Introduction

For the final component of the research, we shift our attention to customer enrollment in current community solar farms. By analyzing tenure length and the presence of an individual departing community solar farms, our research seeks to provide quantitative performance data to add depth to the previously collected survey data.

Given the incipient nature of community solar, this analysis provides an opportunity to quantify the rates of churn or default while controlling for household or demographic attributes. The primary goal of the analysis is to first describe the characteristics of the residents, the prevalence of default and/or churn rates, and any statistically significant differences between groups in default or churn rates. Secondarily, the analysis will seek to determine if enrollment in the community solar farm is associated with any measurable change in credit scores.

Customers in existing community solar projects are evaluated for their subscription and payment status on a monthly basis. Participation in community solar farms is voluntary, hence customers may exit the program at any point, often subject to a cancellation fee set out in the contract. Additionally, customers sometimes fail to pay the monthly solar farm subscription fee. Churn refers to the act of a customer exiting the solar farm, whereas default refers to the customer failing to pay the monthly subscription fee.

Methods

Monthly account level data was collected from two community solar projects from January 2020 to April 2022. Account level information includes monthly payment performance, for which churn and default triggers are captured, along with kWh solar attribute to each account, a tag for which farm an account belongs to, and a tag for if payment method was credit card or direct deposit (ACH).

Monthly, account level performance data is appended using Experian data including Experian’s Income Insight Score and VantageScore, a proxy for FICO scores with the same range of 0,850, along with demographic data such as gender, education, occupation, and homeownership status.

For the primary analysis, a logit model is used to determine likelihood of churn and/or default, taking into account an account’s length of tenure and various demographic and socioeconomic data available.

The secondary analysis is performed to measure the difference in credit scores for customers, to measure if any difference of statistical importance is observed over the length of the enrollment in the farm. Credit scores for each customer are measured in December 2020 and April 2022, offering an approximate pre- and post credit score measurement. A difference of means test (Welch’s t-test) is employed to measure the differences.

Descriptive Statistics

Account level summary statistics by solar farm are provided in Table 4. Note, only 6 instances of default were observed. However, of the 812 accounts, 118 or just over 14.5% of accounts, experienced churn.

Primary Analysis

Secondary Analysis

The secondary analysis sought to determine if a measurable change can be determined in participant credit scores (VantageScore).

The graph below shows boxplots of each VantageScore collected, at December 2020 and April 2022.

The histogram below plots the distribution of the scores by time period, and includes the standard scale and a logged comparison.

Using a Welch t-test, we measure the difference in means between the two groups. From the below, the means of the two groups are not statistically signifcant; therefore we cannot claim that credit scores for individuals in the solar farms were observed to have changed.

## 
##  Welch Two Sample t-test
## 
## data:  value by variable
## t = -0.92527, df = 1051.9, p-value = 0.355
## alternative hypothesis: true difference in means between group April 2022 and group December 2020 is not equal to 0
## 95 percent confidence interval:
##  -16.276750   5.845305
## sample estimates:
##    mean in group April 2022 mean in group December 2020 
##                    781.0366                    786.2523

Discussion

From the primary analysis, we saw that tenure was consistently associated with lower rates of probability of churning. This held constant throughout all five models, suggesting that higher tenure, controlling for demographic and socioeconomic variables such as education and income, corresponds to higher rates of retention in solar farms. This relationship is visualized in Graph A below.

Secondly, female account holders were less likely to churn compared to male account holders as seen in model 2. This effect is observed when controlling for homeownership status and VantageScore

Finally, no other variables were observed to have statistically significant effect on likelihood of churn. Interestingly, this was observed for income and VantageScore Different levels of income or credit scores were not observed with statistically different rates of churn. Further research should be done in the area of LMI retention in solar farm, but from our research we did not see a significant difference in probability of churn between either income or credit data.

In the Default model, as expected, such a low number of observations does not allow for any significant effects to be observed.

The logit curves below show the relationship between the distribution of both tenure and days late with the probability of churning. The first graph shows that the longer the tenure, the lower the likelihood. The second graph shows that the days of late payments are generally not associated with likelihood of churn.