3  Results

Code 
# Data Preprocessing for MacOS
# We found that the column name would be different when reading files for MacOS and Win system, so we imcorporated two block of preprocessing code
library(ggplot2)
library(dplyr)
library(tidyr)
library(forcats)
evp <- read.csv("https://raw.githubusercontent.com/sw547/EVpop/main/data/Electric_Vehicle_Population_Data_Updated.csv")

evp <- evp %>%
  filter(State == "WA") %>%
  select(-Legislative.District, -Postal.Code, -`X2020.Census.Tract`, -`VIN..1.10.`,
         -Model.Year, -`Clean.Alternative.Fuel.Vehicle..CAFV..Eligibility`, -DOL.Vehicle.ID, 
         -Vehicle.Location, -Electric.Utility, -State)

names(evp) <- c("County", "City", "Make", "Model", "Type", "Range", "MSRP")
Code 
# Data Preprocessing for Win
library(readr)
library(dplyr)
library(tidyr)
# EV historical data
hist_url = "https://raw.githubusercontent.com/sw547/EVpop/main/data/Electric_Vehicle_Population_Size_History.csv"
hist <- read_csv(hist_url)
names(hist) <- c("Date", "PHEV_Count", "BEV_Count", "EV_Count")

# EV population data
evpop_url = "https://raw.githubusercontent.com/sw547/EVpop/main/data/Electric_Vehicle_Population_Data_Updated.csv"
evpop <- read_csv(evpop_url)

evpop <- evpop %>%
  filter(`State` == "WA") %>%
  select(-`Legislative District`, -`Postal Code`, -`2020 Census Tract`, -`VIN (1-10)`, -`Model Year`, -`Clean Alternative Fuel Vehicle (CAFV) Eligibility`, -`DOL Vehicle ID`, -`Vehicle Location`, -`Electric Utility`, -State)

names(evpop) <- c("County", "City", "Make", "Model", "Type", "Range", "MSRP")

# EV historical data by county
hist_county_url = "https://raw.githubusercontent.com/sw547/EVpop/main/data/Electric_Vehicle_Population_Size_History_By_County.csv"
hist_county <- read_csv(hist_county_url)

hist_county <- hist_county %>%
  filter(State == "WA")

names(hist_county) = c("Date","County","State","Use","BEV","PHEV","EV","NonEV","V","EV_perc")

3.1 EV Trend and Geological Distribution

3.1.1 How many EVs on the road?

Code 
library(ggplot2)
library(dplyr)
hist <- hist %>%
  mutate(Date = as.Date(Date, format = "%B %d %Y"))

hist_long <- hist %>%
  select(Date, PHEV_Count, BEV_Count) %>%
  gather(key = "VehicleType", value = "Count", -Date)

ggplot(hist_long, aes(x = Date, y = Count, fill = VehicleType)) +
  geom_bar(stat = "identity", position = "stack") +
  scale_fill_manual(values = c("PHEV_Count" = "#beb9db", "BEV_Count" = "#fdcce5")) +
  labs(title = "Total Electric Vehicles Registered in WA Over Time",
       x = "Date",
       y = "Count",
       fill = "Vehicle Type") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1),
        legend.position = "bottom")

Electric vehicle ownership surged remarkably from 22,425 units in January 2017 to 159,467 by October 2023, marking a more than seven fold increase. This trend is evident in the first graph, which illustrates an increasing growth rate each year. Additionally, both Plug-in Hybrid Electric Vehicles (PHEVs) and Battery Electric Vehicles (BEVs) have seen a rise in their numbers. However, the growth in BEV numbers has outpaced that of PHEVs.

Along with surge in overall EV numbers, there has been a notable shift in the composition of EV types. The proportion of BEVs has grown significantly, particularly in 2023. Over the last seven years, BEVs have not only increased in count but also consistently maintained a higher percentage compared to PHEVs.

3.1.2 How many EVs are there in each county?

Code 
library(sf)

ev_count_per_county <- evpop %>%
  group_by(County) %>%
  summarise(EV_Count = n())

counties <- st_read("/Users/wsqia/OneDrive/文档/547files/Columbia/courses/STAT5702 EDAV/Final Project/WA_County_Boundaries.dbf", quiet = TRUE)
# Note: for the shapefile, I tried different methods but I still cannot avoid using local file. Also, the dbf file seems lost some information every time when I upload it. However, I incorporated the data source link in the report for further investigation and reproduction.

counties <- counties %>%
  select(JURISDIC_2, geometry)
names(counties) <- c("County", "geometry")

geo_data <- merge(counties, ev_count_per_county, by = 'County', all.x = TRUE)

ggplot(data = geo_data) +
  geom_sf(aes(fill = EV_Count), color = NA) +
  geom_sf(color = "darkblue", fill = NA, size = 1) +
  geom_sf_text(aes(label = County), size = 3, check_overlap = TRUE) +
  scale_fill_distiller(palette = "Purples", direction = 1) +
  theme_minimal() +
  theme(axis.title.x = element_blank(),
        axis.title.y = element_blank(),
        axis.text.x = element_blank(),
        axis.text.y = element_blank(),
        panel.grid.major = element_blank(),
        panel.grid.minor = element_blank(),
        plot.title = element_text(size = 20)) +
  labs(title = 'Number of EVs by County in Washington State',
       fill = 'EV Count')

link to shapefile source (WA County Boundaries)

Note: Due to the disproportionately high number of EVs in King County, presenting the counts for other counties on the same scale is challenging. To address this, a graph utilizing a logarithmic scale for the counts has been provided for clearer comparison.

Code 
geo_data <- geo_data %>%
  mutate(EV_Count_log = log(EV_Count))

ggplot(data = geo_data) +
  geom_sf(aes(fill = EV_Count_log), color = NA) +  
  geom_sf(color = "darkblue", fill = NA, size = 1) + 
  geom_sf_text(aes(label = County), size = 3, check_overlap = TRUE) +  
  scale_fill_distiller(palette = "Purples", direction = 1) + 
  theme_minimal() +
  theme(axis.title.x = element_blank(),  
        axis.title.y = element_blank(),  
        axis.text.x = element_blank(),   
        axis.text.y = element_blank(),  
        panel.grid.major = element_blank(),  
        panel.grid.minor = element_blank(),
        plot.title = element_text(size = 20)) + 
  labs(title = 'Number of EVs by County in Washington State on Log Scale', 
       fill = 'log(EV Count)')

King County stands out for its high number of electric vehicles (EVs) compared to other counties in Washington. Neighboring Snohomish and Pierce Counties also demonstrate a significant number of EV adoptions, with Clark County following closely in terms of EV counts. Additionally, it is worth noting that Garfield County has a distinctly lower EV adoption, highlighting a sharp contrast within the state.

Outsourced Graph: Population Density in Washington State link to graph source

In terms of population, King County leads with 2,269,675 residents, while Pierce and Snohomish rank second and third with populations of 921,130 and 827,957, respectively. These counties, particularly King, Pierce, and Clark, have population densities exceeding 500 people per square mile. Snohomish County, with a density between 200 and 499.9 people per square mile, also exhibits a notable concentration of population. This pattern suggests a potential correlation between population density and the number of EV adoptions. However, further research is required to establish a definitive correlation or causal relationship between these factors.

3.3 Price

3.3.1 We can’t ignore the price! What is the distribution of EV base MSRP in Washington State?

Code 
# one single entry of very high MSRP causes problem in histogram, thus topcoding
# evp%>%
#   arrange(desc(MSRP)) %>%
#   select(MSRP)
evp_msrp_tc <- evp %>%
  mutate(MSRP_tc = ifelse(MSRP > 100000, 100001, MSRP))

ggplot(evp_msrp_tc, aes(x = MSRP_tc)) +
  geom_histogram(binwidth = 10000, boundary= 10000, fill = "#7eb0d5",color="#115f9a",closed="right") +
  scale_x_continuous(breaks = seq(10000, 110000, by=10000),
                     labels = c("10k","20k","30k","40k","50k",
                                "60k","70k","80k","90k","100k","inf")) +
  theme_minimal() +
  labs(x = "Base MSRP ($)", y = "Count", title = "EV MSRP Frenquency in WA")

More than 70,000 of the Electric Vehicles in Washington State has a Base MSRP ranging from $40,000 to $50,000, making it the most common price range. The second most common price range is $20,000 to $30,000, which has less than a half of count compared to that of the most common price range. Then we have $30,000 to $40,000 and $70,000 to $80,000, which each contains about 20,000 electric vehicles. Note that when price exceeds $50,000, count significantly drops. We can conclude that people lean more towards a more economical choice.

3.3.2 Is there any noticable difference in distribution of prices between PHEV and BEV?

Code 
# since one extreme outlier makes the boxplot hard to read, we adjust y-axis limits(not x-axis because we do coord_flip)
ggplot(evp, aes(x = reorder(Type,MSRP, median), y = MSRP)) +
  geom_boxplot(fill = c("#beb9db", "#fdcce5")) +
  coord_flip()+
  theme_minimal() +
  theme(text=element_text(size=20))+
  scale_y_continuous(breaks = seq(0, 220000, by=20000),
                     labels = c("0","20k","40k","60k","80k","100k",
                                "120k","140k","160k","180k","200k","220k"),
                     limits = c(0,220000)) +
  labs(x = "EV Type", y = "Base MSRP ($)", title = "Multiple Boxplot of MSRP according to EV Type")

There do exist a difference in price pattern between BEV and PHEV. In general, BEVs are more expensive than PHEVs, with respective median of around $41k and $32k. The interquartile range of PHEV is wider than that of BEV, but PHEV has more variability in price overall. There are more outliers for BEV base MSRP than for PHEV base MSRP, especially on the lower side as PHEV has no outliers on the lower side at all. In general, although BEVs are generally more expensive than PHEV, we can find the lowest prices on BEVs and the highest prices on PHEVs. Note that we set limits on the axis because of the existence of an extreme outlier. All of the vehicles except for one has Base MSRP less than $220k, and that one exotic car is Porsche 918 which has a Base MSRP of $845k.

3.4 Electric Range

3.4.1 How’s the electric range for EVs? Are they the same for BEVs and PHEVs?

Code 
ggplot(evpop, aes(x = Range)) +
  geom_histogram(binwidth = 50, boundary = 0, fill = "#7eb0d5", color = "#115f9a", closed="right") +
  scale_x_continuous(breaks = seq(0, 500, by = 50)) + 
  theme_minimal() +
  labs(title = "Histogram of Electric Range of EVs",
       x = "Electric Range",
       y = "Count")

From the histogram above, it’s apparent that the electric range of EVs predominantly falls into two groupings. The first cluster is centered at approximately 50 miles, indicating a significant number of EVs with a lower electric range. The second, more dispersed cluster spans from 200 to 400 miles, suggesting a wide array of EVs offering mid to high electric range capabilities. Apart from these, there is an exceptional category exceeding 500 miles, which is an outlier.

This distribution suggests a trend toward consumer preference for EVs within these two distinct range categories. Notably, there are relatively few purchases of EVs with the highest range capabilities, which may be due to limiting factors such as cost or low supply due to current technological constraints.

Code 
ggplot(evpop, aes(x = reorder(Type, Range, median), y = Range)) +
  geom_boxplot(fill = c("#beb9db", "#fdcce5")) +
  theme_minimal() +
  coord_flip() +
  labs(title = "Distribution of Electric Range for BEVs and PHEVs",
       x = "Vehicle Type",
       y = "Electric Range (miles)")

The boxplot analysis indicates that BEVs primarily constitute the second cluster in the histogram, with a wider range, reflecting higher variation in their electric range. The median range for BEVs is just under 300 miles, significantly surpassing that of PHEVs, which reflects their reliance solely on battery power. Notably, the BEV distribution includes outliers on both the lower and higher ends, pointing to exceptional models deviating from the typical range.

Conversely, PHEVs, which make up the first cluster in the histogram, display a much smaller range, indicative of a narrower spectrum of electric capabilities. With a median of approximately 25 miles, PHEVs generally offer a shorter electric range than BEVs, which aligns with their dual-energy design incorporating gasoline. However, it is quite surprising that half of the hybrid electric vehicles can BARELY drive from Manhattan to Long Beach, solely on its battery. Additionally, the presence of outliers suggests that a few PHEVs have much higher electric range than the majority.

3.5 Price vs. Electric Range

3.5.1 Do we get more Electric Range if we’re paying more for the EV?

Code 
# When dealing with correlation of MSRP and EV Range, we drop duplicates
msrp_range <- evp %>%
  select(Type, MSRP, Range) %>%
  distinct(MSRP, Range, .keep_all=TRUE)

# we adjust limits of x-axis here to prevent that one extreme outlier
# from making the graph hard to read
ggplot(msrp_range, aes(x = MSRP, y = Range, color=Type)) +
  geom_point(size=1.2,alpha=0.8) +
  theme_minimal() +
  scale_x_continuous(breaks = seq(0, 220000, by=20000),
                     labels = c("0","20k","40k","60k","80k","100k",
                                "120k","140k","160k","180k","200k","220k"),
                     limits = c(0,220000)) +
  labs(x = "Base MSRP ($)", y = "Electric Range (miles)", title = "EV MSRP vs. Electric Range") +
  scale_color_manual(values=c("#0d88e6","#e60049"))

The scatterplot of Base MSRP vs. Electric Range for electric vehicles suggest that there’s no clear correlation in the two variables, meaning that we do not get more electric range when we are spending more on an EV, and we observe that we actually get small electric range when we’re spending more than $130k since all the options in that price range are PHEV. On top of the price and electric range analysis in former parts, we can also see how each EV type performs on the both variable together. BEVs form a clear crowded cluster with electric range of 200 to 400 and base MSRP of $20k to $130k. PHEVs are crowded on the lower left side of the scatterplot and spans across the lower side. There are also some BEV points on the lower left side but they are generally seperated clearly with PHEV. Note that here we also set limits on the axis due to the Porsche 918 as in part 3.3.2.