Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=

# Public and Private Transit: Evidence from Lagos

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=mC4ESywpIDc&t=400s


## Talk (00:06:40 – 01:00:40)

[00:06:46] All right, welcome everyone. Uh, and we're like very much on time. I have just a few announcements. Um, so for those of you I haven't met, I'm Simma Jaya Chandran with Ben Olen. I uh co-direct the development economics  
[00:07:00] program and we're happy to have what's I don't know probably the 15th MBR summer institute in in development. Uh so a few announcements. So the first is to remind everyone that in agreeing to attend this  
[00:07:13] conference, you sign the NBR code of conduct uh and are expected to adhere to to it. Uh second I wanted to uh thank the program committee which is is Jing  
[00:07:26] Kai Fred Finean and Melanie Morton who I don't think are here and Doug Golan Doug Golan and Eric Edmmonds and Sandep Sutanker and Victoria Bosi who who are  
[00:07:40] so thanks it's I forget how many papers we had something on the order of 300 so it's a lot of work and uh many great papers on the program and many great papers we weren't able to accept. Um the other announcement is just a reminder  
[00:07:54] that this is live streamed. So what you say as a presenter or as a questioner will be heard by uh more pe people and we'll hear a little bit more about Dean Young's organization. But you know there'll be other people uh hopefully that will encourage more people to watch  
[00:08:09] the live stream. So with questions I think there are some microphones around the room. So if you're near one please use it. Uh I realize it becomes a pain.  
[00:08:17] So the other thing is all the presenters repeat the question you know as you um answer your question at least make the qu question clear to those who um aren't able to hear it. Uh  
[00:08:30] uh we have three of attendees who are coming from low and middle inome countries uh thanks to financial support from the review of economics and statistics and MBR. So  
[00:08:43] there's Alan uh from Indonesia, Vidya from India and Sakib I think probably from Pakistan. So welcome if you have a chance to meet >> Oh Bangladesh. Thank you. Uh  
[00:08:57] I had like a one in two chance he was Rachel's co-author. So uh uh so welcome, you know, to have a chance to chat with him. AND THEN FINALLY ON FORMAT. SO, the 50-minute talks like like Nick, they're going to be no questions for the 10 first 10 minutes  
[00:09:12] and then normal seminar styles after that. We'll also starting later today have 9m minute lightning rounds where there are no questions. So, over to you.  
[00:09:24] >> Yeah. Yeah. Perfect. >> Okay. Thanks very much uh to the organizers for having us on the uh program. Uh this project is the result of years of work with a great team who are all here today. We have Dan Bjork Gran at Colombia and then uh Alice Duho  
[00:09:38] and Gita Kanagpal who at the World Bank and uh today we're going to be talking about transit in uh Africa. African cities are predicted to double by uh 2050 and so they're going to have to figure out how to move large numbers of  
[00:09:51] people around. We're going to be looking at Lagos uh Nigeria. Lagos is subsahara in Africa's largest mega city with around 23 uh million residents. So let me show you some data on how people in  
[00:10:04] Lagos get around. 20% of people uh travel by car. Now despite the focus we have in the economics literature on the impact of expensive capital intensive mass uh rapid uh transit systems run by  
[00:10:18] the public sector. These account for only around 5% of trips in Lagos. The vast majority of trips in this uh enormous mega city are taken using a network of private minibuses. These are decentralized. They're owned and  
[00:10:31] operated by thousands of individuals and they're how people in Lagos and much of the developing world uh move around.  
[00:10:38] Now, in recent years, for a variety of reasons, governments have been making uh investments in public transit infrastructure to try and expand out this blue uh share of the market. And in this paper, we're going to be looking at one uh such investment, which is the  
[00:10:52] Lagos bus reform initiative. And this is the first phase of the government's eventual plan to build out an entire new public transit system for the uh entire city. Uh this phase consisted of around  
[00:11:04] 800 uh buses on 40 new routes which is around the size uh of San Francisco's uh bus network but just jump started from nothing within the space of a couple of years.  
[00:11:16] But next to the existing private network uh it's just dwarfed in size. So using data we collected in this project, we find that this existing private network consists of around 760 routes applied by  
[00:11:30] around 75,000 minibuses. And for a sense of scale, that's 10 times the number of buses or 10 times the number of subway cars that are in New York's uh uh metro system. Okay. And this dwarfing of the public by the private is not something  
[00:11:43] that's unique to Lagos. Many cities in the region have been making similar investments. Darus Salam in Tanzania spent around $400 million on a new BRT system with modern buses on dedicated lanes. Uh but even there it's just  
[00:11:58] dwarfed by the private system. Okay. So it takes up around one and a half% of trips which is just uh uh you know relatively small compared to the private sector 60%. So the kind of starting off point for this paper is is is the image  
[00:12:11] that's in this slide is that a lot of the research that we have are have focused on the direct effects of of what's going on inside this uh circle.  
[00:12:20] But we think there's the potential for this to miss out a big part of the story. If the private sector responds to public entry, right? If they change things that commuters value like prices or service frequency and weight times, then we could miss a big part of the picture by ignoring the response of the  
[00:12:35] uh private sector. And we think that given the you know this interaction is important to study because given the cost of building these kind of systems it's very unlikely that we're going to go from a situation with 0% public  
[00:12:47] penetration to 100% within the matter of years or within even a few decades.  
[00:12:52] Okay. So these public and private systems are going to interact for decades to come. And so we think it's important to understand the nature of this interaction.  
[00:13:01] So that leads us to our uh research question in this paper which is how the private uh transit sector responds to public entry and ultimately whether this matters for thinking about the welfare impacts of new transit infrastructure  
[00:13:14] and of course looking at this interaction between formal and informal market institutions is not something that's new uh in development. We have lots of work looking at this interaction within education, insurance and healthcare. We think uh you know this uh  
[00:13:28] transit application is of independent interest a because of the amount of money that's going to go towards these investments. So understanding the cost benefit is is important. And second, transit networks are by their nature networks and so they involve spillovers  
[00:13:42] across different segments in the market. And so that's something that's a little unique to the transit setting which we want to uh understand in this paper.  
[00:13:51] So we're going to answer this question in four steps. The first part is we're going to conduct a data collection during the staggered roll out of a subset of these roots, 13 of these roots which we managed to uh overlap with our our our data collection period. Now  
[00:14:05] measuring the public system is uh kind of straightforward. So it uses uh swipe cards. So we have all the act we have access to all the same metadata that you might have in uh in in in a city in the US. But the problem is is the private uh  
[00:14:19] sector is kind of invisible to both researchers but as well as to government agencies. So when we started this project with our uh uh government partners, they didn't even know where all of the these private routes in the city were, let alone have the ability to  
[00:14:33] measure their characteristics or how they were changing over time. So we filled this gap uh via large scale data collection effort over a couple of years that consist of three arms. First we kind of measured characteristics of the  
[00:14:46] network itself. So we sent enumerators out to stand at the end of 280 routes and then every 15 to 30 minutes they would collect data on the number of buses that were departing, the price that those buses were charging and also  
[00:14:59] the number of buses that were in the bus cues and for some periods the number of uh commuters that were in the commuter cues as well.  
[00:15:07] We also followed 850 of these drivers over around a year and a half uh across five survey rounds looking at a bunch of their business uh outcomes such as their income, the number of trips they did on the most recent travel day as well as  
[00:15:20] the routes that they uh decided to ply. And then at one point in time, we conducted an entire census of the network that allowed us to measure all every uh all of these 760 routes that we could find uh as well as some of their  
[00:15:32] characteristic. When uh we have this data in hand, we then use in the second step to estimate that how the private market responds to public entry. The main outcomes that we're interested in are the attributes  
[00:15:46] uh that you know commuters value, right? So we're going to be looking at weight times in the private market as measured through the frequency of departures, the prices uh paid by commuters, and then we're also going to measure some supply side outcomes. So, we're going to look  
[00:15:58] at the routes that drivers decide to ply and like Q lengths of buses in those cues, which is going to be indicative for how long drivers have to wait to take a trip and how many trips they'll be able to take in a day. We're going to have a specification that takes  
[00:16:12] spillovers uh seriously. So we're going to break down effects into treated roots where the government provides new service and competes directly with the private sector as well as separating those from connected routes where the government doesn't enter but it's connected to one of these uh new routes  
[00:16:27] where we might uh think there could be spillovers.  
[00:16:30] Um in the paper we have diff and diff we have event studies but I think our most persuasive uh piece of evidence towards these effects being causal are a set of placebo checks that we use where the government had very frequent changes in their opening plans. cancelling uh a lot  
[00:16:45] of routes at the last minute and a subset of these were cancelled after uh a large terminal was burned down in late 2020 due to anti- police brutality protests. And so we can use the fact that there were eight routes where we were collecting data where they planned  
[00:16:57] to open uh uh to to introduce but then cancelled because of this uh uh this this destruction.  
[00:17:05] So in the second part, we're document we're going to be documenting that prices and weight times change in this market. And then if we want to use those to speak to how that matters for commuter welfare, we need to know how commuters value time and price. And we're going to do this by conducting a  
[00:17:18] field experiment. Since this is a context where you have relatively low smartphone penetration, we're going to develop a custom app that works with basic phones. And we're going to recruit people at home. And as I'll explain later, we're going to devise this app  
[00:17:32] and devise this setup so that every time they come to the regular bus stop over the next two weeks, we're going to be able to offer them a different offer each day of, you know, pay you a certain amount to wait a certain amount. And whether or not people reject or accept these offers is going to reveal to us  
[00:17:47] what their value of wait time is. This is going to be somewhat similar to approaches that people have used with ride share apps, right? like on uh people have used people's choices on Uber to infer how willing they are to to save some time and we think that there's  
[00:18:01] a selection bias in these kind of approaches where you know if you're anything like me you know let alone the fact that people who use Uber might be different from the general population even the time that you choose to use Uber you may select on your value of time so if my employer is paying I'm  
[00:18:16] more likely to use Uber if I'm arriving in SFO after a 10-hour flight with two jet lag kids I'm more likely to use uh Uber okay so we're going to design our exper we're going to I'm going to explain to you that there's going to be a similar selection bias in our design, but we're going to be able to design our  
[00:18:31] experiment to directly address this and we're going to show it's quantitatively important in terms of the value time estimates that you get out of that. And then finally, we're going to piece this all together uh using a queueing model of private transit and that's going to allow us to evaluate welfare effects  
[00:18:45] both for commuters but also for drivers who are often a large opposition to these kind of policies. So it's going to be important for us to put a number on uh the exact extent to which they lose relative to commuters gain.  
[00:18:58] Okay. So that's the paper in a nutshell. In the interest of time, I'm going to skip over uh related work. I'm happy to to I think we're basically at the 10 minutes. I'm happy to take questions, but if there's not, I'll get into the  
[00:19:09] background and data collection. Yeah, >> there may be and in that situation that you don't have to the most surplus supply side of  
[00:19:22] that sector to know whether the public entry is a good thing or a bad thing.  
[00:19:27] You can just look are people using public buses if that's covering its cost and then if so we know that that entry will beneficial regardless of what's happening there amongst the buses.  
[00:19:38] >> So the the question was about whether the supply side is perfectly competitive. So I'm going to get into that right now. But the the quick answer is that you should think of this as the drivers have free entry, but there is essentially a firm who's a monopolist, and that's going to be the association.  
[00:19:52] They're the ones who set prices and fees in the market. So, think of it a little like Uber, right? Every driver can choose freely to enter anywhere. But Uber is the one who's setting the price for on the commuter side and the price on the supply side through fees. And so there's going to be, you know, exactly  
[00:20:06] what we label this if we label it driver surplus or association surplus. Yeah, we're not going to take a stand on that, but that's going to be the market structure. Okay. So, I've kind of given you given you half of what's on this slide already. So, how's the transit  
[00:20:20] industry organized? It's exactly kind of how I said to Nick. You don't want to think of it as just drivers and riders.  
[00:20:26] You want to think of there on the supply side being two agents. So, first we have our drivers. And so they're going to take prices pretty much as given at least from the terminals uh you know outside the terminals it's a a little freer but these drivers basically take  
[00:20:40] prices as given and they're going to operate according to free entry. Okay.  
[00:20:44] So if you have the money to have a bus and to pay these entry fees that the association is going to set you can enter uh you can enter a route. Uh 90% of these drivers just drive within a day the same route. So we thought you might have this dynamic problem going on where  
[00:20:59] you go from A to B and then you decide where to go on from A in the data.  
[00:21:02] That's not what we see at all. Almost all drivers go from A to B then B to A then A to B and so on. They change their regular route over time but within a day a day uh that's what they're doing. On top of that there's the drivers association or in other settings these  
[00:21:16] are called unions and this is basically this kind of monopolist or self-regulator but I don't think they're really acting uh in the in the interest of drivers uh per se. They're acting more like a monopolist. So they're the ones who are setting prices which are uh  
[00:21:30] faced by commuters and by drivers. They affect drivers revenue and they also charge fees which affect drivers uh cost.  
[00:21:39] Okay. So that's the kind of private uh market setup and then into this background uh you know on top of this came uh this bus reform initiative from the Lagos uh government. This consisted primarily of these new uh you know  
[00:21:52] modern high-capacity buses almost all of them so 38 of them are not bus rapid transit meaning that they share the same roadway as regular cars regular private minivans. Okay. So when you think of  
[00:22:05] exactly how these uh new this new service is differentiated with the existing one it's not really in travel time it's in price in wait time as well as in amenity comfort and safety uh and  
[00:22:16] so on. They also built a bunch of new uh terminals. This was, you know, after the policy that we study, and we're still collecting data on it, they they've built two new subway lines. So, it's part of the city's overall kind of  
[00:22:29] long-term uh uh vision for providing better public transit, although we're just going to be studying this one uh uh um part, which is this uh bus bus initiative.  
[00:22:40] >> Okay. Um, did you guys track uh the speeds for private vehicles and whether those changed with the introduction of the process?  
[00:22:52] >> Yeah, perfect. So, yeah, I'm I'm going to get to that in a in a second, but we did we did it through Google. So, they don't differ in invehicle travel times because they use the same roads, but it could be true that congestion change and so travel times were affected. So, we  
[00:23:06] collected data uh on travel times on Google Google Maps and you see very precise null effect. And the reason is that these aren't, you know, the service frequency is not super high. It's not shifting around a huge amount of of  
[00:23:18] traffic on these roads. Yeah. Okay. So now I'm going to show you a model. And the purpose of the model is not going to be to do some complicated structural estimation. It's going to be to to retrieve some sufficient statistics that we can use to guide us in our data  
[00:23:33] collection. It's going to it's going to tell us what we want to measure if we want to go out and answer this question of how important are these in is this indirect response from the private market for uh for understanding the welfare effects of of public entry.  
[00:23:47] So what's the setup you should have in mind? You should have in mind a bus terminal where there's many cues to go to to different routes. So we might have terminal I with different routes going to destination J, K, and L.  
[00:24:00] I'm going to show you what's going on under the hood in the model or more specifically in the appendix and then I'm just going to show you the uh equations that come out of that that are going to be pretty intuitive. But what is going on under the hood is a a queueing model that comes from operations research. It's called an MM1  
[00:24:14] queuing model with bulk service. So suppose we have a route from I to J.  
[00:24:19] There's going to be a number of buses already in that queue and a bus might choose uh if it bus chooses to enter that route, it joins at the back of the queue. Passengers are going to arrive dynamically and as they arrive they're going to board the bus which is at the  
[00:24:32] top of the queue. Now in our data we see that around 94% of buses wait until they're completely full before they depart. So we're just going to impose that as a technological constraint.  
[00:24:43] Buses are going to wait until they're completely full. Once they are full, they depart and the next bus is going to continue to load. Okay. And that kind of model going on under the hood gives some pretty intuitive uh expressions for for expected payoffs on the commuter and on  
[00:24:57] the driver's side. Okay. So here we're thinking about if a commuter chooses Minibus M, what what's their payoff going to be? Well, it's going to be an amenity, something like comfort, safety, uh reliability.  
[00:25:10] It's going to depend on price and it's going to depend on total travel time.  
[00:25:15] Now total travel time depends on invehicle travel time as well as uh wait time, right? the time between you getting to the the bus stop and the bus actually departing. So to Marco's question here, we're going to see in the data that travel time doesn't really  
[00:25:28] change. So everything that I'll show you here, that's going to be a fixed uh characteristic. So I want to focus more on weight time, right? So what is wait time determined by? Well, if you arrive at the bus stop, then with some probability there's going to be a bus in  
[00:25:41] the queue. And so how long you're going to have to wait if there's a bus in the queue just depends on how long it takes for that bus to fill. Right? So basically if there is a bus in the queue wait times are completely pinned down by the demand side and that creates a  
[00:25:55] demand side externality. Okay. So if we all agree to take mini buses then these buses are going to fill very quickly and none of us are going to have to wait very long. And as a kind of counterpoint then if a competitor comes in and siphons away half of the demand for mini  
[00:26:09] buses the arrival rates are going to be uh twice as slow. these buses are going to take twice as long to fill and so weight times are going to rise for people who continue to use the service.  
[00:26:19] Okay. So that conceptual point is going to be important for thinking about the results that we'll uh document later if there is yeah mush problem as thank you as uh this one for  
[00:26:34] Cape Town and so in his model the market power I mean this goes back to Nick's question the market power really mattered because it depends on how many private companies are providing like the same route service on the same route so  
[00:26:48] how quickly the um so so just relating it to your paper it just seems like you know when you're looking at the effect of public transit on the private market if private market market power itself changes in the presence of public transit then then you know it might  
[00:27:02] affect some of the parameters here on the >> No no no absolutely so I'm I'm not going to show you this but it this is all in the paper so first of all yeah Lucas has an excellent paper looking at the um private transit market in Cape Town it was it's changed over the years but originally it was more on matching  
[00:27:17] whereas ours is on queuing but conceptually he's looking at just the private market itself whereas the main difference is we're going to be looking at the public private interaction with kind of an a natural experiment but everything that you've just said is going to be baked into our model so I'm  
[00:27:31] going to get to it in a sec well I'm just going to have a line so I'll just I'll just explain it there's going to be the association as I said to um uh Nick right prices are going to be exogenous on the demand side from the driver side they're also going to take them as given  
[00:27:45] and then there's going to be an association who who decid decides uh the price, a per trip fee, and a fixed entry fee. And what they're going to do is be maximizing their own profits, how what are they going to be seeking to maximize? They're going to be seeking to  
[00:27:58] maximize uh like variable revenues. So like post entry revenues because under free entry, they're going to capture all of that through through fixed cost. So exactly as you're saying in the model when a competitor comes in you're actually going to see that they lower  
[00:28:11] they lower prices and the the the way that it's going to become is that the way that's going to come through is demand will become more elastic right under loit preferences as you have more varieties demand elasticity goes up and so they're going to lower their prices that way.  
[00:28:26] Okay. Okay. So this is the commuter side. On the driver side again there's this dynamic model going on but it gives you a pretty intuitive expression for the uh average payoff. So, it's going to depend on the profits that drivers make on every trip. That's just the price  
[00:28:41] times the capacity of the bus net of the cost of uh doing the trip. And then it's going to depend on the number of trips that drivers can make a a day on that route. Okay? And that the number of trips is going to depend on the length of a day relative to total trip time.  
[00:28:55] Total trip time is the length of time it takes to make the trip. That's tcap t.  
[00:29:00] As well as the time it takes to queue, right? of the time from joining the back of the queue to get to the front of the queue. And that's going to be the these this kind of time margin is going to be the key way that this market clears on the uh on the supply side, right? And it creates this margin for business  
[00:29:14] stealing. The more drivers that enter a route, the higher the steady state length of that que is going to be. And so drivers going to face higher queue times and they're going to be able to make less trips within a day. Again, these are all just foreshadowing things that we're going to see in the data.  
[00:29:28] Vice versa, you can think about exactly the same thing as what happens from a demand shock, right? So if a competitor starts providing service on this route, these cues empty uh less quickly. So if the number of drivers are fixed, the length of the queue is going to rise. Q  
[00:29:42] time rises and so drivers are going to be able to make less uh trips in a day.  
[00:29:46] Yeah. >> Are wait until the bus is full or is that a choice variable that the driver could choose to leave with a partially bus in order to, you know, increase the  
[00:30:01] number of trips and decrease. >> Yeah. Yeah, that's a great question. So, the question was, are we imposing that drivers wait until their the bus is full before they before they leave? In this paper, we are. It's just as a as a shortcut, and it's because we see on 94%  
[00:30:15] of our routes, they are waiting until they're full. In a follow-on paper, which I'll kind of foreshadow at the end, we're we're doing an experiment where we're trying to pay drivers to leave earlier to measure that uh margin.  
[00:30:25] We think on less busy routes, you know, I'll talk about route selection in a second, but on less busy routes than the public uh entered here, this may be more of a more of a margin and it's a a form of um another form of kind of potential  
[00:30:38] business stealing as well. Okay, so that's driver payoffs and then drivers going to be free to choose between different routes. So within a terminal, they're going to have idiosyncratic preferences for different routes. And that's just going to give us a an upward sloping supply of drivers at terminal I  
[00:30:52] who choose route J depending on its uh payoff. And then there's going to be free entry across uh terminals as as well. You're going to have this association which sets prices as well as a fixed fee to enter any terminal. And as we've talked about, they're going to  
[00:31:06] be maximizing variable profits. Why? because under free entry they're going to capture all of those variable profits uh via free entry through these fees.  
[00:31:16] Okay, so that's the kind of uh uh setup. So now let me show you what I want to do with this model >> in practice. Are there folks in the association who can take home cash from this or is it a nonprofit and they're not supposed to take money?  
[00:31:28] >> No. So I I want you to think of it as a 100% take-home. So they're maximizing, you know, in principle they're called the union and they represent the drivers. In some settings, you know, it's not all like this. In some countries, they actually do pay back a  
[00:31:42] share of these revenues to drivers. In Lagos, they don't. So, we're working on that in the second paper, trying to infer from the you from the association's fees exactly what they're maximizing. But my answer to you today, which is just anecdotal, is that they're  
[00:31:56] maximizing their own uh profits. They're very important like political uh group as well. They can mobilize like huge numbers of people to come and support candidates that they want. So they're very very uh they're very very powerful.  
[00:32:07] So there's few checks on on on what they can do.  
[00:32:12] >> Where is >> pardon?  
[00:32:16] >> Where the consumers live? They live all over the city and so we are going to take uh choices as if you're are you prefacing like do they would they change where they choose to live?  
[00:32:27] >> I just like model consumers coming from some process.  
[00:32:33] >> Yeah. Exactly. Exact. Exactly. So there there are different uh there is just an arrival rate of commuters who are coming who demand travel from I to J. That's exogenous that we're going to infer from the data. And then what's indogenous is  
[00:32:47] which mode they choose. And that's what I'm going to get.  
[00:33:02] >> Y Yeah.  
[00:33:18] >> No. No. So I would just think of it nests of things that can change. Right.  
[00:33:21] So if if the first place you would expect to look would actually be just all choices fixed and there'll just be some surplus. The next place you would look is that I'm not going to change where I'm where I'm living or where I'm working or the trip, but I might change which mode that I use. And then the next place you would look is that you might  
[00:33:36] change where you live and where you work. So I have to So we don't have a huge amount of data to to uh to to to validate the the the last part is not happening. We have some cell phone data that we looked at some point and we  
[00:33:51] didn't see big movements because so cell phone data would tell us like volume of total trips going from different origins to destinations and we didn't see big movements in that. My prior is this is not you know surprising at all.  
[00:34:10] So that that we're going to that's going to be fine because these stops are not like at so the where you take the public and private is not at exactly the same place. It's just going to be whether you're going the key the the thing that would would be problematic for us is if  
[00:34:24] total number of trips between origin and destination changed because people started changing the types of trips or where they were living or where they were working. And yeah, for me, we don't have really strong evidence of that, but we can just, you know, based on the size  
[00:34:37] of this uh this shock, we don't really think it's a play.  
[00:34:41] >> Yeah. Okay. Yeah. See, >> like destination, but these are like normal buses with stops along the way, different distance trips.  
[00:34:52] >> Yeah. Great. So, I don't So, we also have a a stat on this which I don't have on the top of my head, but so we have data on >> Can you repeat the curs?  
[00:34:59] >> Oh, sorry. Yes. Yes. The the question was uh I'm modeling this or sorry we are modeling this as just origin to destination markets and there's nothing happening at bus stops in between. We have some data for both the public system better data for the private  
[00:35:13] system on for everyone who gets off at the origin where do they get off at and the stat is is something around for the private system it's I think it's over 80% of people travel all the way to the destination. So you should think of this  
[00:35:27] really as kind of like a point-to-point quite dense network instead of a network with lots of transfers and intermediate bus stops. That statistic I think is even higher for the for the for the public system.  
[00:35:38] >> Yeah. >> Sorry if I missed this but so based off of Arun's question as well like so you know is is there anything about because I have to get to this the the bus itself. So are you at all thinking about like my time that it takes to get from  
[00:35:52] home to the the actual bus stop or to the >> so >> transit place because that matters tremendously for >> so so it's go so actually what I'm going to show you right now is that for the public system it actually doesn't matter. Okay. So we're going to need one  
[00:36:06] sufficient statistic to measure the benefit of the public system. So let me just show you that and get back to it.  
[00:36:10] For the p for the for the private system that's not going to change. Right? The time that it takes you to walk to your private bus stop isn't affected by the policy. But let me get back to that when I show you. Okay. So, so I'm now going to we're now going to think about two periods before public entry. You can  
[00:36:24] choose two options, an outside option or a minibus. And you're going to have extreme value preferences over these with theta governing uh how dispersed those draws are. So, how substitutable the two modes are. Then all that's going to happen after public entry is that you  
[00:36:37] have this third option that becomes available. So, let's think about what the change in consumer surplus is. Well, it ends up that you can write it kind of conveniently as a break it down into two margins, right? So, one gets precisely  
[00:36:50] to your uh uh question um which is that if I want to know the direct effect the the direct welfare effect of just providing this new transit option, I don't actually need to know do you dislike it because it's very far to walk  
[00:37:05] to or do you like it because it's very comfortable or do you like it because it's very cheap. this all I need to know is is what the market share is of that in the post- entry period. So that's this SP uh prime term. And the idea is  
[00:37:19] kind of simple, right? It's that I don't need to know exactly why you like it, but under these form of preferences, if I see a 20% market share, there has to be something really uh you know, attractive that induce 20% of people to take it. And in this under these preferences, that's all I need to know.  
[00:37:33] I do need to scale that by the substitution elasticity because if I see a 20% market share and it's really really hard for people to switch then that means right that 20% must have uh you know it must have given people a real uh uh welfare gain in order to get  
[00:37:47] 20% of those people to to switch. Okay, so that's the first part. That's the kind of direct effect. And then there's this indirect effect on consumer surplus which is determined by how the attributes of existing uh modes change  
[00:38:01] in response to entry. And so in our case that's going to be these uh uh how the weight travel times or in our case weight times and prices in the private sector change in response to public entry.  
[00:38:14] Okay. And this kind of tells us that we have a set of four sufficient statistics that we need to go out and measure. We need data on choice shares. We need reduce form elasticities of how public entry uh affects times and prices in the private market. We need to know how  
[00:38:29] people value those changes in prices in time. And that we're going to get at with our weight time experiment. And then we need to know how easy it is to substitute between these different modes. I'm not going to have time to cover this in the talk today, but we're going to estimate that through four  
[00:38:42] different uh quasi experiments in the public system where the government kind of suddenly changed their price and we can look at changes in volumes uh to identify that uh that parameter.  
[00:38:53] We have a similar expression for thinking about impacts on driver surplus. Uh I'm not going to go into this in the interest of time, but I'll I'll show you the the results at the end of the talk.  
[00:39:02] Okay, so now I'm going to get to the main empirics which is just showing you how private transit responds to a public transit roll up the sample of routes uh where we collect our data. So you know just to preface this obviously the the main problem here when when trying to  
[00:39:16] answer this question is that the government is choosing uh which routes to uh to to provide uh service on and so the roots uh may well be different than uh than than non-treated routes. Um our sample overall is going to be routes  
[00:39:31] that the government uh plans eventually to serve. Okay. So government is is spending all this infrastructure.  
[00:39:36] They're building these uh they're buying these large capacity buses. They tend to want to provide them on higher volume routes where you know providing large uh buses makes economic sense. And so we're not just going to compare those routes to random private routes. We're going to  
[00:39:50] look look at their entire network that they wanted to build across five phases.  
[00:39:54] and we're going to uh use as controls uh private transit routes that are included in those later phases.  
[00:40:02] Okay. Of course, still though the government chose to enter on the routes that they did for a reason. And so, you know, when we look in the data to to look at how treated routes compared with untreated roots, we actually failed to reject balance between them. That could be due to the way that we're  
[00:40:16] constructing this sample. Um but we do see that terminals with new public roots are different. So, they tend to look more hubby. they tend to have higher roots and higher higher volumes. And so we have two sets of regressions in the paper. One is we're going to uh where  
[00:40:30] we're going to run within terminal comparisons. So that's going to address this concern. So we're going to have within terminal fixed effects. Uh but it's going to be subject to violations if there's within terminal spillovers.  
[00:40:41] Okay. So if drivers for example switch from treated roots in a terminal to untreated roots, that's going to violate uh uh sva within these specifications.  
[00:40:50] So, we're also going to run a set of spillover specifications where we're going to compare treated roots in a terminal, connected roots, which I'll explain in a second what they are, in the same terminal versus control routes, which don't interact with the public  
[00:41:03] system at any uh at any endpoint. Um, uh, yeah, and so these are what I'm going to show you today. They're a little less precise, but the the treatment effects are kind of similar regardless of which specification we use.  
[00:41:18] Okay, so there aren't any questions. I'll I'll uh I'll I'll go in and show you the specification. So our kind of unit of analysis is going to be a route by 30 minute time interval by survey round t. And so what we're going to think of is different transportation  
[00:41:32] markets, which is really going to be a route by time period uh combination. So we're going to have three time periods, morning peak, afternoon off peak, afternoon peak, right? And so you might think that those transport markets might be very different, right? going from a  
[00:41:46] suburb to the center in the morning is going to be very different route than going from the suburb to the center in the in the afternoon. Okay, so we're going to have fixed effects for each of these transport markets. We're going to look at be looking at changes in our outcomes within them. And we're going to  
[00:42:00] be differentiating between treated routes. So these are where the public system operates from the same origin to the destination. Right? So in this pointto-oint network, they're kind of directly competing.  
[00:42:11] There's going to be our connected routes which you can think of as spillovers. So these are going to be routes where the public system serves either the origin or the destination but not both. Right?  
[00:42:22] So imagine that the public system is running from A to B. You might have another route from A to C where the public system doesn't run. That's going to be a connected route, a kind of spillover route in our in our world.  
[00:42:33] Okay. And then we're going to compare these with control routes. And these will be roots where where the public system isn't present at either the origin or the destination. Okay. So where there's no first order interactions, right? We have some other we have some other checks in the paper  
[00:42:48] to see these could these routes could overlap uh you know not to origin or destination but you know along along the path of these routes to kind of get to see question we see no kind of effects uh on on on those routes. So we really  
[00:43:01] think it is the direct competition of serving exactly the same market from origin to destination uh which is mattering.  
[00:43:09] Okay. Okay. So, let me show you uh what happens, which is a bit pixelated, but um we can still see. Okay. So, what do we see? We see around a 16 uh uh% reduction in departures in the in the private market, meaning that people are  
[00:43:22] having to wait uh longer. On average, this is around 3 minutes uh longer if you continue to use the private service.  
[00:43:29] We see that prices in the private market are falling by around 10%.  
[00:43:34] and Q's driver cues in the private market on treated routes are also falling by around 16%. Okay, so that already in and of itself is telling us that there's probably some spillovers at play. Why? If drivers ent route  
[00:43:48] decisions were completely fixed and then the demand for this service, you know, the number of departures falls, then if the same number of drivers were in the queue, if you're emptying the queue slower, the length of the queue would rise. Okay, so that already tells us that there's some exit on the driver  
[00:44:02] side from these uh uh from these cues either to other routes or uh uh or um or out of the industry.  
[00:44:10] Okay. And then we can also use this data combined with measurements of emissions from the private minibuses and the public buses to uh to infer around a 10% reduction in emissions on these treated routes.  
[00:44:23] >> But the driver needs to wait, not how long a passenger has to wait.  
[00:44:28] >> No, exactly. So in this so nor in most of our routes there is a a queue of buses and so that means there's zero cues of commuters because the commuters are coming and they're just loading that bus at the top and they're waiting in the bus  
[00:44:42] >> that could also like passengers are coming fewer passengers are coming maybe it's like one minute when they fill up but how how long do these buses take to fill up?  
[00:44:52] >> So the the average in our sample I think it's around eight minutes.  
[00:44:56] >> Yeah. Uh, no we haven't we haven't done that yet. The only test No, we haven't done  
[00:45:10] that yet. >> Yes. So it'll be that uh this theta this theta parame well sorry the combination  
[00:45:25] between the theta and the gamma is the is the price elasticity of demand.  
[00:45:34] >> Yeah. So with the way that again we're we're doing more tests of this in the second paper. We haven't done this in the in the first paper but we we think that the association is operating within one terminal. They're setting all of the prices within that terminal and there's  
[00:45:48] a different like subgroup of the association doing it somewhere else.  
[00:45:53] >> Why? >> No. So maybe Uber might do it like that but we don't think the association is doing that but we we going to test this right. So we're collecting data detailed data on all these fees and we can test like these kind of models to test their conduct basically to see what they're doing but this is kind of outside the  
[00:46:07] scope of here. The theory is kind of simple where basic Yeah. Okay. No, that's a fair that's a  
[00:46:20] fair point. Very >> York's entire  
[00:46:37] stop. No, no, the the former the the catchment area.  
[00:46:44] >> The catchment areas. I mean, we use them by name uh by name. But what we do is we go and measure. So the private route, you'll have a private route that says it's going from a place called uh EA.  
[00:46:55] You'll have a public route that says it's going from a place like EA. We went and measured the distance between these two. I think that I I can't remember off the top of my head. It was it was I think that medium was like 130 m or 150  
[00:47:07] mters or something like that. B.  
[00:47:19] >> Yeah. >> A to B because A and B are >> Yeah. When we constructed the sample, the control had some minimum distance criteria. So they were away from any of  
[00:47:34] the treated nodes. I I can't remember the exact distance threshold, but yeah, that was a concern. So controls at least the origins and destinations away are far away from treated nodes. Again, they could interact in the middle. We test to see if we see any effects on roots that  
[00:47:48] overlap more than less. And we see pretty precise errors. Yeah. Okay.  
[00:47:54] Okay. So, um I'm going to go through in kind of two minutes the rest of the empiric so I can go to the uh field experiment. Um but so okay. So already this effect on Q's is kind of prefacing some uh uh potential for spillover  
[00:48:09] effects, right? So on the supply side, we see driver Q's falling. So you know, we have the natural question of where where they're going. On the demand side, you could also see spillovers, right? If new service along one uh public route  
[00:48:22] affects increases demand, for example, from feeder routes that are connected.  
[00:48:26] Um, and so already looking in our driver data, we see first impacts that corroborate these uh these other results that were from the network observation.  
[00:48:35] So we see drivers on treated routes make less trips and earn less revenue, but we also see that they're more likely to switch to other routes in the same uh terminal. Okay, so not we don't see that they're more likely to switch to vastly different uh terminals, right? Where the  
[00:48:49] switching costs might be higher, but we do see they're more likely to switch within the same terminal. Okay. So, and that those are our connected roots, right? Our connected roots are other untreated roots within the same terminal. So, now we're going to I'm  
[00:49:03] going to show you the coefficients for the for these connected terms which are the spillover terms. So, we see uh uh you know no effect on demand uh on the demand side. So, no changes in demand on untreated connected routes suggesting uh  
[00:49:16] demand side spillovers aren't there. on uh connected uh routes we do see uh the association reducing prices. So we see around an 8% reduction in uh in in in in  
[00:49:28] prices and we see cues of drivers uh going up and again this is kind of consistent with all these uh other results right so if we go back here we're seeing drivers on treated m roots more likely to substitute to connected routes we see demand is completely  
[00:49:43] unchanged on connected routes and so if you increase supply to the queue but keep the exit rate from the queue the same you're going to see uh uh the cues u of drivers increasing and That's exactly what we see  
[00:49:57] >> why are the control roots like valid controls >> if you only if you uh so the control routes so the control routes are valid controls if drivers are not substituting away to them and that's what we so we  
[00:50:11] don't see evidence of that. So we see evidence of drivers switching to other other routes within the same terminal but we don't see evidence of them switching to other routes which are you know are control routes ones where neither the origin or the destination is  
[00:50:25] treated by public transit but importantly whether you know back to Aroon's question whether whether terminals are different than the terminals that you're currently serving.  
[00:50:33] So the idea is here there's switching costs, right? If I work a route, I know the route, I know the people who work there. So it's very easy for me to instead of going here, just go over here, I interact with the same people at this terminal. But if I want to switch to another route in a different part of the city, there's much higher switching  
[00:50:48] cost for me, right? None of those things are true for me anymore.  
[00:50:50] >> Yeah. So you're basically saying for them that like change over time is a valid cost effect.  
[00:50:56] >> Yeah. Yeah. >> Yeah.  
[00:50:59] >> Would be very difficult I think. There are lots of control groups, right? would be very dilute be difficult to pick up such an effect even if they were important. Um it would be difficult >> because there's lots of choices right within the same terminal there are small  
[00:51:12] number of choices so you you know the same number that's true but we we just have as an outcome did you move to any route outside of the terminal yeah we see if we looked at for a particular route yeah  
[00:51:24] exactly exactly okay and then the last piece is that you know again this is very suggestive because of power but even in the driver survey we see suggestive uh evidence supportive of what we're seeing in the in the observation, right? So on these  
[00:51:39] connected routes, um drivers are moving in there, Q times are going up. We would expect drivers to complete less routes a day. We see rough evidence of this and uh and they're making less revenue.  
[00:51:50] Okay, so I have seven minutes. I'm going to skip over the placebo checks. These were these these canceled routes uh that were canceled last minute due to these fires. And I'm going to get into the last part um which is how people value  
[00:52:05] uh uh weight time and price. Okay. So let's suppose I have a um a technology that allows me to interact with Sema in the following way. So suppose she chooses to use the mini bus uh and she  
[00:52:18] comes uh to me at the her regular bus stop in the morning and suppose I'm able to offer Sema a payment of S dollars to wait for delta t uh minutes. So, what's Sema's utility if she rejects? Well, she gets a common expected utility for the  
[00:52:32] day as well as a preference on that day to leave at time, right? Maybe she's in a rush. She really wants to leave there.  
[00:52:38] Maybe she's relaxed and and and she's not too fussed about leaving at time.  
[00:52:43] If she accepts my offer, then what's going to change? Well, she's going to dislike the fact that she's having to wait delta t minutes, but she's going to like the fact that I'm paying her s uh dollars. And then she's going to have a preference to leave deltat tow minutes later. Okay? So, we can combine these  
[00:52:57] two to think about whether Sema is going to accept or reject uh my offer. And it, you know, intuitively is going to be does she value the payment that I'm offering more than she dislikes the uh the wait time. Okay. And so, we're going to want to we're going to want to design  
[00:53:12] an experiment that allows us to offer in Sema random combinations of S and delta T and and see how she accepts them, rejects those offers, and use those to infer her value of time and of course do that for for hundreds of people. Okay.  
[00:53:25] But many people in this context don't have smartphones. And so we're going to address this by uh recruiting around 650 people within 1 kilometer of 18 randomly selected private bus stops. Okay, it's important that we recruit them at  
[00:53:39] weekends at home because if we were to just approach people at bus stops and say, "Hey, do you want to play this game with me?" Only people who are very relaxed uh and have a low value of time are going to be willing to to talk to us. Okay? And so they're going to be invited to check in if they uh uh come  
[00:53:54] in are going to be invited to check in with us at bus stops uh during morning weekday travel for the next three or five weeks. And they're going to take place uh in you know take part in the following interaction. So here we have our enumerator who's wearing her yellow Lagos mobility hat to allow her to be  
[00:54:09] kind of discreetly uh identifiable and she's going to be holding a smartphone.  
[00:54:14] Okay. And so every minute, this smartphone is going to have a code uh that changes. Okay, so the participant arrives at 7:12 a.m. and she texts the number that's displayed to a short code  
[00:54:27] that we've uh uh set up and we receive uh this code kind of on our server and we can process it, right? Because these uh codes are changing every minute, every bus stop every day, we know exactly where she's checking in and  
[00:54:40] when. And so we send her back a message thanking her for checking in at 7:12 a.m. We're going to send her a fixed air time uh fixed payment of 200 naira every time she checks in, which I'll talk about in a second. But then I'm going to  
[00:54:53] give her her offer of s and delta t. Right? So today I'm going to offer her 400 naira if she waits for 5 minutes.  
[00:55:01] Okay? Time goes on. The the short the code continues to change. If the participant rejects the offer, she just walks away and we don't receive another text from her. If uh if we do uh if she chooses to accept, how do we verify that  
[00:55:15] she's accepted? Well, she has to wait uh more than those five minutes and then uh she texts the code that's displayed. And again, because these codes are changing every minute every day, we we we verify that she's waited if she texts that code  
[00:55:28] to us. Okay. So then we reply, we thank her for waiting 6 minutes. And because that's longer than the five minute of her offer, we send her a reward of uh 400 naira. Um, I only have three minutes, right? I'm actually just going  
[00:55:41] to hold questions if that's okay. Um, okay. So, let me let me summarize what's in this interaction. So, we have this weight offer that's random for every uh person on every day, but we also have this check-in offer. This is going to be  
[00:55:55] randomized for each person at recruitment. So, some people are going to get 200 naira, other people are going to get a,000 naira. Okay, why do we why do we have this?  
[00:56:04] So, it's going to be because of this selection bias, which in our case is going to be the opposite of the one that I told you about uh Uber. Okay? So, think of what this game represents. It's kind of a hassle, right? If I'm in a rush, I'm not going to bother doing this. I'm just going to go on my way to work. And only in a day when I'm kind of  
[00:56:18] relaxed, I'm going to come in and check in. And so, consistent with this, at baseline, we asked people, "How many days do you plan to travel in the next week?" People report traveling 80% of days, but they only come in and check in with us uh on 55% of days. Okay? So  
[00:56:32] we're worried about exactly the opposite thing that people are going to have negative selection on of on value of time. Okay. So that is why we introduced this uh randomized offer of check-in.  
[00:56:43] Right? So the people that we pay only 200 naira every time they check in. They check in with us about 50% of the time.  
[00:56:50] The people that we offer a,000 naira, we manage to get them to uh check in 20% uh more or 10 percentage points at 60 uh 60% of the time. Right? So this randomized check-in offer is kind of inducing and incentivizing people to  
[00:57:03] participate more in our experiment. What do we see just in the raw data of their accept or reject offers uh when they do this? Well, it's exactly what you would expect with uh this election. So the people that we don't pay very much, they don't check in with us very much. And  
[00:57:17] when they do check in, they're more likely to accept. The people who we pay a lot to check in, we kind of force them to check in more. And when they check in, they're more likely to reject. Okay.  
[00:57:29] So that's our kind of fundamental mechanism at play in the raw data. What do we do? We just extend it's kind of a a Hecman selection model. We just extend our discrete choice model of accepting or rejecting to have a participation  
[00:57:43] equation as well. Right? So your choice to participate depends on how long it takes you to check in which we measure.  
[00:57:48] How long it takes you to go from the exact bus stop to meet the enumerator.  
[00:57:52] this payment which is random as well as your expected payoff uh from the interaction.  
[00:57:58] Okay. So what do we find? We find our main estimate of the value of time is around uh two times the wage. Now that might seem a bit high compared to a a a research that has it between you know  
[00:58:11] equal to the wage or or or half of the wage. And so we also do a naive uh estimation right just to see if this is an artifact of the intricacies of our experiment we or the selection correction we do an alternative  
[00:58:24] estimation where we uh just assume perfect compliance and there we get an estimate of the value of time that's much more in the order of typical estimates of around uh it being equal to the wage. Okay. So this is kind of the  
[00:58:38] headline result of that showing that selection bias matters uh and it can matter a lot for thinking about the value of time.  
[00:58:44] Um in the last minute I'm just going to show you the welfare effect. So you know there's there's kind of three bits of the results. First is thinking about how the private response matters on treated routes. So we find that the direct effect of public transit on treated  
[00:58:59] routes is around 22 cents per individual per day. People dislike the fact that private transit uh you know you have to wait for it more. They like the fact that it's cheaper. And so overall this actually means that you would overestimate the benefit of public  
[00:59:12] transit if you ignore the private response on treated routes. Right? You'd you'd overestimate it by 12% because people dislike these longer weights more than they like the fall in price. Okay, that's what's happening on treated routes. But on connected routes, prices  
[00:59:26] are also falling as the supply of drivers uh shifts out. And so even though this is a small gain, there's a lot of uh connect of of people on these commuted route on these connected routes. And so overall, if I can just have 30 seconds to to to finish this  
[00:59:40] off, overall uh we see that you would underestimate um the consumer surplus gains by around 10% if you ignore the response of the the private system. Okay, we can do the same for drivers. Uh and then you know  
[00:59:54] the punch line there is that we find that losses to drivers are around 50% of the gains to uh uh consumers. All right.  
[01:00:01] And that helps to rationalize something I think which is obvious which is that a lot of these uh drivers oppose uh competition from the private sector because they uh they lose a lot of money and we're just putting numbers on that in this paper. Okay. So I'm going to wrap up there. In this paper we measure  
[01:00:15] the response of private sector to investments in transit. We find that uh me uh uh uh accounting for the private response matters for measuring consumer surplus and we also put a number on losses for drivers and then some of these uh deeper questions were uh taking  
[01:00:29] on in follow on work but thanks very much for all your question to dinner and wants to go to dinner uh