There is an App for that: A randomized experiment on durables and digital financial tools to increase LPG use in Ghana

Our study is designed around two interventions: a free LPG stove and GasPay, a digital financial platform specifically designed to save for and purchase LPG. Based on these two interventions, we have four study arms:
(1) Free LPG stove and GasPay: Participants offered a free LPG stove and access to and encouragement/support to enroll in GasPay, a new, phone-based, mobile money platform where users can save for or finance their LPG purchases
(2) Free LPG stove only: Participants offered a free LPG stove
(3) GasPay only: Participants offered access to and encouragement/support to enroll in GasPay
(4) Control: Participants offered neither a free stove nor access to GasPay

2025-06-16

2026-05-31

Our primary outcomes are LPG adoption, LPG expenditure, charcoal expenditure, firewood expenditure, ownership of any LPG stoves, and ownership of any LPG cylinders.

LPG adoption: This will be a binary variable regarding whether or not the household is using any LPG constructed at every spotcheck. We will construct this by triangulating between participant’s reporting on whether they have spent any amount on LPG in the past month and whether there is any LPG in any cylinder at any of the spotchecks. That is, this variable will be 1 if they have spent on LPG in the past month or if there is LPG in their cylinder.

LPG expenditure: This dependent variable will be a continuous variable regarding the amount the user spent on LPG in the past month.

Charcoal expenditure: This dependent variable will be a continuous variable regarding the amount the user spent on charcoal in the past month.

Firewood expenditure: This dependent variable will be a continuous variable regarding the amount the user spent on charcoal in the past month.

Ownership of any LPG stoves: This will be a binary variable regarding whether or not the household owns any LPG stoves or cylinders.

Ownership of any LPG cylinders: This will be a binary variable regarding whether or not the household owns any LPG stoves or cylinders.

Our secondary outcomes are LPG, charcoal, and firewood purchase frequency and categories of fuel users. We conduct additional investigations into time use, financial agency, mobile money use, digital financial literacy, intra-household conflict.

LPG purchase frequency: This dependent variable will be a continuous variable regarding the amount of times in the past month the user purchased LPG.

Charcoal purchase frequency: This dependent variable will be a continuous variable regarding the amount of times in the past month the user purchased charcoal.

Firewood purchase frequency: This dependent variable will be a continuous variable regarding the amount of times in the past month the user purchased firewood.

Categories of fuel users: We will create five distinct binary variables for exclusive LPG users, LPG - charcoal users, LPG - firewood users, biomass only users, and other fuel users. These will be constructed from survey questions on fuels used, expenditure on specific fuels, and observations of present, warm, and clearly used stoves.

Time use: This will be five continuous variables documenting the time in the last week spent on (1) working for income/pay , (2) domestic work including cooking fuel collection , and (3) leisure.

Financial agency: This will be a binary variable indicating whether the respondent has independent finances from their spouse/household members, has control over their income, and can purchase daily needs without obtaining permission from their spouse or other household members

Digital financial literacy: This will be a binary variable indicating that the respondent has their own mobile money account and can use the mobile money account themselves.

Mobile money use: This will be a binary variable indicating that the respondent has their own mobile money account, uses it more than “almost never”, and has performed one of a list of mobile money activities.

Intra-household conflict: This will be a binary variable that will indicate whether the respondent has quarreled with a spouse or family member over how they or the respondent was spending money.

Through a randomized control trial (RCT) in Techiman, Bono East, Ghana, we will estimate treatment effects on LPG uptake and displacement of polluting fuels, across three treatment arms:
Free LPG stove and GasPay: Participants offered a free LPG stove and access to and encouragement/support to enroll in GasPay, a new, phone-based, mobile money platform where users can save for or finance their LPG purchases
Free LPG stove only: Participants offered a free LPG stove
GasPay only: Participants offered access to and encouragement/support to enroll in GasPay
Control: Participants offered neither a free stove nor access to GasPay

We stratified the study arms across two distinct samples:
Random sample: Participants selected randomly from the most recent KHRC’s Health and Demographic Surveillance Survey (KHDSS)
LPG user sample: Participants targeted based on reporting LPG use in KHDSS. These participants will not be included in the stove treatment groups (arms 1 and 2, above).
After sampling from KHDSS data, we enroll roughly 300 participants in each sample arm (see power calculations). Our methods include household surveys and financial and LPG purchase data from GasPay. We conduct a baseline survey that covers mobile money use, intra-household dynamics, and fuel use, two rounds of spotchecks throughout the follow-up period that covers mainly fuel use, and a final endline that collects additional information on time use, financial agency, mobile money use, digital financial literacy, and intra-household dynamics.

Not available

Treatment assignment was randomly assigned on a computer through R v. 4.3.2.

Our unit of randomization is individual households.

No

1800

1800 households with 4 different responses = 7,200 observations

300 non-LPG Stove and GasPay; 300 non-LPG GasPay only; 300 Stove only; 300 non-LPG control; 300 LPG GasPay only; 300 LPG control

We conducted power calculations based on available data and literature. Our primary outcome of interest was LPG adoption (e.g., any LPG use). We test the null hypothesis that LPG uptake is unaffected by treatment on average. Our focus is on detecting policy-relevant differences in uptake that would justify the intervention at scale. We conducted power calculations using a two-proportion z-test, implemented in R with the pwr package. Since treatment is assigned at the individual household level, observations can be treated as independent. We assume a one-sided test at a 5% significance level, reflecting the expectation that the intervention will increase LPG uptake. We defined a minimum detectable effect size as the difference between control and treatment group rates of “LPG adoption’ by arm. Previous quasi-experimental research on India’s below-the-poverty-line household policies indicates that providing free stoves may increase LPG uptake (defined similarly as a binary indicating any amount of LPG) by 8% points (Gill-Wiehl et al. 2023). In our feasibility study, we found that 10% of participants, all of whom previously did not use LPG, exchanged some amount of LPG through GasPay. Thus, we ensure we have power for both interventions. We calculated the necessary sample size to detect a 10% increase in our binary “LPG adoption” outcome, assuming 13% LPG use at baseline (sd=p(1-p)=0.13*.87=0.34), at a single spotcheck with 80% power (with an alpha of 0.05 and a beta of 0.2) to be 300 participants in each sample arm.