Health awareness and the transition towards clean cooking fuels: Evidence from Rajasthan

Ensuring affordable, reliable, sustainable and modern energy for all by 2030 is part of the internationally agreed Sustainable Development Goals (SDG7). With roughly 3 billion people still lacking access to clean cooking solutions in 2017, this remains an ambitious task. The use of solid biomass such as wood and cow dung for cooking causes household air pollution resulting in severe health hazards. In this context, the Indian government has set up a large program promoting the use of liquefied petroleum gas (LPG) in rural areas. While this has led millions of households to adopt LPG, a major fraction of them continues to rely heavily on solid biomass for their daily cooking. In this paper, we evaluate the effect of simple health messaging on the propensity of these households to use LPG more regularly. Our results from rural Rajasthan are encouraging. They show that health messaging increases the reported willingness to pay for LPG, and substantially increases actual consumption. We measure this based on a voucher, which can only be used if LPG consumption is doubled until a certain deadline. Households exposed to health messaging use the voucher about 30% more often than households exposed to a placebo treatment. We further show that the impact of our very brief, but concrete health messaging is close to the effect of a 10% price reduction for a new LPG cylinder. Finally, our study raises some interesting questions about gender-related effects that would be worth consideration in future research.


Mathematical derivations for the theoretical model
In this paper, we proposed an illustrative model based on the Cobb-Douglas utility function: U (g, x) = g θ x 1−θ (1) where g is the cooking gas LPG, x is a composite good that includes traditional biomass and other goods, and θ ∈ [0, 1] is an indicator for the preference for LPG as compared to the composite good.
For an income B and prices p g and p x , the budget constraint is: Maximizing (1) subject to (2) yields the Marshallian demand function for LPG: Inverting this function we obtain the price a household is willing to pay for this quantity of LPG: (4)

Predictions related to WTP
Imagine we request the household to increase its consumption from g * toḡ = 2g * as we are interested in WTP for regular rather than very sporadic users. Assume that such a doubling of LPG consumption is feasible within the budget constraint (Assumption 1, see section Assumptions). To make this situation again optimal for the household, the new price must be 50% lower than the initial price: While this exact relationship is directly related to the restrictive assumption underlying the Cobb-Douglas utility function that the price-elasticity of demand is equal to 1, even otherwise, we would clearly expect a reduction in WTP with an increase in the requested amount to be consumed.
More interesting in the context of our study, however, is the question to what extent health messaging can compensate some of this reduction in WTP. What is the change in WTP if we increase the decision maker's knowledge about the adverse health effects of cooking with traditional biomass?
Let us consider the preference for LPG θ as a linear function of health knowledge h ∈ [0, 1]: whereθ ∈ [0, 1] is basic preference (e.g., due to the convenience and time savings associated with LPG) and γ ∈ [0, 1] is a factor reflecting the salience of health information, notably due to gender. We can then rewrite p g (ḡ) as: The expected effect of health messaging on WTP is then given by: We are further interested to see if the impact of health messaging is affected by differences related to gender reflected by differences in the salience of the health information. To see this we need to take the cross-derivative with respect to h and γ. To do so, note that factors such as gender already influence the initial value of g * and thus alsoḡ. More formally, we can write: where p m is the original market price andθ =θ +h · γ the initial preference for LPG. In other words,ḡ is fixed as the double of the optimal consumption at the general market price and the initial preference for LPGθ that is based on the initial health knowledge and salience. We keep h fixed at this initial levelh as its change due to the treatment does not influence g * . In contrast, a greater salience of such health knowledge γ already influences the initial g * . Hence,ḡ needs to be considered as a function of γ but not of h when we take the derivatives. We assume that the treatment itself does not affect γ (Assumption 2, see section Assumptions), which is obvious if we think of it as reflecting the gender of the decision maker. Inserting (9) in (8) and taking the derivative with respect to γ, we obtain: Predictions related to the propensity of voucher use The propensity to use the voucher can be expressed as the difference in utility ∆U between a situation in which the voucher is used U 1 and a situation in which it is not used U 0 . Taking into account the conditions for voucher use, namely doubling initial consumption and the discounted offer price p d , we can specify U 1 as In contrast, the utility when the voucher is not used U 0 simply corresponds to Eq (1) evaluated at the optimal level of consumption given the market price p m , without any discount but with the possibility to freely adjust all quantities to changes in θ: ∆U can thus be rewritten as To facilitate the computation of the derivatives we simplify Eq (13) through a monotonous transformation using logs. This transformation will leave the sign of the derivatives unchanged.
Replacing θ by (6) and taking the derivative with respect to h yields: Note that this computation is again based on Assumption 1 (a doubling of consumption is feasible within the budget constraint), or else, we would take the log of a negative quantity in the first term. A further relevant assumption is that the requirement to double LPG consumption in order to use the voucher is a binding constraint (Assumption 3, see section Assumptions). For more extreme preferences for LPG, the model would suggest that the household would forego the voucher in order to be able to consume more LPG. This situation is irrelevant in practice, as the voucher can also be used any time before the deadline, and hence there is no constraint on the maximum use of LPG. For reasons of simplification, the model has not been designed to cover these obvious cases where the health treatment is extremely effective. Finally, remember that 0 < p d pm ≤ 1 since p d is the discounted price while p m is the market price. Considering all these arguments, we obtain the sign of the derivative.
We now examine how the impact of h on ∆u varies for different levels of salience of health information. We use (15) evaluated at the initial preferences for LPG θ =θ. Considering thatθ =θ +h · γ we can take the derivative of ∂∆u ∂h with respect to γ to obtain the cross-derivative: This inequality holds under exactly the same conditions as the inequality in (15). Before concluding this analysis, let us further examine the reaction of ∆u to a change in the discounted offer price p d . Since this price can be obtained only when the household effectively uses the voucher, a lower p d makes voucher use more attractive: This inequality only requires Assumption 1 (see section Assumptions). The negative relationship between WTP and the required consumption is thus also reflected in the lower propensity of voucher use (implying the doubling of consumption) for higher p d .
Finally, note that-as opposed to WTP-the propensity of voucher use is unrelated to the budget B, since it enters in the same way in both U 1 and U 0 and hence cancels out: Assumptions This section provides an overview of the three main assumptions referred to above: