10:11:41	 From  Daniel Sturm  to  Carl Beck (NBER conf. dept.)(Privately) : Carl, could you maybe mute those not currently speaking?
10:12:00	 From  Carl Beck (NBER conf. dept.)  to  matthew turner(Privately) : Matt, I accidently made you the host of the meeting.  Can you click on your name on the participant list and return the host duties to me?  Thanks, Carl
10:13:16	 From  matthew turner  to  Carl Beck (NBER conf. dept.)(Privately) : OK?
10:13:36	 From  Carl Beck (NBER conf. dept.)  to  matthew turner(Privately) : Many thanks!
10:26:25	 From  Jakub Kastl  to  Carl Beck (NBER conf. dept.)(Privately) : Carl, can you please also make Nick Buchholz a co-host? He will be presenting our joint paper in the afternoon. Thanks!
10:36:17	 From  Tanweer AKRAM : Is the paper available online?
10:37:40	 From  Adam Storeygard : Do costs include land takings and/or teardown in dense areas?
10:38:28	 From  Costas Arkolakis : We do not have a version yet online but we plan to incorporate the commuting model in our revision of the paper "On the Welfare Effects of Transportation Infrastructure"
10:39:00	 From  Mona Asudegi : what is the computational cost for solving this model? for example how long it takes to solve for Seattle network?
10:39:44	 From  Laura Grant : Ditto Adam’s question. Can you block/constrain certain links (eg. Downtown) where they are physically impossible (for now)?
10:40:15	 From  Costas Arkolakis : Response to Adam. The highest cost estimates category provided by the FHWA is a rough estimate that includes potential land taking or other major costs (Presumably). That was just included in the latest report
10:40:34	 From  Jeffrey Lin : In estimating the welfare benefits I’d like to see more attention to negative quality of life effects from pollution, etc. These welfare costs may be higher downtown as they affect more people.
10:41:28	 From  Clifford Winston : my hand is raised cliff
10:41:50	 From  Costas Arkolakis : Response to Mona: It is rather fast. About 5-6 hours (though we have not optimized fully the algorithm, probably will be much faster in the future). Seattle has 217 and 1384 bilateral links in our calibration
10:42:56	 From  Costas Arkolakis : Laura: can you clarify please? Do you mean whether you can exclude them for the analysis. 
10:43:33	 From  matthew turner : Is this model going to give us the old Mohring result that the optimal toll on a road will exactly pay for the land rent on the land used by the road?
10:43:41	 From  Costas Arkolakis : To Jeffrey: You are right these are important costs. They can be easily included by changing the population congestion parameter beta
10:44:29	 From  Laura Grant : Clarification: not exclude them, constrain them not to change.
10:45:44	 From  Mona Asudegi : What is the advantage of using such equ. models over more detail oriented  transportation mode choice, route choice, and network models?
10:45:49	 From  Costas Arkolakis : follow up on Laura: The exercise we do now we do the optimization link by link. So we can just ignore it. I think you are referring to a global optimization of trying to optimize all the links at the same time which we have not done yet (but we are considering)
10:46:02	 From  Laura Grant : Thank you
10:46:04	 From  Alejandro Molnar : Great work, thanks. Two related questions: 1. Starting from an equilibrium, does adding a circuitous link (e.g. one strictly dominated for all uses except up to logit error) strictly increase every agent's utilities, before residential choice is re-solved? 2. Does Braess paradox effect happen through residential location choice, or only on the traffic assignment part of the model?
10:47:36	 From  Costas Arkolakis : Follow up on Laura's question: This is related to a question that Treb answered about choice mode. That is not difficult to incorporate as it can be considered as discrete choice across mode (a direct extension of what we did in our 2014 Quarterly Journal of Economics paper that considered multiple routes)
10:48:03	 From  Costas Arkolakis : Sorry that was a follow up on Mona's question
10:48:33	 From  Mona Asudegi : thank you :)
10:49:21	 From  Fabian Eckert : Clarification: as theta —> inf, you said you get back Djikstra. Is this the optimal route taking into account the endogenous congestion, so that there would be different optimal routes for different people on the same route as a certain route gets congested? I am thinking of most people using google maps for route choice which takes into account congestion and recommends two people with same starting and end point different routes at times.
10:49:22	 From  Sari.Radin : In case it's helpful, here is a report on congestion pricing in Seattle: https://rosap.ntl.bts.gov/view/dot/12127.
10:50:24	 From  Costas Arkolakis : Answer to Alejandro's question. 1. That is a good experiment to check. From the Braess paradox cases we get there is a possibility this will reduce welfare, but we have not tried, and we should. 2. The Braess paradox comes from the interaction of the residential location choice, scale effects and traffic assignment. We should try to break it down
10:51:19	 From  Treb Allen : @Fabian: It's optimal cost given traffic. If there are two routes that have the same cost (due to traffic congestion arbitrage), the limiting case of Frechet smooths between the two. So it's as if two people from I to j take two different (but equally costly) routes
10:51:39	 From  Treb Allen : Thanks all for great questions - super helpful! =)
10:53:18	 From  Gopinath Munisamy : Is there a lot of inertia in commuting route choices? Since living and working are medium- to long-term decisions, does the commuting choice align with that time line? Sorry if I misunderstood the model! 
10:55:58	 From  Costas Arkolakis : The model is static so it does not take into account this dynamic nature. But the the analytical solutions we obtain can make it easier to incorporate such choice in a dynamic setup
10:57:22	 From  Costas Arkolakis : thank you for all the comments on my side too
11:00:55	 From  Tanweer AKRAM : The website has the Mehrota et al paper but the link for slides does not have the slides.
11:01:16	 From  Tanweer AKRAM : Could the slides be made available?  Thanks.
11:04:34	 From  Luis Quintero : Is construction and maintenance allocated to builders by a bidding process? Is the number of bidders (or unique builders building these) different across time?
11:04:41	 From  Stephen Redding : Presenters can also upload their papers and slides to the website: https://conference.nber.org/sched/ETf20
11:05:38	 From  neilmehrotra : The data we have does not specify details about the number of bids or competitive v. other bidding process.
11:05:44	 From  Zachary Liscow : Given the low share of the Interstate miles built after 1984 (vs. before), I’m surprised that such dramatic shifts (e.g., in share urban) are possible.  Am I missing something?  Or are those results for new miles?
11:07:07	 From  neilmehrotra : @Zachary - this reflects both increase in new lanes for existing interstate within urban areas and areas being redefined as urban.
11:08:24	 From  Juan Pablo Uribe : We use new lane miles (which includes new segments and new lines in existing segments). In the appendix we have a picture comparing new lane miles vs new miles.
11:10:40	 From  Zachary Liscow : @Neil & Juan Pablo - Thanks. I would have thought that reclassification as urban is something that you’d not want to include as a change, if there’s no new construction.
11:17:47	 From  Ed Glaeser : So I gather the price of asphalt is closely tied to the price of crude oil -- which seems to have tripled nominally since 1999 -- (or about doubled in real terms).   https://fred.stlouisfed.org/series/DCOILWTICO
11:18:35	 From  Ed Glaeser : Asphalt seems to have risen almost perfectly in parallel https://mdasphalt.org/asphalt-index/
11:18:45	 From  neilmehrotra : @Ed - yes there was a sharp rise in the price of asphalt from 2000 to 2008 that coincides with the sharp rise in oil prices.
11:19:23	 From  Clifford Winston : Yes, and highway authorities cannot easily substitute to different surfaces like concrete
11:19:33	 From  Andrew Yates : Environmental Regulations? https://aashtojournal.org/2019/07/26/report-highway-construction-costs-rising-due-to-environmental-regulations/
11:20:34	 From  Ed Glaeser : To Neil -- If you look coefficient of cost on asphalt prices (in some appropriately normalized fashion) is the coefficient one -- or do the contractors seems to raise prices more than one-for-one when asphalt prices go up. 
11:22:26	 From  neilmehrotra : Matt can correct me if I’m wrong, but I don’t think we have looked at the coefficient.
11:22:36	 From  neilmehrotra : But would be something natural to look at.
11:22:51	 From  Luis Quintero : Is the bidding process changing across time? Could larger market power of road builders have any role in explaining increasing costs (rents captured by the builders)?
11:23:06	 From  Matthew Kahn : Are the new roads built in Right to Work states where construction is cheaper?
11:23:57	 From  Stephen Redding : Hi Jim, I have a question if there is time. Steve
11:24:13	 From  Clifford Winston : Jim
11:24:42	 From  neilmehrotra : @Luis - need to investigate further in the case of new road construction but would likely need more disaggregate construction data.
11:25:02	 From  Clifford Winston : JIM my hand is raised for Matt et al.
11:25:12	 From  neilmehrotra : @Matthew - unionization does not explain the trend rise in either construction or unionization cost
11:25:13	 From  Zachary Liscow : @Matt Kahn - We look at right to work laws. They do not correlate with cost increases.
11:25:15	 From  Ed Glaeser : We've done this doing business index measures of public procurement globally (https://www.nber.org/papers/w27188.pdf), it would be nice to do something similar within states and look at whether prices have gone up faster for areas with poor procurement practices
11:25:42	 From  Luis Quintero : Thanks @neilmehrotra
11:26:28	 From  Jonathan Hall : My prior would have been that resurfacing costs had increased in part because we see more resurfacing work done at night. This raises the cost of the work, but lowers the total social cost. Any thoughts? Am I wrong that there is more resurfacing done at night?
11:26:30	 From  Luis Quintero : That is the kind of channel I had in mind @Ed Glaeser
11:29:17	 From  neilmehrotra : @Jonathan - could be the case and would be interested in any data you have in mind on this.
11:29:38	 From  Juan Pablo Uribe : @Andrew Yates. We looked into environmental laws as proximities to water, and it does  not seem to play a big role.
11:30:35	 From  Brad Humphreys : @jonathan & @Neil: We have detailed time stamped data from California on lane closures.  But we are currently having trouble identifying which lane closures are resurfacing projects.
11:31:04	 From  Zachary Liscow : @ Ed Glaeser - I spent a while trying to make progress on the relationship with procurement practices and just couldn’t find decent data.  I think that it’s a great idea —  I hope that someone else or I can work on it.
11:31:39	 From  Jonathan Hall : @Brad, cool! 

Related is the general idea of including the cost of the time spent resurfacing (or closing lanes due to building a new lane).
11:32:31	 From  Ed Glaeser : THANK YOU JIM!
11:32:48	 From  neilmehrotra : Thanks everyone for the helpful comments!
11:33:16	 From  Juan Pablo Uribe : Thanks everyone !
11:34:47	 From  matthew turner : Thank you all!
11:35:39	 From  Daniel Bergstresser : I'd be a little bit careful with the conversations going on at the moment - it's being broadcast over youtube
11:36:43	 From  Ed Glaeser : @Zach Liscow So the American Chamber of Commerce once did a domestic version of the Doing Business Report (about five or six years ago) -- it would be interesting to see whether we could find a domestic funder to follow up on the procurement surveys that we do x-country.
11:39:45	 From  Zachary Liscow : @ Ed Glaeser - Thanks.  That’s a great tip on the existing data (which I didn’t know of) and a great idea on producing a procurement survey.
11:52:32	 From  Ed Glaeser : HURRAH FOR FLINT-- so rarely does it get positive press

11:52:53	 From  Donald Davis : Not sure this is positive press.
11:52:57	 From  Ed Glaeser : Unfortunately, I suspect that fast speeds are a symptom of urban decline and having built an excess of roads relative to current population. 
11:54:26	 From  Gilles Duranton : yes - we agree with that. A city that had an infrastructure developed for a much bigger population
11:55:49	 From  Donald Davis : Speed in the middle of the night is not necessarily maximum feasible uncongested speed in the day.
11:56:22	 From  Luis Quintero : Is the no traffic measure reported by google the speed at night or a theoretical measure assuming road speeds? These two could imply very different things for different cities with different levels of nightlife.
11:57:34	 From  Adam Storeygard : In practice we use google’s definition of uncongested. Alternatively (in India) we pick the fastest speed across all instances of the same trip (origin-destination pair), and get very similar results. We will check this for world sample in the future
11:58:43	 From  Adam Storeygard : Google can use all of its data to estimate the uncongested speed so we think they can do a good job.
11:58:55	 From  Adam Storeygard : “middle of the night” is just a useful shorthand
11:59:11	 From  Ed Glaeser : Is the data public already?  Can we start running mindless cross-city regressions with it?   
11:59:20	 From  michael ostrovsky : I may have missed this - did you get this data by collaborating with Google, or by scrapping Google Maps, or by using their API?
11:59:56	 From  matthew turner : It would be nice to check if the congestion index moves in cities with congestion tolling.  I think there are enough of these cities now that you could do it in the cross-section.
12:00:07	 From  Gilles Duranton : @Ed — we want to have first pass on running mindless regressions. that’s a rare pleasure in life ...
12:00:24	 From  Gilles Duranton : but we’ll make things available after publication.
12:01:23	 From  Treb Allen : thanks victor! I didn't know about this amazing data! <<processing future collaborations.. =) >>
12:01:56	 From  Gilles Duranton : @michael - we scrap google maps. the api would be too expensive...
12:02:02	 From  Adam Storeygard : @Matt, this is a nice idea, thanks.
12:03:04	 From  Luis Quintero : Road building takes time. Many of the congested cities you seem to have found are, anecdotally, places that have recently received significant rural migration (including Bogotá). It would be interesting to see the elasticity to recent population growth.
12:03:35	 From  Gilles Duranton : @Treb and others. We also duplicate the US NHTS and restarted our data collection for the whole world early March.
12:03:51	 From  Treb Allen : Super cool!
12:04:30	 From  michael ostrovsky : @Gilles - in that case, when you share your data after publication, you may also want to share your code as well, possibly with some sort of open-source license (to be able to modify it as Google Maps changes things over time).
12:05:05	 From  Gilles Duranton : @ Treb — for 10 cities in Asia we also collect the full matrix for 1km pixels.
12:05:07	 From  Ed Glaeser : What's the correlation with the tom tom numbers?  
12:05:12	 From  Adam Storeygard : @Luis we would like to try this. In India if anything faster growing cities were faster despite more congestion.
12:05:44	 From  Gilles Duranton : @ Ed - about 0.8 from the top of my head. but it’s a very preliminary number.
12:06:34	 From  Gilles Duranton : @Michael. A complication is that Google Maps will force us to adapt our code once or twice a year.
12:07:05	 From  michael ostrovsky : @Gilles - yes, I understand, and that’s the beauty of open-sourcing it, as this adaptation may become a collaborative effort
12:07:25	 From  michael ostrovsky : Much easier for ambitious PhD students to adapt your code than to re-do it from scratch
12:07:50	 From  Luis Quintero : @Adam interesting. I would have thought that the non monotonicity between actual congestion and income (and the higher numbers in middle income countries), would be explained by higher rural ro urban migration. That would be a great number to have. An important factor hampering agglomeration economies from being higher in developing countries urbanizing cities I believe.
12:09:31	 From  Gilles Duranton : @Michael -maybe a conversation to have off line. A risk is that if there is too much of this, Google may turn the screw much further...
12:09:31	 From  Ed Glaeser : I suspect flint is going to look less good on accessibility.
12:10:05	 From  michael ostrovsky : Fair point
12:10:22	 From  Gilles Duranton : @Ed - that would be my prior as well.
12:11:48	 From  Adam Storeygard : @Luis agreed good reasons to look at correlations with growth, though they will be hard to interpret.
12:12:16	 From  Anjana Susarla : Sorry I joined this talk late. I wonder if adding data from Uber Mobility will help
12:12:21	 From  Ed Glaeser : The slow speeds on developing world roads goes well with a central fact from Gabriel Kreindler's job market paper:  congestion pricing has only modest impacts on speeds in Bangalore.  The reason for that is that the free flow speeds in bangalore are low. 
12:14:59	 From  michael ostrovsky : A suggestion: you may want to continue collecting this data at very regular intervals (e.g., once a quarter)
12:15:41	 From  Adam Storeygard : @Anjana We have looked at the Uber data for the 5 cities it is available for in India. What they report is very hard to interpret for reasons I can get into offline, but they are definitely not “trips”. having said that, under some reasonable assumptions about the effects of that, their data are consistent with ours
12:16:15	 From  Anjana Susarla : Thank you, Adam! Interesting to know
12:37:56	 From  Gilles Duranton : just a detail Caitlin — reporting the size of the datasets would be useful.
12:39:31	 From  Mona Asudegi : Would be great If adding data sources for other modes such as Micromobilty.
12:39:39	 From  Tanweer AKRAM : It would be good to have non-US data as well.
12:39:48	 From  Tanweer AKRAM : thank you for the great effort
12:40:11	 From  Jonathan Dingel : Small detail suggestion for browsing the website would be to post metadata + codebook distinct from downloading the huge datasets would be valuable.
12:40:16	 From  Andrew Waxman : Any possibility of incorporating GTFS shapefuls for public transport?
12:40:24	 From  Chris Severen : How will you bridge potentially confidential data?
12:40:40	 From  Chris Severen : Upvote J Dingel!!!
12:44:17	 From  SimiPhone : Have you worked with the BTS National Transportation Library (NTL) and the DOT DATAHUB
12:44:34	 From  Treb Allen : Love the idea of collecting historical CFS data, fwiw
12:46:10	 From  Jonathan Dingel : @Cliff:  For shipment-level public-use data, see 2017 Commodity Flow Survey (CFS) Public Use File (PUF): https://www2.census.gov/programs-surveys/cfs/datasets/2017/cfs_2017_puf_users_guide.pdf. The 20 variables include metro-area origin + dest, value + weight, and mode of transport.
12:46:53	 From  Tanweer AKRAM : Marx also wrote about time and money well before Becker
12:47:16	 From  Caitlin Gorback : @ SimiPhone, we haven’t worked with those directly yet, instead looking for specific dataset links. Going forward, we can definitely highlight the resources at each
12:48:54	 From  Clifford Winston : Thanks, does the CFS data have firm level information on inventories, commodities, sales? 
12:49:17	 From  Ed Glaeser : Yes -- and Benjamin Franklin was emphasizing the time and money link well before Marx.
12:50:55	 From  Tobias Salz : In previous iterations we had a Franklin reference :)
12:51:13	 From  Brad Humphreys : Has the taxi market in Prague improved?  When I lived there 20 years ago, getting in a cab was effectively asking to be ripped off.
12:51:14	 From  Ed Glaeser :  "He that can earn ten shillings a day by his labour, and goes abroad, or sits idle one half of that day, though he spends but sixpence during his diversion or idleness, it ought not to be reckoned the only expence; he hath really spent or thrown away five shillings besides."  Franklin 1748
12:51:15	 From  Jonathan Dingel : @Cliff: I’ve only used the confidential CFS, so I’m not 100% sure how much you can aggregate up shipments to the establishment level to estimate total value of shipments. [I very much doubt public file has firm identifiers; it’ll only be establishment level.]
12:52:04	 From  Tobias Salz : To Brad’s questions: this is one reason why someone might use the platform
12:53:40	 From  Jonathan Edward Hughes : Regarding Jonathan and Cliff’s discussion,  while the CFS PUM reports shipment level data, you don’t observe firm or establishment.  The finest geographic detail is at something like the MSA level (though differently defined)
12:53:41	 From  Brad Humphreys : Yes, I can see that.
12:56:21	 From  Clifford Winston : if anyone has a source of firm level inventory/logistics data, please let me know. 
12:57:34	 From  michael ostrovsky : If I changed this graph to plot the fraction of times when the bid within, say, 10% of the lowest one was chosen, how close to 100% would I get (at various times)?
12:59:29	 From  Tobias Salz : Good question, we haven’t done this computation, but we could easily augment the graphs for that
13:00:14	 From  Andrew Waxman : Why would we assume that origin or destination values are location specific as opposed to individual specific?
13:00:36	 From  Brad Humphreys : This is similar to the Bento et al paper on the value of urgency, correct?
13:00:54	 From  Victor Couture : This estimates the time spent waiting for a car - so possibly time spent standing on the side walk - which may be more costly than time spent in the relative comfort of a car, which is the relevant value of time for road investment that improve travel speed. Is there any evidence in the paper that those two may be the same (time on side-walk = time in car)? Or different (if wait time more costly in rain, say)?
13:00:54	 From  Tobias Salz : Well, we estimate the full combination location x time of day x person
13:02:54	 From  Tobias Salz : To Victor: in the paper we are clear about the fact that we are estimating the opportunity cost, which is time spend in origin/destination
13:03:25	 From  Tobias Salz : In our setting time in the car is held constant. We therefore have no estimate of the value of time in the car itself
13:05:57	 From  matthew turner : The calculation just described looks a lot like an individual's problem in the bottleneck model. In that model, equilibrium requires that everyone get the same payoff.  Should there not be some similar sort of equilibrium in this problem as well?  If so, does this affect the interpretation of your results?
13:06:15	 From  Tobias Salz : As Nick just pointed, we have a lot of context for our estimates
13:06:41	 From  Tobias Salz : So we can estimate conditional distributions for weather, etc.
13:09:17	 From  Tobias Salz : To Matt’s point: for now we are just measuring, essentially the via revealed preferences. The valuations could be used as inputs to an equilibrium
13:09:34	 From  Tobias Salz : Would be great to follow up on this later, not exactly sure I understand
13:12:42	 From  michael ostrovsky : @Matt - not sure what you mean by “equilibrium requires that everyone get the same payoff” in the bottleneck model. In Vickrey’s 1969 model, people with extreme ideal arrival times (8am and 9am) get higher payoffs than those with ideal arrival times in the middle.
13:14:27	 From  michael ostrovsky : Also, in the bottleneck model, time spent in traffic serves sort of as an equilibrating force. As Tobias said, that force is not present in the current paper.
13:15:27	 From  Laura Doval : @matt our setting differs from the bottleneck model in that the agents have uncertainty of what prices and wait times they will be offered when they make their scheduling choices. This means that even if prior to seeing the price and waiting time draws everyone gets the same expected payoff, conditional on a draw of price and waiting time, they will still express preferences between these two. We are measuring payoffs at this interim level not at the level that the bottleneck model would equalize. Happy to follow up on this later too :)
13:25:17	 From  matthew turner : More precisely: Suppose we did your calculation of willingness to pay for agents who were playing the bottleneck game with symmetric agents. Would your calculation recover the utility parameters correctly?  What about a model with asymmetric agents?
13:28:59	 From  Laura Doval : @matt we show how to identify the parameters of the model when there’s endogenous choice of scheduling, but we have done it in the single agent problem
13:29:44	 From  Tony Choi : 4:
13:37:25	 From  Adam Storeygard : Perhaps relevant (Gertler, Gonzalez-Navarro, Gracner and Rothenberg on Indonesia): https://are.berkeley.edu/sites/are.berkeley.edu/files/user/profile2/main/publications/Indonesia_Roads.pdf
13:40:27	 From  Margaret Bock : @Adam: thanks! will definitely take a look at this 
13:44:46	 From  Ed Glaeser : There is also the endogenous selection of drivers onto rough roads problem-- so people with suvs may be happier driving in rough areas
13:47:29	 From  Alexander Cardazzi : @Ed - this is true.  However, we are able to differentiate truck volume and car volume.  Unfortunately, we cannot identify SUVs, but can identify larger vehicles like tractor trailers
13:49:09	 From  matthew turner : Would it make sense to normalize by number of lanes?  
13:49:23	 From  Ed Glaeser : Nice.   And more generally -- I am very enthusiastic about this -- we (Kreindler, Currier and myself) also have dspeed/droughness estimates from our uber data-- but they are just for Chicago right now-- so it would be good to compare.  
13:51:06	 From  Laura Grant : I might have missed it, what is IRI coefficient interpretation, meaningfully?
13:51:53	 From  Alexander Cardazzi : @Matt this is something we were doing earlier this week, but not in the presentation just yet.
13:52:54	 From  matthew turner : Maybe controlling for design speed, or using design speed as an instrument?  I think this is reported in HPMS.
13:55:01	 From  Alexander Cardazzi : @Laura - IRI is measured in inches of vertical displacement as a car drives over the road, increasing this by one increases Accidents/100K VMT by .001 (for naïve OLS)
13:55:02	 From  Ed Glaeser : What's the r-squared of ground water on elevation for the raw data (not the bins)
13:55:40	 From  Ed Glaeser : I have no intuition about IRI measures other than z-scores -- but we can tell you how much IRI goes up after a repaving event if that is helpful
13:56:42	 From  Ed Glaeser : Might elevation (or changes in elevation) also have a direct impact on accidents or road speed
13:58:54	 From  Margaret Bock : @Matt: thanks for the suggestion, we will look into the design speed variable
13:59:44	 From  Laura Grant : some frame of reference for IRI (and its cost to improve) would be useful context.
14:00:09	 From  Alexander Cardazzi : @Ed: I do not remember the r-squared for the raw data off the top of my head. We are still trying to connect the repaving/construction data, this is a next step for us.
14:02:03	 From  Darren Timothy : Note that CA has relatively few US numbered routes, but has many former US highways that now carry a state route designation (SR 99, SR 60, etc.).
14:02:14	 From  Chris Severen : PPML estimators can accommodate high-dimensioned fixed effects (state <ppmlhdfe>). My experience has been that 0s matter a whole lot, especially in rate settings like this with rare outcomes <-> lots of heteroskedasticity.
14:02:23	 From  Alexander Cardazzi : @Laura: we are hoping to be able to do this once we can connect the construction data (via bidding)
14:02:29	 From  Juan Pablo Uribe : Can you see if what matters is IRI or changes in IRI along a segment? what I have in mind is that what matters is if a good road gets bad suddenntly and that’s where accidents happened.
14:03:29	 From  TFitzgerald : Have you thought about how elevation measure correlates with weather outcomes rather than groundwater depth? You could match roads to NWS precip data as an example. Winter weather is very hard on roads and surely contributes to pavement roughness. My prior is that this may also have interesting implications for the accident data.
14:04:21	 From  Alexander Cardazzi : @Juan Pablo: We can look at SD of IRI along the road.  We also have a variable that tells us what % of the segment is >170 IRI.  I think this might be able to address your question
14:04:32	 From  Elaine Buckberg : I agree with Cliff about the value of this paper--but perhaps for somewhat different reasons.  The paper is very valuable for making the case for investing in road infrastructure.
14:05:12	 From  Juan Pablo Uribe : @Alexander yes that’s exactly what I was thinking
14:07:49	 From  Alexander Cardazzi : Thank you everyone for comments!  I apologize if I missed your question/comment!
14:08:12	 From  Margaret Bock : Thanks for the feedback everyone!
14:10:07	 From  Jonathan Hall  to  Carl Beck (NBER conf. dept.)(Privately) : Any objection to me starting to share my screen?
14:10:18	 From  Brad Humphreys : @chris Thanks.  That's a very helpful point.
14:11:48	 From  Carl Beck (NBER conf. dept.)  to  Jonathan Hall(Privately) : Jonathan, can you wait just a couple of minutes?  YouTube is running.
14:12:10	 From  Jonathan Hall  to  Carl Beck (NBER conf. dept.)(Privately) : Totally fine
14:12:37	 From  Carl Beck (NBER conf. dept.)  to  Jonathan Hall(Privately) : I'll pull my slide down soon, then it is all yours
14:14:24	 From  Laura Grant : I am looking for the link to the datasets Caitlin spoke about, is it live yet or just beta?
14:14:42	 From  Caitlin Gorback : All beta so far!
14:14:49	 From  Caitlin Gorback : We won’t go live until the new NBER website launches
14:15:05	 From  Laura Grant : k, hope to get that announcement!
14:27:13	 From  Tanweer AKRAM : Have authorities responded to the findings of this research?
14:27:13	 From  Christy Zhou : Should we think the placement of the sign and the timing when the sign is turned on are as good as random? 
14:28:33	 From  Joshua M Madsen : We've emailed our findings to all states that ever used this message system, with no response. TxDOT in particular was very helpful in getting data for the project, but has been unresponsive since we wrote the paper. 
14:28:49	 From  Matthew Kahn : How does the % of drunk drivers on the road affect the "treatment effect"?  Was there a change in penalties for drunk driving during the sample period?
14:30:29	 From  Ed Glaeser : Jonathan has such great data on crushes-- it would be nice to link this with repaving events if Texas has data on that -- 
14:30:32	 From  Victor Couture : Is there perhaps a general result that distracting boards - e.g., adds - cause car crashes? Or is there something special about being told that you may crash that cause people to freak out and crash?
14:31:06	 From  Ed Glaeser : @Matthew -- I'm pretty sure that nothing else changes on the treatment week other than the messages -- but the authors should clarify 
14:31:09	 From  Joshua M Madsen : @Christy great question. The week  when it's turned on (week prior to board meeting) can be thought of as exogenous. We have data for a sample of DMS on actual messages shown. We don't use this as the main explanatory variable as there certainly can be endogeneity as to when during that week the message shows. TxDOT instructs engineers to show the message as much as possible during the week, but it still is only shown about 40% of available hours. For this reason, we use a IV approach to estimate the effect of showing the fatality message. 
14:33:07	 From  Joshua M Madsen : @Matthew - we are not aware of any changes for drunk driving penalties or other policy changes in connection with this program. We found no effect on crashes associated with a DUI, possibly because such events are fairly rare in our sample (as a proportion of all crashes). 
14:34:36	 From  Joshua M Madsen : @Victor there certainly are concerns about even a distracting bill board. Our evidence suggest that it is more than just showing any message. The signs are used during the "off" week, so this is not just turning on a sign that isn't use otherwise. Furthermore, we find the effects are increasing in the fatality number on the sign, which Jonathan is discussing now. 
14:34:53	 From  Matthew Kahn : My point is that drunk drivers will be more distracted by the sign.  The rise of automated vehicles will mitigate this.
14:36:09	 From  Ed Glaeser : @Mattew -- that's good -- should be different by time of day then -- although that might be driven by other light related issues as well.  
14:36:42	 From  Ed Glaeser : @Matthew -- I guess there may be other ways of estimating the preponderance of drunk drivers -- proximity to bars, etc.
14:36:54	 From  Joshua M Madsen : @Matthew - I guess this will also be affected by the likelihood a drunk driver notices signs while driving under the influence. I'm not aware of research on this, but your point is certainly worth considering. 
14:38:30	 From  Joshua M Madsen : Our segment fixed effects will take into account any unique characteristics of the segment. However, this could be a possible use of examining differential effects, similar to what Jonathan is discussing now about road characteristics. 
14:38:44	 From  Alexander Cardazzi : Can you use the other half of the highway as a counterfactual as well as upstream/downstream?  If the sign faces one way, the other side of the highway does not get treated.
14:39:01	 From  Alexander Cardazzi : By other half, I mean NB-SB or EB-WB
14:39:01	 From  Matthew Kahn : http://chronicle.uchicago.edu/020110/drunkdriving.shtml
14:39:06	 From  Matthew Kahn : Levitt and Porter
14:40:04	 From  Joshua M Madsen : @Alexander - thank you for the suggestion. We have not yet done anything with the opposite direction. We can certainly look into this as a further falsification test.
14:40:29	 From  Matthew Kahn : “The peak hours for drinking and driving are between 1 a.m. and 3 a.m., when as many as 25 percent of drivers are estimated to have been drinking,” Levitt said. During those hours, about 60 percent of the fatal crashes are caused by drivers who have been drinking, the research shows.
14:42:57	 From  Ed Glaeser : @Matt-- so that's great-- he needs to just isolate 1-3 am from other time periods.  
14:43:48	 From  Joshua M Madsen : @Matthew Thank you for sharing! We have looked at differential effects by time of day, and found generally stronger effects during rush hour and less effects during the night. Our research design also controls explicitly for time of day differences--so we would need an increase in drunk driving on a specific segment during the Tuesday 2am hour during the third week of the month that isn't occurring during the other 3 Tuesday 2am hours in the same month.  
14:43:55	 From  Ed Glaeser : @Not Matt -- It is not that there is some general optimal salience -- it is that salience is dangerous if it distracts you from doing a high risk avcitivity.  
14:44:29	 From  Ed Glaeser : @Matt Given Josh's response-- it maybe that alcohol makes you less likely to pay attention to these signs -- 
14:45:38	 From  Joshua M Madsen : @ Ed that's my conjecture. I would love to see research on how effective messages are when delivered to a drunk driver. 
14:46:31	 From  Matthew Kahn : Building on Cliff's question, what would be evidence of learning by the authority? 
14:47:10	 From  Sari.Radin : Is there human factors literature running experiments with different messages or different types of signs that address similar issues?
14:48:04	 From  Matthew Kahn : can't talk.
14:48:13	 From  Matthew Kahn : UCLA faculty meeting going on and dora is chairing it
14:52:25	 From  Joshua M Madsen : Thank you everyone!
15:09:27	 From  Ed Glaeser : If it weren't for the fact that 300 different things were going on at the same time, reintroducing the fare would be a nice natural experiment to get the price elasticity.  
15:17:04	 From  Matthew Kahn : Will the public sector union negotiations get tougher to slow pay increases?  Who bears the incidence of the COVID shock?
15:17:19	 From  Chris Severen : @Ed lucas davis has new wp doing that (fare change elasticity) in mexico pre-Covid
15:18:27	 From  Clifford Winston : there are many issues before voters to provide additional funding for transit
15:26:29	 From  matthew turner : 1. One of the big messages of the empirical literature on transit is that people readjust their locations and other behavior in response to transit.  This suggests picking routes where you want to provide service, and letting people adjust their locations. Have you considered this? 2. What about raising fares?
15:51:57	 From  Victor Couture : A key result of Kreindler's paper in Bangalore is that the "congestion function" is linear, i.e. there is a linear relationship between change in travel speed and change traffic volume over the city. If this result is externally valid it would help predict how reduction in traffic flows affect congestion.
15:58:29	 From  matthew turner : We are in the middle of a big, and probably partly transitory labor market readjustment: from Starbucks to Amazon.  Getting people back to work seems important relative transit budgets. Doesn't this suggest that we want just subsidize transit and let it run at loss until the MUCH more important labor market is settled, before we worry about changes to transit service?
15:59:06	 From  Steve Poftak - PPT : Many thanks for having me.  I have to move on to yet another meeting.  I appreciated the questions and everyone's engagement in transit.  Look forward to hosting all of you (especially Cliff!) on the MBTA at some point in the future.