In this study, we were interested in finding out the factors that influenced people’s preferences or aversion for electric vehicles. With the used auto-vehicle market significantly outweighing the new vehicle market, our focus was directed towards used electric vehicles and the various conditions that has limited its adoption. Our goal was to identify how consumers’ preferences change with varying attributes such as price, mileage, brand, model year, range, warranty and power/energy sources. Using these attributes, we carried out a survey providing consumers with different options in order to gain insights into how their choices changed with each attribute. A total of 170 respondents were considered for the survey in order to provide a comprehensive/balanced data source. Ultimately, we discovered that consumers prefer to have free charging stations from the dealer. Participants who did not possess EV preferred are less sensitive to charging station promotion and Finally, price and range are the main driving force for used EV demand.
The Transportation sector is the biggest source of Greenhouse Gas emissions in the United States and besides reducing the number of fuel engines by promoting public transportation, pushing the adoption of Electric Vehicles is a great way to cut tail-pipe emission
While the adoption of Electric vehicles seems like a simple response and the EV market has been booming since 2017, the reality remains that market size of electric vehicles is still small despite the government incentives and environmental benefits. To paint a clearer picture, the used auto market is much bigger than the new auto market. However, for electric vehicles, the used EV market is only just starting to find its feet.
This market situation is also described by Rogers’ Technology Adoption Theorem. Rogers (2003) describes the adoption categories in five parts according to his technology diffusion theory namely Innovators, Early Adopters, Early Majority, Late Majority and Laggards. The distribution of these categories follows a normal distribution with people of high socioeconomic status adopting new technology earlier than the vast majority of early and late users which make two-thirds of the total members.Given the side of EV market today, we can assume assume that majority of potential buyers are waiting for core attributes such as lower prices or more reliable technology . This makes the exploration of the used EV market critical to the adoption of this technology.
!
The share of the EV second-hand market in the second-hand car market is slowly rising with the sales of electrified vehicles jumping from 4.8% of the market in the first quarter of 2020 to 7.8% in the same period of 2021. Used electric vehicle marketing is gaining more attention as people are realizing its benefits in comparison to the cars on gas. The goal of this project is to explore the factors that influence the choices of car buyers especially with those inclined towards electric vehicles and the used vehicle market. Our focus is to create and analyze data sets to give facts and help individuals make decisions.
Currently many incentives are given by the public sector and private sector on used and new EVs to reduce carbon emission. One of the private sector companies pays its employees $2000 if they opt in to buy an electric vehicle if they have worked for the company for at least one year.
Our main focus is to look into used EV vs the normal gas car and compare and capture insights by doing a survey of what is the thought process of people looking to buy cars.
Some of the studies have shown that if some sort of incentive or rebate is given, it really has an effect on the customer’s decisions , especially if the customer is considered between a low to average income household.(Legislators propose $2,500 federal tax credit ) One of the things we can’t quantify is the features coming in with used EV so that’s one area we can’t incentivize the customer.
Looking into the behavior of our stakeholders i.e potential used EV buyers would give us a comprehensive insight on how generally used EV is perceived by the general public and what this market would be like in the near future.
Out of scope is the market for new EV for this current project since we won’t be asking our stakeholder anything related to the new EV in the survey. The reason for this is to just focus on the used EV market vs the gas cars. Looking into new EVs won’t be beneficial since there are a lot of options on how a customer can customize their cars and in comparison it’s not possible to do that in used EVs.
There are two eligibility requirements in our design choices. The first is whether the survey subject is over 18 years old. The second requirement is to finish understanding the above information. Our project is about the purchase of used vehicles, where people over 18 years old have spending power, and by reading the information, the respondent has a basic idea of the direction of the survey.
We only collect two pieces of information that interviewees live in the DMV area or not and the current zip code at the beginning. These two pieces of information were collected to determine whether the interviewees lived in a DMV area. Our survey was limited to the DMV area. Our interview population was also limited to people living in the DMV area. We also asked interviewees if they had any intention of purchasing a vehicle in the next six months. If yes it would meet our interview requirements because the price of the vehicle is time sensitive.
At the end of the survey, we will ask six more questions, if the interviewer finished our survey. We will use it for analyzing our survey data. Five questions that “In what year were you born?”, “How many cars are currently present in your household?”, “ Do you currently or have you owned an electric vehicle?” , “What is your current gender identity?”, “I identify my race as” , and “What is your annual household income (from all sources) before taxes and other deductions from pay?”. If the interviewees are not part of our interview scope, they will not see these questions.
We provided educational material about the vehicle’s functions, such as “Price”, “Mileage”, “Model Year”, “Brand”, “Range”, “Warranty”, “Charging Station”, and “Fast Recharge”. The “Price” refers to the price to buy this car. The “Mileage” that it refers to is how much the car has been driven. The “Model Year” that it is the year when the car was manufactured. The “Brand” shows the brand of this car. The “Range” that it’s the number of miles a car can drive in one complete charge. The “Warranty” is whether the car comes with a warranty or not. The “Charging Station” is whether the dealer offers a free home charging port when you buy the car. The “Fast Recharge” that the car can be charged faster than its normal charging time
We noticed that our doe contains some unreasonable situations such as a 1 year used car with 50k miles and 5 year car with 5k miles. This kind of extreme cases need to be fixed and we will make sure the mileage are within a reasonable range according to the model year. Besides that, we also increased the minimum price and mileage to $10,000 and 5,000 from $5,000 and 1,000 respectively. They are now both increasing in a 5,000 increment instead of 1,000 increment. In this way, we could significantly reduce the total number of doe under a full factorial design. We are also planning to increase the minimum EV range from (100,250) to (200,400) interval, which is much reasonable representation of current EVs’ specification.
We also noticed the chart we have is very misleading for some participants. Showing a Tesla image and put only 200 miles in the range attribute is inconsistent for participants who are familiar with EVs. To achieve the generality of the estimation result, we will exclude the image and only present attributes in text. To account for the endogeneity of our model, we will ask the participants to assume all cars are a black sedan in a good condition.
However, we understand the limitation in ways to distribute our survey and since we are having a very specific geographic requirement on out participants, we will not get sufficient observations suggested by the power analysis. What we will do is increase the number of questions that each participant have to answer so we could get the error bar narrower.
The price range is from $10,000 to $50,000. The “FuelEconomy” range is from 5 mpg to 35 mpg. The “modelyear” is from 2014 to 2019. The “powertrain” has two options: “Gas” and “Electric”. The “Warranty” has two options: “Yes” and “No”. The EV driving range is from 200 to 400. The “Mileage” range is from 10,000 to 70,000. For the “Charging Station”, we give three options: “Free”, “Discounted”, and “No Promotion”. We tried to make each option logical and relatable. For example, a 2020 car can’t have driven 50,000 miles, and the higher the “modelyear” (new vehicle), the more expensive the car is likely to be. We need to rationalize all the elements inside each option and balance it.
Attributes | Range/Level |
---|---|
Price | $ 10,000 to $50,000 |
Mileage | 10,000 to 70,000 |
Model Year | 2014 to 2020 |
Range | 200 to 400 |
Fuel Economy | 15 to 35 |
Powertrain | Gas/Electric |
Warranty | Yes/No |
Charging Station | No Promotion/Discounted/Free |
Sample Survey is listed below:
Option 1
Price: $ 10,000 Mileage: 24,000 ModelYear: 2017 powertrain: Electric range: 400 MPG: NA Warranty: Yes Charging Station Promotion: Free
Option 2
Price: $ 30,000 Mileage: 14,000 ModelYear: 2019 powertrain: Gas range: NA MPG: 25 Warranty: No Charging Station Promotion: NA
Option 3
Price: $ 30,000 Mileage: 14,000 ModelYear: 2019 powertrain: Electric range: 300 MPG: NA Warranty: No Charging Station Promotion: No Promotion
The data was collected from late November to early December 2021. At the beginning we have 335 participants took the survey and 213 eligible results were received. The survey results were then went through a data cleaning process to drop bad responses. The criteria we used for this process are:
Did the participant finish all questions?
Did the participant choose same options across all eight questions?
Since our survey questions are randomly picked for each participant, it is unlikely to choose same options for all 8 questions. This kind of behavior make us believe that the participant was not paying attention at all to our survey.
Participant could also randomly clicking through our survey to avoid being easily detected by the previous criteria. To identify such bad responses, we will use time spent during the survey to filter them out.
To determine whether participants were going to fast, we first calculated the total time spend on survey for each participant. Then we get the distribution of total time and identify a reasonable cut-off point. Participants who spend less time than the cut-off time will be identified and dropped.
To determine the cut-off point, our group members randomly clicking through the survey and record the time spent. After multiple tries, we found that it took about 2.5 minutes to finish if you are not paying attention to our questions at all. To include participants that are slightly higher than 2.5 minutes, we decided to use the 10% quantile, which is 2.8 minutes, as the cut-off point.
The first step drops 21, the second step drops 3 and the third step drops 19 participants. In total 43 participants were dropped, so out final dataset contains 170 participants. The summary statistics is shown in the following table:
#Summary Statistics
choiceData <- read_csv(here("data","choiceData.csv"))
summary_stat <- choiceData %>%
mutate(age=2021-yearOfBirth,
gender=if_else(gender=="male",1,0),
ev=abs(ev-2),
dc=abs(dc_resident-2),
carbuyer=abs(car_buyer-2)) %>%
distinct(respID,.keep_all=TRUE) %>%
select(age,gender,ev,dc,carbuyer) %>%
summarise_each(funs(mean=mean,
sd=sd,
min=min,
max=max))
df <- summary_stat %>%
gather(stat,val) %>%
separate(stat, into = c("var", "stat"), sep = "_") %>%
spread(stat, val) %>%
select(var,mean, sd, min, max) %>%
mutate(var=fct_recode(var,
"dc resident"="dc",
"car_buyer"="carbuyer",
"has_ev"="ev"))
df %>%
kbl() %>%
kable_styling()
var | mean | sd | min | max |
---|---|---|---|---|
age | 34.3647059 | 13.7478106 | 18 | 73 |
car_buyer | 0.3411765 | 0.4755051 | 0 | 1 |
dc resident | 0.3823529 | 0.4873977 | 0 | 1 |
has_ev | 0.0588235 | 0.2359892 | 0 | 1 |
gender | 0.4411765 | 0.4979946 | 0 | 1 |
The utility mode is constructed as the following.
\(U_{ij}=\beta_1x^{price}_j+\beta_2x^{fuel\_econ}_j+\beta_3x^{model\_year}_j+\beta_4x^{range}_j+\beta_5x^{mileage}_j+\beta_6\delta^{gas}_j+\beta_7\delta^{have\_warranty}_j+\beta_8\delta^{free\_cs}_j+\beta_9\delta^{discounted\_cs}_j+\epsilon_{ij}\)
where \(\delta^{gas}\), \(\delta^{have\_warranty}\), \(\delta^{free\_cs}\) and \(\delta^{discounted\_cs}\) are dummies for being gas car, have warranty, dealer provide free charging station and dealer provide discounted price for charging station respectively.
We have estimated a multinomial, mixed logit and create a subgroup based on whether participants have owned an EV before. The preference space results are shown in the following table
source("modeling.R")
mxl_result <- mxl_result %>%
filter(str_detect(vars,"sigma",negate = TRUE)) %>%
mutate(vars=fct_recode(vars,
"price"="price_mu",
"Mileage"="Mileage_mu"
))
results <- cbind(mnl_result,mxl_result[,2:3],sub_result[,2:3])
names(results) <- c("Vars","MNL", "MNL sd","MXL", "MXL sd","Has EV", "Has EV sd")
results %>%
kbl() %>%
kable_styling()
Vars | MNL | MNL sd | MXL | MXL sd | Has EV | Has EV sd |
---|---|---|---|---|---|---|
price | -0.08 | 0.004 | -0.11 | 0.008 | -0.09 | 0.019 |
fuelEconomy | 0.05 | 0.008 | 0.06 | 0.010 | 0.07 | 0.042 |
modelyear | -0.10 | 0.026 | -0.13 | 0.030 | -0.35 | 0.134 |
powertrain_Gas | -0.69 | 0.339 | -0.59 | 0.381 | 0.57 | 1.609 |
Warranty | 0.54 | 0.080 | 0.63 | 0.092 | 1.05 | 0.392 |
range | 0.00 | 0.001 | 0.01 | 0.001 | 0.01 | 0.004 |
Mileage | -0.02 | 0.003 | -0.03 | 0.004 | -0.03 | 0.014 |
ChargingStation_Free | 0.21 | 0.133 | 0.21 | 0.151 | 0.99 | 0.628 |
ChargingStation_Discounted | -0.05 | 0.135 | 0.02 | 0.152 | 0.41 | 0.653 |
WTP Simulation
The WTP model is computed and estimated in Preference Space and WTP Space. The data frames have been created for each WTP variable and then plotted. For the purpose of analysis, the baseline for ‘fuel economy’ is set to 15, the ‘model year’ is computed as how old the car is when compared to the model year 2021, the ‘price’ is scaled down to thousand dollars, and the ‘mileage’ is also scaled down by 1000. The Willingness to pay per thousand dollars for the attributes - “Range”, “Mileage”,”Fuel Economy”, “Powertrain Gas”, “Warranty”, “Model Year” “Free Charging Station” and “Discounted Charging Station” as follows,
!