Analysis of Sales of Electric Vehicles

Github Link

University of Washington  Jan 2020 - Mar 2020

Electric Vehicles (EV) has been growing in the last decade. Understanding the influential factors contributing to the increasing EV adoptions is critical not only to the car manufacturers but also the government policy makers in supporting this trend. This is our aim of this study to determine the factors that affect the EVs adoption.There are potentially dozens of factors that could impact the EVs adoption. Although, performing a study with all the potential factors will yield a comprehensive picture of EV adoption, given the limitations of our current dataset, we have constrained our study and analysis to question the factors pertaining to the available dataset.There are three main questions for our current study:

  1. Does the electricity price of a state affect the count of registered EVs in the state?
  2. Do economic indicators have an effect on this EV adoption at the US level?
  3. Do economic indicators have a similar effect on EV adoption in WA?

Data Collection

Our study is conducted on the following independent datasets:

  • EV registration count in the US and in the Washington State in 2017
  • Electricity price of different states in US in 2017
  • Economic indicators dataset for the US states and WA counties in 2017
  • Population dataset for the US states and WA counties in 2018
Since our study assesses EV adoption at the US level with a focus on WA state. Therefore, we group our datasets into two separate group levels:
  1. US-Level
    1. Electric Vehicle Registration Counts by State: Each row of the dataset represents one US state with its total approximate count of EV registration in 2017.URL: https://afdc.energy.gov/data/10962. Source: IHS Markit light-duty vehicle registration in 2017
    2. Electricity Price by State:Each row of the dataset represents one US state with its average retail electricity price (cents/kWh) in 2017.URL: https://www.eia.gov/electricity/state/.Source: U.S. Energy Information Administration
    3. Economic Indicators by State:Each row of the dataset represents one US state with its economic indicators: unemployment rate, GDP, and per capita personal income.URL: https://fred.stlouisfed.org/categories/27281. Source: Federal Reserve Bank of St. Louis
    4. Population by State:Each row of the dataset represents one US state with its population estimates in 2018.URL: https://www.census.gov/quickfacts/geo/chart/US/PST045218. Source: U.S. Census
  2. County-Level
    1. Electric Vehicle Population by WA:Each row of the dataset represents one unique EV Vehicle Identification Number (VIN) and its information (city, state, model year, electric range, sale date, etc.) within the Washington state.URL: https://data.wa.gov/Transportation/Electric-Vehicle-Title-and-Registration-Activity/rpr4-cgyd. Source: WA Department of Licensing
    2. Economic Indicators by WA counties:Each row of the dataset represents one WA county with its economic indicators: unemployment rate, GDP, and per capita personal income.URL: https://fred.stlouisfed.org/categories/30336. Source: Federal Reserve Bank of St. Louis
    3. Population by WA counties:Each row of the dataset represents one WA county with its population estimates in 2018. URL: https://www.census.gov/quickfacts/fact/map/WA/PST045218. Source: U.S. Census Bureau

Experiments

There were three questions that we were interested in exploring regarding the sales of EVs in the US.

Question 1: Is there a relationship between the electricity prices of a state and the number of registered EVs in the state? Further, is there a positive or a negative relationship between the two?Since both count of registered EVs and electricity prices are quantitative variables, we used linear regression to answer the question. Furthermore, the coefficients of the linear regression tell us about the strength of the relationship and whether there is a positive or negative effect.We used the electricity and registered EV counts data from 2017 for 50 US states (+ District of Columbia) for the analysis. Since the data for the states is highly dependent on the demographic population, we used the population data for the states from 2018, to standardize the registered EV counts.

Question 2: Do the economic factors of a state have an effect on the number of EV sales in a state? Which factors have the most significant effect? To answer this question, we studied the following indicators at a state level:

  • GDP
  • Per capita personal Income
  • Unemployment Rate
To study the effect of each of these quantitative indicators we used linear regression. We also analyzed whether there was a significant interaction between any of the indicators.We used the economic indicators and registered EV counts data from 2017. We used the population data for the 50 states (+ District of Columbia) from 2018, to standardize the GDP and EV counts since they are dependent on the size of the state.

Question 3: Do the same economic factors (from Question 2) have a similar effect on the EV sales in the counties of Washington State? To answer this question, we studied the following indicators at a county level:
  • GDP
  • Per capita personal Income
  • Unemployment Rate


To study the effect of each of these quantitative indicators we used linear regression. We used the economic indicators and EV sales data from 2017. We used the population data for 39 counties in Washington from 2018, to standardize the GDP and EV counts since they are dependent on the size of the counties.
While answering all three questions, we used linear regression and thus we will state the assumptions of linear regression before we proceed.
  1. Independent observations: Based on the data collection methods, the data observations in all our datasets are independent.
  2. Constant variance of residuals: This will be observed based on the residual plots of the models
  3. Normality or large sample size : For all three questions, our data was limited to the number of demographic regions (50 states in the US + District of Columbia, 39 counties in Washington). Thus we needed to check for the normality and residual assumptions for our data. While analyzing our response variable, we found that the data was not normally distributed. On trying a few different transformations, we found that the log-transform made the data quite normal. This was affirmed by the residual plots
  4. Linearity of the data points: Although linearity is not a strict requirement for fitting a linear regression, it is good to have and we can see from the scatterplots that the relationship between the predictors i.e. electricity, per capita personal income, etc., have a linear relationship between the variables

Result and Conclusion

Question 1 Conclusion: We reject the null hypothesis that βElectricity_Price = 0 at 5% significance level. Thus we conclude that there is a significant effect of Electricity Price on the EV Population. Based on the coefficient of regression, we see that the relationship is positive i.e. for a difference of 1¢/kWh of electricity prices there is an 11.8% increase in the EV sales per 100,000 people. However, we expected a drop in the EV sales with an increase in the electricity prices. The data shows us otherwise. While trying to examine this result we found that there is a positive correlation between the Per Capita Personal Income and the Electricity Price of a state. Also there is a positive correlation between the Per Capita Personal Income and EV sales of a state. Based on this we infer that states with higher electricity prices tend to have a higher Per Capita Personal Income. Per Capita Personal Income has a significant effect on EV sales which makes sense since people would buy more EV cars when they have a higher purchasing power. Thus, electricity prices also seem to have a positive effect on the EV sales. The correlation between Per Capita Personal Income and Electricity Prices creates a confounding effect.

Question 2 Conclusion:PCPI and Per Capita GDP have an effect on the EV Sales by Population of states. Unemployment Rate does not. There are no significant interactions between the indicators. We use WLS method for fitting the models due to the limitations of size and high leverage points.For a difference of $1000 in the Per Capita Personal Income, the number of registered EVs per 100,000 people is 4.2% higher.

Question 3 Conclusion: PCPI and Unemployment Rate have a significant effect on the EV Sales by Population of counties. Per Capita GDP does not. Note that at a country level, PCPI and Per Capita GDP were significant but the Unemployment Rate was not.

From the WLS coefficients we conclude:

  • For a difference of $1000 in the Per Capita Personal Income, the number of registered EVs per 100,000 people in Washington State is 6.9% higher.Compare this to the country level where % difference of the registered EVs per 100,000 people was 4.2%.
  • For a 1% change in Unemployment Rate, the number of registered EVs per 100,000 people in Washington State is 26.4% lower.Compare this to the country level where there is no significant effect of Unemployment rate on the EV sales.