Workplace Charging Network Expansion

Last registered on November 15, 2024

Pre-Trial

Trial Information

General Information

Title
Workplace Charging Network Expansion
RCT ID
AEARCTR-0014731
Initial registration date
October 30, 2024

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
November 15, 2024, 1:02 PM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

Affiliation
The University of Melbourne

Other Primary Investigator(s)

Additional Trial Information

Status
In development
Start date
2024-01-01
End date
2025-04-01
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
We provide estimates of the impact of a workplace EV network expansion on driver charging behavior by leveraging the staggered installation of new chargers at parking lots. Our analysis couples data on charging behavior around the networks’ expansion with a generalized difference-in-differences empirical strategy.
External Link(s)

Registration Citation

Citation
Tebbe, Sebastian. 2024. "Workplace Charging Network Expansion." AEA RCT Registry. November 15. https://doi.org/10.1257/rct.14731-1.0
Experimental Details

Interventions

Intervention(s)
The UCSD charging network is currently undergoing a substantial expansion, with increases in infrastructure projected between December 2023 and June 2024, and again from November 2025 to January 2026. According to the Transportation Office, the number of Level 2 chargers is expected to increase from approximately 331 Level-2 charging stations, including 250 ChargePoint and 72 PowerFlex stations, in December 2023 to around 480 by June 2024.We exclude eight Level-2 chargers operated by SemaConnect. Following this period of rapid growth, the network was anticipated to remain relatively stable until November 2025, when the expansion will accelerate sharply, reaching an estimated 1,250 chargers by January 2026. However the initial phase of expansion is currently behind schedule and there may be an additional expansion in Q4 of 2024. While the posted expansion plans available at [https://transportation.ucsd.edu/initiatives/drive-electric/projects.html] provide a general timeline (with projections from December 14, 2023, to February 23, 2024), they are not fully precise due to phased project deliveries and varying award timelines. Thus, our analysis relies on internally tracked charger expansion dates, which accurately reflect the exact dates when old chargers are decommissioned and replacements are brought online. This internal data allows us to align planned and actual operational dates and identify when chargers are taken out of service based on the termination of their reporting histories.
Intervention Start Date
2024-12-01
Intervention End Date
2025-04-01

Primary Outcomes

Primary Outcomes (end points)
Net benefits of charger installation, and how drivers react to a supply side shift in chargers.

This will be measured by 6 outcome variables for each driver, both the log and level. Session duration indicates the total time plugged in, irrespective of charging, and is the sum of charging duration (when the EV is plugged in and charging) and idle duration (when the EV is plugged in but not charging).

1. Share of total charging done on campus

2. The number of campus sessions initiated

3. Energy consumed

4. Session duration

5. Charging duration

6. Idle duration.
Primary Outcomes (explanation)
A driver's share of total charging on campus is calculated as the ratio of energy consumed from campus charging to the expected energy consumed from total driving. The latter is calculated per driver-reported odometer readings and Department of Energy vehicle efficiency estimates. Odometer readings depend on responses from divers within our on campus EV club; because readings are not reported by all drivers we have a sample size smaller than the group size. The remaining five variables are collected directly from charging session data collected by chargers and encompass all drivers on campus who charge. However our sample is largely restricted to drivers within our EV club where we can observe demographic variable in addition to all of their on campus charging behavior. We can also observe the time for each of these five remaining variables allowing us to also look for changes in timing of sessions.

Our primary interest is how the expansion of an EV charging network changes drivers' charging behavior as well as the policy implications of such changes. “Behavior” has three main components: 1) decisions about where to charge, measured as the share of charging done on campus; 2) when on campus, decisions about when to charge, measured as the hours of the day over which charging occurs; and 3) the “depth” of charging sessions, measured as the energy consumed relative to the EV's battery capacity. These three components of charging behavior have implications for the greenhouse gas emissions from charging, cost recovery to the EV network host, and congestion on the electric grid.

In California, daytime charging is associated with substantially fewer emissions compared to overnight charging as well as lower grid congestion. During the night, California electricity is derived primarily from natural gas power plants, while solar generation peaks during the daytime. Yet the majority of EV charging nationally and in California occurs at home overnight. Shifts toward greater campus charging, induced for example by installing new chargers, therefore generate social benefits through avoided damages from emissions. As with any such positive externality, a common policy response is to subsidize the good in question.

Workplace charging generates revenues for the site host through the sale of electricity and, in California, through the California's Low Carbon Fuel Standard (LCFS) program. At the same time, shifting charging to hours in the day when site's electric load is low reduces the site's demand charges. The size of these private benefits depends on the total charging done at the site as well as the timing and depth of charging. Drivers who recoup energy via fewer longer charging sessions (compared to more frequent shallower sessions) increase overall network efficiency because they minimize the time that vehicles sit idle in EV stalls and prohibit others from charging.

Lastly, shifting charging to hours in the day when system electric demand is low reduces grid congestion both locally, potentially deferring costly capacity upgrades, as well as on the transmission system, thereby reducing congestion costs in the wholesale electricity market.

Our interest in the effects of network expansion therefore cover both the intensive and extensive margins of charging. The intensive margin measures if drivers who currently charge on campus increase their use of the network post-expansion; the extensive margin measures if drivers who did not previously charge on campus are induced to do so. From these shifts in behavior, we can calculate social benefits derived from avoided emissions and the social cost of carbon. Furthermore, we can calculate private benefits to the network host using equivalent annual costs of installing and maintaining charger infrastructure to calculate the net annual benefits/costs of the expansion.

Secondary Outcomes

Secondary Outcomes (end points)
Charger network congestion and Substitution effects
Secondary Outcomes (explanation)
Our secondary interest are the general equilibrium effects of expanding the charger network. Adding chargers decreases the cost to drivers, in the short term, of finding an available charger. However, expansion may induce more drivers to charge on campus, negating the short-term benefits of lower congestion. To measure network congestion, and how it evolves over time, we will calculate network utilization, measured hourly as the fraction of chargers in use across the network.

We are also interested in whether changes in charging behavior are driven by substitution effects. A driver who charges often in multiple lots, for example, may shift their charging toward a treated lot they use, even if it is not near their original preferred charging lot, which does not increase total campus charging.

Experimental Design

Experimental Design
This paper takes advantage of a natural experiment—the staggered roll-out of EV chargers on the UCSD campus. We also provide e-mail notifications to EV drivers on campus that enumerate the number and location of newly installed chargers sent once per month.
Experimental Design Details
Not available
Randomization Method
As a natural experiment, we do not directly employ a randomization method. However, we show that the roll-out of new chargers followed requirements in grants that partially funded the chargers, as well as on engineering and technical feasibilities, rather than on the basis of demand for charging. We therefore apply a difference in difference (DiD) method common in natural experiments. The technical details of the method, its assumptions, and how we meet them are covered in the attached pre-analysis plan.
Randomization Unit
We use two levels of randomization—by driver and by parking lot—based on pretreatment charging behavior.

At the driver level, we analyze the subset of drivers who charge primarily at a single lot and a second sample that inlcudes the remaining drivers who spread their charging activity across multiple lots. (We observe every campus charging session initiated by drivers and thus know where and how they charge.) Based on charging data collected before the roll-out \parencite{garg_electric_2024}, 60% of our EV drivers charge at a single lot at least 90% of the time. Our first driver-level analysis is on these single-lot drivers. These drivers are counted as treated if and when their primary lot is treated, ; otherwise they are untreated. The behavioral responses of these single-lot drivers form a good analog for EV network expansion by smaller institutions with only a single parking lot.

Our second treatment includes drivers who charge at multiple lots (i.e., the remaining 40% of drivers who charge less than 90% of the time at any single lot). We will utilize a continuous or dose-level version of treatment in which a driver's treatment dose is a function of the lots they charge in ttimes the number of new chargers received by that lot. For example, a driver who charges 50% of the time at two lots will be counted as gaining 5 new chargers if one of those lots gains 10.

At the lot level treatment, we analyze changes in the average of our outcomes of interest for each lot. This method has the disadvantage of reduced power, but the advantages of a cleaner mapping of treatment, as well as enabling us to explore the spatial nature of spillovers from other nearby garages. The lot averages include information from drivers who We include more details about each analysis in our pre-analysis plan.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
Treatment will naturally be assigned at the lot level. There are approximately 40 lots on the UCSD campus, of which we have reliable data on 35 with have approximately 1060 drivers.
Sample size: planned number of observations
Our sample consists panel data of 1060 drivers, across 35 lots.
Sample size (or number of clusters) by treatment arms
530 control/530 treatment
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
We could find no similar papers with a similar intervention (network expansion) to calculate the standard deviation. However, we do have a paper that used the same population of drivers focused on the same outcome variables (Garg, 2024). Using data from that paper, we have the following minimum possible effect sizes for our outcomes of interest: 1. Share of total charging done on campus: 3.6% 2. The number of campus sessions initiated: 0.093 sessions 3. Energy consumed : 1.19 kWh 4. Session duration 16.6 minutes 5. Charging duration 13.4 minutes 6. Idle duration. 10.15 minutes
IRB

Institutional Review Boards (IRBs)

IRB Name
UC San Diego Office of IRB Administration
IRB Approval Date
2023-06-01
IRB Approval Number
805222