19:53,380 And they concluded that they would 413 00:19:53,380 --> 00:19:56,200 raise fares to repay bonds that they 414 00:19:56,200 --> 00:20:00,110 had taken for capital projects. 415 00:20:00,110 --> 00:20:03,795 So it's been applied successfully. 416 00:20:06,530 --> 00:20:09,390 The next method or family of methods 417 00:20:09,390 --> 00:20:11,600 is cross-sectional models. 418 00:20:11,600 --> 00:20:15,550 So here we develop a model of demand 419 00:20:15,550 --> 00:20:18,910 as a function of route and demographic data 420 00:20:18,910 --> 00:20:20,530 to explain ridership. 421 00:20:20,530 --> 00:20:25,060 So we're looking at a snapshot at a single point in time 422 00:20:25,060 --> 00:20:29,830 or, say, we pick a period of three months altogether 423 00:20:29,830 --> 00:20:32,500 and we look at the ridership over those three months 424 00:20:32,500 --> 00:20:35,860 across different bus routes in the region. 425 00:20:35,860 --> 00:20:38,380 And then we observe that different bus routes have 426 00:20:38,380 --> 00:20:42,670 different demand and they have different frequencies 427 00:20:42,670 --> 00:20:45,550 and they run through sectors with different population 428 00:20:45,550 --> 00:20:48,250 densities and different employment densities. 429 00:20:48,250 --> 00:20:50,710 And we tried to relate the differences in demand 430 00:20:50,710 --> 00:20:54,080 on those routes to the ridership to those characteristics. 431 00:20:54,080 --> 00:20:56,440 This is called a cross-sectional model. 432 00:20:56,440 --> 00:20:58,630 So again, we explained the differences in demand 433 00:20:58,630 --> 00:21:01,420 as a function of the differences in characteristics. 434 00:21:01,420 --> 00:21:05,320 And there are different approaches here. 435 00:21:05,320 --> 00:21:08,230 Rules of thumb is the simplest. 436 00:21:08,230 --> 00:21:13,120 So we might say if it's key bus routes, 437 00:21:13,120 --> 00:21:14,820 then it has this much demand. 438 00:21:14,820 --> 00:21:19,630 That's a very sort of simple, crude example. 439 00:21:19,630 --> 00:21:22,120 The similar routes method is I'm going 440 00:21:22,120 --> 00:21:24,950 to convert this bus route to-- 441 00:21:24,950 --> 00:21:26,345 I'm going to add frequency to it, 442 00:21:26,345 --> 00:21:28,150 say, increase its frequency. 443 00:21:28,150 --> 00:21:30,580 And then it's going to look a lot more like this other bus 444 00:21:30,580 --> 00:21:32,621 route that I have in another part of the network. 445 00:21:32,621 --> 00:21:35,260 So ridership should look a lot more like that. 446 00:21:35,260 --> 00:21:38,170 Maybe I separate my bus routes in the network 447 00:21:38,170 --> 00:21:42,460 into high ridership key bus routes 448 00:21:42,460 --> 00:21:48,040 that have certain high frequency and other less frequent bus 449 00:21:48,040 --> 00:21:50,980 routes that run on less dense parts of the network. 450 00:21:50,980 --> 00:21:52,750 And I take those averages and I apply 451 00:21:52,750 --> 00:21:54,400 them to extrapolate, right? 452 00:21:54,400 --> 00:21:55,540 So that's a similar route. 453 00:21:55,540 --> 00:21:57,010 You find similar routes. 454 00:21:57,010 --> 00:22:00,220 You use those to make a prediction. 455 00:22:00,220 --> 00:22:02,920 Then there's the multiple factor trip rate model. 456 00:22:02,920 --> 00:22:05,050 You, again, look at-- 457 00:22:05,050 --> 00:22:06,970 you do a regression, maybe. 458 00:22:06,970 --> 00:22:10,930 You can do this with regression or by separate factors. 459 00:22:10,930 --> 00:22:13,450 So those are the last two bullets here. 460 00:22:13,450 --> 00:22:17,590 So you essentially run regression where you have-- 461 00:22:17,590 --> 00:22:19,420 maybe you load up a GIS layer and you 462 00:22:19,420 --> 00:22:23,290 have population density, employment density, frequency. 463 00:22:23,290 --> 00:22:28,250 And out of the regression comes these factors 464 00:22:28,250 --> 00:22:30,192 where you can control for multiple factors 465 00:22:30,192 --> 00:22:30,900 at the same time. 466 00:22:34,530 --> 00:22:37,676 Any questions on cross-sectional methods? 467 00:22:37,676 --> 00:22:38,540 [INAUDIBLE] 468 00:22:38,540 --> 00:22:40,790 AUDIENCE: Because these models generally 469 00:22:40,790 --> 00:22:42,500 focus only on endogenous factors and not 470 00:22:42,500 --> 00:22:46,060 exogenous, so they don't physically 471 00:22:46,060 --> 00:22:48,500 capture the demand, how the demand is [INAUDIBLE] 472 00:22:48,500 --> 00:22:50,900 with difference to other factors. 473 00:22:50,900 --> 00:22:54,035 So how can this be generalized so that second point is 474 00:22:54,035 --> 00:22:55,610 that similar routes-- 475 00:22:55,610 --> 00:22:57,440 like, it can be generalized. 476 00:22:57,440 --> 00:23:00,480 One that uses less data or is shallow in its nature 477 00:23:00,480 --> 00:23:03,069 could not generally be generalized, right? 478 00:23:03,069 --> 00:23:03,860 PROFESSOR: Correct. 479 00:23:03,860 --> 00:23:04,359 Yes. 480 00:23:04,359 --> 00:23:08,104 These models are hard to generalize because they don't-- 481 00:23:08,104 --> 00:23:10,520 actually, any cross-sectional model is hard to generalize, 482 00:23:10,520 --> 00:23:11,020 right? 483 00:23:11,020 --> 00:23:15,470 Because you are looking at a specific time 484 00:23:15,470 --> 00:23:17,030 window of analysis. 485 00:23:17,030 --> 00:23:20,240 So you are already, in doing so, excluding 486 00:23:20,240 --> 00:23:23,600 any longer-term trends that are often exogenous, 487 00:23:23,600 --> 00:23:28,340 like fuel prices and population changes and things like that. 488 00:23:28,340 --> 00:23:31,610 So yes, by doing a model, even the most 489 00:23:31,610 --> 00:23:34,970 sophisticated of these, which is a multivariate regression 490 00:23:34,970 --> 00:23:38,870 model, if you have any factors that you leave out, 491 00:23:38,870 --> 00:23:41,780 that can lead to biases in the estimators, right? 492 00:23:41,780 --> 00:23:46,790 So will you actually see that bias? 493 00:23:46,790 --> 00:23:49,810 Well, that depends on how short run of a prediction 494 00:23:49,810 --> 00:23:51,260 are you trying to do? 495 00:23:51,260 --> 00:23:54,410 Is your prediction going to be applied 496 00:23:54,410 --> 00:23:57,020 to a period in the short term that 497 00:23:57,020 --> 00:23:58,430 is very similar to the period you 498 00:23:58,430 --> 00:24:03,260 have now where these exogenous factors have been out, 499 00:24:03,260 --> 00:24:05,510 have not changed, and therefore there 500 00:24:05,510 --> 00:24:11,360 is no impact on the bias of this model. 501 00:24:11,360 --> 00:24:13,820 Or are you going to apply this-- 502 00:24:13,820 --> 00:24:15,820 abuse this, let's say, misapply it-- 503 00:24:15,820 --> 00:24:20,630 abuse it to predict outside of that window, 504 00:24:20,630 --> 00:24:23,300 outside of that domain, where you do have changes 505 00:24:23,300 --> 00:24:25,280 that you did not include in the model, 506 00:24:25,280 --> 00:24:27,560 and they have affected ridership? 507 00:24:27,560 --> 00:24:30,300 Yes. 508 00:24:30,300 --> 00:24:32,710 Any other questions about cross-sectional methods? 509 00:24:36,890 --> 00:24:38,076 OK. 510 00:24:38,076 --> 00:24:39,750 Let's move on. 511 00:24:39,750 --> 00:24:41,060 A little bit about regression. 512 00:24:41,060 --> 00:24:47,630 So this is an example of how an agency might respond 513 00:24:47,630 --> 00:24:51,710 to increases in demand by increasing frequency and vise 514 00:24:51,710 --> 00:24:52,440 versa. 515 00:24:52,440 --> 00:24:53,900 So you look at-- 516 00:24:53,900 --> 00:24:57,170 you start with an observation that ridership is quite high 517 00:24:57,170 --> 00:25:00,680 and the frequency is some amount right now. 518 00:25:00,680 --> 00:25:03,320 And then you react to that. 519 00:25:03,320 --> 00:25:06,390 So as an agency, you might say, well, this bus route 520 00:25:06,390 --> 00:25:09,330 is quite crowded, therefore, let's increase 521 00:25:09,330 --> 00:25:11,696 the frequency, right? 522 00:25:11,696 --> 00:25:13,070 And let's increase the frequency. 523 00:25:19,685 --> 00:25:21,810 That's going to have an effect on the demand curve. 524 00:25:21,810 --> 00:25:26,130 So all of these lines, d, are demand curves. 525 00:25:26,130 --> 00:25:31,920 And this demand curve shows the response of ridership 526 00:25:31,920 --> 00:25:34,440 to a change in frequency, assuming 527 00:25:34,440 --> 00:25:36,720 everything else stays the same. 528 00:25:36,720 --> 00:25:39,960 But there are also other changes that are not included here. 529 00:25:39,960 --> 00:25:42,930 So you might increase frequency but there 530 00:25:42,930 --> 00:25:46,620 might be other exogenous factors, exogenous in this case 531 00:25:46,620 --> 00:25:49,500 because frequency is the only variable being shown here. 532 00:25:49,500 --> 00:25:51,480 Any other variable that increases demand 533 00:25:51,480 --> 00:25:54,420 will not result in a change along the demand curve. 534 00:25:54,420 --> 00:25:56,880 It will actually shift the demand curve, right? 535 00:25:56,880 --> 00:26:01,830 So you have demand curves going in a third dimension, 536 00:26:01,830 --> 00:26:06,280 increasing demand in a third dimension due to other factors. 537 00:26:06,280 --> 00:26:10,450 So typically, again, agencies are reactive. 538 00:26:10,450 --> 00:26:13,800 They sort of observe an increase then they move to the next one 539 00:26:13,800 --> 00:26:15,750 and they're always thinking about one demand 540 00:26:15,750 --> 00:26:17,680 curve, if at all. 541 00:26:17,680 --> 00:26:21,850 And so it's better for this reason 542 00:26:21,850 --> 00:26:24,840 to look at multiple factors at the same time instead of just 543 00:26:24,840 --> 00:26:28,680 one factor, such as frequency. 544 00:26:28,680 --> 00:26:35,440 OK, so these are some typical transit elasticities. 545 00:26:35,440 --> 00:26:38,580 We have elasticities to fare, to headway, 546 00:26:38,580 --> 00:26:39,980 and to total travel time here. 547 00:26:39,980 --> 00:26:41,840 And so these are the typical values 548 00:26:41,840 --> 00:26:44,154 if you look at many studies and take averages, 549 00:26:44,154 --> 00:26:45,070 this is what you find. 550 00:26:45,070 --> 00:26:48,070 So elasticity to fare, negative 0.3. 551 00:26:48,070 --> 00:26:50,360 Let's do a quick recap. 552 00:26:50,360 --> 00:26:51,546 What is elasticity? 553 00:26:54,621 --> 00:26:55,120 Henry? 554 00:26:55,120 --> 00:26:57,890 AUDIENCE: Percent change in demand 555 00:26:57,890 --> 00:26:59,980 over percent change in price. 556 00:26:59,980 --> 00:27:00,650 PROFESSOR: Yes. 557 00:27:00,650 --> 00:27:03,520 That's-- so the answer is percent change in demand over 558 00:27:03,520 --> 00:27:05,030 percent change in price. 559 00:27:05,030 --> 00:27:08,090 I want to make two adjustments to that definition. 560 00:27:08,090 --> 00:27:09,620 That's often how it's calculated. 561 00:27:09,620 --> 00:27:15,230 But it's actually the derivative of the curve, not necessarily 562 00:27:15,230 --> 00:27:16,420 a percent change. 563 00:27:16,420 --> 00:27:19,340 That is an approximation of the derivative. 564 00:27:19,340 --> 00:27:21,519 And the other is that it's not just to price. 565 00:27:21,519 --> 00:27:23,060 In this case, we're looking at change 566 00:27:23,060 --> 00:27:27,110 to fare towards headway or to other disutilities, so 567 00:27:27,110 --> 00:27:29,220 any other factor that affects disutility. 568 00:27:29,220 --> 00:27:33,220 So what is the difference between-- 569 00:27:33,220 --> 00:27:36,226 what is negative elasticity? 570 00:27:39,800 --> 00:27:41,874 [INTERPOSING VOICES] 571 00:27:41,874 --> 00:27:43,790 AUDIENCE: As all those variables are going up, 572 00:27:43,790 --> 00:27:45,017 your ridership is going down. 573 00:27:45,017 --> 00:27:45,600 PROFESSOR: OK. 574 00:27:45,600 --> 00:27:49,760 So as fare increases, the ridership goes down. 575 00:27:49,760 --> 00:27:51,000 Great. 576 00:27:51,000 --> 00:27:56,860 What is the meaning of an elasticity 577 00:27:56,860 --> 00:27:59,840 with an absolute value between 0 and 1 instead 578 00:27:59,840 --> 00:28:00,590 of greater than 1? 579 00:28:04,454 --> 00:28:05,870 Or, rather, what is the difference 580 00:28:05,870 --> 00:28:09,960 between an elasticity that has absolute value less than one 581 00:28:09,960 --> 00:28:12,770 versus one that has an absolute value greater than one? 582 00:28:12,770 --> 00:28:14,610 AUDIENCE: Relatively inelastic? 583 00:28:14,610 --> 00:28:15,402 PROFESSOR: OK, so-- 584 00:28:15,402 --> 00:28:16,485 AUDIENCE: So for example-- 585 00:28:16,485 --> 00:28:18,260 PROFESSOR: --inelastic versus elastic. 586 00:28:18,260 --> 00:28:21,320 Those are the terms used to explain the differences, right? 587 00:28:21,320 --> 00:28:24,159 So what is an elastic change? 588 00:28:24,159 --> 00:28:26,200 AUDIENCE: A change in price would not necessarily 589 00:28:26,200 --> 00:28:32,600 diminish the demands from the quantities demanded by people. 590 00:28:32,600 --> 00:28:33,730 You could have-- 591 00:28:33,730 --> 00:28:36,260 PROFESSOR: Is this fair value of [INAUDIBLE] 592 00:28:36,260 --> 00:28:37,760 inelastic or elastic? 593 00:28:37,760 --> 00:28:38,840 AUDIENCE: Elastic. 594 00:28:38,840 --> 00:28:40,220 PROFESSOR: This is inelastic. 595 00:28:40,220 --> 00:28:40,570 AUDIENCE: I'm sorry. 596 00:28:40,570 --> 00:28:41,530 Yeah, inelastic, yes. 597 00:28:41,530 --> 00:28:42,988 PROFESSOR: Yeah, this is inelastic. 598 00:28:42,988 --> 00:28:49,574 So what does that mean in practice? 599 00:28:49,574 --> 00:28:50,740 AUDIENCE: A change in fare-- 600 00:28:50,740 --> 00:28:51,290 PROFESSOR: So I know that-- 601 00:28:51,290 --> 00:28:52,123 [INTERPOSING VOICES] 602 00:28:52,123 --> 00:28:54,930 AUDIENCE: --would not lead to an equivalent change in ridership. 603 00:28:54,930 --> 00:28:55,513 PROFESSOR: OK. 604 00:28:55,513 --> 00:28:58,499 AUDIENCE: It would lead to less than the equivalent change 605 00:28:58,499 --> 00:28:59,040 in ridership. 606 00:28:59,040 --> 00:28:59,623 PROFESSOR: OK. 607 00:28:59,623 --> 00:29:07,040 Therefore if you increase fares, you 608 00:29:07,040 --> 00:29:10,460 will lose some ridership, but not enough 609 00:29:10,460 --> 00:29:12,700 to lose total [INAUDIBLE]. 610 00:29:12,700 --> 00:29:13,430 AUDIENCE: Right. 611 00:29:13,430 --> 00:29:14,300 PROFESSOR: Right? 612 00:29:14,300 --> 00:29:17,120 If this volume were greater than-- have an absolute value 613 00:29:17,120 --> 00:29:19,611 greater than 1, then you might actually 614 00:29:19,611 --> 00:29:21,360 decrease-- the response in the [INAUDIBLE] 615 00:29:21,360 --> 00:29:24,740 might be so strong that you might end up loosing revenue 616 00:29:24,740 --> 00:29:27,940 because the amount is so much lower. 617 00:29:27,940 --> 00:29:30,980 So if you overprice an item by a lot and nobody buys it, 618 00:29:30,980 --> 00:29:33,650 you might end up getting less, right? 619 00:29:33,650 --> 00:29:36,090 In transit, typical [INAUDIBLE] are inelastic. 620 00:29:36,090 --> 00:29:37,740 So this is important. 621 00:29:37,740 --> 00:29:41,090 This means that any transit agency 622 00:29:41,090 --> 00:29:42,920 who needs more federal revenue can probably 623 00:29:42,920 --> 00:29:46,340 increase fares and know that they will get more money. 624 00:29:46,340 --> 00:29:49,100 That's sort of one that keeps conclusions of this. 625 00:29:49,100 --> 00:29:51,380 The range of these values is from negative 0.1 626 00:29:51,380 --> 00:29:54,330 to negative 0.5. 627 00:29:54,330 --> 00:29:58,710 The elasticity of headway is a little larger in magnitude. 628 00:29:58,710 --> 00:30:01,310 So the response of demand to frequency 629 00:30:01,310 --> 00:30:04,890 is stronger a little bit than the response to fares, 630 00:30:04,890 --> 00:30:09,620 and often overlooked in ridership spikes. 631 00:30:09,620 --> 00:30:12,740 And then total travel time, people respond very strongly 632 00:30:12,740 --> 00:30:13,310 to that. 633 00:30:13,310 --> 00:30:18,000 So 634 00:30:18,000 --> 00:30:20,460 OK, then there are these important other points 635 00:30:20,460 --> 00:30:21,660 that I want to go through. 636 00:30:21,660 --> 00:30:26,850 Small cities have larger fare elasticities than large cities. 637 00:30:26,850 --> 00:30:29,520 Let's talk about each one and suggest the reasons for them. 638 00:30:29,520 --> 00:30:34,888 So why would small cities have larger elasticities? 639 00:30:34,888 --> 00:30:36,609 AUDIENCE: The absolute value? 640 00:30:36,609 --> 00:30:37,900 PROFESSOR: Yes, absolute value. 641 00:30:37,900 --> 00:30:42,099 So you're saying that they are more elastic, right? 642 00:30:42,099 --> 00:30:43,140 AUDIENCE: More sensitive. 643 00:30:43,140 --> 00:30:43,950 PROFESSOR: Yeah, more sensitive. 644 00:30:43,950 --> 00:30:46,260 So small cities-- ridership in small cities 645 00:30:46,260 --> 00:30:51,000 is more sensitive than ridership in large cities. 646 00:30:51,000 --> 00:30:52,620 AUDIENCE: There's not as much of a-- 647 00:30:52,620 --> 00:30:55,217 there's not as much utility lost by switching 648 00:30:55,217 --> 00:30:56,050 two different modes. 649 00:30:56,050 --> 00:30:58,440 There might not be as much traffic if you drive instead. 650 00:30:58,440 --> 00:31:02,640 Some fare might be more of a reason that people that use it. 651 00:31:02,640 --> 00:31:05,799 And also that parking might not be as costly 652 00:31:05,799 --> 00:31:07,590 so that if you raise the fare a little bit, 653 00:31:07,590 --> 00:31:08,990 you might get more expensive than parking. 654 00:31:08,990 --> 00:31:11,430 PROFESSOR: OK, so you have other sort of characteristics 655 00:31:11,430 --> 00:31:13,800 in large cities that you don't find in small cities, 656 00:31:13,800 --> 00:31:15,714 such as parking costs. 657 00:31:15,714 --> 00:31:17,130 AUDIENCE: That would be a way of-- 658 00:31:17,130 --> 00:31:17,550 [INTERPOSING VOICES] 659 00:31:17,550 --> 00:31:19,830 PROFESSOR: I'm trying to summarize it, and also 660 00:31:19,830 --> 00:31:21,900 repeat for posterity. 661 00:31:21,900 --> 00:31:22,570 [INAUDIBLE] 662 00:31:22,570 --> 00:31:26,630 AUDIENCE: Also in urban cities, since these elasticities are 663 00:31:26,630 --> 00:31:29,280 being computed for the population as a whole, 664 00:31:29,280 --> 00:31:32,824 there are more people dependent on the mass transit system. 665 00:31:32,824 --> 00:31:34,240 PROFESSOR: That is very important. 666 00:31:34,240 --> 00:31:36,630 This is one of the key factors affecting this, actually. 667 00:31:36,630 --> 00:31:39,150 So you have a much larger dependence 668 00:31:39,150 --> 00:31:42,900 on public transportation in large cities. 669 00:31:42,900 --> 00:31:44,329 And also-- yeah. 670 00:31:44,329 --> 00:31:46,245 AUDIENCE: Is it really the size of the city or 671 00:31:46,245 --> 00:31:47,700 is it the density of the city-- 672 00:31:47,700 --> 00:31:49,750 PROFESSOR: It's probably density. 673 00:31:49,750 --> 00:31:50,680 These are correlated. 674 00:31:50,680 --> 00:31:53,900 So maybe this could have been more carefully chosen. 675 00:31:53,900 --> 00:31:56,435 And you could say older cities, as well. 676 00:31:56,435 --> 00:32:01,140 I would argue older pre-auto industry, pre-highway expansion 677 00:32:01,140 --> 00:32:04,930 cities as another way of putting footings. 678 00:32:04,930 --> 00:32:06,660 Yes? 679 00:32:06,660 --> 00:32:08,721 More ideas about this? 680 00:32:08,721 --> 00:32:09,220 Henry? 681 00:32:09,220 --> 00:32:10,740 AUDIENCE: I don't know how you're defining this-- 682 00:32:10,740 --> 00:32:11,740 [INTERPOSING VOICES] 683 00:32:11,740 --> 00:32:12,573 Sorry, sorry, sorry. 684 00:32:12,573 --> 00:32:13,290 Someone first-- 685 00:32:13,290 --> 00:32:14,950 PROFESSOR: Henry. 686 00:32:14,950 --> 00:32:16,096 OK. 687 00:32:16,096 --> 00:32:17,812 You decide. 688 00:32:17,812 --> 00:32:19,270 AUDIENCE: I was just curious about, 689 00:32:19,270 --> 00:32:21,920 how do you define small cities? 690 00:32:21,920 --> 00:32:24,930 Because for example, I'm thinking in my own country, 691 00:32:24,930 --> 00:32:31,385 a small city would be a city of around 200,000, 100,000 people. 692 00:32:31,385 --> 00:32:35,984 And over there, for example, an increase in the fare 693 00:32:35,984 --> 00:32:38,150 would have this characteristic because you can still 694 00:32:38,150 --> 00:32:39,630 walk [INAUDIBLE]. 695 00:32:39,630 --> 00:32:42,020 So walking is still an option. 696 00:32:42,020 --> 00:32:44,090 And I think it's kind of correlated with what 697 00:32:44,090 --> 00:32:45,310 you were saying about 698 00:32:45,310 --> 00:32:47,030 [INTERPOSING VOICES] 699 00:32:47,030 --> 00:32:48,590 PROFESSOR: So all of these things 700 00:32:48,590 --> 00:32:50,170 have to do with where you apply them. 701 00:32:50,170 --> 00:32:52,760 And maybe they're not completely generalizable. 702 00:32:52,760 --> 00:32:55,580 But I think this observation has more 703 00:32:55,580 --> 00:32:59,750 to do with the North American and European context. 704 00:32:59,750 --> 00:33:01,730 So but that's a good observation. 705 00:33:01,730 --> 00:33:02,619 Henry? 706 00:33:02,619 --> 00:33:04,160 AUDIENCE: I think smaller cities also 707 00:33:04,160 --> 00:33:06,727 have less extensive coverage. 708 00:33:06,727 --> 00:33:08,060 PROFESSOR: Yes, so our imports-- 709 00:33:08,060 --> 00:33:08,210 [INTERPOSING VOICES] 710 00:33:08,210 --> 00:33:10,590 AUDIENCE: --are already trained to not rely on-- 711 00:33:10,590 --> 00:33:12,506 PROFESSOR: So the public transportation supply 712 00:33:12,506 --> 00:33:16,310 might be quite undesirable or inefficient. 713 00:33:16,310 --> 00:33:17,215 And so therefore-- 714 00:33:17,215 --> 00:33:18,840 AUDIENCE: And if you jack up the price, 715 00:33:18,840 --> 00:33:19,970 that will make it less desirable. 716 00:33:19,970 --> 00:33:20,761 PROFESSOR: Exactly. 717 00:33:20,761 --> 00:33:21,559 OK, great. 718 00:33:21,559 --> 00:33:22,850 OK, let's move to the next one. 719 00:33:22,850 --> 00:33:25,460 Bus travel is more elastic than commuter rail 720 00:33:25,460 --> 00:33:26,370 rapid rail travel. 721 00:33:30,680 --> 00:33:34,070 AUDIENCE: Simply, bus travel is smaller traveled distance 722 00:33:34,070 --> 00:33:38,635 and there are possible another alternatives. 723 00:33:38,635 --> 00:33:39,260 PROFESSOR: Yes. 724 00:33:39,260 --> 00:33:40,190 That's the key reason. 725 00:33:40,190 --> 00:33:42,140 So if you have a shorter distance, 726 00:33:42,140 --> 00:33:45,410 and bus travel tends to be shorter distance, 727 00:33:45,410 --> 00:33:48,050 then you are more likely to have other options, 728 00:33:48,050 --> 00:33:53,000 whereas commuter rail, you might be able to drive maybe 729 00:33:53,000 --> 00:33:54,960 your own car, so et cetera. 730 00:33:54,960 --> 00:33:58,570 So there are fewer options in general to the public 731 00:33:58,570 --> 00:34:00,800 for a longer distance travel. 732 00:34:00,800 --> 00:34:02,630 And commuter rail and rapid rail travel 733 00:34:02,630 --> 00:34:04,396 tends to be longer distance. 734 00:34:04,396 --> 00:34:06,770 AUDIENCE: Commuter rail, rapid rail, certainly in Russia, 735 00:34:06,770 --> 00:34:10,130 have a time advantage, tend to have a time advantage 736 00:34:10,130 --> 00:34:11,227 over the private vehicle. 737 00:34:11,227 --> 00:34:12,560 PROFESSOR: That's true, as well. 738 00:34:12,560 --> 00:34:13,393 [INTERPOSING VOICES] 739 00:34:13,393 --> 00:34:14,469 This is another reason-- 740 00:34:14,469 --> 00:34:15,400 AUDIENCE: The bus never has-- 741 00:34:15,400 --> 00:34:16,233 [INTERPOSING VOICES] 742 00:34:16,233 --> 00:34:18,949 PROFESSOR: --competitiveness of the mode in rush hour 743 00:34:18,949 --> 00:34:22,159 where buses might be stuck in traffic and rail 744 00:34:22,159 --> 00:34:23,639 is not stuck in traffic, right? 745 00:34:23,639 --> 00:34:26,840 So that drives the elasticity more 746 00:34:26,840 --> 00:34:30,620 because so many more people are traveling in rush hours. 747 00:34:30,620 --> 00:34:31,310 Great. 748 00:34:31,310 --> 00:34:34,040 Let's move on to the third one, off peak fare elasticities 749 00:34:34,040 --> 00:34:36,800 are double the size of peak fare elasticities. 750 00:34:36,800 --> 00:34:37,520 Lee. 751 00:34:37,520 --> 00:34:40,145 AUDIENCE: The difference between leisure and business travelers 752 00:34:40,145 --> 00:34:42,125 where you can change your trip decision 753 00:34:42,125 --> 00:34:44,570 if it's a leisure trip, but not if you're going to work. 754 00:34:44,570 --> 00:34:45,949 PROFESSOR: Trip purpose, right? 755 00:34:45,949 --> 00:34:53,449 So peak travel tends to have a higher proportion of business 756 00:34:53,449 --> 00:34:56,120 and you don't have the option of not doing the trip. 757 00:34:56,120 --> 00:34:58,230 So you must do the trip. 758 00:34:58,230 --> 00:34:59,360 What else? 759 00:34:59,360 --> 00:35:01,920 Something related to that. 760 00:35:01,920 --> 00:35:02,420 Yes? 761 00:35:02,420 --> 00:35:04,610 AUDIENCE: Again, congestion has a role. 762 00:35:07,210 --> 00:35:10,580 To consider the idea of commuting by Uber in New York 763 00:35:10,580 --> 00:35:14,894 City, it might not make sense to do that during a peak period 764 00:35:14,894 --> 00:35:17,060 if you're going to be sitting in traffic on the road 765 00:35:17,060 --> 00:35:20,720 and the subway has, then, a time advantage. 766 00:35:20,720 --> 00:35:26,300 But returning home late in the evening, that time advantage 767 00:35:26,300 --> 00:35:28,910 of transit could be eroded because there's 768 00:35:28,910 --> 00:35:29,930 less congestion. 769 00:35:29,930 --> 00:35:32,320 PROFESSOR: Right, so congestion can have a major effect. 770 00:35:32,320 --> 00:35:35,000 And congestion is larger at peak. 771 00:35:35,000 --> 00:35:37,910 And therefore, it could make modes 772 00:35:37,910 --> 00:35:41,300 that are segregated have their own right of way more-- 773 00:35:41,300 --> 00:35:43,400 so related to the previous one. 774 00:35:43,400 --> 00:35:46,220 What else? 775 00:35:46,220 --> 00:35:46,721 Eli. 776 00:35:46,721 --> 00:35:49,220 AUDIENCE: [INAUDIBLE] I think it was discretionary ridership 777 00:35:49,220 --> 00:35:49,730 during-- 778 00:35:49,730 --> 00:35:51,260 PROFESSOR: Very important, yeah. 779 00:35:51,260 --> 00:35:52,610 So can you elaborate? 780 00:35:52,610 --> 00:35:53,950 AUDIENCE: Yeah. 781 00:35:53,950 --> 00:35:57,050 During the off peak hours, you have 782 00:35:57,050 --> 00:35:59,000 more people who have no other choice 783 00:35:59,000 --> 00:36:00,690 to get around, use transit. 784 00:36:00,690 --> 00:36:03,080 PROFESSOR: And why not? 785 00:36:03,080 --> 00:36:05,084 Why might these people not have other choices? 786 00:36:05,084 --> 00:36:06,750 AUDIENCE: Because they are lower income. 787 00:36:06,750 --> 00:36:08,041 PROFESSOR: Lower income, right. 788 00:36:08,041 --> 00:36:13,400 So lower income people who work may have no other options. 789 00:36:13,400 --> 00:36:15,950 This may be the only option they can afford. 790 00:36:15,950 --> 00:36:20,280 And therefore, they are less sensitive. 791 00:36:20,280 --> 00:36:23,460 So if you raise fares, well, it doesn't look good for them. 792 00:36:23,460 --> 00:36:26,210 But that's the only alternative for them. 793 00:36:26,210 --> 00:36:28,350 So they'll react. 794 00:36:28,350 --> 00:36:30,290 Great. 795 00:36:30,290 --> 00:36:32,850 Short distance trips are more elastic than long distance 796 00:36:32,850 --> 00:36:33,410 trips. 797 00:36:33,410 --> 00:36:35,368 We already covered that in the context of rail, 798 00:36:35,368 --> 00:36:36,610 so let's skip that. 799 00:36:36,610 --> 00:36:38,590 Fare elasticities rise with income. 800 00:36:43,550 --> 00:36:46,772 This should be straightforward. 801 00:36:46,772 --> 00:36:48,730 AUDIENCE: More options if you have more income. 802 00:36:48,730 --> 00:36:50,110 PROFESSOR: If you have more income, you have more options. 803 00:36:50,110 --> 00:36:51,890 You can afford all the other options. 804 00:36:51,890 --> 00:36:52,840 So there you go. 805 00:36:52,840 --> 00:36:56,310 And what about fare elasticities fall with age? 806 00:36:56,310 --> 00:36:59,550 So older people, senior people are less 807 00:36:59,550 --> 00:37:02,130 sensitive to changes in fair. 808 00:37:02,130 --> 00:37:04,270 Why is that? 809 00:37:04,270 --> 00:37:04,770 Emily? 810 00:37:04,770 --> 00:37:07,110 AUDIENCE: Higher income [INAUDIBLE].. 811 00:37:07,110 --> 00:37:08,705 They tend to have fewer options. 812 00:37:08,705 --> 00:37:09,330 PROFESSOR: Why? 813 00:37:09,330 --> 00:37:13,220 AUDIENCE: Because, well, walking might not be an option. 814 00:37:13,220 --> 00:37:16,780 PROFESSOR: So the active modes might be less desirable? 815 00:37:16,780 --> 00:37:19,244 AUDIENCE: They might not have a driver's license anymore. 816 00:37:19,244 --> 00:37:21,410 PROFESSOR: They may not be-- have a driver's license 817 00:37:21,410 --> 00:37:22,140 anymore. 818 00:37:22,140 --> 00:37:23,910 AUDIENCE: They might have failing vision 819 00:37:23,910 --> 00:37:27,827 and therefore not be able to drive or have other-- 820 00:37:27,827 --> 00:37:29,160 PROFESSOR: That's exactly right. 821 00:37:29,160 --> 00:37:30,130 OK. 822 00:37:30,130 --> 00:37:33,330 Of all trip purposes, the work trip is the most inelastic. 823 00:37:33,330 --> 00:37:34,910 We've covered that. 824 00:37:34,910 --> 00:37:37,110 And then promotional fare elasticities 825 00:37:37,110 --> 00:37:39,420 are slightly larger than short term fare 826 00:37:39,420 --> 00:37:43,111 elasticities following permanent fare additions. 827 00:37:43,111 --> 00:37:43,610 So-- 828 00:37:43,610 --> 00:37:45,110 AUDIENCE: Can you explain that? 829 00:37:45,110 --> 00:37:46,030 What does promotion-- 830 00:37:46,030 --> 00:37:50,380 PROFESSOR: So if you do a free fare day 831 00:37:50,380 --> 00:37:55,840 and you put ads everywhere or something like this. 832 00:37:55,840 --> 00:37:57,670 The response that you will get is 833 00:37:57,670 --> 00:38:02,170 going to be much stronger than if you are on a routine basis 834 00:38:02,170 --> 00:38:06,440 raise fares and-- 835 00:38:06,440 --> 00:38:08,895 yeah. 836 00:38:08,895 --> 00:38:10,295 So why does that happen? 837 00:38:10,295 --> 00:38:13,160 AUDIENCE: Maybe the increase in demand due to the lower-- 838 00:38:13,160 --> 00:38:15,380 [INTERPOSING VOICES] 839 00:38:15,380 --> 00:38:17,480 PROFESSOR: Increase if you reduce them or decrease 840 00:38:17,480 --> 00:38:19,460 if you increase them, right? 841 00:38:19,460 --> 00:38:21,100 AUDIENCE: If you increase fares, they-- 842 00:38:21,100 --> 00:38:23,396 Better promotional if you have lower fare, right? 843 00:38:23,396 --> 00:38:24,020 PROFESSOR: Yes. 844 00:38:24,020 --> 00:38:29,720 But we can extend this to how frequently this change happens 845 00:38:29,720 --> 00:38:36,090 and how much buzz is there about the spare change. 846 00:38:36,090 --> 00:38:37,080 Yes? 847 00:38:37,080 --> 00:38:39,292 AUDIENCE: People feel like, oh, it's happening once. 848 00:38:39,292 --> 00:38:40,750 I've got to do it now, or I won't-- 849 00:38:40,750 --> 00:38:42,870 PROFESSOR: Yeah, so the psychological effects 850 00:38:42,870 --> 00:38:46,150 of marketing, the effectiveness of marketing on-- 851 00:38:46,150 --> 00:38:50,640 so if you look at agencies that every year at a certain dates 852 00:38:50,640 --> 00:38:54,090 raise their fares more or less with inflation, 853 00:38:54,090 --> 00:38:58,290 there's much less sort of the reaction-- the public reaction 854 00:38:58,290 --> 00:39:00,760 is much lower. 855 00:39:00,760 --> 00:39:01,840 So-- 856 00:39:01,840 --> 00:39:04,620 AUDIENCE: Are those promotional fare-- 857 00:39:04,620 --> 00:39:08,600 free fare day, free fair weekend, are they effective? 858 00:39:08,600 --> 00:39:09,880 Does anyone-- 859 00:39:09,880 --> 00:39:10,830 [INTERPOSING VOICES] 860 00:39:10,830 --> 00:39:12,690 PROFESSOR: That's what your goal is, right? 861 00:39:12,690 --> 00:39:13,815 AUDIENCE: No, the goal is-- 862 00:39:13,815 --> 00:39:14,957 [INTERPOSING VOICES] 863 00:39:14,957 --> 00:39:17,040 PROFESSOR: But it might be effective at increasing 864 00:39:17,040 --> 00:39:20,910 ridership temporarily if you have some event on the city 865 00:39:20,910 --> 00:39:21,990 and you want to-- 866 00:39:21,990 --> 00:39:25,140 if you want people to use those modes instead of drive, 867 00:39:25,140 --> 00:39:28,390 that might be a success. 868 00:39:28,390 --> 00:39:32,550 So you want to increase the efficiency of vehicles 869 00:39:32,550 --> 00:39:34,860 by decreasing load times so people 870 00:39:34,860 --> 00:39:36,500 don't have to interact with a fare box, 871 00:39:36,500 --> 00:39:37,458 you can open all doors. 872 00:39:37,458 --> 00:39:39,120 People go in and out. 873 00:39:39,120 --> 00:39:40,120 That might be a success. 874 00:39:40,120 --> 00:39:40,620 So-- 875 00:39:40,620 --> 00:39:42,690 AUDIENCE: Paris once did that for exposure, 876 00:39:42,690 --> 00:39:44,910 to try to lure people in because they already-- 877 00:39:44,910 --> 00:39:45,440 yeah. 878 00:39:45,440 --> 00:39:47,030 But the question is, is it effective? 879 00:39:47,030 --> 00:39:48,780 PROFESSOR: Well, again, it depends on what 880 00:39:48,780 --> 00:39:50,868 your measures of success are. 881 00:39:50,868 --> 00:39:52,740 Emily? 882 00:39:52,740 --> 00:39:55,700 AUDIENCE: Also, promotional fares 883 00:39:55,700 --> 00:40:00,642 might attract more discretionary trips than people-- 884 00:40:00,642 --> 00:40:03,100 PROFESSOR: People might say, oh, I want to try this, right? 885 00:40:03,100 --> 00:40:03,725 AUDIENCE: Yeah. 886 00:40:03,725 --> 00:40:08,430 Whereas permanent fare revisions tend 887 00:40:08,430 --> 00:40:11,460 to effect people who are just doing their commute. 888 00:40:11,460 --> 00:40:12,570 PROFESSOR: And therefore-- 889 00:40:12,570 --> 00:40:14,562 AUDIENCE: Who are therefore less-- 890 00:40:14,562 --> 00:40:15,270 PROFESSOR: Right. 891 00:40:15,270 --> 00:40:17,070 So if you do any kind of promotion, 892 00:40:17,070 --> 00:40:21,820 you might make people that tend not to even be aware of transit 893 00:40:21,820 --> 00:40:24,390 fares be aware of this option. 894 00:40:24,390 --> 00:40:26,050 And so you might lure them in. 895 00:40:26,050 --> 00:40:26,960 Yeah, Henry? 896 00:40:26,960 --> 00:40:28,626 AUDIENCE: There are also a lot of cities 897 00:40:28,626 --> 00:40:30,780 that do free transit on New Years Eve 898 00:40:30,780 --> 00:40:33,725 to reduce drunk driving-- 899 00:40:33,725 --> 00:40:34,350 PROFESSOR: Yes. 900 00:40:34,350 --> 00:40:35,560 So that's another example. 901 00:40:35,560 --> 00:40:38,540 So on New Year's Eve, to reduce drunk driving, 902 00:40:38,540 --> 00:40:39,620 offer free fares. 903 00:40:39,620 --> 00:40:40,170 Yes? 904 00:40:40,170 --> 00:40:42,711 AUDIENCE: So it's safe to say that promotional fares are only 905 00:40:42,711 --> 00:40:44,530 useful for short-term effects but they 906 00:40:44,530 --> 00:40:46,530 don't have any effect on long-term [INAUDIBLE].. 907 00:40:46,530 --> 00:40:49,380 PROFESSOR: I don't want to make that broad claim, 908 00:40:49,380 --> 00:40:52,410 but I think I agree with the premise of it. 909 00:40:54,960 --> 00:40:56,710 OK. 910 00:40:56,710 --> 00:41:00,990 OK, let's revisit a model that we saw on an earlier lecture. 911 00:41:03,570 --> 00:41:17,050 This is the TTC elasticity model, which we saw is used 912 00:41:17,050 --> 00:41:22,670 or has been used to predict the rider changes to service 913 00:41:22,670 --> 00:41:26,490 changes, so ridership response to service changes. 914 00:41:26,490 --> 00:41:29,240 So this is the-- let's recount. 915 00:41:29,240 --> 00:41:32,900 The total weighted travel time is 916 00:41:32,900 --> 00:41:36,290 a sum of four components-- the in-vehicle travel 917 00:41:36,290 --> 00:41:40,070 time, the waiting time, the walking time, which 918 00:41:40,070 --> 00:41:42,440 is the access and egress and transfer time, 919 00:41:42,440 --> 00:41:45,080 and a penalty for the number of transfers that you have 920 00:41:45,080 --> 00:41:47,216 to take for that [INAUDIBLE]. 921 00:41:47,216 --> 00:41:48,590 And then there are these weights. 922 00:41:48,590 --> 00:41:55,445 1.5, so what does that mean, 1.5 in this equation? 923 00:41:58,790 --> 00:42:01,060 I'll generalize this utility equation, 924 00:42:01,060 --> 00:42:02,830 a systematic disutility equation. 925 00:42:02,830 --> 00:42:06,350 It has some factors before each of the explanatory variables. 926 00:42:06,350 --> 00:42:07,031 Sonia? 927 00:42:07,031 --> 00:42:08,280 AUDIENCE: People hate waiting. 928 00:42:08,280 --> 00:42:09,910 So it is extra-- 929 00:42:09,910 --> 00:42:12,120 PROFESSOR: How much more do they hate waiting? 930 00:42:12,120 --> 00:42:15,651 AUDIENCE: They hate waiting 1.5 times as much 931 00:42:15,651 --> 00:42:17,150 as they hate being in their vehicle. 932 00:42:17,150 --> 00:42:17,800 PROFESSOR: Yes. 933 00:42:17,800 --> 00:42:20,380 Exactly. 934 00:42:20,380 --> 00:42:24,340 So now these numbers are nice and round. 935 00:42:24,340 --> 00:42:27,730 So TTC did not conduct a revealed preference study 936 00:42:27,730 --> 00:42:30,280 to estimate it from data. 937 00:42:30,280 --> 00:42:34,930 But they align somewhat well with revealed preference 938 00:42:34,930 --> 00:42:35,620 studies. 939 00:42:35,620 --> 00:42:38,620 So the order of magnitude, at least, is correct. 940 00:42:38,620 --> 00:42:41,550 I've seen factors as high as 3 for waiting 941 00:42:41,550 --> 00:42:47,980 time, as low as one, in cases of rush hour and, say, metro. 942 00:42:47,980 --> 00:42:49,570 So that gives you a range. 943 00:42:49,570 --> 00:42:52,330 Walking, I've seen 5, 10. 944 00:42:52,330 --> 00:42:53,402 I've seen higher. 945 00:42:53,402 --> 00:42:56,440 But penalties for number of transfers, 946 00:42:56,440 --> 00:42:59,120 I've seen five minutes, 10 minutes. 947 00:42:59,120 --> 00:43:01,219 So people hate transferring. 948 00:43:01,219 --> 00:43:02,260 This is the street bound. 949 00:43:02,260 --> 00:43:08,860 So this is 10 minutes for every transfer you have to make. 950 00:43:08,860 --> 00:43:10,270 Great, so what happens? 951 00:43:10,270 --> 00:43:13,996 You compute the total weighted travel time before and after. 952 00:43:13,996 --> 00:43:15,370 So you have the current situation 953 00:43:15,370 --> 00:43:18,010 and then you have a proposal that will alter these values 954 00:43:18,010 --> 00:43:21,100 and result in a new total weighted travel time. 955 00:43:21,100 --> 00:43:23,510 And then we apply some elasticities. 956 00:43:23,510 --> 00:43:26,320 So these elasticities defer by period. 957 00:43:26,320 --> 00:43:30,470 So during the peak, we have negative 1.5. 958 00:43:30,470 --> 00:43:33,460 Mid-day, we have a higher response. 959 00:43:33,460 --> 00:43:36,460 This aligns with what we just said. 960 00:43:36,460 --> 00:43:38,830 Peak period ridership tends to be more inelastic. 961 00:43:38,830 --> 00:43:42,490 So we have a stronger response during the middle of the day. 962 00:43:42,490 --> 00:43:45,370 And if you look at early in the morning or evening, 963 00:43:45,370 --> 00:43:46,250 even stronger. 964 00:43:46,250 --> 00:43:49,300 So you have higher values. 965 00:43:49,300 --> 00:43:53,358 Question, why are these values have 966 00:43:53,358 --> 00:43:56,230 a [INAUDIBLE] greater than one? 967 00:43:56,230 --> 00:43:59,140 This is a trick question, in some sense. 968 00:44:01,870 --> 00:44:04,880 AUDIENCE: Travel time is-- 969 00:44:04,880 --> 00:44:06,590 ridership is dependent on travel time. 970 00:44:06,590 --> 00:44:08,737 It's very elastic. 971 00:44:08,737 --> 00:44:09,320 PROFESSOR: OK. 972 00:44:09,320 --> 00:44:11,179 So that's one way of putting it. 973 00:44:11,179 --> 00:44:13,220 So I guess the observation is like, these are now 974 00:44:13,220 --> 00:44:15,470 fair elasticities. 975 00:44:15,470 --> 00:44:17,570 And they are not frequency elasticities. 976 00:44:17,570 --> 00:44:21,680 They are elasticities of total weighted travel time. 977 00:44:21,680 --> 00:44:25,660 And these are four components coming together 978 00:44:25,660 --> 00:44:27,440 and we can't really think about it 979 00:44:27,440 --> 00:44:31,970 in terms of fare elasticity and single variable, really. 980 00:44:31,970 --> 00:44:34,550 These are multiple effects coming together. 981 00:44:34,550 --> 00:44:36,530 So great. 982 00:44:36,530 --> 00:44:39,350 So do we understand how this applies? 983 00:44:39,350 --> 00:44:41,150 We have a service proposal. 984 00:44:41,150 --> 00:44:45,452 We estimate the effects on riderships 985 00:44:45,452 --> 00:44:47,660 in [INAUDIBLE] waiting time, walking time, and number 986 00:44:47,660 --> 00:44:49,220 of transfers. 987 00:44:49,220 --> 00:44:53,219 We know how many people would ride before and after, or we-- 988 00:44:53,219 --> 00:44:54,260 well, we don't know that. 989 00:44:54,260 --> 00:44:56,660 We want to predict that, but we know 990 00:44:56,660 --> 00:44:59,180 for each [INAUDIBLE] how each of these components 991 00:44:59,180 --> 00:45:01,170 would be affected. 992 00:45:01,170 --> 00:45:03,650 And then we use the [INAUDIBLE] elasticities 993 00:45:03,650 --> 00:45:06,240 to predict the change. 994 00:45:06,240 --> 00:45:07,870 OK? 995 00:45:07,870 --> 00:45:08,370 Questions? 996 00:45:08,370 --> 00:45:08,953 AUDIENCE: Yes. 997 00:45:08,953 --> 00:45:10,680 Is the 10 in front of in trans, does 998 00:45:10,680 --> 00:45:13,560 that mean the average transfer is 10 minutes, 999 00:45:13,560 --> 00:45:15,820 or does this mean people just really, really 1000 00:45:15,820 --> 00:45:18,070 don't like transfers because they want a simple route? 1001 00:45:18,070 --> 00:45:20,630 PROFESSOR: So this excludes waiting time. 1002 00:45:20,630 --> 00:45:25,590 This is the fact that I have to transfer is so annoying that I 1003 00:45:25,590 --> 00:45:26,700 would rather spend-- 1004 00:45:26,700 --> 00:45:29,070 or it would be the same as spending extra 10 minutes 1005 00:45:29,070 --> 00:45:30,540 in a vehicle. 1006 00:45:30,540 --> 00:45:35,550 If you would give me a situation where I have a single ride 1007 00:45:35,550 --> 00:45:38,210 option that takes 10 minutes longer 1008 00:45:38,210 --> 00:45:41,220 and you gave me another one that takes 10 minutes less 1009 00:45:41,220 --> 00:45:43,770 but I have to transfer, I wouldn't 1010 00:45:43,770 --> 00:45:46,970 know which one to take. 1011 00:45:46,970 --> 00:45:51,381 They would be equally desirable or undesirable. 1012 00:45:51,381 --> 00:45:51,880 Yeah. 1013 00:45:51,880 --> 00:45:53,546 AUDIENCE: Would you say that in a more-- 1014 00:45:53,546 --> 00:45:55,350 if this was a more distinct model, 1015 00:45:55,350 --> 00:45:58,599 then the penalty for transferring rail to rail would 1016 00:45:58,599 --> 00:46:00,390 be less than the penalty for transferring-- 1017 00:46:00,390 --> 00:46:01,515 PROFESSOR: Yeah, of course. 1018 00:46:01,515 --> 00:46:03,990 You could make this much more sophisticated. 1019 00:46:03,990 --> 00:46:09,480 And you could say, are there escalators or not? 1020 00:46:09,480 --> 00:46:15,420 You know, you could argue, are some parts of the city, 1021 00:46:15,420 --> 00:46:19,470 say, less safe, and therefore walking time 1022 00:46:19,470 --> 00:46:23,030 is very undesirable, especially at night? 1023 00:46:23,030 --> 00:46:27,120 So you could really go disaggregate and apply 1024 00:46:27,120 --> 00:46:29,310 all these behavioral impacts. 1025 00:46:29,310 --> 00:46:31,470 So yeah, you can make this more sophisticated. 1026 00:46:31,470 --> 00:46:33,290 But you have the idea, right? 1027 00:46:33,290 --> 00:46:35,590 This is an elasticity-based model. 1028 00:46:35,590 --> 00:46:38,050 OK. 1029 00:46:38,050 --> 00:46:41,122 Direct demand regression models, again, we 1030 00:46:41,122 --> 00:46:42,330 covered a little bit of this. 1031 00:46:42,330 --> 00:46:45,910 We match census tract data to route service characteristics. 1032 00:46:45,910 --> 00:46:48,650 This is hard without GIS tools. 1033 00:46:48,650 --> 00:46:51,030 GIS is a geographical information system 1034 00:46:51,030 --> 00:46:52,500 where you have a map and you load 1035 00:46:52,500 --> 00:46:55,470 layers of data showing things geographically 1036 00:46:55,470 --> 00:46:58,740 like population density and a point of density. 1037 00:46:58,740 --> 00:47:01,470 Your bus stops, your bus lines, your train lines. 1038 00:47:01,470 --> 00:47:05,160 So we use GIS tools to apportion population and employment 1039 00:47:05,160 --> 00:47:07,050 data to different bus routes. 1040 00:47:07,050 --> 00:47:09,780 So you know that the bus route goes along these centers that 1041 00:47:09,780 --> 00:47:12,240 have certain properties. 1042 00:47:12,240 --> 00:47:16,260 Then we run a regression of trying 1043 00:47:16,260 --> 00:47:17,880 to explain ridership as a function 1044 00:47:17,880 --> 00:47:19,410 of those characteristics. 1045 00:47:19,410 --> 00:47:20,940 And often what happens is that you 1046 00:47:20,940 --> 00:47:25,600 have large or high significance of dummy variables 1047 00:47:25,600 --> 00:47:29,600 and track-specific constants. 1048 00:47:29,600 --> 00:47:34,620 So the general variables like frequency, 1049 00:47:34,620 --> 00:47:37,140 the variables that apply to all the bus routes together, 1050 00:47:37,140 --> 00:47:40,940 tend to be maybe not as significant statistically 1051 00:47:40,940 --> 00:47:45,210 as these very specific variables that apply only 1052 00:47:45,210 --> 00:47:47,034 to one or two bus routes. 1053 00:47:47,034 --> 00:47:48,450 And that means that in some sense, 1054 00:47:48,450 --> 00:47:51,600 we are specifying a model that is over fitted 1055 00:47:51,600 --> 00:47:53,820 and there are some issues with this. 1056 00:47:53,820 --> 00:47:57,720 So the other issue is that these models tend 1057 00:47:57,720 --> 00:48:00,420 not to recognize network interactions, 1058 00:48:00,420 --> 00:48:04,500 or they never do, rather, because if you have two bus 1059 00:48:04,500 --> 00:48:08,940 lines running through the same employment center 1060 00:48:08,940 --> 00:48:11,610 and going to two different places, 1061 00:48:11,610 --> 00:48:14,815 both at the same frequency, then presumably 1062 00:48:14,815 --> 00:48:17,190 they should share the demand from that employment center, 1063 00:48:17,190 --> 00:48:18,060 right? 1064 00:48:18,060 --> 00:48:20,880 If you only had one bus route and you 1065 00:48:20,880 --> 00:48:23,100 leave the characteristics the same, 1066 00:48:23,100 --> 00:48:26,706 then that ridership would be a lot higher. 1067 00:48:26,706 --> 00:48:28,080 This model doesn't really account 1068 00:48:28,080 --> 00:48:32,100 for that, the complementarity or competition 1069 00:48:32,100 --> 00:48:34,570 of the supply in the network. 1070 00:48:34,570 --> 00:48:36,490 So that's an issue. 1071 00:48:36,490 --> 00:48:38,860 And then the relationship between supply and demand 1072 00:48:38,860 --> 00:48:39,760 is not-- 1073 00:48:39,760 --> 00:48:41,760 the interaction is not captured. 1074 00:48:41,760 --> 00:48:46,470 So we can-- there are alternative approaches. 1075 00:48:46,470 --> 00:48:51,480 Simultaneous equation models are capable of addressing 1076 00:48:51,480 --> 00:48:55,260 the relationship between supply and demand. 1077 00:48:55,260 --> 00:48:58,590 And then some other models are able to capture 1078 00:48:58,590 --> 00:49:01,830 the relationship between competing and complementary 1079 00:49:01,830 --> 00:49:02,330 routes. 1080 00:49:02,330 --> 00:49:04,410 And we'll look at an example. 1081 00:49:04,410 --> 00:49:07,220 That's the logical next step beyond direct demand models. 1082 00:49:07,220 --> 00:49:09,840 And then the full network models which sort of 1083 00:49:09,840 --> 00:49:11,090 have-- you have the GIS. 1084 00:49:11,090 --> 00:49:12,690 You have all the network. 1085 00:49:12,690 --> 00:49:16,350 Those deal explicitly with competing complementary routes 1086 00:49:16,350 --> 00:49:20,130 because people in that mall have the option of any bus route 1087 00:49:20,130 --> 00:49:23,780 and they see all their options. 1088 00:49:23,780 --> 00:49:25,740 And in these models you can include 1089 00:49:25,740 --> 00:49:27,850 [INAUDIBLE] distribution of model split effect. 1090 00:49:27,850 --> 00:49:30,660 So that's the next step beyond-- 1091 00:49:30,660 --> 00:49:33,840 the logical next step beyond the TTC-type model-- 1092 00:49:33,840 --> 00:49:36,210 the elasticity model that we just discussed. 1093 00:49:36,210 --> 00:49:40,110 Both of these require GIS-- 1094 00:49:40,110 --> 00:49:42,480 some computer representation of the transit network 1095 00:49:42,480 --> 00:49:43,530 and the service area. 1096 00:49:43,530 --> 00:49:45,220 This used to be a challenge. 1097 00:49:45,220 --> 00:49:47,700 Not so much anymore. 1098 00:49:47,700 --> 00:49:52,236 Most agencies put out TTFS data and write layers 1099 00:49:52,236 --> 00:49:53,610 showing where their bus stops are 1100 00:49:53,610 --> 00:49:56,050 and so this is now commonplace. 1101 00:49:56,050 --> 00:50:00,120 This is not an obstacle anymore. 1102 00:50:00,120 --> 00:50:01,000 Questions about this? 1103 00:50:01,000 --> 00:50:01,499 Henry? 1104 00:50:01,499 --> 00:50:02,824 AUDIENCE: What is TTC? 1105 00:50:02,824 --> 00:50:04,740 PROFESSOR: Oh, the Toronto Transit Commission. 1106 00:50:04,740 --> 00:50:08,280 It's the bus agency in [INAUDIBLE].. 1107 00:50:08,280 --> 00:50:10,780 Other questions? 1108 00:50:10,780 --> 00:50:15,430 OK, so let's look at a simultaneous equation model. 1109 00:50:15,430 --> 00:50:23,580 So here we have ridership of some route, I, on some segment, 1110 00:50:23,580 --> 00:50:26,590 Z. So we divide up bus routes into segments. 1111 00:50:26,590 --> 00:50:29,350 And we have many bus routes and many segments. 1112 00:50:29,350 --> 00:50:35,230 And we say that is a function of the level of service provided. 1113 00:50:35,230 --> 00:50:39,310 So think of that as frequency, the level 1114 00:50:39,310 --> 00:50:43,690 of service provided on that bus route on that zone, 1115 00:50:43,690 --> 00:50:45,400 and other external factors. 1116 00:50:45,400 --> 00:50:47,720 Those external factors are demographics, 1117 00:50:47,720 --> 00:50:50,780 socioeconomic characteristics of ridership, 1118 00:50:50,780 --> 00:50:55,640 the population density, things like that, so maybe exogenous. 1119 00:50:55,640 --> 00:50:59,730 Then you say supply, on the other hand, 1120 00:50:59,730 --> 00:51:04,980 so my level of service, is a function of the ridership I 1121 00:51:04,980 --> 00:51:07,650 have right now because if I have more riders, 1122 00:51:07,650 --> 00:51:09,870 I will provide more service. 1123 00:51:09,870 --> 00:51:12,320 Also, it is a reaction to ridership 1124 00:51:12,320 --> 00:51:14,260 as it was in the past. 1125 00:51:14,260 --> 00:51:18,670 And it's also a reaction to other external factors. 1126 00:51:18,670 --> 00:51:24,750 So these two equations are simultaneous. 1127 00:51:24,750 --> 00:51:27,420 You have to solve them at the same time. 1128 00:51:27,420 --> 00:51:29,670 And they could take different functional forms. 1129 00:51:29,670 --> 00:51:34,240 We are specifying here at a sort of conceptual level. 1130 00:51:34,240 --> 00:51:34,780 [INAUDIBLE] 1131 00:51:34,780 --> 00:51:37,640 AUDIENCE: Yeah, is that in the second version, is it a typo 1132 00:51:37,640 --> 00:51:40,450 or do we actually consider the overall ridership 1133 00:51:40,450 --> 00:51:44,280 in the previous time period? 1134 00:51:44,280 --> 00:51:45,960 PROFESSOR: It could be a function 1135 00:51:45,960 --> 00:51:48,510 of the ridership in the past. 1136 00:51:48,510 --> 00:51:51,245 In other words, this is a way of saying, my supply right now, 1137 00:51:51,245 --> 00:51:54,100 my supply today, is a function of what ridership used to be. 1138 00:51:54,100 --> 00:51:56,842 AUDIENCE: No, what I mean is, should that be segment 1139 00:51:56,842 --> 00:51:59,550 dependent or do we then scale it up to the entire [INAUDIBLE]---- 1140 00:51:59,550 --> 00:52:04,690 PROFESSOR: Oh, it could be segment dependent or not. 1141 00:52:04,690 --> 00:52:07,560 So you're saying why not have z? 1142 00:52:07,560 --> 00:52:08,910 You could. 1143 00:52:08,910 --> 00:52:11,010 If you want to add it in, go ahead. 1144 00:52:11,010 --> 00:52:12,190 Again, this is conceptual. 1145 00:52:12,190 --> 00:52:16,410 So what I'm saying is this is not a specific model. 1146 00:52:16,410 --> 00:52:18,600 Rather, this is a conceptual description 1147 00:52:18,600 --> 00:52:21,030 of how you set up a simultaneous equation model. 1148 00:52:21,030 --> 00:52:23,220 So the point of that variable is to say 1149 00:52:23,220 --> 00:52:27,750 that you could include the effects of previously 1150 00:52:27,750 --> 00:52:30,540 observed ridership. 1151 00:52:30,540 --> 00:52:34,420 So don't think about this as a specification of a model. 1152 00:52:37,200 --> 00:52:38,232 I'll put this on Cellar. 1153 00:52:38,232 --> 00:52:40,190 This is the paper that describes this in detail 1154 00:52:40,190 --> 00:52:44,950 if you want to look at the specifics of such a model. 1155 00:52:44,950 --> 00:52:48,840 Here's an example from Portland Tri-Met on that paper. 1156 00:52:48,840 --> 00:52:54,480 You have route 19 and route 20 which were on, in this diagram, 1157 00:52:54,480 --> 00:52:56,580 kind of east-west. 1158 00:52:56,580 --> 00:53:01,650 So route 19 runs along this line right here. 1159 00:53:01,650 --> 00:53:05,160 And for each bus route we have a [INAUDIBLE] area. 1160 00:53:05,160 --> 00:53:09,630 So route 19 captures this part on top and also 1161 00:53:09,630 --> 00:53:12,640 this middle part that is shaded both ways. 1162 00:53:12,640 --> 00:53:15,060 Another is route 20, which runs along this line. 1163 00:53:15,060 --> 00:53:19,780 And it captures people here and also people in this area. 1164 00:53:19,780 --> 00:53:23,670 So this area in the middle is-- 1165 00:53:23,670 --> 00:53:27,240 people from that area go to route 19 or route 20. 1166 00:53:27,240 --> 00:53:30,360 So in that area, route 19 and 20 are competing. 1167 00:53:33,900 --> 00:53:37,720 Then they kind of merge and go through zone 0 1168 00:53:37,720 --> 00:53:38,970 and they are fully competing. 1169 00:53:38,970 --> 00:53:42,264 They are-- maybe it's a branch and trunk system. 1170 00:53:42,264 --> 00:53:44,430 And then there are these other bus routes, route 70, 1171 00:53:44,430 --> 00:53:49,350 route 75, 21, and 22, that are across South-North, 1172 00:53:49,350 --> 00:53:52,420 North-South, that corridor. 1173 00:53:52,420 --> 00:53:54,120 And we say that these bus routes are 1174 00:53:54,120 --> 00:54:02,010 complementary because people might transfer, 1175 00:54:02,010 --> 00:54:03,680 people going from somewhere on the south 1176 00:54:03,680 --> 00:54:07,170 might transfer to one of these route 19 or route 20. 1177 00:54:07,170 --> 00:54:09,810 So the fact that route 70 or 75 is there, 1178 00:54:09,810 --> 00:54:11,940 it means that people get-- more than not 1179 00:54:11,940 --> 00:54:14,130 will be attracted to the corridor. 1180 00:54:14,130 --> 00:54:18,570 So then we quantify the degree to which bus routes compete 1181 00:54:18,570 --> 00:54:23,070 with this overlapping population fraction 1182 00:54:23,070 --> 00:54:24,540 percentage or proportion. 1183 00:54:24,540 --> 00:54:31,890 So this is the percentage of the fraction of population 1184 00:54:31,890 --> 00:54:34,740 that is on the overlapping portion. 1185 00:54:34,740 --> 00:54:36,450 So we look at route length in route 20. 1186 00:54:36,450 --> 00:54:40,830 We're saying how much relation is on this overlapping 1187 00:54:40,830 --> 00:54:44,980 area divided by the sum of the population captured by route 19 1188 00:54:44,980 --> 00:54:49,200 and the sum of population captured by route 20? 1189 00:54:49,200 --> 00:54:54,990 So that gives you an idea of the degree of overlap. 1190 00:54:54,990 --> 00:54:58,230 We also measure population in each catchment area 1191 00:54:58,230 --> 00:55:00,930 for each bus route. 1192 00:55:00,930 --> 00:55:04,960 And we can then capture inter-route effects. 1193 00:55:04,960 --> 00:55:08,780 So we do that by modifying the equation two slides ago 1194 00:55:08,780 --> 00:55:10,630 and adding this third question here. 1195 00:55:10,630 --> 00:55:14,370 So before we have ridership as a function of level of service 1196 00:55:14,370 --> 00:55:15,790 and external factors. 1197 00:55:15,790 --> 00:55:26,720 Now we're adding the ridership on competing bus routes. 1198 00:55:26,720 --> 00:55:31,580 So these are bus routes that are competing on the same segment. 1199 00:55:31,580 --> 00:55:35,090 And they run with overlapping population. 1200 00:55:35,090 --> 00:55:37,430 So to the extent that this increases, 1201 00:55:37,430 --> 00:55:40,460 the ridership on Route I decreases. 1202 00:55:40,460 --> 00:55:43,070 And then we also consider the ridership 1203 00:55:43,070 --> 00:55:47,090 on complementary routes across this catchment area. 1204 00:55:47,090 --> 00:55:49,430 And to the extent that this increases, 1205 00:55:49,430 --> 00:55:52,310 the ridership on route I increases. 1206 00:55:52,310 --> 00:55:56,950 We also consider the degree of population overlap 1207 00:55:56,950 --> 00:55:58,207 and the external factors. 1208 00:55:58,207 --> 00:55:59,290 Again, this is conceptual. 1209 00:55:59,290 --> 00:56:00,760 If you want to look at the specifics, 1210 00:56:00,760 --> 00:56:02,218 I suggest you look at the readings. 1211 00:56:02,218 --> 00:56:03,280 I will post the paper. 1212 00:56:05,870 --> 00:56:11,980 So this is a way of modeling the effects of complementarity 1213 00:56:11,980 --> 00:56:15,510 and competition. 1214 00:56:15,510 --> 00:56:20,480 OK, a different approach, moving down 1215 00:56:20,480 --> 00:56:23,600 the list of sophistication of modeling approaches 1216 00:56:23,600 --> 00:56:27,210 is the network-based modeling analysis approach. 1217 00:56:27,210 --> 00:56:30,920 So here we have a transit origin destination matrix as an input. 1218 00:56:30,920 --> 00:56:33,980 We measure how people are using the system right now. 1219 00:56:33,980 --> 00:56:39,260 We have layer like a DTFS layer with the base transit network. 1220 00:56:39,260 --> 00:56:41,304 And we put that into a model. 1221 00:56:41,304 --> 00:56:43,720 And then we tell that model, we're going to make a change. 1222 00:56:43,720 --> 00:56:44,736 We're going to add a bus line or we're 1223 00:56:44,736 --> 00:56:46,236 going to increase frequency or we're 1224 00:56:46,236 --> 00:56:48,290 going to increase fares, whatever it is. 1225 00:56:48,290 --> 00:56:51,770 And that model will look at the current transit demand 1226 00:56:51,770 --> 00:56:53,070 and predict changes. 1227 00:56:53,070 --> 00:56:53,570 OK? 1228 00:56:53,570 --> 00:56:55,580 So that's the general idea. 1229 00:56:55,580 --> 00:56:58,220 It'll also output not just ridership and revenue 1230 00:56:58,220 --> 00:57:00,770 but things like rider attributes. 1231 00:57:00,770 --> 00:57:05,960 You might have information about income and things in GIS layers 1232 00:57:05,960 --> 00:57:09,650 that can be allocated to different services. 1233 00:57:09,650 --> 00:57:15,620 And there are three levels of analysis for how to use-- 1234 00:57:15,620 --> 00:57:18,260 three levels of detail on the OD matrix. 1235 00:57:18,260 --> 00:57:20,120 The first one is fixed transit flows. 1236 00:57:20,120 --> 00:57:23,000 So again, in order of sophistication, 1237 00:57:23,000 --> 00:57:25,690 starting with the least sophisticated, fixed transit 1238 00:57:25,690 --> 00:57:26,190 flow. 1239 00:57:26,190 --> 00:57:31,520 So we look at the current transit relief flow. 1240 00:57:31,520 --> 00:57:33,875 This could be from a telephone survey 1241 00:57:33,875 --> 00:57:35,000 or from any kind of survey. 1242 00:57:35,000 --> 00:57:37,880 Or we have, now, modern techniques 1243 00:57:37,880 --> 00:57:40,560 based on AFC and inference models 1244 00:57:40,560 --> 00:57:44,790 that we will discuss shortly in the next few lectures. 1245 00:57:44,790 --> 00:57:48,600 So we have the OE matrix, current OE matrix for transit. 1246 00:57:48,600 --> 00:57:50,480 And we assume that the demand for transit 1247 00:57:50,480 --> 00:57:52,730 won't change as a result of the service 1248 00:57:52,730 --> 00:57:55,480 changes in the short run. 1249 00:57:55,480 --> 00:57:57,800 So that could be a strong assumption. 1250 00:57:57,800 --> 00:57:59,330 That's typically what we do. 1251 00:57:59,330 --> 00:58:01,160 That's typically how these models work. 1252 00:58:01,160 --> 00:58:02,960 Again, we're more interested in the impacts 1253 00:58:02,960 --> 00:58:11,630 on sort of current riders than on big changes. 1254 00:58:11,630 --> 00:58:15,470 The next approach is to say the total demand is fixed 1255 00:58:15,470 --> 00:58:17,670 if I look at the demand across all modes, 1256 00:58:17,670 --> 00:58:22,320 including car and bike and TNCs and all these things. 1257 00:58:22,320 --> 00:58:26,240 So that you have to get from surveys, the kinds of surveys 1258 00:58:26,240 --> 00:58:30,830 that are used to input data into a four-step model. 1259 00:58:30,830 --> 00:58:32,880 But then you allow the model split to change. 1260 00:58:32,880 --> 00:58:36,290 So as a result of service changes, 1261 00:58:36,290 --> 00:58:38,300 people might switch between modes. 1262 00:58:38,300 --> 00:58:41,760 They might go from transit to car or from car to transit. 1263 00:58:41,760 --> 00:58:46,730 So that's preferable, especially if you have significant service 1264 00:58:46,730 --> 00:58:48,350 changes. 1265 00:58:48,350 --> 00:58:51,720 But it's not often put in place. 1266 00:58:51,720 --> 00:58:54,290 And then there's the variable total [INAUDIBLE].. 1267 00:58:54,290 --> 00:58:59,720 Now you require all the steps of modeling, 1268 00:58:59,720 --> 00:59:01,940 the four-step modeling. 1269 00:59:01,940 --> 00:59:08,060 And you are essentially allowing the total demand 1270 00:59:08,060 --> 00:59:10,880 to increase or decrease as a result of service changes 1271 00:59:10,880 --> 00:59:12,470 or fare changes. 1272 00:59:12,470 --> 00:59:14,960 Typically no one does this. 1273 00:59:14,960 --> 00:59:18,361 And one could argue this is unnecessary. 1274 00:59:18,361 --> 00:59:18,860 Why? 1275 00:59:18,860 --> 00:59:21,470 Because we are looking at short-run changes. 1276 00:59:21,470 --> 00:59:25,580 And total demand shouldn't change 1277 00:59:25,580 --> 00:59:31,220 in a very short horizon as a result of at least 1278 00:59:31,220 --> 00:59:33,330 the typical service changes. 1279 00:59:33,330 --> 00:59:37,040 So this kind of model is great if you 1280 00:59:37,040 --> 00:59:40,430 want to capture the influence of land use patterns 1281 00:59:40,430 --> 00:59:43,710 on long term demand, for example. 1282 00:59:43,710 --> 00:59:46,640 If, however, you were to bring in a new rail line that 1283 00:59:46,640 --> 00:59:50,590 connects two employment centers, that's a change in land use. 1284 00:59:50,590 --> 00:59:52,970 If you somehow think that some change 1285 00:59:52,970 --> 00:59:55,550 that you will do in your city, on your transit network, 1286 00:59:55,550 --> 00:59:58,250 will bring in a change in population 1287 00:59:58,250 --> 01:00:01,460 or a change in where people live significantly, 1288 01:00:01,460 --> 01:00:02,930 then you have to do this. 1289 01:00:02,930 --> 01:00:05,750 That's the only case where this would be useful. 1290 01:00:05,750 --> 01:00:09,530 Otherwise, you probably are thinking in the short run 1291 01:00:09,530 --> 01:00:11,330 total the amount is the same. 1292 01:00:11,330 --> 01:00:14,970 People might switch between modes. 1293 01:00:14,970 --> 01:00:17,860 So you will get step two. 1294 01:00:17,860 --> 01:00:19,400 And if it's a very small change, you 1295 01:00:19,400 --> 01:00:21,830 could even get away with step one, where 1296 01:00:21,830 --> 01:00:26,120 you look at current demand and people might change 1297 01:00:26,120 --> 01:00:29,480 when they make the trip or what bus route or rail line 1298 01:00:29,480 --> 01:00:35,690 they choose, but not necessarily the level of ridership. 1299 01:00:35,690 --> 01:00:38,810 OK, there's different modeling and analysis packages. 1300 01:00:38,810 --> 01:00:40,840 Let's show three of them. 1301 01:00:40,840 --> 01:00:46,430 MADITUC, EMME/2, and TrainsCAD. 1302 01:00:46,430 --> 01:00:48,410 Have anybody heard of any of these? 1303 01:00:48,410 --> 01:00:50,700 Raise your hand. 1304 01:00:50,700 --> 01:00:52,180 OK, about half of you. 1305 01:00:52,180 --> 01:00:52,700 Great. 1306 01:00:52,700 --> 01:00:55,670 So let's start with MADITUC. 1307 01:00:55,670 --> 01:00:59,470 It was developed in Montreal. 1308 01:00:59,470 --> 01:01:01,850 It was a common technique in Montreal. 1309 01:01:01,850 --> 01:01:05,270 It requires only survey data that includes route choice 1310 01:01:05,270 --> 01:01:06,270 information. 1311 01:01:06,270 --> 01:01:12,410 So this is a line level model, essentially. 1312 01:01:12,410 --> 01:01:15,140 It's designed for transit service planning 1313 01:01:15,140 --> 01:01:17,000 and it uses all or nothing assignments. 1314 01:01:17,000 --> 01:01:20,240 So it looks at, for a given [INAUDIBLE],, 1315 01:01:20,240 --> 01:01:22,580 which is the best path on the network. 1316 01:01:22,580 --> 01:01:25,340 And then it puts all of the people on the [INAUDIBLE] 1317 01:01:25,340 --> 01:01:27,470 into that one path. 1318 01:01:27,470 --> 01:01:30,050 So everybody makes the same decision kind of thing. 1319 01:01:30,050 --> 01:01:33,860 And it doesn't have built-in data analysis or a graphics 1320 01:01:33,860 --> 01:01:34,610 capability. 1321 01:01:34,610 --> 01:01:38,780 So you take the output of it and you put it somewhere else. 1322 01:01:38,780 --> 01:01:42,740 These people were using SAS and other software 1323 01:01:42,740 --> 01:01:46,220 to generate graphics and to generate plots 1324 01:01:46,220 --> 01:01:48,050 and calculated statistics. 1325 01:01:48,050 --> 01:01:50,900 It's been used in four of the Canadian cities-- 1326 01:01:50,900 --> 01:01:53,510 Montreal, Quebec, Toronto, and Winnipeg. 1327 01:01:53,510 --> 01:01:57,200 And here's an output from of that data, plotted 1328 01:01:57,200 --> 01:01:58,640 with a different software. 1329 01:01:58,640 --> 01:02:08,360 So this is-- you can sort of see the business plot of OD. 1330 01:02:08,360 --> 01:02:12,440 I'm not sure, actually, how it's-- 1331 01:02:12,440 --> 01:02:15,870 if it's destinations for a given origin or what it is. 1332 01:02:15,870 --> 01:02:18,660 It might just be-- 1333 01:02:18,660 --> 01:02:21,550 oh, this is number of entries. 1334 01:02:21,550 --> 01:02:22,117 Yeah. 1335 01:02:22,117 --> 01:02:22,950 I don't know French. 1336 01:02:22,950 --> 01:02:24,420 But if anybody-- 1337 01:02:24,420 --> 01:02:28,429 AUDIENCE: It's number of entries by foot or-- 1338 01:02:28,429 --> 01:02:29,220 PROFESSOR: On foot. 1339 01:02:29,220 --> 01:02:29,970 AUDIENCE: On foot? 1340 01:02:29,970 --> 01:02:31,545 PROFESSOR: Yeah. 1341 01:02:31,545 --> 01:02:32,580 What is that? 1342 01:02:32,580 --> 01:02:34,100 I don't know that last word. 1343 01:02:34,100 --> 01:02:37,950 AUDIENCE: [INAUDIBLE] 1344 01:02:37,950 --> 01:02:40,340 PROFESSOR: Anyway. 1345 01:02:40,340 --> 01:02:45,370 And then it says AMP 629, right? 1346 01:02:45,370 --> 01:02:47,850 OK, great. 1347 01:02:47,850 --> 01:02:48,450 EMME/2. 1348 01:02:48,450 --> 01:02:51,750 So this is multi-modal equilibrium. 1349 01:02:51,750 --> 01:02:54,210 It was also developed in Montreal. 1350 01:02:54,210 --> 01:02:55,890 It's a little more sophisticated. 1351 01:02:55,890 --> 01:02:58,265 It was developed with the general regional transportation 1352 01:02:58,265 --> 01:02:59,430 modeling package. 1353 01:02:59,430 --> 01:03:01,950 It can generate transit OD flows from a travel demand 1354 01:03:01,950 --> 01:03:08,280 model, which the other package needed that as an input. 1355 01:03:08,280 --> 01:03:12,780 It has a link node oriented approach instead of a line 1356 01:03:12,780 --> 01:03:13,620 level approach. 1357 01:03:13,620 --> 01:03:15,360 So you can have bus stops and you can-- 1358 01:03:15,360 --> 01:03:18,240 it has more detail, [INAUDIBLE]. 1359 01:03:18,240 --> 01:03:21,520 There are two options for a transit assignment. 1360 01:03:21,520 --> 01:03:25,750 There's aggregate zone to zone, flow, multi-path assignment. 1361 01:03:25,750 --> 01:03:28,620 So that's usually not precise enough 1362 01:03:28,620 --> 01:03:29,910 because it's zone to zone. 1363 01:03:29,910 --> 01:03:33,150 And then there's a disaggregate point-to-point trip assignment 1364 01:03:33,150 --> 01:03:35,700 procedure, which you could know which bus stops 1365 01:03:35,700 --> 01:03:38,160 and lines and everything. 1366 01:03:38,160 --> 01:03:41,350 It is probabilistic or multi-path. 1367 01:03:41,350 --> 01:03:47,160 So now you have an OD pair and different alternatives of how 1368 01:03:47,160 --> 01:03:51,540 to get from both O to D. And one may be preferable, 1369 01:03:51,540 --> 01:03:55,870 but people might be split on which lines they take. 1370 01:03:55,870 --> 01:03:58,469 And more, maybe 2/3 of people go one way, 1371 01:03:58,469 --> 01:03:59,510 and a third go the other. 1372 01:03:59,510 --> 01:04:02,070 And some equilibrium is reached this way. 1373 01:04:02,070 --> 01:04:03,360 It's a standalone package. 1374 01:04:03,360 --> 01:04:06,670 This has all the graphics built in. 1375 01:04:06,670 --> 01:04:07,680 And here's a picture. 1376 01:04:07,680 --> 01:04:12,310 So yeah. 1377 01:04:12,310 --> 01:04:15,980 The TransCAD, this is much more local. 1378 01:04:15,980 --> 01:04:19,530 It's over caliber develops in Newton. 1379 01:04:22,910 --> 01:04:28,730 It has good tools to edit transit networks. 1380 01:04:28,730 --> 01:04:31,890 It has an API that allows you to add your own models. 1381 01:04:31,890 --> 01:04:36,200 So if you have your own model split model, you can put it in. 1382 01:04:36,200 --> 01:04:40,430 It has good interactive computer graphics, 1383 01:04:40,430 --> 01:04:42,320 database, built-in database. 1384 01:04:42,320 --> 01:04:44,720 It has a network assignment procedure, 1385 01:04:44,720 --> 01:04:47,300 lots of display options and output formats 1386 01:04:47,300 --> 01:04:49,430 and very general purpose. 1387 01:04:49,430 --> 01:04:53,730 So typically, you use a transit network database for this. 1388 01:04:53,730 --> 01:04:57,350 So that means geocoded transit links and nodes. 1389 01:04:57,350 --> 01:04:58,940 You have mappings of transit lines 1390 01:04:58,940 --> 01:05:00,470 onto network links and nodes. 1391 01:05:00,470 --> 01:05:05,390 So you have essentially a street network, right? 1392 01:05:05,390 --> 01:05:08,240 And you say, this bus route goes from-- 1393 01:05:08,240 --> 01:05:12,230 along these road links, crossing these intersections. 1394 01:05:12,230 --> 01:05:14,750 And transit lines are modeled with some attributes. 1395 01:05:14,750 --> 01:05:17,330 So headway is by service period. 1396 01:05:17,330 --> 01:05:19,970 Travel times, models, mode of service or bus, 1397 01:05:19,970 --> 01:05:22,010 subway, et cetera. 1398 01:05:22,010 --> 01:05:23,480 And there's also system attributes, 1399 01:05:23,480 --> 01:05:26,660 like operating cost data, energy consumption data, and fares. 1400 01:05:26,660 --> 01:05:27,670 Some of this is-- 1401 01:05:27,670 --> 01:05:29,390 well, these things are put in. 1402 01:05:29,390 --> 01:05:31,920 And they then are used to compute outputs. 1403 01:05:31,920 --> 01:05:35,340 So hybrid outputs. 1404 01:05:35,340 --> 01:05:41,940 So you get ridership by link and by line, boardings by node, 1405 01:05:41,940 --> 01:05:43,230 by line, by link. 1406 01:05:43,230 --> 01:05:45,450 You can have OD travel times, so in-vehicle 1407 01:05:45,450 --> 01:05:49,320 time and also including the access and egress portions, 1408 01:05:49,320 --> 01:05:51,480 transfers, et cetera. 1409 01:05:51,480 --> 01:05:53,850 Revenue predictions, operating cost predictions, 1410 01:05:53,850 --> 01:06:01,110 energy consumption predictions, even CO2 emission predictions, 1411 01:06:01,110 --> 01:06:03,810 revenues, operating costs, characteristics. 1412 01:06:03,810 --> 01:06:06,330 And you have all sorts of ways of displaying that-- tables, 1413 01:06:06,330 --> 01:06:08,410 reports, plots, et cetera. 1414 01:06:08,410 --> 01:06:13,800 So how about transit route assignment? 1415 01:06:13,800 --> 01:06:14,830 How do we go about that? 1416 01:06:14,830 --> 01:06:18,390 So transit route assignment is the process 1417 01:06:18,390 --> 01:06:22,380 by which we assign origin-destination flows 1418 01:06:22,380 --> 01:06:24,570 to specific paths on the transit network. 1419 01:06:24,570 --> 01:06:28,109 So there are two ways of doing that. 1420 01:06:28,109 --> 01:06:29,650 There's an all or nothing assignment, 1421 01:06:29,650 --> 01:06:31,080 which we talked about. 1422 01:06:31,080 --> 01:06:32,940 That's the example of MADITUC. 1423 01:06:32,940 --> 01:06:35,760 And then there's the multipath assignment, which 1424 01:06:35,760 --> 01:06:39,930 we gave an example for EMME/2. 1425 01:06:39,930 --> 01:06:43,690 When would all or nothing be OK to apply? 1426 01:06:43,690 --> 01:06:46,950 If you have not two dense of a network and not too many 1427 01:06:46,950 --> 01:06:47,960 alternatives, right? 1428 01:06:47,960 --> 01:06:50,640 Then some networks, there's only one sort 1429 01:06:50,640 --> 01:06:52,590 of feasible, logical path that one 1430 01:06:52,590 --> 01:06:55,630 would take for a given OD pair. 1431 01:06:55,630 --> 01:06:58,650 But if you look at a dense network, say, London, 1432 01:06:58,650 --> 01:07:01,630 there are often for an OD pair multiple good ways 1433 01:07:01,630 --> 01:07:04,020 of traveling that OD fare. 1434 01:07:04,020 --> 01:07:06,060 And therefore, that would not be a good model 1435 01:07:06,060 --> 01:07:10,980 to use because the flows on each link might be way off. 1436 01:07:10,980 --> 01:07:13,470 Is that understood, difference between multipath 1437 01:07:13,470 --> 01:07:14,700 and all or nothing? 1438 01:07:17,382 --> 01:07:18,280 OK. 1439 01:07:18,280 --> 01:07:20,670 Another way of dividing these or categorizing these 1440 01:07:20,670 --> 01:07:22,210 is aggregate and disaggregate. 1441 01:07:22,210 --> 01:07:27,070 So we also looked at these in the context of EMME/2. 1442 01:07:27,070 --> 01:07:31,660 So aggregate is zone to zone based on zone centroids. 1443 01:07:31,660 --> 01:07:35,200 So you have [INAUDIBLE] of transportation and ultra zones 1444 01:07:35,200 --> 01:07:36,640 or census tracts. 1445 01:07:36,640 --> 01:07:39,040 And you look at demand at that block level. 1446 01:07:39,040 --> 01:07:40,090 Or you have disaggregate. 1447 01:07:40,090 --> 01:07:42,700 So if you load a DTFS layer, you have 1448 01:07:42,700 --> 01:07:45,880 sort of stop level, line level, much more precise, 1449 01:07:45,880 --> 01:07:50,200 high resolution output on where the demand would 1450 01:07:50,200 --> 01:07:51,510 be assigned to, OK? 1451 01:07:54,490 --> 01:07:56,650 And how do we go about more choice 1452 01:07:56,650 --> 01:08:00,190 and sometimes how do you go about even 1453 01:08:00,190 --> 01:08:03,940 assignment to specific lines or options within the transit 1454 01:08:03,940 --> 01:08:05,260 mode? 1455 01:08:05,260 --> 01:08:07,940 Often with logit mode choice model. 1456 01:08:07,940 --> 01:08:10,740 So I think most-- well, some of you-- 1457 01:08:10,740 --> 01:08:14,440 anyone who took 201 or is taking 202 has seeen this? 1458 01:08:14,440 --> 01:08:17,329 Raise your hands if you've seen what a logit mode choice 1459 01:08:17,329 --> 01:08:19,620 model, if you're familiar with discreet choice modeling 1460 01:08:19,620 --> 01:08:22,760 at all, have heard about it? 1461 01:08:22,760 --> 01:08:24,040 OK. 1462 01:08:24,040 --> 01:08:25,270 So it was something what-- 1463 01:08:25,270 --> 01:08:29,149 I could see some shy hands that I know are more exposed to it. 1464 01:08:29,149 --> 01:08:32,390 So in this kind of model, the-- 1465 01:08:32,390 --> 01:08:35,830 when you specify systematic utility equations 1466 01:08:35,830 --> 01:08:38,140 where we say for each of the alternatives 1467 01:08:38,140 --> 01:08:40,819 that this person has, there are some variables. 1468 01:08:40,819 --> 01:08:47,260 These variables include travel time, cost, reliability, fare, 1469 01:08:47,260 --> 01:08:48,160 anything, right? 1470 01:08:48,160 --> 01:08:50,560 And the differences between the alternatives 1471 01:08:50,560 --> 01:08:52,779 are explained in terms of those variables. 1472 01:08:52,779 --> 01:08:58,635 And then what we get out of-- so we run a regression. 1473 01:08:58,635 --> 01:08:59,760 It's not linear regression. 1474 01:08:59,760 --> 01:09:01,260 It's a different kind of regression. 1475 01:09:01,260 --> 01:09:04,310 And what we get out of that is the probability-- 1476 01:09:04,310 --> 01:09:05,720 so we get some system-- 1477 01:09:05,720 --> 01:09:08,569 we get some parameter estimates, beta. 1478 01:09:08,569 --> 01:09:15,319 And we can use those to calculate the probability 1479 01:09:15,319 --> 01:09:18,410 that an individual will take some of the options 1480 01:09:18,410 --> 01:09:21,830 that we listed rather than the others that were listed. 1481 01:09:21,830 --> 01:09:24,439 So there's a very good course taught here 1482 01:09:24,439 --> 01:09:27,020 at MIT, 1202, demand modeling. 1483 01:09:27,020 --> 01:09:30,590 If you're interested in this topic, I recommend you take it. 1484 01:09:30,590 --> 01:09:33,649 And it's taught in the spring only. 1485 01:09:33,649 --> 01:09:37,100 So yeah, so this is obviously [INAUDIBLE] probability 1486 01:09:37,100 --> 01:09:39,109 that [INAUDIBLE] you would choose 1487 01:09:39,109 --> 01:09:40,819 one mode and not the other. 1488 01:09:40,819 --> 01:09:43,640 Often, this is the structure that 1489 01:09:43,640 --> 01:09:46,724 is underneath the TransCAD mode choice 1490 01:09:46,724 --> 01:09:47,890 model or anything like that. 1491 01:09:47,890 --> 01:09:50,760 And it's great if you understand how it works. 1492 01:09:50,760 --> 01:09:52,520 And if you can build your own and put it 1493 01:09:52,520 --> 01:09:53,910 in there, even better. 1494 01:09:53,910 --> 01:09:56,800 So all right. 1495 01:09:56,800 --> 01:09:58,900 And what are typical variables included 1496 01:09:58,900 --> 01:10:02,200 in these systematic utility equations for, say, the work 1497 01:10:02,200 --> 01:10:02,790 trip? 1498 01:10:02,790 --> 01:10:06,350 Well, it would be travel time, out of pocket travel costs, 1499 01:10:06,350 --> 01:10:10,390 vehicle travel time, income, gender, auto availability, 1500 01:10:10,390 --> 01:10:11,700 and occupation. 1501 01:10:11,700 --> 01:10:15,550 So some of these are traveler characteristics 1502 01:10:15,550 --> 01:10:17,940 which vary by person surveyed. 1503 01:10:17,940 --> 01:10:20,470 So typically, you will survey and you collect information 1504 01:10:20,470 --> 01:10:23,380 on income, gender, things that vary by person. 1505 01:10:23,380 --> 01:10:27,700 And then you also collect or put in characteristics 1506 01:10:27,700 --> 01:10:30,990 of the choices or the options other person has. 1507 01:10:30,990 --> 01:10:33,040 And both of these go into the model, 1508 01:10:33,040 --> 01:10:35,230 both the decision maker characteristics 1509 01:10:35,230 --> 01:10:37,720 and the options characteristics. 1510 01:10:37,720 --> 01:10:41,110 And sometimes we defer those two by calling one attribute 1511 01:10:41,110 --> 01:10:44,100 and the other characteristic. 1512 01:10:44,100 --> 01:10:48,830 OK, that's it for ridership prediction. 1513 01:10:48,830 --> 01:10:50,830 But if you have questions, let's discuss. 1514 01:10:56,570 --> 01:10:57,350 Yes? 1515 01:10:57,350 --> 01:11:02,680 AUDIENCE: What level of modeling is the T doing for their-- 1516 01:11:02,680 --> 01:11:06,910 and do they change their bus network to match their modeling 1517 01:11:06,910 --> 01:11:08,890 or is it sort of a longer-term-- 1518 01:11:08,890 --> 01:11:12,170 PROFESSOR: The T has a model called [INAUDIBLE],, 1519 01:11:12,170 --> 01:11:22,280 which is an elasticity-based model with market segmentation. 1520 01:11:22,280 --> 01:11:24,080 So we talked about-- 1521 01:11:24,080 --> 01:11:26,630 that was mentioned earlier in the lecture. 1522 01:11:26,630 --> 01:11:34,200 So they split population by type of fare product. 1523 01:11:34,200 --> 01:11:36,560 So people with passes might-- 1524 01:11:36,560 --> 01:11:39,110 that's a proxy for, say, frequent riders and those 1525 01:11:39,110 --> 01:11:40,040 who might have-- 1526 01:11:40,040 --> 01:11:42,230 those people might have different elasticities 1527 01:11:42,230 --> 01:11:44,240 than people that are less frequent 1528 01:11:44,240 --> 01:11:46,460 or people that pay cash or people, right? 1529 01:11:46,460 --> 01:11:49,490 So you split people into-- you segment the market, 1530 01:11:49,490 --> 01:11:51,410 then you look at-- 1531 01:11:51,410 --> 01:11:53,810 it looks at previous fare increases 1532 01:11:53,810 --> 01:11:56,120 and it calculates the elasticities 1533 01:11:56,120 --> 01:11:59,900 for each of those market segments 1534 01:11:59,900 --> 01:12:03,320 across several time periods and everything. 1535 01:12:03,320 --> 01:12:08,740 So then it calculates the total change in ridership and revenue 1536 01:12:08,740 --> 01:12:13,430 and also any changes across would somebody now take, 1537 01:12:13,430 --> 01:12:16,080 let's say, would some people switch to paths 1538 01:12:16,080 --> 01:12:18,750 or not switch to paths. 1539 01:12:18,750 --> 01:12:21,050 The Transit Hub has done some research on it. 1540 01:12:21,050 --> 01:12:22,810 And some of the people who-- 1541 01:12:22,810 --> 01:12:28,510 one of the persons who worked on their research here at MIT 1542 01:12:28,510 --> 01:12:31,090 now works at CTPS. 1543 01:12:31,090 --> 01:12:34,060 And he's one of the people who applied the model. 1544 01:12:36,700 --> 01:12:40,870 And then I think for other products, 1545 01:12:40,870 --> 01:12:42,790 it depends on the product. 1546 01:12:42,790 --> 01:12:47,290 So certainly, there is kind of the regional long term 1547 01:12:47,290 --> 01:12:52,510 planning, which is more four-step modeling approach. 1548 01:12:52,510 --> 01:12:55,870 If there is any specific project, 1549 01:12:55,870 --> 01:12:59,102 I haven't seen what they've done for the green line expansion. 1550 01:12:59,102 --> 01:12:59,810 Ari, do you know? 1551 01:12:59,810 --> 01:13:02,420 You always know something. 1552 01:13:02,420 --> 01:13:06,230 AUDIENCE: I mean, I-- it's the [INAUDIBLE].. 1553 01:13:06,230 --> 01:13:08,776 PROFESSOR: So that's the professional judgment? 1554 01:13:08,776 --> 01:13:11,150 AUDIENCE: That is a completely non-professional judgment. 1555 01:13:11,150 --> 01:13:12,914 I don't-- 1556 01:13:12,914 --> 01:13:14,830 PROFESSOR: Does anybody know what they've done 1557 01:13:14,830 --> 01:13:16,420 for the green line extension? 1558 01:13:16,420 --> 01:13:18,587 AUDIENCE: I mean, I'm sure they had to do something. 1559 01:13:18,587 --> 01:13:20,336 PROFESSOR: Often these capital [INAUDIBLE] 1560 01:13:20,336 --> 01:13:21,670 require demand analysis. 1561 01:13:21,670 --> 01:13:24,220 So I'm sure they did some modeling. 1562 01:13:24,220 --> 01:13:28,030 I'm not sure to what degree. 1563 01:13:28,030 --> 01:13:29,659 AUDIENCE: [INAUDIBLE] 1564 01:13:29,659 --> 01:13:30,700 PROFESSOR: Yeah, exactly. 1565 01:13:30,700 --> 01:13:38,260 So whether the modeling of the FTA requires 1566 01:13:38,260 --> 01:13:41,800 is the best one is subject to discussion. 1567 01:13:41,800 --> 01:13:45,340 And I think a lot of transportation experts 1568 01:13:45,340 --> 01:13:48,240 have issues with it. 1569 01:13:48,240 --> 01:13:51,370 It tends to put a lot of weight on-- 1570 01:13:51,370 --> 01:13:55,270 if you save many people one second, then that's great. 1571 01:13:55,270 --> 01:13:59,640 And instead of being sort of goal-oriented, and-- 1572 01:14:02,310 --> 01:14:04,230 AUDIENCE: Yeah, I feel like often agencies 1573 01:14:04,230 --> 01:14:06,206 will do one analysis for their grant 1574 01:14:06,206 --> 01:14:08,500 and then their own analysis because they 1575 01:14:08,500 --> 01:14:10,074 think they have benefits. 1576 01:14:10,074 --> 01:14:10,740 PROFESSOR: Yeah. 1577 01:14:10,740 --> 01:14:16,490 So, any other comments or questions on ridership? 1578 01:14:16,490 --> 01:14:17,920 This is an important topic, right? 1579 01:14:17,920 --> 01:14:21,825 It's one of the key topics in public transportation. 1580 01:14:21,825 --> 01:14:24,090 If it's a new system, how many people will ride it? 1581 01:14:24,090 --> 01:14:26,130 And therefore, what mode will I choose? 1582 01:14:26,130 --> 01:14:31,080 How much will it cost, or many, many applications. 1583 01:14:31,080 --> 01:14:31,740 Ari? 1584 01:14:31,740 --> 01:14:33,406 AUDIENCE: Well, I guess, maybe could you 1585 01:14:33,406 --> 01:14:35,770 speak to how well models do? 1586 01:14:35,770 --> 01:14:39,710 Do we have evidence that models are good, models are-- 1587 01:14:39,710 --> 01:14:40,614 how-- what are some-- 1588 01:14:40,614 --> 01:14:41,280 PROFESSOR: Yeah. 1589 01:14:41,280 --> 01:14:44,790 I haven't actually seen-- 1590 01:14:44,790 --> 01:14:47,490 I'm not very familiar with any study that 1591 01:14:47,490 --> 01:14:48,900 systematically looked at that. 1592 01:14:48,900 --> 01:14:49,525 AUDIENCE: Yeah. 1593 01:14:49,525 --> 01:14:56,920 PROFESSOR: But again, what we have seen a lot 1594 01:14:56,920 --> 01:15:00,340 of is the earlier approaches. 1595 01:15:00,340 --> 01:15:02,470 And these sort of fudge factors put 1596 01:15:02,470 --> 01:15:06,790 in, so the 50% stated preference without conjoint analysis, 1597 01:15:06,790 --> 01:15:07,940 things like that. 1598 01:15:07,940 --> 01:15:10,450 And those are all over the map in terms 1599 01:15:10,450 --> 01:15:16,690 of whether they were accurate or not, often because often they 1600 01:15:16,690 --> 01:15:18,420 overshoot the prediction. 1601 01:15:18,420 --> 01:15:21,535 And you wonder, to what extent is that 1602 01:15:21,535 --> 01:15:24,240 a bias because you want funding and you 1603 01:15:24,240 --> 01:15:29,410 want to show that there will be ridership for it if it's 1604 01:15:29,410 --> 01:15:31,990 a capital project? 1605 01:15:31,990 --> 01:15:36,070 But I haven't really seen-- and maybe it exists-- 1606 01:15:36,070 --> 01:15:39,100 a systematic look at recent models and how accurate 1607 01:15:39,100 --> 01:15:40,050 they were. 1608 01:15:40,050 --> 01:15:43,140 AUDIENCE: You mostly see it for new systems and extensions, 1609 01:15:43,140 --> 01:15:45,200 not for existing service changes. 1610 01:15:45,200 --> 01:15:47,530 PROFESSOR: Yeah. 1611 01:15:47,530 --> 01:15:51,580 AUDIENCE: I'm sort of interested not in comparing models 1612 01:15:51,580 --> 01:15:54,430 to what happens in reality but what we talked about an hour 1613 01:15:54,430 --> 01:16:00,070 ago about surveys, how stated preference surveys compared 1614 01:16:00,070 --> 01:16:02,990 to real preference surveys, if-- 1615 01:16:02,990 --> 01:16:04,990 has anyone done any significant research 1616 01:16:04,990 --> 01:16:07,330 about a project that was open? 1617 01:16:07,330 --> 01:16:10,270 What did people say before about the moment 1618 01:16:10,270 --> 01:16:12,220 that they were going to use it? 1619 01:16:12,220 --> 01:16:15,400 What did people then say after about what mode did I use? 1620 01:16:15,400 --> 01:16:17,260 PROFESSOR: Yeah. 1621 01:16:17,260 --> 01:16:20,740 I can't site specific results because I don't remember them. 1622 01:16:20,740 --> 01:16:22,690 But that has been looked at. 1623 01:16:22,690 --> 01:16:28,070 And there is research on the literature on that topic. 1624 01:16:28,070 --> 01:16:30,700 AUDIENCE: Isn't there generally pretty bad follow-up, though, 1625 01:16:30,700 --> 01:16:31,240 when-- 1626 01:16:31,240 --> 01:16:31,865 PROFESSOR: Yes. 1627 01:16:31,865 --> 01:16:34,754 AUDIENCE: --these demand models are done, there is very poor-- 1628 01:16:34,754 --> 01:16:36,170 PROFESSOR: But I think generally-- 1629 01:16:36,170 --> 01:16:37,300 AUDIENCE: They don't actually predict-- 1630 01:16:37,300 --> 01:16:38,466 PROFESSOR: Yes, that's true. 1631 01:16:38,466 --> 01:16:43,160 So usually once you do a study, you predict the demand. 1632 01:16:43,160 --> 01:16:46,300 If the party goes forward, nobody 1633 01:16:46,300 --> 01:16:49,894 bothers to compare how bad it was often, 1634 01:16:49,894 --> 01:16:50,810 at least not formally. 1635 01:16:50,810 --> 01:16:54,620 I'm sure people comment, gee, that was way off 1636 01:16:54,620 --> 01:16:56,750 or that was great. 1637 01:16:56,750 --> 01:17:02,530 But do we have sort of documented studies showing 1638 01:17:02,530 --> 01:17:06,310 how well or how accurate were these predictions? 1639 01:17:06,310 --> 01:17:08,270 I haven't really seen many of them. 1640 01:17:08,270 --> 01:17:09,910 AUDIENCE: [INAUDIBLE] newspapers. 1641 01:17:09,910 --> 01:17:11,560 And [INAUDIBLE] what we do with that? 1642 01:17:11,560 --> 01:17:13,630 Should we require follow-up and then if-- 1643 01:17:13,630 --> 01:17:14,940 [INTERPOSING VOICES] 1644 01:17:14,940 --> 01:17:16,740 --portion of the grant you got? 1645 01:17:16,740 --> 01:17:17,884 PROFESSOR: Yeah. 1646 01:17:17,884 --> 01:17:19,300 AUDIENCE: Well, and maybe it would 1647 01:17:19,300 --> 01:17:22,594 allow us to say, OK, which models have-- 1648 01:17:22,594 --> 01:17:24,010 where have the models worked well? 1649 01:17:24,010 --> 01:17:25,010 Where have they not worked well? 1650 01:17:25,010 --> 01:17:25,676 PROFESSOR: Yeah. 1651 01:17:25,676 --> 01:17:27,030 Yeah. 1652 01:17:27,030 --> 01:17:29,890 Well, the idea of having a disincentive for overshooting 1653 01:17:29,890 --> 01:17:30,782 is interesting. 1654 01:17:30,782 --> 01:17:31,990 AUDIENCE: The price is right. 1655 01:17:31,990 --> 01:17:33,490 If you consistently overshoot, we'll 1656 01:17:33,490 --> 01:17:36,080 penalize you in the future. 1657 01:17:36,080 --> 01:17:38,770 PROFESSOR: So that would maybe put you on a position 1658 01:17:38,770 --> 01:17:41,304 where you want to be as accurate as possible. 1659 01:17:41,304 --> 01:17:45,174 AUDIENCE: But then you might be too conservative, maybe. 1660 01:17:45,174 --> 01:17:45,840 PROFESSOR: Yeah. 1661 01:17:45,840 --> 01:17:47,240 It could be either way, right? 1662 01:17:47,240 --> 01:17:49,200 You could be penalized for undershooting it. 1663 01:17:49,200 --> 01:17:52,500 AUDIENCE: [INAUDIBLE] 1664 01:17:52,500 --> 01:17:55,230 PROFESSOR: So, yeah. 1665 01:17:55,230 --> 01:17:59,350 Any other comments or questions? 1666 01:17:59,350 --> 01:18:01,200 All right.