Effect of Worldwide Oil Price Fluctuations on Biomass Fuel Use and Child Respiratory Health: Evidence from Guatemala

Abstract

Objectives. We examined the effect of worldwide oil price fluctuations on household fuel use and child respiratory health in Guatemala.

Methods. We regressed measures of household fuel use and child respiratory health on the average worldwide oil price and a rich set of covariates. We leveraged variation in oil prices over the 6-month period of the survey to identify associations between fuel prices, fuel choice, and child respiratory outcomes.

Results. A $1 (3.4% point) increase in worldwide fuel prices was associated with a 2.8% point decrease in liquid propane gasoline use (P < .05), a 0.75% point increase in wood use (P < .05), and a 1.5% point increase in the likelihood of the child reporting a respiratory symptom (P < .1). The association between oil prices and the fuel choice indicators was largest for households in the middle of the income distribution.

Conclusions. Fluctuations in worldwide fuel prices affected household fuel use and, consequently, child health. Policies to help households tide over fuel price shocks or reduce pollution from biomass sources would confer positive health benefits. Such policies would be most effective if they targeted both poor and middle-income households.

Acute respiratory illnesses are the leading cause of death among children in the developing world and account for nearly 20% of child deaths.^1–4 Past research has found a strong association between respiratory health and household use of biomass fuels,^5–15 leading to the conclusion that exposure to indoor air pollution from cooking and heating with biomass fuels such as wood, dung, or crop residues causes at least one third of childhood respiratory illnesses.^3,4 These findings are supported by results from studies that used quasi-experimental conditions to more rigorously establish causality.^16–18

The design of effective policies to reduce indoor air pollution requires an understanding of how families choose fuels. Certainly, socioeconomic status is important: poorer households are more likely to use biomass fuels, with high start-up costs and lack of access preventing adoption of cleaner alternatives.^2-4,7 The relative price of clean fuels, such as liquid propane gasoline, vis-à-vis dirty fuels could also affect fuel choice.^19,20 The price of liquid propane gasoline is determined not only by local supply and demand but also by the worldwide market price of crude oil, which is used to produce liquid propane gasoline. Anecdotal evidence indicates that increases in worldwide crude oil prices may cause families to substitute away from cleaner alternatives, particularly in the short run.²¹ However, no research has explored the effect of fluctuations in worldwide crude oil price on fuel use and health outcomes.

The great variation in household fuel use in Guatemala—around 60% of households use wood, 40% use liquid propane gasoline, 35% use some combination of the two, and 8% use other types of fuels, such as kerosene, coal, or electricity²²—makes it an ideal setting to study the impact of price changes on fuel choice. As with other developing countries, previous research on Guatemala has suggested that education, socioeconomic position, and clean fuel availability play a large role in driving liquid propane gasoline adoption. The effect of price fluctuations, however, has not been well elucidated.²³ Therefore, we addressed this gap by using a rich data set from Guatemala to examine the association between fluctuations in the worldwide crude oil price, household fuel choice, and childhood respiratory illness.

METHODS

We used data from the 2000 Encuesta Nacional sobre Condiciones de Vida, a nationally representative cross-sectional survey covering more than 7200 households.²⁴ The National Statistical Institute of Guatemala (Instituto Nacional de Estadística) administered the survey with the assistance of international agencies, including the World Bank. Fieldwork was conducted over the 6-month period between July and December 2000. The data included information on fuel usage, morbidity, asset ownership, expenditures, anthropometrics, and education. The fuel use data indicated whether the household used wood, coal, kerosene, electricity, or liquid propane gasoline in the previous month and what the fuel was used for (e.g., cooking, lighting). These questions were asked separately for each type of fuel.

Measures

The main outcomes of interest were fuel use and respiratory health. As discussed, most households used either liquid propane gasoline or wood, so we focused on these 2 options. We created binary indicators for whether the household used liquid propane gasoline or wood, respectively, in general and specifically for cooking. Because more than 99.5% of households who reported using either fuel in general used it for cooking, our analysis focused on the latter. We also incorporated into our analysis information about whether the household collected firewood in the month before the survey. Anecdotal evidence has suggested that families respond to price increases by manually gathering biomass fuel.²¹

Regarding respiratory health, we created a binary variable that indicated whether children younger than 6 years experienced a “cold, cough, whooping cough, bronchitis, breathing trouble, or any respiratory infection” in the month before the interview, as reported by the child's mother. Similar information for older children and adults was not collected.

For our main independent variable, we used weekly worldwide crude oil price data from the US Government Energy Information Administration,²⁵ which we adjusted for inflation. For each household, we matched worldwide oil prices during the week of the interview and for each of the previous 3 weeks. We then averaged worldwide oil prices over this 4-week period.

Control variables were taken from the 2000 Encuesta Nacional sobre Condiciones de Vida. We gathered information on parental educational attainment (specifically, indicators of whether the parent received some primary education or completed primary schooling), whether the household belonged to an indigenous group, whether a household member purchased cigarettes in the month before the survey, and household asset ownership. In the last case, we generated a cardinal asset index by combining indicators for whether the household owned a microwave, refrigerator, personal computer, radio, television, telephone, washing machine, car, motorcycle, or bicycle. Additional controls included housing characteristics (binary variables for whether the household had a nonthatched roof or hard material floor), season of interview (an indicator of whether the interview occurred during the May to October wet season), the Guatemalan department of residence (binary variables for each of the 22 departments), and residence in an urban area. Given the high prevalence of migrant labor (17% of sample fathers, predominantly from higher income households, were reported as missing), we imputed data for paternal education by using multiple imputation techniques²⁶ and created a binary variable to denote imputed data.

Our choice of controls was driven by theory and the extant literature. Our inclusion of controls for parental schooling, indigenous identity, and household assets reflected the importance of education, culture, and socioeconomic status in determining both fuel choice and child health. Because past research has shown that consumption variables are less prone to transitory disturbances and measurement error in developing countries, we used an asset index to indicate socioeconomic status.²⁷ Housing characteristics both indicate wealth and act as a proxy for environmental exposures relating to dwelling quality. Binary indicators for department and area (urban vs rural) of residence were used to control for spatial factors, such as the availability of health facilities, fuel markets, or ambient outdoor pollution, that may influence health or fuel choice. The dummy variable for season of interview captured any drivers of fuel use or disease risk that followed seasonal economic or weather patterns. Finally, we included a dummy variable for whether someone in the household purchased cigarettes to control for children's exposure to environmental tobacco smoke.

To achieve a more lucid interpretation of our results, we created indicators for whether a child experienced nonrespiratory morbidities in the previous month, whether the household's kitchen was outdoors, and whether the household was in a community where liquid propane gasoline was sold. We also used information on scale measurements of child weight (conducted by a trained enumerator). We describe the rationale for using these variables in the next section.

Empirical Strategy

To investigate the link between worldwide crude oil price and household fuel use, we estimated versions of the following probit specification:

where h represents the household, F_h is the fuel choice variable (whether the household used liquid propane gasoline for cooking, used wood for cooking, or collected firewood in the month before the survey), and Φ(β × AvgWCOP + α × X_h) is the evaluation of the standard normal cumulative density function. AvgWCOP is the average worldwide crude oil price in the 4 weeks preceding the household's interview date, X_h is a vector representing the control variables described in the previous section, and β and the vector α are the parameters to be estimated.

To further explore the policy implications of this relationship, we estimated Equation 1 for samples stratified by socioeconomic status. We expected middle income households (specifically, the third and fourth quintiles of the consumption expenditure distribution) to be most sensitive to changes in fuel price because previous research has shown that these households are most likely to use firewood and liquid propane gasoline concurrently. Poor households (first and second quintile) do not use liquid propane gasoline, and richer households (fifth quintile) generally do not use wood and are less affected by price shocks.^22,23

We then used a similar approach to examine the association between fuel prices and child respiratory health:

where the new index i represents each child, R_ih is a binary indicator for whether the child experienced a respiratory symptom within the month preceding the survey, and the vector X now includes indicators for child gender interacted with a third-order polynomial for child age in months and household member smoking, in addition to the parent and household characteristics described above.

Several issues related to our empirical strategy warrant further discussion. First, our model examined contemporaneous associations between fuel use behavior and average worldwide oil prices both measured for the month preceding the interview, that is, the period referenced by the survey questions about fuel use and respiratory symptoms. The implicit assumptions behind our specification were, first, that worldwide oil prices affect local fuel prices within a short time frame and, second, that fuel use decisions are made within the span of a single month. We demonstrate the validity of the former assumption in Figure 1, with local prices closely tracking worldwide price fluctuations. Regarding the second assumption, fuel use decisions are made frequently, because households typically purchase or collect firewood on a daily to weekly basis and refill liquid propane gasoline canisters on at least a monthly basis.^22,28,29 Indeed, households can acquire firewood whenever necessary, thus allowing them to conserve previously purchased liquid propane gasoline when faced with high prices. (Consistent with this idea, we found that the average worldwide crude oil price, lagged by ≥ 1 month, did not predict fuel use or respiratory health relative to contemporaneous prices. These results are available from A. S. V.).

FIGURE 1.

Worldwide crude oil and Guatemalan liquid propane gasoline canister prices, July–December 2000.

A second set of issues, which stem from the cross-sectional nature of the survey and the inherent limitations in the data, involve the interpretation of the coefficient estimates on average worldwide crude oil price. It is possible that other factors correlated with worldwide crude oil price, rather than fluctuations in the worldwide crude oil price itself, drive the association between oil prices, fuel choice, and respiratory health. Additionally, even if worldwide crude oil price was the main driver, it may have affected child health through channels other than household fuel choice. To assess these possibilities, we conducted a series of mechanism checks. First, we examined the relationship between worldwide crude oil price, nonrespiratory illnesses, and child weight. An association with these variables would indicate that fluctuations in worldwide crude oil price were associated with factors such as access to nutrition or health care that may affect child health. We also assessed whether the relationship between worldwide fuel prices and respiratory health existed for the sample of households living in communities where liquid propane gasoline was not available for purchase. If worldwide prices primarily influenced health through effects on household fuel choice, we should not see an association between these prices and respiratory health in areas where substitution across fuels is more difficult. Last, we examined whether children in houses with outdoor kitchens were differentially affected by fuel price changes as opposed to those with indoor kitchens. We would expect a stronger association between worldwide crude oil price and respiratory outcomes for houses with indoor kitchens versus outdoor ones because the produced smoke cannot escape as easily. However, a different set of findings would point to the importance of other mechanisms, such as more general changes in pollution driven by worldwide crude oil price or alternate macroeconomic effects.

Finally, we clustered standard errors in all regressions by dividing the 22 Guatemalan departments into 44 urban and rural clusters. Fuel choice decisions and respiratory symptoms likely respond to common shocks at the local area level (e.g., changes in access to fuel markets or health care, changes in weather, and disease contagion within and across households), thus creating a correlation in the error terms of Equations 1 and 2. In addition, fuel choices and health outcomes in 1 week could be correlated with outcomes and choices in previous or forthcoming weeks at the regional or national level. Clustering standard errors at the department × area level accounted for both spatial and serial correlation: department × area units represented relevant spatial units and the survey sampled individuals residing in these units at various time points, thus helping to account for the time dimension.³⁰ This strategy was more conservative than alternate approaches. For example, clustering at the household level to account for multiple children per family or using a multilevel model specifying random effects at the household and community levels produced smaller confidence intervals.

RESULTS

Our final sample included 5665 children younger than 6 years residing in 3490 households. Descriptive statistics are provided in Table 1. A total of 83% of the households used wood, and 37% used liquid propane gasoline in the month before the interview; nearly 20% of households used both fuels. The average household reported having 1.6 of the 10 assets used to create the asset index, and only 15% of the fathers and 21% of the mothers reported completing primary school. Roughly 40% of households were situated in urban areas. The average worldwide crude oil price fluctuated between $25 per barrel to over $31 per barrel. The relationship between the worldwide crude oil price and the week of the survey is shown in Figure 1. The oil price fluctuations during this period, although meaningful, were moderate and typical of the surrounding 5-year window (Figure A, available as a supplement to the online version of this article at http://www.ajph.org). Also plotted in Figure 1 are the weekly liquid propane gasoline canister prices for a single Guatemalan department for which we had measurements for each week of the study period. These local liquid propane gasoline prices closely tracked the pattern of the worldwide oil prices, with a lag of about 1 week (confirmatory regression results provided in Table A, available as a supplement to the online version of this article at http://www.ajph.org). Finally, regarding health outcomes nearly 48% of children were reported to have had some respiratory symptom in the past month; 34% experienced other illnesses.

TABLE 1.

Descriptive Statistics for the Study Population: Guatemala, 2000

Variables	Mean (SD)
Household
LPG for cookinga	0.373 (0.484)
Wood for cooking	0.828 (0.378)
Collected wood	0.429 (0.495)
AvgWCOP,b $	29.175 (1.454)
Asset indexc	1.587 (1.650)
Hard material roofd	0.803 (0.398)
Hard material floore	0.520 (0.500)
Indigenous languagef	0.419 (0.493)
Smoker in householdg	0.110 (0.313)
Father missingh	0.171 (0.369)
Father's education
No formal schooling	0.385 (0.487)
Some primary schooling	0.462 (0.499)
≥ primary school	0.153 (0.360)
Mother's education
No formal schooling	0.255 (0.436)
Some primary schooling	0.531 (0.499)
≥ primary school	0.214 (0.410)
Mother's age, y	29.708 (7.568)
Urban area	0.403 (0.491)
Rainy seasoni	0.663 (0.472)
Children
Respiratory symptomsj	0.475 (0.499)
Other illnessesk	0.342 (0.474)
Weight,l kg	28.753 (8.568)
Age, mo	39.772 (18.203)
Boys	0.507 (0.500)

Note. AvgWCOP = average worldwide crude oil price; LPG = liquid propane gasoline. The sample size for households was n = 3490. The sample size for children was n = 5665. Binary variables were scored as 0 (no) or 1 (yes) and then averaged.

^aLPG for cooking = 1 if the household used liquid propane gasoline for cooking anytime in mo preceding the interview date. “Wood for cooking” and “collected wood” were interpreted similarly.

^bAvgWCOP refers to the average worldwide crude oil price in the 4 wk preceding the date a given household was interviewed.

^cThe asset index was formed by adding binary variables for whether a household reported owning microwave, refrigerator, personal computer, radio, television, telephone, washing machine, car, motorcycle, or bicycle (min = 0, max = 10).

^dConcrete, foil, and asbestos cement roofs, excluding thatched roofs.

^eCement, wood, or granite floors, excluding sand or dirt surfaces.

^fIndigenous language = 1 for households speaking indigenous languages within the home.

^gSmoker in household = 1 if the anyone in the household reported purchasing cigarettes in the month before the survey.

^hFather missing = 1 if no data on the child's father were reported; father schooling variables include imputations for missing data.

ⁱRainy season = 1 if the household was interviewed between July and October, 0 if interviewed in November or December.

^jRespiratory symptoms = 1 if child was reported by the mother to have had “cold, cough whooping cough, bronchitis, breathing trouble, or any respiratory infection” in the last month.

^kOther illness = 1 if child was reported by the mother to have a nonrespiratory morbidity.

^lAs measured by a trained survey enumerator.

The probit marginal effects estimates for Equation 1 are presented in Table 2, which shows the association between the use of liquid propane gasoline for cooking and average fuel prices. The results show that a $1 (3.4%) increase in average worldwide crude oil price was associated with 2.8% point decrease (−0.0282 × 100) in the probability of using liquid propane gasoline in the month before the survey (P < .05). The same increase in worldwide crude oil price was associated with a 0.75% point increase in the probability of using wood for cooking and a 1.9% point increase in the probability of manually collecting firewood in the past month. Estimates for the other covariates yielded intuitive results: indicators of higher socioeconomic status, such as higher asset scores and the presence of solid roofs and floors, were positively associated with liquid propane gasoline use and were negatively associated with wood use and collection. Many of these coefficients were statistically significant. Increased parental education and urban residence exhibited a similar pattern. The results in Table 3 indicate that middle-income households and those with gas stoves were most sensitive to fuel price shocks, as seen by their greater propensity to alter fuel use.

TABLE 2.

Association Between Worldwide Fuel Prices and Covariates and Household Fuel Use, Probit Marginal Effects: Guatemala, 2000

Independent Variables	LPG for Cooking, B (95% CI)	Wood for Cooking, B (95% CI)	Collected Wood, B (95% CI)
AvgWCOPa	−0.028** (−0.054, −0.002)	0.008*** (0.003, 0.013)	0.019* (−0.003, 0.041)
Asset indexb	0.060** (0.001, 0.133)	−0.018*** (−0.024, −0.012)	−0.066*** (−0.089, −0.043)
Hard material floorc	0.248*** (0.193, 0.303)	−0.057*** (−0.081, −0.034)	−0.122*** (−0.187, −0.057)
Hard material roofd	0.106*** (0.031, 0.181)	−0.008 (−0.046, 0.030)	−0.061 (−0.145, −0.023)
Indigenous languagee	−0.119*** (−0.188, −0.050)	0.014 (−0.031, 0.058)	0.022 (−0.067, 0.112)
Mother's Education
Some primary schooling	0.114*** (0.039, 0.189)	−0.015 (−0.036, 0.006)	−0.067* (−0.136, 0.001)
≥ primary schooling	0.254*** (0.070, 0.438)	−0.062*** (−0.109, −0.016)	−0.168*** (−0.249, −0.087)
Father's Education
Some primary schooling	0.124*** (0.034, 0.213)	−0.063** (−0.125, −0.000)	0.026 (−0.0122, 0.065)
≥ primary schooling	0.286*** (0.185, 0.388)	−0.168** (−0.316, −0.020)	−0.065** (−0.118, −0.013)
Father missingf	−0.007 (−0.073, 0.059)	0.014 (−0.005, 0.033)	−0.034 (−0.107, 0.040)
Mother's age	−0.005*** (−0.008, −0.001)	0.002*** (0.001, 0.003)	0.002 (−0.001, 0.005)
Urban area	0.334*** (0.262, 0.407)	−0.111*** (−0.136, −0.086)	−0.244*** (−0.292, −0.197)
Rainy season	0.042 (−0.036, 0.120)	−0.011 (−0.043, 0.020)	−0.042 (−0.117, 0.033)

Note. AvgWCOP = average worldwide crude oil price; CI = confidence interval; LPG = liquid propane gasoline. The sample size was n = 3490. Coefficients shown are probit marginal effects, which can be interpreted as the change in the percent probability of observing the outcome of interest from an infinitesimal change in the value of a given continuous independent variable (evaluated at this variable's mean), or from moving from 0 to 1 for a given binary independent variable. All regressions include binary variables denoting department of residence. The estimates on these variables are available upon request. CIs were computed by using standard errors corrected for clustering at the department × area level. The relevant comparison group for the “maternal schooling” and “paternal schooling” variables is “no formal schooling.”

^aAvgWCOP refers to the average worldwide crude oil price in the 4 wk preceding the date a given household was interviewed.

^bThe asset index was formed by adding binary variables for whether a household reported owning a microwave, refrigerator, personal computer, radio, television, telephone, washing machine, car, motorcycle, or bicycle (min = 0, max = 10).

^cConcrete, foil, and asbestos cement roofs, excluding thatched roofs.

^dCement, wood, or granite floors, excluding sand or dirt surfaces.

^eIndigenous language = 1 for households speaking indigenous languages within the home.

^fFather missing = 1 if no data on the child's father were reported.

^*P < .1; **P < .05; ***P < .01.

TABLE 3.

Household Fuel Use Models Stratified by Socioeconomic Group, Probit Marginal Effects: Guatemala, 2000

	LPG for Cooking		Wood for Cooking		Collected Wood
Socioeconomic Group	B (95% CI)	No. of Households	B (95% CI)	No. of Households	(B; 95% CI)	No. of Households
Bottom 2 quintiles	−0.002 (−0.006, 0.002)	1431	−0.001 (0.002, 0.00082)	1021	0.004 (−0.018, 0.026)	1570
Middle 2 quintiles	−0.030* (−0.062, 0.002)	1384	0.022*** (0.0063, 0.038)	1317	0.038** (0.00148, 0.068)	1384
Top quintile	−0.027*** (−0.039, −0.015)	446	0.012 (−0.045, 0.069)	527	…	…

Note. CI = confidence interval. Each cell represents a separate regression estimate. The sample was stratified by the household's position in the consumption expenditure distribution. The division of the quintiles was based on the fuel use data reported in Ahmad et al.²² The coefficients displayed are probit marginal effects for the AvgWCOP variable. Probit estimates could not be computed for wood collection for the richest quintile because of the very small percentage of households in this subgroup that reported engaging in this activity. All models include household asset index, floor and roof materials, whether indigenous languages were spoken at home, mother and father's schooling, whether the father lived at home at the time of survey, residence in an urban area, and whether the household was interviewed during the rainy season. The full set of coefficient estimates is available upon request. CIs were computed by using standard errors corrected for clustering at the department × area level. Ellipses indicate that estimation was not possible due to small sample size.

^*P < .1; **P < .05; ***P < .01.

The results for child respiratory symptoms in the previous month are presented in Table 4, which shows that a $1 (3.4%) increase in average worldwide crude oil price was associated with a 1.5% point increase in the probability of reporting a respiratory symptom (P < .1). Household cigarette smoking and rural residence were both associated with increased respiratory symptoms; few significant associations were seen with the other parent or household characteristics.

TABLE 4.

Association Between Worldwide Fuel Prices and Covariates and Child Respiratory Symptoms, Other Morbidities, and Weight, Probit Marginal Effects: Guatemala, 2000

Independent Variables	Respiratory Symptoms, B (95% CI)	Other Illnesses, B (95% CI)	Child Weight, B (95% CI)
AvgWCOPa	0.015* (−0.002, 0.033)	−0.008 (−0.027, 0.011)	0.015 (−0.245, 0.275)
Asset indexb	0.015 (0.004, 0.034)	−0.001 (−0.015, 0.013)	0.474*** (0.276, 0.672)
Hard material floorc	−0.033 (−0.080, 0.013)	0.015 (−0.028, 0.059)	0.722** (0.086, 1.358)
Hard material roofd	0.015 (−0.043, 0.073)	0.003 (−0.056, 0.062)	0.092 (−0.806, 0.990)
Indigenous languagee	0.044 (−0.052, 0.140)	0.027 (−0.046, 0.099)	−0.648** (−1.153, −0.143)
Smoker in household	0.082** (0.019, 0.145)	0.039 (−0.040, 0.117)	−0.552 (−1.476, 0.373)
Mother's Education
Some primary schooling	0.017 (−0.031, 0.064)	−0.020 (−0.097, 0.057)	0.302 (−0.088, 0.693)
≥ primary schooling	−0.041 (−0.137, 0.055)	−0.089** (−0.173, −0.005)	0.705 (−0.173, 1.582)
Father's Education
Some primary schooling	0.0061 (−0.044, 0.057)	0.005 (−0.029, 0.039)	0.378 (−0.215, 0.972)
≥ primary schooling	0.093 (−0.028, 0.215)	−0.00995 (−0.069, 0.049)	0.994** (0.004, 1.985)
Father missingf	−0.048 (−0.111, 0.015)	0.032 (−0.017, 0.081)	0.401 (−0.414, 1.217)
Mother's age	−0.002 (−0.005, 0.001)	−0.002 (−0.006, 0.002)	0.020 (−0.017, 0.058)
Urban area	−0.114*** (−0.168, −0.060)	−0.0563** (−0.110, −0.003)	0.786*** (0.344, 1.229)
Rainy season	−0.009 (−0.081, 0.063)	0.029 (−0.018, 0.078)	−0.030 (−0.976, 0.916)

Note. AvgWCOP = average worldwide crude oil price; CI = confidence interval. The sample size was N = 5665. Coefficients shown are probit marginal effects. All models included controls for child gender, a cubic of child age, and their interactions, as well as dummy variables for department of residence. The estimates on these variables are available upon request. CIs were computed by using standard errors corrected for clustering at the department × area level.

^aAvgWCOP refers to the average worldwide crude oil price in the 4 wk preceding the date a given household was interviewed.

^bThe asset index was formed by adding binary variables for whether a household reported owning a microwave refrigerator, personal computer, radio, television, telephone, washing machine, car, motorcycle, or bicycle (min = 0, max = 10).

^cConcrete, foil, and asbestos cement roofs, excluding thatched roofs.

^dCement, wood, or granite floors, excluding sand or dirt surfaces.

^eIndigenous language = 1 for households speaking indigenous languages within the home.

^fFather missing = 1 if no data on the child's father were reported.

^*P < .1; **P < .05; ***P < .01.

The remainder of Table 4 assesses the association between average worldwide crude oil prices, other illnesses, and child weight. The estimate for the former was not statistically significant, and the association was in the wrong direction (a $1 increase in worldwide crude oil price was associated with a 2% point decrease in the probability of having experienced nonrespiratory morbidity). The estimate on the latter was positive but was very small in magnitude and not statistically significant.

The results from the additional “mechanism checks” are presented in Table 5. Among households living in communities with liquid propane gasoline markets, a $1 increase in worldwide crude oil price was associated with a 2.2% point increase in the probability of reporting respiratory symptoms in the previous month (P < .05). By contrast, the association for children living in communities where liquid propane gasoline was not sold was insignificant both statistically and clinically. The association between worldwide crude oil price and the probability of reporting a respiratory illness was large and statistically significant among children whose families cooked indoors but was substantively and statistically insignificant among those living in houses with outdoor kitchens.

TABLE 5.

Models for Child Respiratory Symptoms Stratified by Community Availability of Liquid Propane Gasoline and Location of Kitchen: Guatemala, 2000

	Communities With LPG Markets		Communities Without LPG Markets		House Has Outdoor Kitchen		House Has Indoor Kitchen
	B (95% CI)	No. of Households	B (95% CI)	No. of Households	B (95% CI)	No. of Households	B (95% CI)	No. of Households
AvgWCOPa	0.022** (0.003, 0.042)	4173	0.001 (−0.045, 0.043)	1483	0.004 (−0.030, 0.037)	1455	0.019** (0.002, 0.036)	4208

Note. AvgWCOP = average worldwide crude oil price; CI = confidence interval; LPG = liquid propane gasoline. The dependent variable in all models was respiratory symptoms. Coefficients shown are probit marginal effects. Each column is a separate regression estimate for the sample of children stratified by the variable indicated at the top of the column. All models included control for household asset index as well as floor and roof materials whether indigenous languages were spoken at home, whether someone in the household smoked, whether the mother and father were educated, whether the father lived at home at the time of survey, whether residing in an urban area, and whether the household was interviewed during the rainy season. CIs were computed by using standard errors corrected for clustering at the department × area level.

^aAvgWCOP refers to the average worldwide crude oil price in the 4 wk preceding the date a given household was interviewed.

^**P < .05.

DISCUSSION

Exposure to smoke from biomass fuel use poses a significant public health risk in the developing world. Understanding the factors that induce households to use biomass fuels is an important step toward designing policies that encourage households to choose cleaner alternatives. We considered one such factor: changes in worldwide fuel prices. To our knowledge, ours is the first study to address this research question. We found that increases in worldwide oil prices were associated with a reduced likelihood of using liquid propane gasoline to cook and an increased likelihood of using wood to cook. These associations were strongest for those households in the middle of the distribution of socioeconomic status.

We then examined the relationship between worldwide crude oil price and child respiratory health. We found that increases in the average worldwide crude oil price were associated with a higher probability of reporting respiratory symptoms. Although our estimate was statistically significant at only the 10% level, its magnitude was substantively large: moving from the lowest worldwide crude oil price in the study period to the highest was associated with a 9% point (6 × 1.5) increase in respiratory symptoms, an effect that was comparable in magnitude with the estimates on household smoking and rural residence. Furthermore, these effects were even larger in magnitude and more precisely estimated for the subsample of children living in areas where liquid propane gasoline was sold, which made switching between fuels less difficult, and in houses with an indoor kitchen.

Limitations

Several limitations to our study should be addressed in future work. First, our data came from a cross-sectional survey. Having multiple measurements for a given household over time, particularly at a high frequency, would allow for a better assessment of the (temporal) relationships between prices, biomass fuel use, and health outcomes. Second, although we had access to a rich set of control variables information on other potentially confounding environmental factors, such as ambient outdoor air pollution from vehicular or industrial sources, would have been useful in further characterizing and interpreting our results. Along these lines, our measures of fuel use and child health were based on self-reports and can only be viewed as approximations of the true indoor air pollution burden and respiratory illness, respectively. Furthermore, self-reports may be subject to a variety of nonrandom reporting biases. Although our use of a rich set of control variables helps to address this issue, objective measures of household fuel use, pollutants, and respiratory illness, as used in some studies, would be more desirable for future work.^8,31–33

Conclusions

Despite these potential limitations, our study makes important contributions. First, although previous studies established a correlation between biomass fuel use and respiratory health, few assessed the determinants of fuel choice. Our results clearly indicated the role of fuel prices on household fuel use. Our study also adds to the body of work examining the impact of price shocks,^34,35 economic shocks, and insecurity more generally^36–40 on household food purchases, decision-making, and health outcomes. Additionally, our results add to a broader literature that investigates the impact of indoor and outdoor air pollution on health outcomes.^3,41–44

Our findings also have policy implications. Most efforts to reduce biomass fuel use and indoor air pollution, such as fuel subsidies, the provision of cleaner-burning biomass cookstoves, and educational campaigns, have aimed to increase the long-run adoption of clean fuels, particularly among poor individuals.³ Although such efforts continue to warrant attention, the results presented here indicate that policies aimed at influencing fuel choice behavior in the shorter run also may be important for public health.

Strategies that encourage long-run adoption should target those who cannot afford clean fuels. However, policies aimed at undesirable fuel choice decisions in response to price shocks should focus on those households that are on the margin of switching between biomass and clean options. As seen in our analysis, these households likely are better off because they can afford to purchase clean fuels, and the necessary equipment to cook with them, but not so well off as to be unaffected by price changes. This suggests that policies such as fuel subsidies or price controls, which in some contexts have been shown to be more beneficial to middle- and upper-income families,⁴⁵ may be more cost-effective at discouraging short-run use of dirty fuels than promoting long-run clean fuel adoption. Similarly, programs providing improved biomass cookstoves, which typically target the poor, could be expanded to cover middle-income households so that pollution levels are reduced when these households are forced to use biomass fuels in response to price shocks. Ultimately, choosing the appropriate set of policies to address the impact of fuel price fluctuations on household fuel use requires careful consideration of the costs of such interventions as well as the value of all potential benefits.