Context of cooking energy use in Nairobi and Dar es Salaam

There are key differences between Nairobi and Dar es Salaam that inform interpretation of the comparative analysis of cooking fuel use in these two large and important East African capitals. In 2019, average gross annual income per capita in Nairobi County was 5,497 USD, compared to 1,941 USD in the Dar es Salaam Region (assuming 1 USD = 2,333.3 TZS, and 1 USD = 108.5 KS) [51,52]. Despite having similar urban use of clean fuels in 2000 (4% in urban Kenya and 2% in urban Tanzania), adoption in urban Kenya has far outpaced that in urban Tanzania over the past twenty years (24% in urban Kenya and 7% in urban Tanzania) (S1 Fig) with apparent rapid changes in the primary cooking fuel identified by households in both cities between 2015 and 2020 [53]. Over the same period, the LPG market has also been expanding in both cities (in Nairobi, reported primary LPG use was 40% in 2015 and 65% in 2020; in Dar es Salaam, reported primary LPG use was 12% in 2015 and 36% in 2020), with the proportion of households using LPG more than doubling in Dar es Salaam even as charcoal dependence remains high (in Nairobi, reported primary charcoal use was 5% in 2015 and 3% in 2020; in Dar es Salaam, reported primary charcoal use was 76% in 2015 and 60% in 2020) (S2 Fig) [54–56]. In both locations and in other similar settings, most households perceive that electricity is too expensive to use for cooking purposes [57].

Policies that affect cooking fuel prices have likely played some role in influencing these trends. The pro-clean cooking policy stance of the Kenyan government, and the dynamism of the private sector response to rising demand for clean fuels, for example, are well known [58]. Throughout the 2000’s, Kenya eliminated excise duties on kerosene, facilitating adoption of that fuel at the expense of solid fuels [59]. In 2016, Kenya zero-rated the value-added tax (VAT) on LPG, and introduced subsidized access to electricity for low-income households [60]. The Government-led Mwananchi gas project launched in 2017 had an ambitious target of increasing nation-wide LPG penetration from 10–70% in three years, by subsidizing purchases of LPG canisters and stoves by low-income households [61]. That project was suspended in 2018 owing to many issues, however, including high rates of LPG cylinder defects, poor targeting of beneficiaries, and low preparedness of the project implementer, the National Oil Corporation of Kenya, for the distribution and refilling of cylinders [62]. In keeping with its 2030 target of 35% LPG cooking fuel adoption [58], though, the Kenyan government’s LN 121 reforms of 2019 included better enforcement against illegal refilling of LPG cylinders, enhanced safety protocols, and creation of a structure of single LPG cylinder brand ownership that unified and overcame the opposition of majority members of the Energy Dealers Association (EDA, a trade group comprising small-scale LPG marketers) [62]. As a result of these reforms, new market entrants like Proto Energy have been able to drive LPG prices down in the Nairobi market, by about 25%, and these effects have begun to spread to other parts of Kenya as well. Finally, though it was criticized as regressive and ineffective, a ban on the production and transportation of charcoal into Nairobi was introduced in 2018, which increased the effective price of charcoal [63–65].

In contrast, while the Tanzanian government has long embraced similar goals for households transitioning to modern fuels (in its national energy policy of 1992, amended in 2003 and 2015) [66], it has taken fewer specific policy actions to support those goals, and most urban households continue to use charcoal as their main cooking fuel. In 2008, the government did exempt LPG stoves and fuel from VAT and excise duties in order to encourage uptake [67]. In both Nairobi and Dar es Salaam, the VAT is not collected on charcoal transactions given the informality of this industry, though various other royalties and license fees are collected from producers and transporters in Tanzania [68]. Kenya and Tanzania also differ substantially in their LPG distribution and storage infrastructure investments. The capacity of Kenya’s existing supply chain infrastructure is adequate to meet its 2030 demand: 92% of Kenya’s LPG comes through two terminals at the Mombasa port; the publicly-owned terminal’s storage capacity is 3,000 metric tons and the privately-owned terminal’s bulk storage capacity is ~26,000 metric tons and the temporary floating facility’s capacity is 14,000 metric tons [62]. Tanzania, on the other hand, has historically had lower storage capacity (8,050 metric tons as of 2016, though two new facilities constructed and commissioned as of 2019 added about 9,000 metric tons) [58].

Energy use characteristics of sampled households

In 2019, we interviewed 354 households in four informal settlements in Nairobi (we focused on lower income areas in Nairobi since these are the areas where the city’s population continues to rely on polluting fuels [69]), and in 2020 we interviewed a representative sample of 1,100 households (largely residing in informal settlements) across Dar es Salaam (S3 Fig). In the Materials and Methods section, we elaborate on our sampling strategy. A detailed summary of the characteristics of sampled households in both locations is presented in S1 Table. Here, we specifically describe the energy profile of the sampled households. In the Nairobi sample, LPG is the most common cooking fuel (54%), followed by kerosene (29%) and charcoal (11%). In Dar es Salaam, charcoal is the most common cooking fuel (61%), followed by LPG (32%) and kerosene (4%). In Nairobi, fuel procurement times for all major cooking fuels are similar, ranging from 10–14 minutes per purchase. Average reported daily fuel collection times in Dar es Salaam are much higher than in Nairobi, with firewood collection time being the highest (44 minutes per trip) and kerosene collection time being the lowest (24 minutes). All households in the Nairobi sample have electricity, and electricity access is slightly lower in the Dar es Salaam sample (85%). Households in the Nairobi sample are of slightly higher socio-economic status than those in the Dar es Salaam sample. Per capita monthly expenses are 94 USD in Nairobi, compared to 78 USD in the Dar es Salaam sample, and roughly 51% of the Nairobi sample had completed secondary schooling, compared to 33% in Dar es Salaam.

Primary cooking fuel choices in response to price increases

For each of the three main cooking fuels in Nairobi (charcoal, kerosene and LPG) and Dar es Salaam (firewood, charcoal and LPG), we assessed demand over randomly-assigned price increases that ranged from 25% to 200% (S2 Table presents the full range of prices offered for the three fuels in both cities). The derived demand curves from responses to initial bids show a mostly linear relationship between WTP for cooking fuel and price (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Demand graph for cooking fuels in Nairobi and Dar es Salaam (initial bids only).

Note: This figure shows the percentage of households in the Dar es Salaam and Nairobi samples reporting that they would continue using their primary cooking fuel when faced with a given initial price increase. Price increases are randomly assigned across survey respondents and baseline prices are converted into USD for comparison in this figure. The initial bids in the price increases and willingness to maintain use responses are presented here.

https://doi.org/10.1371/journal.pstr.0000077.g001

Among respondents in Nairobi that use charcoal as their primary cooking fuel, nearly half the respondents were willing to maintain their primary fuel use under the lowest price increase of 25%, but only 10% were willing to pay 200% more for the fuel. The WTP probabilities among primary kerosene-using respondents were similar: ranging from 62% to 16% for these initial lowest and highest bids, respectively. For primary LPG fuel respondents, the range of WTP probabilities was somewhat higher, dropping from 93% to 30%. Regression analyses further show that the price elasticities of maintaining use of each of these primary fuels are somewhat similar, ranging from -0.5 to -0.7 (Table 1, columns 1, 4, and 7). Yet, controlling for fuel stacking (i.e., use of multiple cooking fuels), which represents an important strategy for coping with high fuel costs or unreliable fuel alternatives [70,71], adds important nuance to these findings. Specifically, we find that accounting for stacking leads to higher estimates of the price elasticity for charcoal and LPG (Table 1, columns 2 and 8), while kerosene use appears less price-sensitive (Table 1, column 5). Primary LPG-using households are 13 percentage points more likely to switch away from that fuel when they already use other cooking fuels (Table 1, column 8). Moreover, the higher the proportion of LPG consumed in total cooking fuel use, the more likely a household is to maintain use of this option (Table 1, column 9). In S3 Table, we further explore the determinants of cooking fuel stacking.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Probability of maintaining primary cooking fuel use in Nairobi and Dar es Salaam: average marginal effects.

https://doi.org/10.1371/journal.pstr.0000077.t001

For primary charcoal users in Dar es Salaam, 70% were willing to pay the lowest increase, and 30% the highest price (Fig 1). For primary firewood users, half of the respondents facing the lowest bid level were willing to pay, and only 14% were willing to pay the highest price. Finally, for LPG 72% of primary LPG users were willing to continue using LPG at the lowest price increase, which dropped to 27% willing to continue at the highest price. Unlike in Nairobi, the regression analyses indicate very different price elasticities for maintaining use of charcoal (-1.1) vs. LPG (-0.3) in Dar es Salaam (Table 1, columns 10 and 13), and controlling for fuel stacking does not substantively alter these estimates (Table 1, columns 11, 12, 14 and 15). However, primary charcoal and LPG users are more likely to maintain their primary use when they rely more heavily on these as their primary fuels, as measured by proportion of total cooking fuel use (Table 1, columns 12 and 15).

In models that pool across all fuels in each location, based on conversions of quantities in kg or L to energy equivalents (S4 Table, Columns 1–6), we find similar negative fuel price elasticities in the two cities (ranging from -0.5 to -0.6). The relationship between the stacking variables and the WTP probability is markedly different in the two cities, however. Stacking is associated with a greater likelihood of switching away from one’s primary fuel in Nairobi. In Dar es Salaam, stacking has the opposite relationship with switching away from one’s primary fuel. This may be due to the differing stage of the energy transition in these cities and how it relates to fuel preferences. That is, given the relative early stage of LPG adoption in Dar es Salaam compared to Nairobi, LPG users (who tend to be stackers) there may perceive charcoal as their primary fuel despite reporting, aspirationally perhaps, that LPG is their primary fuel.

Percentage decreases in cooking

Rather than responding on the extensive margin of primary fuel use (that is, switching away from it entirely), households may choose only to reduce their use of a more expensive primary cooking fuel when its price increases. In the restricted sample of households that accepted the first proposed price increase and maintained primary use of their preferred fuel, we analyze the extent of the cooking reduction they predicted they would make. In Nairobi (S5 Table), we find that a 1 USD increase in LPG price per kg would reduce cooking with LPG by 9–12 percentage points. In Dar es Salaam (S6 Table), we find that a 1 USD increase in charcoal per kg and LPG price per kg faced by primary users of those fuels would reduce cooking with those fuels by 25–30 percentage points, and 8 percentage points, respectively.

Willingness to pay for the various primary cooking fuels

We provide three different measures of WTP for each of the primary fuels considered: non-parametric a) Turnbull lower-bound estimates and b) Kristrom mid-point estimates, as well as c) estimates obtained from application of maximum likelihood regression estimation to the double-bounded dichotomous choice responses to the CV questions (see the Materials and Methods section for details on the differences in these methods).

In Nairobi, the measures range from 0.14–0.2 USD/kg charcoal, 1.1–1.4 USD/liter kerosene, and 2.3–2.9 USD/kg LPG (Table 2). These measures are generally not sensitive to controlling for stacking behavior in the regression models, though the estimate declines slightly for LPG in Nairobi, from 2.7 to 2.3 USD/kg. In Dar es Salaam, the measures range from 0.05–0.13 USD/kg firewood, 0.3–0.4 USD/kg charcoal, and 1.0–1.2 USD/kg LPG. Controlling for stacking measures slightly increases WTP for firewood and LPG. In the pooled analysis, despite the substantial difference in the fuel mix relative to that in the Nairobi sample, WTP for 1 MJ of cooking fuel in Dar es Salaam is similar to that in Nairobi, at 0.1 USD.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Willingness to pay estimates (in USD) for cooking fuels in Nairobi and Dar es Salaam.

https://doi.org/10.1371/journal.pstr.0000077.t002

These estimates provide valuable insight on the nature of demand for cooking fuels in these two cities. First, in both cities, the WTP for users’ primary fuels is somewhat higher than currently observed prices. For LPG, WTP is much higher (by roughly over 65%) than current prices, reflecting a strong preference among users of this clean fuel for the benefits that it provides. Policy may need to focus on better targeting of incentives and other approaches to foster uptake and increase access among households who persist in their use of polluting cooking alternatives.

Second, WTP for LPG in Nairobi is roughly double that in Dar es Salaam, despite the low-income Nairobi sample only being slightly richer on average than the representative sample that was drawn in Dar es Salaam. Third, WTP for charcoal among primary charcoal users is substantially lower and different in Nairobi (where it is less than 1.5 times the prevailing market price) and Dar es Salaam (where it is twice the prevailing market price). Thus, households clearly have higher demand for charcoal in Dar es Salaam, which may be related to the less vigorous clean cooking policy agenda there relative to that in Nairobi. From a clean cooking transition perspective, it is important to know how much relative charcoal prices would need to change to induce more substantial switching towards clean fuels.

Distributional aspects of price change policies

It is critical to also understand the distributional impacts that fuel pricing policies might have across the income distribution. Low-income households are likely to respond differently to price increases than high-income households, especially when the switching costs are high. There is also a risk that price increases on some fuels could be regressive. To explore such aspects, we restrict the Dar es Salaam sample to households who cook mainly with charcoal, given that over 60% households (n = 672) use charcoal as their primary cooking fuel. In Nairobi, we restrict the sample to households who cook mainly with kerosene, who constitute 30% of the sample (n = 104). These two subsamples represent households that policy makers might target for transitioning to cleaner fuels. We then divide the income distribution into relatively low- and high-income households, based on whether households fall below or above the median of per capita monthly expenditures. Finally, we group the low price (25–50%) and high price increases (100–200%) together.

In Dar es Salaam, high-income charcoal users are more likely than low-income charcoal users to switch up the energy ladder to LPG for any given price increase, especially when the price increase is large (Fig 2). Lower-income households are less likely to switch away from charcoal, but those switching more often move to kerosene and firewood. Thus, large charcoal price increases in Dar es Salaam at this time could be regressive, in the sense that low-income households either maintain the use of charcoal, or switch down the energy ladder. In Nairobi, in contrast, where policies are more supportive of clean options, low-income households appear more likely to switch to LPG than high-income households. High-income kerosene-using households are instead more likely to switch to charcoal, suggesting that these households’ current preference for polluting fuel may reflect a resistance to using LPG that should be further explored.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Stated responses to hypothetical price increases for low- and high-income households in Dar es Salaam and Nairobi.

Note: This figure restricts the samples to households who mainly cook with charcoal in Dar es Salaam (n = 672), and households who mainly cook with kerosene in Nairobi (n = 104), as these are the most commonly used polluting primary cooking fuels in our study sample. These respondents are divided into relatively low-income and relatively high-income households, based on whether they fall below or above the median of per capita monthly expenditure in the sample. The hypothetical price increases are then categorized into low increases (25–50%) and high price increases (100–200%). The figures present stated responses to the question of whether households continue to use their primary cooking fuel, or switch to another cooking fuel.

https://doi.org/10.1371/journal.pstr.0000077.g002

Similar results are obtained from regression analyses (S4 Table). In the Dar es Salaam sample, higher expenditure households are more likely to switch fuels, while the income elasticity of fuel switching is much lower in the Nairobi sample. The Nairobi results may be driven by vertical differentiation in the quality of kerosene stoves (between wick and kerosene stoves); in Dar es Salaam, charcoal stoves are comparatively homogenous. Thus, low-income kerosene users in Nairobi may switch up the energy ladder because the low utility of using a low-quality kerosene stove constitutes an additional cost for the poor. Alternatively, these results may be driven by differences in the market for alternative fuels like LPG and charcoal in these two contexts, as influenced by historical and existing policy actions.

Sampling strategy

We draw on data collected in mid-2019 in Nairobi, Kenya and early-2020 in Dar es Salaam. The two surveys were part of distinct energy access studies conducted less than a year apart in the two locations. Their respective budget constraints determined the final sample size in each location. All data were gathered before the COVID-19 pandemic began in these countries. In Nairobi, the sample comprises 354 households living in four informal settlements, with sampling in each area following a probability proportional to size (PPS) sampling methodology (S3 Fig, Panel A). In each informal settlement, households were randomly selected using a field-based counting method to fulfil the study sample requirement. In Dar es Salaam, the fieldwork was conducted in January and February 2020, and a total of 1,100 households were interviewed (S3 Fig, Panel B). A similar multi-stage stratified random sampling design was applied for selection of final wards, streets and households, also using a PPS sampling methodology.

To meet our research goal of obtaining a distribution of cooking fuel users in each city, particularly polluting cooking fuel users, we used somewhat different sampling approaches in the two cities. In Dar es Salaam, the sampling strategy aimed for a representative sample of cooking fuel use in the city. Our Dar es Salaam study data are comparable with the Household Budget Survey (HBS) (2017–2018) for the Dar es Salaam Region. For example, the average age in the HBS 2017–2018 is 26 years (N = 3,272 household members in Dar es Salaam Region) and 28 years in our survey (N = 4,393 household members). In addition, 52.4% of HBS 2017–2018 household members in Dar es Salaam Region are female, while this figure is 55.2% in our survey. The proportion of household heads whose highest level of education was primary school is 47.8% in the HBS 2017–2018 and 48.4% in our survey. In terms of fuel use, 61.5% of respondents in our survey cook mainly with charcoal, 32.2% with LPG, 3.7% with kerosene and 2.4% with firewood in our survey; in the 2017–2018 HBS these figures for Dar es Salaam region are 62% of households cook mainly with charcoal, 10.4% with LPG, 6.9% with kerosene and 4.5% with firewood. On the other hand, in Nairobi’s primary cooking fuel mix, LPG has the highest share (65%, as of 2020) and use of polluting fuels is relatively low (compared to Dar es Salaam). Based on our extensive interviews with public and private sector stakeholders in the energy landscape in Nairobi, discussions with the local field partner and prior empirical literature on energy use in Nairobi, we targeted informal settlements where polluting cooking fuel use remains much higher. Therefore, the Nairobi sample was explicitly designed to cover lower-income households in the city residing in informal settlements. While the share of LPG in the primary cooking fuel mix in our sample is lower (54%) compared to that for Nairobi in the Kenya Continuous Household Survey Programme 2020 (65%) [54], it is comparable with that for Nairobi in the Kenya Household Cooking Sector Study 2019, where LPG was the most common primary cooking fuel (56%), followed by kerosene (27%) and charcoal (6%) [79].

Informed consent

The Campus Institutional Review Board (IRB) at Duke University reviewed and approved the research protocol for the Nairobi survey (Campus IRB Protocol Number: 2019–0330). Research permits were obtained from the University of Dar es Salaam (UDSM) and study district officials for the Dar es Salaam survey (UDSM Reference Number: AB3/12(B)). In both Nairobi and Dar es Salaam, we obtained oral informed consent from the household head and the primary cook in our sampled households, prior to administering the questionnaire. Oral consent was deemed acceptable because the research took place in settings where requesting people to sign a document can cause distress and mistrust. Field officers read out the consent script to the respondents, and their response to the consent question was recorded in the survey form.

Survey

Some of the key issues in administering CV experiments in LMICs are poor development of CV scenarios, poor survey implementation and oversight of not testing the effects of survey design variations on the CV experiments’ results [43]. As part of the survey elaboration process, we reviewed pertinent reports and documents on the cooking sector in both countries. For the survey design and implementation, we worked with local partners (research institutions and survey firms), conducted scoping field visits, and prepared a sampling framework (using maps in both locations).

The survey instruments in Nairobi and Dar es Salaam were similar, save for minor adjustments to ensure suitability to the local context. Comprehensive data were collected on household demographics, cooking practices and fuel preferences, household consumption and wealth and access to credit. To assess households’ WTP for cooking fuels (namely, firewood, charcoal, kerosene and LPG), each CV experiment included a double-bounded, dichotomous choice design, thereby avoiding incentive-incompatibility problems [43]. Testing of the CV experiment was especially important given the dearth of prior work in resource-constrained settings that aimed to value cooking fuels. In addition to helping frame the CV scenario, pilot testing helped determine the percentage price increases that would be most relevant to selected wards in Nairobi and Dar es Salaam. Following the empirical literature [43], we randomly assigned four bidding games or four price increases to different respondents in each study location. As expected, these different starting points brought out different responses.

Experienced field partners implemented the field work, where enumerators underwent rigorous training on the survey instrument. Two co-authors of this study extensively trained enumerators on the CV experiment module of the household survey. Local field teams’ thorough pilot testing of the household survey and the CV experiment informed the final questionnaire. Surveys were completed using tablet-based, in-person enumeration.

In both the Nairobi and Dar es Salaam household surveys, in the CV experiment module, the hypothetical situation was described to the respondent in detail. S1 Text and S2 Text include the CV experiment modules administered in Nairobi and Dar es Salaam, respectively. Enumerators first described how different cooking fuels affect households in a multitude of ways. They then asked respondents if they would consider switching their primary cooking fuel should there be an increase in its price. In both settings, respondents received randomized initial bids for cooking fuel price increases (in percentage terms) from a set of four different prices increases (25%, 50%, 100% and 200%). S2 Table shows the randomized price increases offered to households in both locations. If respondents responded positively to the initial bid, they received a follow up question with a payment option that was double the initial bid; if respondents replied in the negative to the initial bid, they received a follow up question of a payment option that was half the initial bid. In addition, respondents who declined to switch their main cooking fuels were asked whether, and to what extent, they would reduce their cooking in response to the price increase. This design allows us to assess both intensive and extensive margin responses to hypothetical fuel price changes.

Relationship between cooking fuel preferences and price

We examine households’ propensity to maintain their primary fuel use in the face of the randomly-assigned price increases. In the basic analysis, we examine the role of price alone (Model 1); more sophisticated regression analysis then controls for fuel stacking behavior and a range of household characteristics (Models 2 and 3). Fuel stacking behavior is operationalized as both binary and continuous (proportion energy use) variables. (In S3 Text, we describe the construction of these variables and in S3 Table, we examine the determinants of cooking fuel stacking among surveyed households in Nairobi and Dar es Salaam). In determining the relationship between cooking fuel preference and stacking, WTP is a binary variable that indicates whether the respondent agreed to the first price increase randomly offered. Using a probit specification, we estimate a household’s demand for cooking fuel, wherein the functional form assumes that: (1) where Φ is the cumulative distribution function of the standard normal distribution.

In Eq 1, the outcome variable is the binary answer of whether (1) or not (0) the household is willing to pay the first randomly allocated price increase, P is the price variable of the randomized increase (categorical), S is the fuel stacking variable (run separately for binary and the continuous variables), and Z is the vector of all other variables included in the model.

The Z variables included in the regressions are: household size, dependency ratio (defined as the ratio of the younger (ages 14 and under) and older (ages 65 and above) population in the household to the working age population (between 15–64 years) in the household), the age of the household head (in years), the education level of the household head (categorical), an indicator for whether the household head is female, the log of monthly per capita total expenditures (in USD), whether the household has saved money anywhere in the past year, whether the household has an electricity connection, and weekly time taken (in minutes) to acquire various cooking fuels (firewood, charcoal, kerosene, LPG).

We also analyze whether households that said they would continue with the same primary fuel use, would change their amount of cooking given the price increase. For this analysis, we run a conditional regression on those that responded positively to the initial bid (in other words, they would continue using their primary fuel even if its unit price increased), wherein the outcome variable is the percentage of normal cooking that would be reduced, and the explanatory variables are the same right-hand side variables as in Eq 1.

Estimating WTP for cooking energy

We adopt a double-bounded, dichotomous choice CV experiment to elicit respondents’ willingness to maintain both primary reliance on their main cooking fuel (namely, charcoal, kerosene and LPG in Nairobi, and charcoal, LPG and firewood in Dar es Salaam) [41,80]. We examine the sensitivity of that main fuel use to price increases that were randomly assigned to survey respondents. We estimate average WTP in the sample of primary users of each fuel, both in terms of units purchased, and their useful energy content.

Taking into account the double-bounded design of the CV experiment, we use a maximum likelihood estimator that includes both the initial and second bids to estimate the WTP for each cooking fuel–charcoal, kerosene and LPG in Nairobi, and firewood, charcoal and LPG in Dar es Salaam. The user-generated STATA command ‘doubleb’ is used for this calculation [81]. The independent variables controlled for are the same as those used in Eq 1. For comparison, we also derive non-parametric WTP estimates, namely the conservative Turnbull lower-bound estimates of WTP and the Kristrom mid-point estimates [82,83]. These alternative estimates only leverage the data on response to the first experimentally assigned price increase, which maximizes the incentive compatibility of the CV design, and do not control for household-specific factors that might influence demand [84].

We also pool responses across the three cooking fuel categories in each city to examine the links between household characteristics and cooking fuel valuation, in addition to eliciting WTP for cooking fuels. We use a similar probit regression approach as in Eq 1 to estimate this association, where the left-hand side variable is the probability that a household responds positively to the CV questionnaire. We normalize across fuels in the pooled models to account for the different useful energy content of each. Specifically, prices were pooled and normalized by dividing by calorific value for each fuel (MJ/kg) and again by fuel efficiency (%); the final unit is in KES/MJ in Nairobi and TZS/MJ in Dar es Salaam, which have further been converted into USD/MJ.