Weather Data

The ERA5 dataset provides information on climate variables, including hourly temperature, total precipitation, and humidity from 1970 onwards. This data was used to derive the heat index from climate variables using the weathermetrics package in R for both 2019 and 2024 at the daily and hourly level from May- September. Once these heat indexes were calculated, we identified different thresholds to better understand extreme and dangerous heat levels. As per the National Weather Service (NWS), a heat index between 80°F and 90°F could result in fatigue under prolonged exposure and a heat index between 90-105°F falls under the extreme caution category, possibly resulting in heat disorders such as sunstroke and heat exhaustion. Using these definitions, we described Hot as days when the heat index is equal to or greater than the historical 90th percentile (to better understand local climate with historical trends) and the heat index is greater than or equal to 86°F; Very Hot Days as days when the maximum heat index is greater than or equal to 90°F. Lastly, Extremely Hot Days includes days when the heat index is greater than or equal to 100°F, which falls under heat advisory and heat warning categories (Figure 4).

The ERA5 dataset was also used to calculate precipitation to identify heavy rain days that may alter regular travel decisions on a given day. During the data exploratory phase, we noticed an unusual decrease in the overall number of trips on certain days with no particular association with the heat index. Upon further investigation, we found these days to have received precipitation at different times of the day or throughout. One of the key decisions made during this phase was to identify a heavy precipitation day from a regular rainy day. A heavy precipitation event is defined as days where the total amount of precipitation is in the top 1 percent of all precipitation days during the 1958-2022 reference period. Based on this definition, each day’s total daily precipitation was calculated for Chicago, and days in the top 1 percent category for each year were marked as rainy days, which were later dropped from model estimations.

Figure 4: Heat Index Thresholds

Trip Data

The Chicago Open Data Portal publishes trip record data for various services, including yellow taxis and TNCs. These trip records provide date, time, pickup, and drop-off census tracts, trip distance, and fare. The date and time columns provide the date and timestamp of when a trip was undertaken. This data on the website rests in a different timezone; therefore, we converted it to EST zone (America/ New York) for this analysis, as heat index data was also retrieved for the same time zone.

For yellow taxis and TNCs, the trip records were cleaned to prepare the data for analysis. Some holidays that occurred within the selected study period in Chicago were removed from analysis, as they are expected to generate unusual demand for taxi/TNC usage at certain locations. These events include the Memorial Day, Juneteenth, Independence Day, and Labor Day. Furthermore, weekend trips, i.e. Saturday and Sunday, were also removed from the analysis as weekend travel patterns tend to differ from those on regular weekdays.

Socio-economic and Demographic Data

To understand how people make travel decisions on extremely hot days, we looked at different spatial demographics and socio-economic patterns across the City that could influence these decisions. The analysis used ACS 5-year summary from 2019-2023 data to look at median household income and racial and ethnic background. The ACS data was filtered to only return census tracts with at least 50 housing units.

Using ACS summary data, the Area Median Income (AMI) for Chicago was calculated using household-weighted median household income across all tracts. Income categories were then defined by setting the low-income threshold at 80% of AMI and the high-income threshold at 200% of AMI. For racial analysis, we defined categories by calculating the share of residents who identify as Non-White (Hispanic or any Non-Hispanic race other than White). Tracts were then grouped into racial composition categories using percentile: Tracts were then grouped into racial composition categories using percentiles:

• Low Non-White tracts: Less than or equal to the 25th percentile.
• Medium Non-White tracts: Between the 25th and 75th percentiles.
• High Non-White tracts: Greater than or equal to the 75th percentile.

Analysis

We set out to understand the effects of extreme heat on travel behavior—specifically, taxi/TNC usage on hot days—and to analyze how that usage varies over time, by mode availability, and across neighborhoods defined by different socio‑economic and demographic characteristics. As discussed earlier, there were three different thresholds considered to define hot days. These days were selected after filtering weekends and rainy days from the weather dataset. For each hot day, we examined the heat index on the same day in the preceding and following weeks (Figure 5), which were considered as control days in the statistical analysis. If the heat index of the day selected in the prior week or week after fell under one of the hot day categories, then these days were removed from the set of days defined as control. Figure 6 shows the final number of different hot (Hot, Very Hot Day, and Extremely Hot Day) and control days selected from each year.

Figure 5: Identifying Hot and Control Days

Figure 6: Count of Hot and Control Days in Different Categories

1. More Trips Being Taken

The average number of trips for yellow taxis and TNCs shows a similar trend for all three definitions of hot days across all three years (Figure 7). On hot days (regardless of the category), we see that there is an increase in the average number of trips taken via yellow taxi/TNCs. Between yellow taxis and TNCs, we observe that TNCs experience a greater increase in the average number of trips taken on hot days as compared to control days, especially in 2024. This is an expected result, as it is much easier and more convenient to request a TNC trip while staying indoors on a hot day than to step outside and hail a yellow taxi on the street.

#hourly plot processing 
df_daily <- read.csv(here("Processed Data", "df_daily_hot_control.csv"))
#---------------------------------------------------
### Create daily summary plots

df_daily_summary <- sqldf('
    SELECT 
        mode,
        year,
        day_type,
        category_type,
        SUM(total_dist) / SUM(num_trips_all) AS avg_dist,
        AVG(num_trips_all) AS avg_trips
    FROM df_daily
    GROUP BY 
        mode, year, day_type, category_type;
')

df_daily_summary$year_fac <- factor(df_daily_summary$year, levels = c(2019, 2024))
df_daily_summary$day_type_fac <- factor(df_daily_summary$day_type, levels = c('hot', 'control'), labels = c('Hot Days', 'Control Days'))
df_daily_summary$mode_fac <- factor(df_daily_summary$mode, levels = c('Taxi', 'TNC'), labels = c('Yellow Taxi', 'TNC'))
df_daily_summary$hot_day_def_fac <- factor(df_daily_summary$category_type, levels = c(2, 1, 3), labels = c('HI > HI_hist & HI >= 86F', 'HI >= 90F', 'HI >= 100F'))

#----------------------------------------------------
###########################################################################################################

# Plot average number of trips

daily_taxi <- ggplot(data = df_daily_summary, aes(x = year_fac, y = avg_trips / 1000, fill = day_type_fac)) +
  geom_bar(stat = "identity", position = "dodge") +
  geom_text(aes(x = year_fac, y = avg_trips / 1000, label = round(avg_trips / 1000, 1)),
            position = position_dodge(width = 1),
            vjust = 2, size = 4, color = "black", fontface = "bold") +
  theme(text = element_text(size = 32), axis.text.x = element_text(size = 28), axis.text.y = element_text(size = 20),
        legend.position = "top", legend.title=element_blank()) +
  xlab("") +
  ylab("Average number of trips (000's)\n") +
  scale_alpha_manual(values = c(0.8, 1.0)) +
  scale_fill_manual(values = c("#ca0020", "#bababa")) +
  facet_grid(mode_fac ~ hot_day_def_fac, scales = "free_y")

#----------------------------------------------------

# Plot average trip distance (miles)

daily_tnc <- ggplot(data = df_daily_summary, aes(x = year_fac, y = avg_dist, fill = day_type_fac)) +
  geom_bar(stat = "identity", position = "dodge") +
  geom_text(aes(x = year_fac, y = avg_dist, label = round(avg_dist, 2)),
            position = position_dodge(width = 1),
            vjust = 2, size = 4, color = "black", fontface = "bold") +
  theme(text = element_text(size = 32), axis.text.x = element_text(size = 28), axis.text.y = element_text(size = 20),
        legend.position = "top", legend.title=element_blank()) +
  xlab("") +
  ylab("Average trip distance (miles)\n") +
  scale_alpha_manual(values = c(0.8, 1.0)) +
  scale_fill_manual(values = c("#ca0020", "#bababa")) +
  facet_grid(mode_fac ~ hot_day_def_fac, scales = "free_y")

Figure 7: More Trips Being Taken

#hourly plot processing 
df_hourly <- read.csv(here("Processed Data", "df_hourly.csv"))

#---------------------------------------------------
#Prepare dataframe for plots 

df_hourly_hi_summary <- sqldf('SELECT mode, year, start_hour, day_type, category_type,
                          AVG(num_trips_all) AS avg_trips
                          FROM df_hourly
                          GROUP BY mode, year, start_hour, day_type, category_type;')

df_hourly_hi_summary <- df_hourly_hi_summary %>% drop_na()
df_hourly_hi_summary$year_fac <- factor(df_hourly_hi_summary$year, levels = c(2019, 2024))
df_hourly_hi_summary$day_type_fac <- factor(df_hourly_hi_summary$day_type, levels = c('hot', 'control'), labels = c('Hot Days', 'Control Days'))
df_hourly_hi_summary$mode_fac <- factor(df_hourly_hi_summary$mode, levels = c('Taxi', 'TNC'), labels = c('Yellow Taxi', 'TNC')) 
df_hourly_hi_summary$category_type_fac <- factor(df_hourly_hi_summary$category_type, levels = c(1, 2, 3), labels = c('HI > HI_hist & HI >= 86F', 'HI >= 90F', 'HI >= 100F'))

###########################################################################################################

# Plot average number of trips (Taxi) - hourly 
taxi_plot <- ggplot(data = df_hourly_hi_summary[df_hourly_hi_summary$mode == 'Taxi',],
                    aes(x = start_hour, y = avg_trips/1000, color = day_type_fac)) +
  geom_line(linewidth = 1.25) +
  theme(text = element_text(size = 32),
        axis.text.x = element_text(size = 28),
        axis.text.y = element_text(size = 20),
        legend.position = "top") +
  scale_color_manual(values = c("#ca0020", "#bababa"), name = "Yellow Taxi") +
  xlab("Start Hour") +
  ylab("Average number of trips (000's)\n") +
  facet_grid(year_fac ~ category_type_fac, scales = "free_y")


# Plot average number of trips (TNC) hourly 
tnc_plot <- ggplot(data = df_hourly_hi_summary[df_hourly_hi_summary$mode == 'TNC',],
                   aes(x = start_hour, y = avg_trips/1000, color = day_type_fac)) +
  geom_line(linewidth = 1.25) +
  theme(text = element_text(size = 32),
        axis.text.x = element_text(size = 28),
        axis.text.y = element_text(size = 20),
        legend.position = "top") +
  scale_color_manual(values = c("#ca0020", "#bababa"), name = "TNC") +
  xlab("Start Hour") +
  ylab("Average number of trips (000's)\n") +
  facet_grid(year_fac ~ category_type_fac, scales = "free_y")

2. Shorter Trips Being Taken:

We set out to test our hypothesis that shorter walk and bike trips are being switched to taxi/TNC trips on hot days and as it gets hotter during the day. Figure 8 shows how trip distances for both years and modes. As mentioned earlier, these results are more pronounced for TNCs, the Extremely Hot category.

Figure 8: Shorter Trips Being Taken

3. When are They Being Taken:

As we zoom into the effects, we get more insights into the hourly trip distribution as the heat index changes within a hot day vs. a control day. We see little to no difference in the average number of yellow taxi trips between hot and control days when the day starts. However, as the day progresses, we observe the difference to increase, especially during 10 am- 7 pm, with some of the biggest differences in afternoon peaks observed in early mornings when the day begins. However, as the day progresses, the gap becomes more profound and peaks during the hottest time of the day in the afternoon between 3 pm -7 pm.

The trends for TNCs follow the same direction as well, but they are a lot more significant than what we observed for yellow taxis. While the overall average number of trips is much higher for TNCs than for yellow taxis, we also observe a much larger difference between average trip counts between hot and control days for TNCs as compared to yellow taxis, especially between 3 pm- 7 pm in 2019 as well as 2024.

Figure 9a: When are They Being Taken (Yellow Taxi)

Figure 9b: When are They Being Taken (TNC)