“It’s Too Nice Out to Take a Cab”: Weather and Taxi Usage in New York, 2019–2024

1 Introduction

Do you ever wonder how weather influences our everyday choices? In a city like New York, even simple behavioral changes can have macro-level ripple effects. This is what we set out to explore—starting with taxi data.

At first, Moacir was interested in seeing if there is a relationship between “unseasonably” warm weather and New York and drought-like conditions, but Sophie suggested taking it in a different direction: what if we crossed this data with some human behavior, like taxi usage? What might this look like? After a bit of discussion, we landed on a fun hypothesis: people use cabs less often when it is “nice” out in Manhattan – they would rather walk than hail an expensive cab.

Proving this turned out to be tricky: what is even is “nice” weather? Does it just mean sunny skies, or does it have a relationship to a temperature threshold? How might relative temperature come into play, such as an unusually warm day after a cold spell, impacting people’s inclination to walk? And does the effect wear off if there are multiple nice days in a row, as the novelty of walking gives way to taking cabs again? These questions struck us as more amusing and speculative, so we decided to pursue them, instead. Instead of proving the hypothesis, we are using taxi data to see if we can define what “nice” weather is.

In short, our project aims to uncover how weather influences small, everyday decisions in urban life. Using taxi data provides a unique lens into how people adapt their transportation preferences due to the weather, serving as a microcosm for understanding human responses to environmental factors. In a city as dynamic as NYC, the answers could be particularly relevant and reveal fascinating insights about how we adapt to our surroundings.

1.1 A High-Level Look at Weather and Taxi Trends: Can We Spot a Pattern from Here?

Let’s take a quick peek at the data. Below we have weekly averages for daily taxi rides and temperature from January 2019 to June 2024.

df <- read_parquet("data/complete_weather_and_taxi_data.parquet")

df |> 
  group_by(date) |> 
  summarize(
    total_trips_day = sum(trip_count), .groups = "drop",
    daily_temp = mean(temperature, na.rm = T)
  )  |> 
  mutate(week_start = floor_date(date, unit = "week")) |> 
  group_by(week_start) |> 
  summarize(
    avg_trips_day = mean(total_trips_day),
    avg_temp_day = mean(daily_temp), .groups = "drop"
  ) |> 
  pivot_longer(
    cols=c(avg_trips_day, avg_temp_day), 
    names_to="avg_day_metric", 
    values_to="avg_day_value"
  ) |>
  mutate(avg_day_metric=as.factor(avg_day_metric)) |> 
  mutate(avg_day_metric=fct_recode(
    avg_day_metric, 
    "Average Daily Trips"="avg_trips_day", 
    "Average Weekly Temperature (°C)"="avg_temp_day")
  ) |> 
  mutate(avg_day_metric=fct_rev(avg_day_metric))  |> 
  ggplot(aes(x = week_start, y = avg_day_value)) +
    geom_line(color = base_color, linewidth = 0.7) +
    facet_wrap(~ avg_day_metric, scales = "free_y", ncol=1) +
  scale_x_date(date_labels = "%b %Y", date_breaks = "1 year") +
  scale_y_continuous(labels=double_scale) +
  labs(title = "Ridership and temperature trends, Jan 2019 – June 2024",
       x = "Date",
       y = "") +
  theme_minimal() +
  theme(
    plot.title = element_text(size = 16),
    strip.text = element_text(size = 12, hjust = 0, vjust = 0.5, face="bold"),
    strip.placement = "outside" 
  )

🚕 Considering the taxi data, the most notable observation is the dramatic drop in ridership in March 2020, coinciding with the beginning of the COVID-19 Pandemic. While ridership has increased since, it has not nearly returned to pre-Pandemic levels, reflecting increased work-from-home and other behavioral shifts. We can also observe seasonal fluctuations; for example, it appears that there are dips and peaks around January of each year, possibly tied to holiday travel patterns, weather, or other behavioral changes.

⛅ Average weekly temperature follows predicted seasonal trends with peaks in the summer and dips in the winter. There may be a subtle trend of slightly higher average temperatures in more recent years, but nothing too definitive. Temperature alone provides a limited slice into what may distinguish a “nice day.” In the next chapter, additional metrics like temperature change and weather categories (e.g. cloud cover, rain) will be explored to refine the relationship between weather and ridership.

From this broad view, it is really hard to see any clear relationship between ridership and weather. This is what makes our project so interesting–how can we hammer out the data to unveil micro-patterns? We’ll need to conduct a much more granular analysis to get to the bottom of things.