Weathering the Ride: Experimental Evidence on Transport Pricing, Climate Extremes, and Future Travel Demand

Abstract

The future of travel will be characterized by changes in weather patterns and changes in transportation technology. How will these forces interact? We explore this question by utilizing a unique randomized experiment with Uber riders in Cairo, Egypt. We consider how very hot days ($>$35\degree C/95\degree F) affect transportation choices, how a sizeable price decrease (simulating a future with autonomous vehicles and access to cheaper transportation) changes travel, and how the interaction of these two elements affect choices. We find that while travel will increase significantly in response to the price decrease, extreme weather dampens this effect by 21%. Individuals receiving subsidies also shift away from public transportation modes and towards private transportation modes, except when the public transit option is air-conditioned. These results provide important insights for policymakers when considering optimal travel policy in the face of climate change.

Introduction

Understanding how people respond to extreme weather is important for policymakers, academics, and businesses. In transportation markets, responses to weather shocks will depend on the cost, comfort, and other characteristics of travel options available to consumers for a given trip. The literature has shown that extreme weather events will become more common in the future (alimonti_mariani_prodi_ricci_2022; box_2022; swenson_2023) and that private transport could become substantially less costly as technologies such as autonomous vehicles mature (bagloee_tavana_asadi_oliver_2016; BOSCH201876; reid_2021). How might these two forces interact? What will the effects on overall travel and on substitution across transport modes be?

We combine data from a unique randomized experiment (Christensen and Osman (2021)) with detailed data on extreme weather events (days where the high temperature is greater than 95\degree F\slash 35\degree C) to examine how transport decisions respond to exogenous variation in weather in the context of reductions in the cost of personalized transport. Using 1,373 Uber riders in Egypt, we randomize individuals into two groups who are provided a 25% or 50% discount, respectively, on their Uber rides for 3 months and a group that serves as control and receives no discount.^[1] We collect data on Uber utilization and overall mobility using Google’s Timeline feature on participant smartphones and regular follow up surveys that provide information about the other modes of transportation taken during travel.

First, we find that increases in mobility resulting from price reductions are dampened by extreme heat. While new technologies that decrease the price of private transport will increase overall travel, our findings suggest that increases in the frequency of hot days in certain regions could greatly mitigate the effect on travel. This decrease in travel is primarily driven by adjustments during the hottest part of the day (from 11am-6pm). We find that travel in the control group does not decrease on hot days. This indicates that riders forgo trips that they only considered when the price of personalized travel had gone down. Hence, while cheaper travel leads to more trips, hotter weather will make those trips less enticing.

Second, we find evidence of a pronounced effect on substitution away from public transport (which does not have air conditioning in Egypt) to Uber in the context of fare reductions. With lower prices on private modes, high ambient temperatures result in disproportionately larger reductions in travel on public transport modes (relative to private transport). This mode substitution effect increases the emissions intensity of a kilometer of travel, creating a potentially important positive feedback between temperature and greenhouse gas emissions that could be mitigated through efforts to ensure that public transport modes are properly temperature-controlled.

We use these estimates to simulate what could happen to travel demand in several large developing country cities with similar numbers of hot days at baseline. Using forecasts of the growth in the frequency of hot days from current climate models, our estimates indicate that travel demand could increase more in cities such as Dehli, India, where extreme temperature projections are more mild. The behavioral effects of technology-induced price changes could be dampened in cities such as Bangkok, Thailand, which are subject to higher projected growth in the fraction of days that are extremely hot. While long-run adaptation related to incremental changes in temperature could directly affect transportation markets and produce general equilibrium effects that lead to impacts that differ from the specific magnitudes estimated from our short-run experiment (Bento et al. (2023); Carleton et al. (2022); Hsiang (2016); Kahn (2016)), responses captured in the present setting provide information about the behavioral mechanisms that underlie public/private transportation choices and potential emissions feedback in the context of changes in temperature and price.

We contribute to the literature at the intersection of transport, environmental, and urban economics. A large set of studies have shown that climate affects transport behavior, but they rely on observational data and typically focus on a single mode of transportation (Calvert and Snelder (2016); Hymel (2009); Leard and Roth (2015); Lin et al. (2020); Singhal et al. (2014)). While the economics literature has shown that extreme weather can reduce the utilization of public transport, it is unclear how extreme weather affects substitution toward private modes. We contribute to this literature in three ways. First, we provide experimentally-identified estimates of the behavioral impacts of reductions in the price of transport in the context of different temperature regimes. Second, we use detailed trip-level data on travel and hourly temperatures to examine how the impacts differ during the hottest parts of the day. With these data, we show that travel is not simply shifting to cooler parts of the day, but is decreasing overall. Third, by collecting detailed data on total mobility, we are able to estimate the effects of the temperature-price interaction to examine shifts to the overall proportion of travel taken by transport mode. While both public and private transport use decreases on hot days, public transport use decreases significantly more.

This study also contributes to the literature on the impacts of climate on decision making and beliefs. For example, (Espanol (2022)) and (rose_dolega_2021) both look at the impact of weather on retail sales. Similarly, Colelli et al. (2023) and rode2021 investigate the impacts of climate on energy use via air conditioning decisions. There is a wide literature on the impacts of weather and climate on investment decisions and consumer beliefs (Anderson et al. (2013); Cao and Wei (2005); Makridis and Schloetzer (2023)), in addition to evidence that changes in climate impact global economic growth (Burke2015; Dell2012).

While projections indicate that the cost of urban transport and the frequency of hot days could both shift in the latter half of the 21st century, empirical work to date has fueled a limited discussion about the interaction between these important forces. Our results illustrate that extreme weather will mitigate the impacts of technological advances in transport and that this effect will differ across space and time. This is important for policymakers to consider as they anticipate changes in transportation technology and a changing climate. Our results suggest that the impacts of autonomous vehicles may be attenuated in locations that are more exposed to extreme weather and that substitution from public to private vehicles will depend on existing infrastructure, including whether public transport options are temperature controlled (i.e. have air conditioning). Together, these results suggest important impacts on emissions, congestion, and revenues from public transit services.

Background: Transportation & Weather in Cairo

\par Cairo, the capital city of Egypt, is one of the largest cities in the world. The greater Cairo metropolitan area contains 20 million residents (Hudec (2023)). Traffic congestion in Cairo is high due to the large volume of traffic and the relative lack of infrastructure. This congestion greatly reduces travel speed and increases travel time, leading to an estimated loss of approximately 4% of Egypt’s GDP (Group (2014)).

\par As in many major cities, public transport in Cairo is a major contributor to mobility. Cairo has several forms of public transport, including public bus services, private shuttle bus services, and metro lines (News (2023)). The bus system in Cairo serves more than 350 million passengers per day and is quite affordable for riders. Cairo does not have a public bus map, though a recent grassroots movement has attempted to create a map of bus routes in the city (Cairo (2022)). The metro system consists of nearly 100 km of tracks across three lines with 50 stations. Daily metro ridership is around 1.5 million, and trains run every 2-5 minutes, depending on the time of day (Metro (2023); Hudec (2023)). Notably, both the public buses and metro lack air-conditioning, making them very hot on high-temperature days. On the other hand, private shuttle bus services, such as Swvl, have air conditioning. This is one way that temperature could impact individuals’ transportation decisions in Cairo.

Cairo also has options for private, single-occupancy transportation. Private cars are plentiful and traditional taxis are seemingly ubiquitous. Uber has been present in Cairo since 2014 and has held a large market share in ride-hailing from 2014 until the present (espanol_2022).

The climate in Cairo is warm and dry year-round. fig:AppendixFig1 illustrates daily maximum temperatures from 2019-2020, as recorded by Weather Underground (Underground (2023)). The maximum temperatures during this period reached a low of around 60\degree F during the winter months and climbed to between 95-100\degree F during the summer months, with a handful of days surpassing 100\degree F. The average high temperature in Cairo across this time period was 82.7\degree F. Due to the lack of variation in humidity or precipitation in Cairo, variation in high temperatures serves as the primary climatic determinant of comfort in travel decisions.

Research Design & Data

We partnered with Uber to implement a randomized experiment that provided prolonged price subsidies to a sample of riders. Riders were sent a text message inviting them to join a “study on mobility behavior,” and those who opted in were asked to answer our surveys (described below), to provide access to their activities on the Uber platform, and to turn on their Google Maps timeline, which provided us with detailed data on total daily travel (including, but not limited to, Uber trips).

Individuals who enrolled in the sample were randomized into three groups: (1) a group that received a 50% subsidy for 3 months, (2) a group that received a 25% subsidy for 3 months, and (3) a control group that did not receive any price subsidies. Our dataset combines this random variation with the exogenous variation in extreme weather^[2], which we define as days where the maximum temperature reaches 35\degree C/95\degree F.^[3]

Survey Data

\par Baseline and follow-up surveys were conducted throughout the period of study and provide information about the rider and the totality of their transport behavior, on and off Uber.^[4] In addition to asking about the number of trips taken on each mode of transport on the day prior to the survey, we also collect information on the total distance traveled (as recorded in their Google Timeline) over the three days prior to the survey. We use the date that the survey was completed to match temperature data with the survey outcomes for each individual.

fig:AppendixFig1 shows the days for which we have Uber data (in Panel (a)) and survey data (in Panel (b)), with kernel density plots for the number of observations on across the sample period.

Uber Admin Data

\par Administrative data from Uber are available for all individuals at the trip level. These data include time (rounded to the nearest hour), location (rounded to 4 digit latitude\slash longitude), distance traveled, and fare information. We match these data with the daily temperature data from Cairo.

Temperature Data

\par Temperature data come from Weather Underground, which reports data from the Cairo International Airport Station. The data include the daily minimum and maximum temperature, precipitation, and wind speed. We define a hot day as one that has a maximum temperature of at least 95\degree F (35\degree C), though we also include a cutoff of 90\degree F as a robustness check for our results. fig:AppendixFig1 shows the maximum temperatures in Cairo each day in 2019 and 2020, covering the time period when the study took place. The red dashed line represents 95\degree F, so days with maximum temperatures at or above that line are defined as hot days for the purposes of the analysis. \par Weather data are also available at the hourly level for each day in 2019 and 2020. The hourly-level data are used to determine which times of the day are generally the hottest in Cairo. Average hourly temperatures by month are plotted in fig:AppendixFig2, which shows that the hottest time of day is generally between 11am and 6pm, and this appears to be consistent throughout the year.

Methodology

Our empirical strategy utilizes two sources of exogenous variation. The first is the explicit random assignment of riders into treatment and control groups. In order to increase power, we combine individuals in the two treatment arms who received the 25% and 50% subsidies into a single treated group.^[5] tab:BLBalance shows that the randomization was successful and the treatment and control groups are balanced on baseline characteristics. The second source of exogenous variation comes from daily temperature, which is generally accepted as being plausibly exogenous (Schultz2019).^[6]

We investigate the impact of hot days and access to a subsidy on several outcomes related to overall travel and mode substitution. Using the Uber data, we estimate the effect on the number of trips and the distance traveled, both for the entire day and at certain times of day.^[7] Using data from the survey, we estimate the effect on the total number of trips per day, trips per day by mode, total distance traveled, and the number and proportion of trips taken on public and private modes of transportation.^[8] Our treatment variable is defined as an indicator for having received any subsidy towards the price of Uber travel.^[9] Hence, we estimate the following equation, which gives us the impact of experiencing a hot day, of receiving any subsidy, and of the interaction between the two:

where $Y_{id}$ is the outcome of interest (e.g. number of trips taken or distance traveled), $HotDay_d$ is an indicator for whether a given day was at least 95\degree F (35\degree C), $Subsidy_i$ is an indicator for whether an individual receives any subsidy (either 25% or 50% of the price of an Uber ride), $\delta_m$ is a month fixed effect, $\delta_c$ is a cohort fixed effect, $\delta_f$ is a survey round fixed effect (for the regressions using survey data only), and $\delta_d$ is a day-of-the-week fixed effect.^[10] We use two-way clustering to calculate our standard errors at the individual and day levels (Cameron et al. (2011)).

Results

Effects on Overall Travel

We begin by looking at the changes in overall trips taken and distance traveled on Uber. tab:Table2 reports estimates of the effect of a hot day on the number of Uber trips and distance traveled on Uber. We estimate both aggregate effects and intra-day impacts, which we generate by splitting the day into 3 time periods: early morning (3am-10am), midday (11am-6pm) and evening (7pm-2am). Intra-day variation in temperature is substantial in Cairo. fig:AppendixFig2 shows that the temperature is usually relatively mild in Cairo until around 11am and again after 6pm. We use the detailed time values in our data to assign each Uber trip taken to one of these times of the day.

tab:Table2 reports the results, showing that travel on Uber seems to decrease during hotter times of the day for individuals who receive the subsidy. Columns 1 and 2 show the impacts for trips taken and distance traveled throughout the entire day. Receiving a subsidy significantly increases the number of trips by 0.423 and also significantly increases the distance traveled. Both interaction terms have negative but insignificant coefficients of -0.042 and -0.067, respectively. This suggests that heat may attenuate the responses to discounts on ride-hailing services. When we estimate the effects separately for each time period, we find that the effect of temperature on the Uber price elasticity is driven by reduction during the midday hours (11am-6pm), when Uber trips and distance traveled decline by 0.040 and 0.082, respectively.

There is little evidence that temperature has any effect on travel choices on Uber during the cooler parts of the day, as shown in columns 3-4 and columns 7-8. This setup allows us to directly examine intertemporal substitution within the day. If participants were choosing to reallocate their trips across the hours of a day to avoid travel during hot times, we would expect to find a compensatory effect during the other time intervals (3am-10am and 7pm-2am). While we cannot rule out a modest level of substitution to the evening interval (7pm-2am), the estimates suggest that participants who face lower Uber prices are taking fewer trips on Uber when the day is hot.

Next, we use our survey data to assess how overall travel behavior responds to extreme temperatures. Results reported in column 1 of tab:Table3 indicate that price reductions for Uber services result in a positive but non-significant treatment effect on the total number of trips taken in a given day, the magnitude of which aligns with the results found in (Christensen and Osman (2021)). We do not find evidence of any effect of a hot day on travel when travelers face current market prices for all modes. However, the interaction term in column 1 indicates that hot days do have an impact on the additional travel taken in the context of a price reduction. Individuals who receive the treatment dramatically reduce the number of trips that they take when experiencing a day above 95\degree F (the magnitude of the reduction is 0.795).

\par Since not every day is a hot day, we scale the magnitudes of the effects up to the month level in order to calculate how large the effects are over the period of a 30-day month. These scaled-up coefficients are shown in Panel B of tab:Table3. Column 1 shows that, across a month with the average number of hot days (5.47% of days in our sample), the total number of trips increases by a total of 3.689 trips, which is smaller than the 4.680 trip increase that we would estimate if we only looked at the main effect of the subsidy without taking temperature into account. Hence, experiencing a hot day reduces the treatment effect of the subsidy by 21%.

In addition to looking at the number of trips, it is also possible to examine the impact of subsidies and hot days on the total distance traveled. We estimate these impacts in tab:TableA5, but because total distance traveled has much higher variance than total number of trips, we have lower statistical power to detect effects. Column 1 uses the inverse hyperbolic sine transformation of the distance as the outcome, and column 2 uses distance in levels as the outcome. The larger standard errors imply that we cannot rule out large increases or decreases in overall travel, but the point estimates suggest that the effect on distance is likely similar to the effect on total trips.

\par Overall, we find that for both the Uber data and the survey data, the interaction between receiving a subsidy and experiencing hotter weather results in a decrease in overall travel. This suggests that warmer temperatures mitigate a potential future increase in trips taken as a result of cheaper private transport.

\par If individuals are not reallocating the timing of their trips within the day, it seems likely that the people who receive a subsidy just take more low-value trips on days that are not hot. The negative coefficient on the interaction term for total trips in tab:Table3, then, could be a reduction in these types of trips. Individuals who do not receive a subsidy may not be taking as many unnecessary trips, meaning that they have fewer non-essential trips that they can eliminate. This, then, could be why individuals in the control group do not respond in the same way to a hot day.

\par The interaction effect of receiving a subsidy and experiencing a hot day leads to a 24% reduction in total number of trips on hot days relative to the control group mean in column 1 of tab:Table3. This effect is large relative to effects found in the previous literature, which capture the direct impact of weather on transportation without including changes in transportation price. For comparison, (Changnon1996) finds that rain in Chicago caused a 9% reduction in highway traffic volume on weekends and 3-5% reduction in public transit ridership. Similarly, (Singhal et al. (2014)) find that adverse weather caused a 3% decline in subway ridership at a station in New York. This suggests that including both potential changes in temperature and potential changes in transportation pricing is important when considering future travel demand.

Effects on Mode Substitution

\par Our findings above show that individuals who are receiving a subsidy decrease their trips taken when the weather is hotter. To understand the impacts of this decrease, it is important to understand whether individuals are decreasing travel proportionately for all modes or are engaging in substitution across modes. Using our survey data we are able to disentangle the change in travel by mode to assess the potential of mode substitution. In Panel A of tab:Table3, column 2 reports estimates from equation (1) with the proportion of private trips (number of private trips divided by the number of trips taken) as the outcome.^[11] Estimates reported in column 2 indicate that, when an individual receives a subsidy, the proportion of trips that they take on private modes of transportation increases by 0.078. A hot day does not have a significant effect on the proportion of private trips. However, the interaction between receiving a subsidy and experiencing a hot day increases the proportion of private trips taken by 0.107. This effect is equal to 18% of the control group mean.

\par Columns 3 and 4 of tab:Table3 show how the impact on the number of trips taken varies by transport mode. The change in the proportion of trips taken on private transport is driven by a decrease of 0.562 public transit trips on a hot day for individuals receiving a subsidy as shown in column 3, with a smaller (and statistically insignificant) decrease in trips taken on private transport modes in column 4. Combined with our findings from the previous section, the increase in the proportion of private trips implies that people are changing their transportation decisions on both margins: in some cases, they are deciding not to travel, and in some cases, they seem to be substituting towards private transportation.^[12]

\par When effects are scaled up to the monthly level in Panel B, the total effect on public transportation is a decrease of 5.0 trips (an 11% increase in the treatment effect on hot days), and the total effect on private transportation over the month is an increase of 8.7 trips (a 5% decrease in the treatment effect on hot days).^[13]

Our data allow us to break down the results further and consider how travel changes for each separate transport mode. This is done in tab:Table4. In this table, we see that the main decreases in travel come from decreases in metro trips in column 1 and bus trips in column 2. Another interesting result is that the coefficients for the bus and for Swvl have opposite signs in columns 2 and 3. This is important because a defining difference between Swvl (a private bus service) and the other types of shared transportation is the fact that Swvl has air conditioning, while the public buses and metro generally do not.

\par This evidence on mode substitution has important implications for policy. If private transportation becomes cheaper and temperatures become hotter, individuals may shift away from public transportation, which could increase the emissions intensity of travel (in addition to effects on congestion and other externalities). Strategies such as implementing air conditioning on buses could potentially mitigate impacts on substitution away from public transport and attenuate the feedback between ambient temperatures and greenhouse gas emissions in the transport sector.

Robustness Checks and Additional Results

\par We conduct several robustness checks and additional analyses: (1) comparing different temperature cut-offs, (2) splitting effects by subsidy level, (3) considering differences by weekends vs. weekdays, (4) exploring potential impacts of cold temperatures, (5) comparing our sample to the average Cairene.

First, we check that our results are not driven by choosing the cutoff of 95\degree F (35\degree C). This cutoff aligns with that used for defining hot days in several climate projections models (Bank (2021)). However, we re-estimate our main results for days above 90\degree F. tab:90d_uber and tab:90d_survey show that redefining the cutoff as temperatures above 90\degree F produces results that are generally similar to our main results for Uber trips and all trips, respectively. While the signs are the same as those in our main results, the magnitudes are attenuated because we lose some power to detect effects. This is unsurprising, as days above 90\degree F are quite common, making up 36.6% of overall observations. This means that these days are less likely to represent an exogenous shock and more likely to represent a seasonal trend that individuals gradually adjust to. \par While it would be nice to also investigate how higher temperature cutoffs above 95\degree F impact these outcomes, this is not feasible because of the limited number of days with higher temperatures. As can be seen in Figure 1, many of the days with high temperatures above 95\degree F are still below 100\degree F, meaning that the right tail of the maximum temperature distribution drops off quite quickly.

\par Second, our results are robust to separating the two subsidy amounts (the 25% and 50% subsidy). tab:separate_uber shows the effects of hot days and the two separate treatments on Uber trips and distance traveled for the whole day and during the different times of day. The results are very similar to those found in our main specification, with slightly stronger effects for individuals who receive the 50% subsidy. tab:separate_survey shows the effects of private and public transportation for the two separate treatment arms. Again, the estimated coefficients are quite similar, though in this case, the interaction effects are stronger for recipients of the 25% subsidy.

\par Third, it is possible that effects could differ on weekdays relative to weekend days, so we run our main specification interacted with an indicator for weekend days. The results for daily Uber trips are presented in tab:tab1int, and the results for trips by mode are presented in tab:tab2int. In most cases, we cannot reject that the effects on weekdays and weekend days are the same. The exception to this is the effect on the proportion of trips taken on private transportation modes in column (2), where the interaction effect is much higher on weekend days. This difference is in line with the previous literature, which tends to find stronger effects of weather on weekend days relative to weekdays (Changnon1996; Guo et al. (2007); Singhal et al. (2014); Tao et al. (2018)).

Fourth, our study has focused on responses to extreme heat, but travel decisions may also vary in response to unusually cold days. We test for this in tab:tab1tempbins and tab:tab2tempbins by estimating a set of regressions with 10\degree \space temperature bins, leaving 65-74\degree F as the omitted category. We find suggestive evidence of an inverted-U-shaped response to temperature for Uber trips and all private transit trips. For public transit trips and all trips, we do not find evidence of an inverted-U-shaped response. All together this suggests that responses to climate extremes may not be linear.

\par Finally, to check the external validity of our results, we compare the individuals in our sample with a representative sample of Cairo residents from the 2018 Egypt Labor Market Panel Survey in tab:CairoComp. Individuals in our sample are more likely to be male, younger, currently employed, single, college educated, car owners, and looking for work. Additionally, they likely work longer hours and have a higher monthly income. This is due to our recruitment procedure, since our sample is comprised of existing Uber riders, who tend to be younger and richer than average. Since the sample is quite different from the overall population of Cairo, caution should be taken when generalizing our results.

Discussion, Policy Simulation, & Conclusion

Our results show that extreme weather will affect transportation choices in a future with autonomous vehicles. As transportation becomes cheaper, people will increase the number of trips that they take; however, as temperatures increase, this effect will be dampened.

Next, we consider the implications of these results across time and space. In order to do this, we examine how future projected changes in temperatures will change people’s transportation decisions as travel becomes cheaper, holding constant the treatment effect of cheaper transportation. In addition to looking at projections for Cairo, we also show estimates for four other cities: Delhi, India; Bangkok, Thailand; Dhaka, Bangladesh; and Kano, Nigeria. These cities were chosen based on having city sizes and temperatures that are loosely comparable to the setting of Cairo. Data comes from the World Bank’s Climate Change Knowledge Portal (Bank (2021)). fig:projection shows the percent reduction over time that high temperatures would have on the treatment effect if a similar subsidy was enacted in each of the cities.

In this figure, we consider the effects of reduced prices for private transportation and assume that the treatment effect of the subsidy, the effect of a hot day, the interaction effect between the two, and the number of hot days are the same across cities at baseline. By ensuring that all cities start out at the same point along the y-axis, we are able to examine the implications of variation in the effects of temperature projections by city. The cities that experience larger percentage increases in the number of hot days will also experience a faster decline in additional trips taken as a result of the price reductions.

The figure suggests that the impact of cheaper transportation on trips taken could be greatly influenced by climate change. This is important because it has implications for congestion and emissions in these settings. Places like Bangkok will experience more warming, which will substantially mute responses to cheaper private transport and in turn will affect the use of public transport services. Places like Delhi will react similarly to Cairo, with decreases as described above, but not as drastic as what is predicted for Bangkok.

An important limitation of these projections is that our experiment only provides a well-identified estimate of the short-run responses to these price changes. In the longer run, sustained changes to prices and climate would likely lead to general equilibrium effects and potential adaptation effects that would likely differ from the effects we estimate (Kahn (2016)). Long-run changes will also likely cause agents to update their beliefs and adapt their behaviors in a way that short-run changes do not (Bento et al. (2023); deschenes2011; Hsiang (2016); M\’erel and Gammans (2021)). Nonetheless, the short-term behavioral responses captured here provide new information about how transportation choices adjust in response to concomitant shifts in price and temperature. In particular, they illustrate that reductions in the cost of private transport can result in substantially different effects across cities that vary in their temperature pathways.

\par Understanding how people respond to extreme weather and changes in transportation prices is important given future predictions for rising temperatures and falling prices. Overall, as temperatures rise and the price of private transportation falls, we find that individuals change their transportation decisions. People travel more when transportation is cheaper, but the increase in travel is smaller on hot days. Utilizing the setting of a randomized experiment and the exogenous nature of hot temperatures, we are able to capture the effect of each of these forces and the interaction between them.

\par These results are important to consider when evaluating the impact of cheaper autonomous travel in conjunction with the potential of a warming planet. Additional investments in public transit infrastructure (e.g. air conditioning) could prove a useful strategy to keep travelers using modes like buses and the metro in the face of rising temperatures. This could help mitigate potential substitution from public transit towards private cars.

\par Further research is needed to show how these two forces --- falling prices and rising temperatures --- will impact transportation decisions in other contexts. This paper provides a motivation and framework for understanding this interaction more fully in a broad range of settings.

Exhibits

Appendix

Online Appendix

Notes

1. This experiment is first reported and analyzed in Christensen and Osman (2021). That paper builds a research framework for estimating the price elasticity of demand for mobility (i.e. total distance traveled) and uses it to estimate the change to welfare and external costs (e.g. congestion and emissions) in response to a technologically induced price change. ↩

2. Temperature fluctuations are generally considered to be exogenous. There is concern over endogeneity in temperature measurement in cases with large amounts of missing data (see (Schultz2019)). Since we are not missing temperature data for any days in our sample period, this is not an issue for our setting. ↩

3. See Christensen and Osman (2021) for extensive details about the experiment and data collection. ↩

4. Multiple rounds of follow-up phone surveys were implemented with each participant in the sample, with four attempts per participant. Follow-up surveys collected data on recent travel, counterfactual expectations about a participant’s longest trip using alternate modes, and Google Timeline data over the past three days using the summary feature in the mobile application. Follow-ups surveys were implemented within the study period. fig:AppendixFig3 shows the timing of the study period for each of the 13 cohorts in our sample. ↩

5. Our results are robust to splitting the treatment groups apart, which we do in Appendix tab:separate_uber and Appendix tab:separate_survey. ↩

6. While we choose 95\degree F\slash35\degree C as the cutoff for extremely hot days, we show in tab:90d_uber and tab:90d_survey that our results are robust to choosing 90\degree F as the cutoff. ↩

7. For distance, we use the inverse hyperbolic sine transformation. ↩

8. Public modes of transportation are defined as the metro, buses, and SWVL. Private modes of transportation are defined as personal car, Uber, taxi, and Toktok (i.e. auto-rickshaw). In order to calculate the proportion of trips taken on private modes of transportation, we drop days where individuals do not take any trips, but the results are robust to multiple ways of handling these missing values, as seen in tab:prop_private. ↩

9. Appendix tab:separate_uber and Appendix tab:separate_survey report estimates when we split the treatments. The results are qualitatively similar but less statistically precise. ↩

10. All results are robust to the inclusion of individual baseline controls. ↩

11. Private trips include trips taken via personal car, Uber, taxi, or Toktok. Public trips include trips taken via metro, bus, or SWVL. When no trips are taken, meaning that the denominator is equal to zero, the values are treated as missing for these observations, though the results are robust to different ways of treating these values, as seen in tab:prop_private. ↩

12. Since several of our outcomes are in the form of count data, we also run both Poisson and Negative Binomial specifications for those outcomes. The effects on daily Uber trips can be seen in tab:tab1poisson, and the effects on number of trips by transit type can be seen in tab:tab2poisson. Standard errors are clustered only at the individual level, which is a limitation of these methods. Though many of the coefficients are no longer statistically significant, they remain consistent with our findings in tab:Table2 and tab:Table3. ↩

13. Note that in our sample, 5.5% of days are above 95\degree F. However, over a full year in Cairo, the total number of hot days will be closer to 30%. The difference between these two numbers is due to the period over which the survey data was collected since we do not have as much data during the summer, which is the hottest time of the year. ↩

References

• A. Colin Cameron and Jonah B. Gelbach and Douglas L. Miller (2011). Robust Inference With Multiway Clustering. Journal of Business \& Economic Statistics.
• Antonio M. Bento and Noah Miller and Mehreen Mookerjee and Edson Severnini (2023). A unifying approach to measuring climate change impacts and adaptation. Journal of Environmental Economics and Management.
• Cairo Metro (2023). Cairo Metro | Passenger Services.
• Carleton, Tamma and Jina, Amir and Delgado, Michael and Greenstone, Michael and Houser, Trevor and Hsiang, Solomon and Hultgren, Andrew and Kopp, Robert E and McCusker, Kelly E and Nath, Ishan and others (2022). Valuing the global mortality consequences of climate change accounting for adaptation costs and benefits. The Quarterly Journal of Economics.
• Christensen, Peter and Osman, Adam (2021). The Demand for Mobility: Evidence from an Experiment with Uber Riders.
• Christos A. Makridis and Jason D. Schloetzer (2023). Extreme local temperatures lower expressed sentiment about U.S. economic conditions with implications for the stock returns of local firms. Journal of Behavioral and Experimental Finance.
• Colelli, Francesco Pietro and Wing, Ian Sue and Cian, Enrica De (2023). Air-conditioning adoption and electricity demand highlight climate change mitigation–adaptation tradeoffs. Scientific Reports.
• Espanol, Marc (2022). Uber faces new competition in Egypt. Al-Monitor.
• Hsiang, Solomon (2016). Climate Econometrics. Annual Review of Resource Economics.
• Hudec, Mikuláš (2023). Metro in Cairo - Map, lines, stations and tickets | Tour guide 2023.
• Hymel, Kent (2009). Does traffic congestion reduce employment growth?. Journal of Urban Economics.
• Kahn, Matthew E (2016). The climate change adaptation literature. Review of Environmental Economics and Policy.
• Leard, Benjamin and Roth, Kevin (2015). Weather, Traffic Accidents, and Climate Change.
• M\'erel, Pierre and Gammans, Matthew (2021). Climate Econometrics: Can the Panel Approach Account for Long-Run Adaptation?. American Journal of Agricultural Economics.
• Melanie Cao and Jason Wei (2005). Stock market returns: A note on temperature anomaly. Journal of Banking & Finance.
• Pengfei Lin and Jiancheng Weng and Quan Liang and Dimitrios Alivanistos and Siyong Ma (2020). Impact of Weather Conditions and Built Environment on Public Bikesharing Trips in Beijing. Networks and Spatial Economics.
• Simeon Calvert and Maaike Snelder (2016). Influence of Weather on Traffic Flow: An Extensive Stochastic Multi-effect Capacity and Demand Analysis. European Transport - Trasporti Europei.
• Singhal, Abhishek and Kamga, Camille and Yazici, Anil (2014). Impact of weather on urban transit ridership. Transportation Research Part A: Policy and Practice.
• Soren T. Anderson and Ryan Kellogg and James M. Sallee (2013). What do consumers believe about future gasoline prices?. Journal of Environmental Economics and Management.
• Sui Tao and Jonathan Corcoran and Francisco Rowe and Mark Hickman (2018). To travel or not to travel: ‘Weather’ is the question. Modelling the effect of local weather conditions on bus ridership. Transportation Research Part C: Emerging Technologies.
• Transport for Cairo (2022). Transport for Cairo.
• US News (2023). How to Get Around Cairo..
• Weather Underground (2023). Cairo, Egypt Weather Conditions | Weather Underground.
• World Bank (2021). World Bank Climate Change Knowledge Portal.
• World Bank Group (2014). Cairo Traffic Congestion Study Executive Note.
• Zhan Guo and Nigel H. M. Wilson and Adam Rahbee (2007). Impact of Weather on Transit Ridership in Chicago, Illinois. Transportation Research Record.