Cash for Coolers or Sustainable Lighting? Assessing Different Components of a Large-Scale Energy Efficiency Program in Mexico

Abstract

In one of the largest demand-side energy efficiency programs at the time, the Mexican government supported more than 11 million, mostly low-income households in replacing their old light bulbs and appliances with more efficient models. Previous evaluations of this program focused exclusively on appliances, which made up almost 90 per cent of the total program cost, and found modest benefits in terms of energy savings. This study compares the respective effects of replacing light bulbs and appliances simultaneously in a single econometric framework, using data from nationally representative household surveys and a difference-in-differences approach which exploits geographical variation in treatment intensities. Despite using different sources and types of data, our results for the appliances replacement intervention are largely in line with the estimates of previous studies. In addition, we find that the impact on energy consumption (proxied by the amount paid for electricity) of replacing light bulbs was of comparable magnitude as that of replacing appliances, although the average cost per participating household was much smaller. Overall, our results suggest that low-cost investments that help poor households reduce their energy consumption for lighting can have high returns on energy efficiency in the residential sector.

Keywords:

• Energy efficiency
• demand-side management
• electricity
• developing countries
• Mexico

Acknowledgements

We thank Guillermo Hernández González, Giuseppe Iarossi, Jevgenijs Steinbuks, Jos Vaessen, members of the evaluation team, and various World Bank internal referees for valuable comments and suggestions.

Disclosure statement

A previous version of this paper was prepared as a background paper for the World Bank’s Independent Evaluation Group’s evaluation study ‘World Bank Group Support to Energy Efficiency: An Independent Evaluation of Demand-side Approaches’. The views expressed in this paper are solely those of the authors and do not necessarily reflect the views of the World Bank Group.

Notes

1 The information and numbers in this section are mostly based on the respective Project Appraisal Document and Implementation Completion and Results Report, which are publicly available from the World Bank’s website.

2 To limit the potential role of outliers we winsorize the outcome variable at the 99^th percentile (that is, we set all values above the 99^th percentile equal to the value at the 99^th percentile).

3 These authors also show that it is generally difficult to interpret estimates obtained from these models across different groups if there is treatment effect heterogeneity (see also de Chaisemartin & D’Haultfœuille, 2023). This limits our ability to calculate the potential impact of alternative treatment intensities (for example, to inform a scaling up of the program). However, it does not crucially affect the conclusions of our main analysis which focuses on comparing the intention-to-treat estimates of different project components for a given set of treatment intensities. Note also that the program we assess features no (or only neglectable) variation in treatment timing, as it was implemented at the national level (across all Mexican states) at the same time. It is thus not necessary to consider a ‘staggered DiD’ model (Callaway & Sant’Anna, 2021; Goodman-Bacon, 2021).

4 In addition, one may investigate the mechanisms behind the observed variation in treatment intensities. While we do not have enough information to characterise these perfectly, the fact that the official targeting strategy favored low and medium-income households (recall the Background Section), suggests that differences across states in the share of these households (as opposed to high-income households which were not eligible for the program) are likely an important factor. Our empirical strategy accounts for this fact by including income as a control variable in the regression analysis.

5 Another concern might be that our identification strategy is compromised by other factors impacting electricity prices during our study period. While we cannot completely rule out this possibility, our DiD approach largely addresses this concern. Specifically, in order to make our estimates spurious, those price changes would have to be highly correlated with our treatment variable (both across states and across time). In contrast, any price changes that either affected all Mexican states in a similar way, or that affected some states more than others but unrelated to the program’s treatment intensities, will be controlled for in DiD (in the same way as differences in baseline characteristics between states are controlled for, and subject to the parallel trends assumption).

6 A possible contributing factor to this result are rebound effects, which have been argued to be stronger for some appliances (such as air conditioners) than for other appliances (refrigerators) and lighting (Davis et al., 2014; Schleich et al., 2014).

7 Our dataset does not allow us to discern the specific impacts of different activities within each program component. In particular, we are unable to quantify the extent to which energy savings can be attributed to economic incentives (provision of free CFLs, appliances vouchers) versus the influence of information and awareness campaigns, which may have helped households in adopting adequate technologies and using them in energy-efficient ways by addressing binding information constraints (Naeher, 2022; Sallee, 2014).

Additional information

Funding

This work was supported by World Bank Group.