Electric Cooking After the Trial: Three-Year Impacts, Perceptions, and Preferences

This pre-analysis plan covers a +3-year follow-up of an RCT in Goma (Democratic Republic of Congo) that randomized offers of free Electric Pressure Cookers (EPCs). This intervention started in August 2022. The data collection for the +3 year follow-up began in December 2025 and may extend beyond March 2026. This pre-analysis plan was developed prior to data collection and registered while data collection was ongoing, but before any analysis was conducted.
The core treatment consisted of a free EPC, a demonstration session, house visits, and a tailor-made cookbook. We designed two sub interventions to explore adoption mechanisms: a 20 kWh (~ $5) electricity transfer credited directly to the household electricity meter (the equivalent of about 80 EPC-cooked dishes, or 32 full meals) to test the hypothesis that a financial nudge overcomes (uncertainty about) the EPCs variable usage costs; and an environmental education session (presented by Virunga Park rangers) to test whether awareness about the peace and environmental effects of charcoal production may boost EPC use. To remind beneficiaries about this education session, they received a sticker on their EPC carrying the same message.

2022-08-01

2022-09-30

EPC usage
Charcoal reliance
Preference for EPC-like attributes
Perception accuracy

EPC usage
Average monthly electricity consumption in the 12 months prior to the survey: Monthly electricity consumption proxied by monthly electricity purchases (in USD). The outcome will be computed using the universe of electricity transactions from all VE clients. Transactions will be summed for every month to obtain monthly consumption.

Charcoal reliance
A continuous variable indicating monthly charcoal expenditure in USD. A recall question will be asked during the survey: “How much did your household spend on charcoal last month?”. Values reported in Congolese Francs will be converted to values in USD.

Preference for EPC-like attributes
Treatment heterogeneity in AMCEs for EPC-like attributes.We expect the main cook in treated households to have higher estimated Average Marginal Component Effects (AMCEs) for lower operating cost, forest protection, smoke reduction, and time saving. We will conduct a joint test of treatment heterogeneity across these attributes.

Perception accuracy
Perception Bias Index = mean(|N−N*|/σ, |C−C*|/σ, |E−E*|/σ, |T−T*|/σ ). From the main cook’s survey responses, elicit perceived weekly EPC uses (N), monthly charcoal saving (C), monthly electricity expenditure increase (E), and daily time saved (T). For each metric, we will compute standardized absolute deviations from the measured study average(*). The index is the mean of the four standardized deviations. Lower = more accurate.

EPC functionality status
Off-peak EPC use
Within-meal stacking
WTP for EPC-like attributes
Within-household heterogeneity in attribute preferences
Overall attribute preferences
Preference for forest protection
Perception bias for each metric separately
Within-Household Perception Gap

EPC functionality status
A dummy variable taking 1 if EPC is functional at the time of the survey, and zero otherwise. Derived from a survey question on EPC functionality: “What is the current status of your cooker (EPC)?”

Off-peak EPC use
Off-peak-share = (# EPC heat events outside peak time) / (total EPC events). For SUM subsample; constructed based on timestamped SUM logs. The peak window is defined using VE’s hour-of-day load profiles as the 2–3 consecutive hours with the highest average system load. Off-peak measures use all SUM activity outside this window.

Within-meal stacking
Share of meals with overlapping EPC–charcoal heat events. For SUM subsample; constructed based on timestamped SUM logs.

WTP for EPC-like attributes
Treatment heterogeneity in WTP for EPC-like attributes. We will translate estimated attribute effects into implied WTP for EPC-like attributes: lower operating cost, forest protection, smoke reduction, time saving, and lower cooking flexibility. We report WTP using (i) a common price slope pooled across treatment arms, and (ii) treatment-specific price slopes that allow price sensitivity to differ between treated and control households.

Within-household heterogeneity in attribute preferences
Within-household heterogeneity in AMCEs and WTP. We will explore whether AMCEs and WTP for attributes vary by cooking responsibility and financial roles in the household.

Overall attribute preferences
Estimated AMCEs. We expect that households will positively value forest protection, smoke reduction, time saving, and cooking flexibility; while purchase price and operating cost negatively impact choice for the cooking device.

Preference for forest protection
Estimated AMCEs and WTP We will explore whether preferences for forest protection are larger for households that received the environmental message in the original RCT

Perception bias for each metric separately
Perception Bias = |X−X*|/σ For each metric (N, C, E, T), compute the standardized bias. Lower = more accurate.

Within-Household Perception Gap
Within-household perception gap, by metric (pairwise). For each metric (N, C, E, T), and pair of roles (e.g., cook vs electricity payer), compute the difference in standardized biases. Lower = more aligned perceptions within the household.

The core treatment consisted of a free EPC, a demonstration session, house visits, and a tailor-made cookbook. Sub interventions consisted of a 20 kWh (~ $5) electricity transfer credited directly to the household electricity meter; and an environmental education session (presented by Virunga Park rangers), with a reminder in the form of a sticker on their EPC carrying the same message.

In this follow-up study, more than three years after the EPC distribution, we assess whether the impacts on EPC use and fuel substitution, measured at 12 months after the distribution, persist after the three-year mark.

Additionally, this study adds two modules, administered separately to the main cook and their spouse:
1. A conjoint experiment to identify preferences and willingness to pay for stove attributes and to test whether EPC exposure shifts these preferences;
2. A perceptions module, to benchmark beliefs about EPC use, time and money savings against objective study averages, and analyse whether these beliefs are affected by EPC exposure, and by intra-household financial roles—who pays for charcoal and electricity.

Finally, in a subsample, Stove Use Monitors (SUMs) measure on/off-peak EPC use and characterize stacking (using EPC and charcoal in the same meal) versus substitution (replacing charcoal with electricity).

Randomization done in office by a computer

City neighbourhoods: households within 150 m were grouped into “neighborhoods” and all households in a neighborhood received the same assignment

Yes

We randomized the core treatment across 387 neighborhoods; however this 3-year follow-up only treats with about half of the sample size, divided across 198 neighborhoods.

In the original RCT, Virunga Energies distributed EPCs to a random sample of 1,034 residential clients in two waves (Wave 1: August 2022; Wave 2: August 2023). The present +3-year follow-up focuses on Wave 1 (N = 762).

N = 762; 462 treatment, 300 control. The 462 treatment observations were assigned to 'EPC only' (N=89), 'EPC + nudge' (N=137), 'EPC + transfer' (N=122), and 'EPC + nudge & transfer' (N=114)

For EPC usage, proxied through monthly electricity consumption, we will have data for 46 time periods (6 months before and 40 months after the distribution), including 28 months since the end of the original RCT. Focussing on the 12 months preceding the new survey, Monte Carlo simulations that account for cluster-level dependence and within-household serial correlation show that the study is fully powered to detect an ITT effect of the same magnitude as that estimated in the original RCT, and has a power of approximately 95% to detect an effect equal to half of the original effect size. For charcoal spending, Monte Carlo simulations calibrated to the original panel data and cluster structure similarly indicate that the study is fully powered to detect an ITT effect of the same magnitude as that estimated in the original RCT, and has approximately 95% power to detect an effect equal to one-third of the original effect size.

Pre-analysis Plan for 'Electric Cooking After the Trial: Three-Year Impacts, Perceptions, and Preferences'