Widespread Clean Cooking Fuel Scale-Up and under-5 Lower Respiratory Infection Mortality: An Ecological Analysis in Ecuador, 1990–2019

Background: Nationwide household transitions to the use of clean-burning cooking fuels are a promising pathway to reducing under-5 lower respiratory infection (LRI) mortality, the leading cause of child mortality globally, but such transitions are rare and evidence supporting an association between increased clean fuel use and improved health is limited. Objectives: This study aimed to investigate the association between increased primary clean cooking fuel use and under-5 LRI mortality in Ecuador between 1990 and 2019. Methods: We documented cooking fuel use and cause-coded child mortalities at the canton (county) level in Ecuador from 1990 to 2019 (in four periods, 1988–1992, 1999–2003, 2008–2012, and 2015–2019). We characterized the association between clean fuel use and the rate of under-5 LRI mortalities at the canton level using quasi-Poisson generalized linear and generalized additive models, accounting for potential confounding variables that characterize wealth, urbanization, and child health care and vaccination rates, as well as canton and period fixed effects. We estimated averted under-5 LRI mortalities accrued over 30 y by predicting a counterfactual count of canton-period under-5 LRI mortalities were clean fuel use to not have increased and comparing with predicted canton-period under-5 LRI mortalities from our model and observed data. Results: From 1990 to 2019, the proportion of households primarily using a clean cooking fuel increased from 59% to 95%, and under-5 LRI mortality fell from 28 to 7 per 100,000 under-5 population. Canton-level clean fuel use was negatively associated with under-5 LRI mortalities in linear and nonlinear models. The nonlinear association suggested a threshold at approximately 60% clean fuel use, above which there was a negative association. Increases in clean fuel use between 1990 and 2019 were associated with an estimated 7,300 averted under-5 LRI mortalities (95% confidence interval: 2,600, 12,100), accounting for nearly 20% of the declines in under-5 LRI mortality observed in Ecuador over the study period. Discussion: Our findings suggest that the widespread household transition from using biomass to clean-burning fuels for cooking reduced under-5 LRI mortalities in Ecuador over the last 30 y. https://doi.org/10.1289/EHP11016


Table of Contents
. Description of data sources. Table S2. Comparison of canton-level outcome, exposure, and potential confounding variables by cantons with any observed under-5 LRI mortalities and those without over the full study period. Table S3. ICD 9 and 10 codes and causes included in under-5 lower respiratory infection mortalities. Table S4. Covariates in the final model, and alternative specifications. Table S5. Descriptive statistics of cantonal under-5 lower respiratory infection mortality, clean fuel use, and covariates in the overall sample and by time period. Table S6. Summarizing results from subset analyses and testing for heterogeneity of effect. Table S7. Estimated averted under-5 LRI mortalities for preferred specification and alternative potential confounder specifications as based on generalized additive mixed models. Table S8. Estimated thresholds and mortality rate ratios for preferred specification and alternative potential confounder specifications in segmented generalized linear model regressions. Table S9. Summary of estimated linear associations across all alternative potential confounder combinations.   Figure S3. Cantons that reached 61% clean fuel use in each study period. 61% CF is the threshold found by breakpoint/segmented regression. Cantons below 61% clean fuel use are shown in grey, and the remainder are illustrated in the color palette used in Figure 1. In the period centered at 1988-1992, 15% of cantons were at 61% primary clean fuel use or above ; 1999-2003: 69%; 2008-2012: 92%; and 2015-2019: 98%. Figure S4. Sex-stratified linear and non-linear associations between %CF and under-5 LRI mortality. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. For sex-stratified analyses, sexspecific under-5 canton-period population estimates are used as an offset. We directly count sexspecific canton-period under-5 population in 1990, 2001, and 2010 based on the census. We can estimate overall under-5 population in the 2015-2019 period as discussed above in Table 1. We use the average sex ratio from 1990, 2001, and 2010 to then estimate sex-specific canton-period under-5 population by multiplying the sex ratio with the overall population. Sex-specific splines are restricted to three degrees of freedom.

Figure S5. Period-stratified linear and non-linear associations between %CF and under-5 LRI mortality.
In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. Period-specific models do not contain canton or period fixed effects. Otherwise, they mirror the preferred specifications. Figure S6. Region-specific linear and non-linear associations. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between cantonlevel %CF and under-5 LRI mortalities among subsets. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period, colored by region. Models mirror the preferred specifications. Figure S7. Estimating the proportion of the decline in under-5 LRI mortalities observed between 1990 and 2019 attributable to increased clean fuel use. Kinks in the plot are associated with period fixed effects (1988-1992 used between 1990 and 1996; 1999-2003 between 1997 and 2006; 2008-2012 between 2007 and 2013; 2015-2019 between 2014 and 2019). Only the 1990 fixed effect was used in the 'Counterfactual' prediction. The counterfactual prediction is associated with 68,972 under-5 LRI mortalities total across the 30-year time period. The "No %CF increase" prediction yields an estimated 37,145 under-5 LRI mortalities and the "True" prediction yields 29,905 under-5 LRI mortalities over the 30-year time period. Figure S8. Ambient air pollution, clean fuel use, and under-5 LRI mortality. Panel a shows raw ambient PM2.5 concentrations from van Donkelaar et al. (2021) overlaid with canton-province borders in the years designated. Canton ambient PM2.5 concentrations were estimated by taking the average of the pixels that intersected or fell within each canton in each year of the appropriate period. Then, those five estimates were averaged to produce the canton-period estimates. Panel b summarizes the distribution of canton ambient PM2.5 concentrations across periods with violin and box plots (box lines at 25 th , 50 th , and 75 th percentiles and whiskers extend to the maximum and minimum concentrations). World Health Organization interim-1 and guideline shown for annual PM2.5 exposure. Panel c shows the association between canton-period %CF and ambient PM2.5 concentrations per 10 percentage point increase in %CF in micrograms per cubic meter in an empty model with only canton and period fixed effects and an adjusted model that includes: percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; and an individual in the household or the respondent speaks an indigenous language. Panel d shows results from models with an outcome of under-5 LRI mortalities per canton-period (1) with an empty model that includes only ambient PM2.5 and canton and period fixed effects; (2) with an empty model that includes only %CF and canton and period fixed effects (recall this is only the most recent three periods); (3) an empty model that includes ambient PM2.5, %CF, and canton and period fixed effects; and (4) a fully adjusted model that includes ambient PM2.5, %CF, the full range of potential confounding variables in our preferred specification, and canton and period fixed effects. Figure S9. Alternative adjusted linear and non-linear associations between clean fuel use and under-5 LRI mortality rate. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. See Table S4 for full characterization of covariates included.  Table S4 for full characterization of covariates included. Plots are ordered by increasing rate ratio (closer to the null). The empty model (covariates are fixed effects for period and canton only) and preferred specification mortality rate ratios are shown as horizontal lines. The plots show a point estimate and 95% confidence intervals. Figure S12. Full adjusted non-linear association shown where potential confounders are only included as linear covariates and where potential confounders that display a non-linear association with the outcome are modeled non-linearly using penalized splines with three knots. The top panel shows non-linear associations between canton-level %CF and under-5 LRI mortalities for each specification. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. Figure S13. Linear and non-linear association shown for the preferred outcome specification (quasi-Poisson) and an alternative approach, the negative binomial model. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities by model outcome type. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. Figure S14. Linear and non-linear association shown for the preferred specification (fixed effects for canton) and an alternative specification using random intercepts for canton. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities by model type. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. Figure S15. Robustness of the preferred linear and non-linear specifications and empty models (i.e., only canton and period fixed effects) to the inclusion of a regional fixed effect. Models here mirror the main specification but we additionally include a regional fixed effect for which region each given canton observation is in (Amazon, Andes, or Coast). Associations do not meaningfully differ from models without a regional fixed effect, suggesting that our results are not driven by unmeasured regional level confounding. Figure S16. Robustness of the full model and Coastal region subset linear and non-linear associations to the exclusion of the Galapagos islands. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the seconddarkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. In the lower panel, despite the appearance of only two lines there are four plotted. Models that omit the Galapagos islands are nearly identical to those that include the full study sample. Figure S17. Robustness of the full model non-linear associations to the exclusion of the cantons containing Quito and Guayaquil. Non-linear empty model (i.e., unadjusted, only including canton and period fixed effects) and preferred specifications mirror the full model but remove all canton-period observations from the cantons containing Quito and Guayaquil. Figure S18. Province-level linear and non-linear association between percent of households primarily using a clean cooking fuel and under-5 LRI mortality. Models mirror the approach of the main analysis, but aggregate data to the province rather than the canton level. Note that there is no variation in the measure of completeness over the four study periods because it is from a single time point.
Correlations between all potential confounding variables and the exposure and outcome, panels A to T Shading is as follows: Blue are data sources used to construct %CF, covariates, and under-5 population; Orange are data sources used for child health and healthcare (namely, vaccinations and prenatal care); and Green is an overview of mortality data.   16.9 (14.7, 19.5) a Under-5 lower respiratory mortalities represent the yearly average of the years covered in the time period (1990-1992; 1999-2003; 2008-2012; 2015-2019). Therefore, it is possible for a canton-period estimate to not be a whole number. b We estimate under-5 population in the 1990-1992, 1999-2003, and 2008-2012 time periods by counting the number of children under 5 years in the 1990, 2001, and 2010 censuses, respectively. Because there has not been a survey with national coverage since 2010, we rely on age-specific population estimates produced by the Ecuadorian National statistical agency (INEC) that are based on the most recent Census, the national birth and death registries, and data on migration and immigration, among other factors. These can be found freely at https://sni.gob.ec/proyecciones-y-estudios-demograficos. We average the yearly estimates from 2015-2019 to produce the canton-period estimates. c Cooking fuel options included: piped/centralized gas, gas cylinders, electricity, kerosene (locally referred to as kerex), firewood, charcoal, and gasoline. Clean fuel options included piped gas, gas cylinders, and electricity. d Roof material options included: mixed concrete, asbestos-reinforced concrete sheets, zinc corrugated sheets, ceramic shingles, straw, and other. We specified a "nicer" roof material to be anything other than straw. e Wall material options included: mixed concrete / bricks / concrete blocks, mud bricks / mud walls, wooden slats, reinforced bamboo, non-reinforced bamboo, and other. We specified the nicest wall material to be mixed concrete / bricks / concrete blocks. f Floor material options included: tiled floors (broad category that may include other options), ceramic tiles / stone tiles / vinyl flooring, laminate / wood laminate, bricks / concrete, bamboo, and dirt. We specified the "nicer" floor materials to be anything other than bamboo or dirt flooring. g Indices are produced using the first component from principal components analysis. Canton-period indices are produced by subtracting a given canton-period estimate from the overall parameter mean, dividing by the scaling factor, and multiplying by the first principal component. Then, all parameters are summed to produce the index. The household materials index is comprised of roof, wall, and floor materials as specified in d-f; positive values indicate higher quality materials. The household hygiene index is comprised of the household water source, household toilet and solid waste disposal, household trash removal, and household exclusive shower; more negative values indicate more hygienic practices. The vaccine index is comprised of all vaccines other than the pneumococcal conjugate vaccine -3; more negative values indicate greater overall vaccination coverage. h The pneumococcal conjugate vaccine -3 did not exist prior to the 2010 time period in Ecuador. Given that there was no similar vaccine, we assign a 0% coverage value to all cantons in the 1990 and 2001 period. We do not have data on which of the multiple pneumococcal conjugate vaccines was administered in the 2010 and 2015-2019 periods in Ecuador.  a We do not estimate mortality rate ratios from 45% to 55% and 75% to 85% from the non-linear association in the region-specific analyses because of our preferred interaction approach, which renders it difficult to generate these statistics. The reader is referred to the Figure S6 for a visual of the associations.

Supplementary tables
Models adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized. Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized.

Figure S4. Sex-stratified linear and non-linear associations between %CF and under-5 LRI mortality.
In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. For sex-stratified analyses, sex-specific under-5 canton-period population estimates are used as an offset. We directly count sex-specific canton-period under-5 population in 1990, 2001, and 2010 based on the census. We can estimate overall under-5 population in the 2015-2019 period as discussed above in Table 1. We use the average sex ratio from 1990, 2001, and 2010 to then estimate sex-specific canton-period under-5 population by multiplying the sex ratio with the overall population. Sex-specific splines are restricted to three degrees of freedom.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized.

Figure S5. Period-stratified linear and non-linear associations between %CF and under-5 LRI mortality.
In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. Period-specific models do not contain canton or period fixed effects. Otherwise, they mirror the preferred specifications.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized. Figure S6. Region-specific linear and non-linear associations. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities among subsets. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period, colored by region. Models mirror the preferred specifications.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized.  (1988-1992 used between 1990 and 1996; 1999-2003 between 1997 and 2006; 2008-2012 between 2007 and 2013; 2015-2019 between 2014 and 2019). Only the 1990 fixed effect was used in the 'Counterfactual' prediction. The counterfactual prediction is associated with 68,972 under-5 LRI mortalities total across the 30-year time period. The "No %CF increase" prediction yields an estimated 37,145 under-5 LRI mortalities and the "True" prediction yields 29,905 under-5 LRI mortalities over the 30-year time period. Canton ambient PM 2.5 concentrations were estimated by taking the average of the pixels that intersected or fell within each canton in each year of the appropriate period. Then, those five estimates were averaged to produce the canton-period estimates. Panel b summarizes the distribution of canton ambient PM 2.5 concentrations across periods with violin and box plots (box lines at 25 th , 50 th , and 75 th percentiles and whiskers extend to the maximum and minimum concentrations). World Health Organization interim-1 and guideline shown for annual PM 2.5 exposure. Panel c shows the association between canton-period %CF and ambient PM 2.5 concentrations per 10 percentage point increase in %CF in micrograms per cubic meter in an empty model with only canton and period fixed effects and an adjusted model that includes: percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; and an individual in the household or the respondent speaks an indigenous language. Panel d shows results from models with an outcome of under-5 LRI mortalities per canton-period (1) with an empty model that includes only ambient PM 2.5 and canton and period fixed effects; (2) with an empty model that includes only %CF and canton and period fixed effects (recall this is only the most recent three periods); (3) an empty model that includes ambient PM 2.5 , %CF, and canton and period fixed effects; and (4) a fully adjusted model that includes ambient PM 2.5 , %CF, the full range of potential confounding variables in our preferred specification, and canton and period fixed effects. Figure S9. Alternative adjusted linear and non-linear associations between clean fuel use and under-5 LRI mortality rate. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. See Table S4 for full characterization of covariates included.  Table S4 for full characterization of covariates included.

Figure S11. Specification plots showing main effect estimate for a 10-percentage point increase in %CF in quasi-Poisson GLMs with an offset for under 5 population and fixed effects for canton and period.
Specification plots show all potential confounder combinations. Plots are ordered by increasing rate ratio (closer to the null). The empty model (covariates are fixed effects for period and canton only) and preferred specification mortality rate ratios are shown as horizontal lines. The plots show a point estimate and 95% confidence intervals. Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized.

Figure S13. Linear and non-linear association shown for the preferred outcome specification (quasi-Poisson)
and an alternative approach, the negative binomial model. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities by model outcome type. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized.
Figure S14. Linear and non-linear association shown for the preferred specification (fixed effects for canton) and an alternative specification using random intercepts for canton. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. The middle panel shows alternative non-linear associations between canton-level %CF and under-5 LRI mortalities by model type. The bottom panel shows a histogram of the distribution of canton-period %CF estimates over the full time period. Figure S15. Robustness of the preferred linear and non-linear specifications and empty models (i.e., only canton and period fixed effects) to the inclusion of a regional fixed effect. Models here mirror the main specification but we additionally include a regional fixed effect for which region each given canton observation is in (Amazon, Andes, or Coast). Associations do not meaningfully differ from models without a regional fixed effect, suggesting that our results are not driven by unmeasured regional level confounding. Figure S16. Robustness of the full model and Coastal region subset linear and non-linear associations to the exclusion of the Galapagos islands. In the top panel, estimates are shown in white diamonds, the innermost and darkest band is the 66% confidence interval, the second-darkest band is the 80% confidence interval, and the widest and lightest band is the 95% confidence interval. In the lower panel, despite the appearance of only two lines there are four plotted. Models that omit the Galapagos islands are nearly identical to those that include the full study sample. Figure S17. Robustness of the full model non-linear associations to the exclusion of the cantons containing Quito and Guayaquil. Non-linear empty model (i.e., unadjusted, only including canton and period fixed effects) and preferred specifications mirror the full model but remove all canton-period observations from the cantons containing Quito and Guayaquil.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized. Figure S18. Province-level linear and non-linear association between percent of households primarily using a clean cooking fuel and under-5 LRI mortality. Models mirror the approach of the main analysis, but aggregate data to the province rather than the canton level.
Preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women's literacy; under 18 years of age girls' school attendance rate; an individual in the household or the respondent speaks an indigenous language; an index of vaccines administered among children under 5 years; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 years; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if utilized. on province-level %CF in each study period, with fixed effects for province and period, with standard errors clustered at the province level and found no association between the two. Results from the analysis are presented below: Observations: 80 Fixed-effects: province: 20, period: 4 Standard-errors: Clustered (province) Estimate Std. Error t value Pr(>|t|) cf 1.28e-28 8e-29 1.6023 0.12558 ---Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 RMSE: 9.533e-15 Adj. R2: 1 Within R2: 0