2.1. Household survey

The first survey in the sampled households was administered to the head of the household and included questions eliciting demographic information, household income, housing characteristics (such as number of rooms, whether the kitchen has windows or exhaust fan), fuel and stove types (see the online appendix for the questionnaires and health diaries). Table 2 presents the summary statistics of the household characteristics used in the empirical analysis.

Table 2.

Summary of variables used in regressions

2.2. Individual survey

The individual survey was conducted with each member of the household to gather detailed information on smoking status, occupation, time spent in each micro-environment and health status. We used the standardized respiratory health questions of the British Medical Research Council. For children (those aged 15 or below) the individual surveys and time activity information was collected from their mothers (or primary caretakers). The surveys used the recall method to ascertain the specific health problems in the last three months that were self-reported by the individuals, including doctor visits and fees. Given the focus on respiratory health in this study, illnesses reported in the individual survey were classified into three groups by the cardio-respiratory specialist, namely: (1) upper respiratory (illnesses and symptoms related to the upper respiratory tract that could be linked to air pollution, but not necessarily prolonged exposure); (2) lower respiratory (chronic illnesses related to the lower respiratory tract that are likely to occur as a result of prolonged exposure to air pollution); and (3) all other illnesses. In our main estimations, we use the sick days attributed to upper respiratory illness as respiratory sick days and the sick days from lower respiratory illness as chronic respiratory sick days (Cooper et al., Reference Cooper, Sridharan, Uma, D'Souza, Murugesan and Dayal2006).

The time budget (or time spent in the various micro-environments) of these individuals in a day was collected through the individual questionnaire. Responses were further verified by a field assistant when making household measurements. In addition, subjects in each household were provided health diaries (in Marathi, the local language) and asked to record details on type and days of illness, visits to the doctor, doctor fees, work lost and cost of treatment. Table 2 summarizes the key individual level information collected.

2.3. Air pollution measurement

The air pollution monitoring component of the study measured the exposure to both outdoor and household air pollution of the individuals from the sampled households. Environmental monitoring and the time budget survey of individuals for the exposure assessment were carried out for the study (between May 2003 and April 2004). A preliminary survey was conducted which aided in identifying the essential micro-environments necessary for estimating daily exposure. Four micro-environments were selected for the study: (1) indoor or living room, (2) cooking area during cooking, (3) outdoor or ambient, and (4) work area (including mining workers and truck drivers). The assessment of daily exposure entailed measuring concentrations of $PM_{10}$ (respirable suspended particulate matter or RSPM) in these micro-environments. RSPM in cooking and living room micro-environments was collected on a conditioned and pre-weighed filter paper using low volume universal pump (SKC, UK). In the living room, sampling was done for a period of 24 hours in all the sampled households. In the cooking micro-environment, monitoring was carried out for a subset of households during the cooking period (covering 2 or 3 meals cooking in a day) which typically was about 2 to 3 hours in a day.

Outdoor air samples were collected through high volume air samplers (Envirotech, India). The outdoor concentrations were measured in three locations in each of the four mining clusters. One location was chosen for outdoor concentration measurement in the control cluster. The sampling in each location was continuous for three days in two seasons, and the filters were replaced every 8 hours. After sampling, RSPM levels were calculated by the gravimetric method (difference in the weight of filter paper after sampling divided by volume of air sampled). The daily 24-hour average concentration was derived for each cluster from this data. RSPM sampling in the workplace was carried out for working hours in a day (about 8 hours) with a low volume personal air sampler (SKC, UK) for a sub-sample of 18 subjects working in mining-related occupations.

2.4. Health tests and diagnosis

The clinical measures were conducted by trained technicians in local clinics for a sub-sample of individuals from the sampled households. We collected data on the chest X-rays for 769 adults (900 including children) and pulmonary lung function test (PFT) for 668 adults (782 including children). The chest X-ray and PFT reports were analyzed and diagnosed by a cardio-respiratory health specialist for chronic respiratory symptoms. The X-rays are expected to highlight the impacts of long-term exposure while PFT measures lung efficiency/capacity at the time of the test. We use the specialist's interpretation of the reports by creating dummy variables: X-ray symptom (equals 1, if diagnosed “not normal”) and PFT symptom (equals 1, if PFT results were diagnosed as “not okay”). The X-ray reports were provided to the subjects after the radiologist's and specialist's diagnosis.Footnote ²

Table 2 includes summary statistics of the individual characteristics, average 24-hr pollution exposure to $PM_{10}$, respiratory sick days, clinical tests and medical diagnosis. In our sample, the mean age was 32 years and 50 per cent were male. Eleven per cent of the X-ray reports were diagnosed with respiratory problems and just over 4 per cent had below normal PFT measurements.

2.5. Fuel usage

In the overall sample, the fuel categories of biomass only, liquefied petroleum gas (LPG) only, and biomass and LPG account for almost equal proportions (table 3). However, there are sharp contrasts in the shares of fuels among the clusters. As expected, the control cluster, which is a relatively less connected region, has a very high proportion of households (79 per cent) that use biomass fuels only. In contrast, the corridor cluster, with better road connectivity and where we would expect the highest LPG availability, has the highest proportion of LPG only users (68 per cent). Clusters 1, 2 and 3 exhibited lower LPG usage than the corridor (but higher than the control region) and lower biomass only use compared to the control cluster (but higher than the corridor). The control cluster also had the highest number of kitchens located outside the house, while the corridor had the least. The mean income in the corridor was the highest (lowest in the control region) and the corridor correspondingly has the highest percentage of separate kitchens inside the household (the control the lowest). The income distributions among clusters observed in table 3 partly explain the fuel usage patterns, where the households with higher income (mining activity regions) had higher usage of cleaner fuels compared to the control cluster which had the lowest mean income.Footnote ³

Table 3.

Fuel usage, income and concentration across clusters (%)

Note: Indoor concentration was measured in each household; outdoor in three locations per cluster.

Cooking concentration was measured in a sub-sample for each fuel type which was used to estimate the household concentration based on the fuels used. The fuel use percentages do not add up to 100% as some households did not have a kitchen (or do not report cooking).

Table 3 also shows that the outdoor air quality (discussed in detail in the next subsection) was the worst in the corridor, more than seven times higher than the control. Due to high LPG usage, the cooking concentration is the lowest among households in the corridor. Note that the indoor concentration will be affected both by outdoor air quality (due to infiltration) as well as cooking. The high concentration of $PM_{10}$ indoors among households in the corridor region (despite having the lowest cooking concentration) suggests that infiltration of pollutants from the outside can affect indoor air quality.

2.6. Air quality and exposure

We construct the total 24-hr exposure for each individual by computing the exposure in each micro-environment (share of the day spent in the micro-environment $\times$ concentration in the micro-environment) and summing it over all the micro-environments. We measured outdoor at the village level and indoor in the living area of all households, while cooking measurements from a subset were used with information about the fuel choice in the household to get the cooking concentration.

Table 4 illustrates the data and calculations for one of the individuals (anonymized) in the sample. We multiply the concentration in each micro-environment by the time spent by the individual in each micro-environment in a day, to arrive at the 24-hr exposure (see the last column in table 4). We then divide the total 24-hr exposure by 24 hours to arrive at the average 24-hr exposure. Thus, the units of concentration and total exposure (this is a weighted average of concentrations, with weights being the fraction of time spent in each micro-environment) are the same, $\mu g/m^{3}$. Although the workplace exposure was measured for those working in the mines, mining offices or driving, for most individuals in the sample, workplace exposure was not applicable (as in the case of the individual in table 4).

Table 4.

Illustration of total exposure calculation for an individual

The average 24-hr exposure for this individual is equal to:

\[ \frac{\sum_{m} \text{Concentration}_{m} \times \text{time spent}_{m}}{24} = 5732 / 24 = 239\,\mu g/m^3.\]

2.7. Cumulative exposure

The cumulative exposure to air pollution is the total 24-hr exposure to pollutants summed up over the years of residence for the individual in the regionFootnote ⁴ as:

Cumulative exposure$_{i}$ = total 24-hr exposure$_{i}\times 365 \times$ exposure years$_{i},$ which captures the accumulated exposure to air pollution over the years for each individual living in a particular environment which determines respiratory health. Therefore by construction, time spent in polluting micro-environments and their concentration will have a positive relationship with cumulative exposure. Biomass fuel usage will directly enter cumulative exposure via higher concentrations in the kitchen and indoor environment and correspondingly affect health (Das et al., Reference Das, Pedit, Handa and Jagger2018; Jeuland et al., Reference Jeuland, Soo J-S and Shindell2018; Pattanayak et al., Reference Pattanayak, Jeuland, Lewis, Usmani, Brooks and Bhojvaid2019).

Figure 1 captures the key argument of this paper. In the top panels of figure 1, we see that the distribution of outdoor concentration is very different from that of indoor concentration and cumulative exposure. In the far left scatter plot in the bottom panels of figure 1, in which we have plotted indoor concentration on the y-axis and outdoor concentration on the x-axis, we can see that there is a very low correlation between the two. Some observations are characterised by high values of indoor concentration and low values of outdoor concentration. This reinforces the claim that using either as a measure of exposure is inadequate. Outdoor concentrations only vary by cluster, and would be particularly inadequate, though their measurements would be reasonably accurate. Studies which focus on ambient (outdoor) concentration or household (indoor) air pollution in isolation may also fail to document the relationship between outdoor and indoor air quality in such a setting. In the other two scatterplots in the bottom panels of figure 1, we see that cumulative exposure has a weak positive relationship with outdoor concentration and a relatively stronger positive relationship with indoor concentration.

Figure 1.

Relationship among concentrations (${\mu} g/m^{3}$) and cumulative exposure (million ${\mu} g/m^{3}$ hours). Top row shows distributions of outdoor, indoor concentrations and cumulative exposure; bottom row shows scatterplots.

Table 5 presents the two sample t-test for difference of means in exposure for the four mining clusters compared to the control cluster. Outdoor exposure in column (1) is higher in all the clusters (with mining activity) compared to the control region (with no mining activity), with Cluster 4 (the mining corridor) recording the highest outdoor exposure. On the other hand, Cluster 3 and the Corridor have lower cooking exposure (column (3)) compared to the control region, due to a higher proportion of LPG usage. The average 24-hr exposure in column (4) is a weighted measure of exposure to different micro-environments and is higher for all four clusters compared to the control.

Table 5.

Difference in individuals’ exposure to $PM_{10}$ (vs. control)

Standard errors in parentheses; **$p<0.05$, ***$p<0.01$

Table 6 reports the time spent in micro-environments as elicited in the individual recall survey. The field assistants were able to verify the reported time spent during the household air quality measurements, but this would not completely address the issues with recall methods. In the empirical results section, we discuss how we try to address this concern.

Table 6.

Time spent in micro-environments (hrs/day) and exposure

Standard deviation in parentheses; $^{\ast \ast }$ $p<0.05$, $^{\ast \ast \ast }$ $p<0.01$.

Men (adult males) spend 8.4 hours outdoors on average, and women (adult females) about half of that. Time spent by women in the kitchen on average is about 3.4 hours, while men on average spend less than half an hour. And yet, on average, the 24 hour average exposure is 280 micrograms per cubic metre for males and 277 for females, so total exposure balances out on average, in line with this paper's argument that we need to consider micro-environments together rather than separately.

3.1. Theoretical model

We examine a household which consists of a child, an adult male and an adult female. We assume that a household aims to maximize its utility ($U$) which is a function of sickness ($S$) experienced by the child (indexed by $C$), the adult male (indexed by $AM$) and the adult female (indexed by $AF$), and non-food consumption $(C^{NF})$, so $U=U(S^{C},\,S^{AM},\,S^{AF},\,C^{NF}).$

We assume that sickness is a function of total exposure to air pollution ($E$ ), consumption of cooked food ($CF$), doctor-visits ($D$) and individual characteristics ($Z$), so $S^{i}=S^{i}(E^{i},\,CF^{i},\,D^{i};Z^{i}),$ where $i=C,\,AM,$ and $AF$.

The kinds of sickness that result from poor nutrition and from household pollution are different. The knowledge of or beliefs in causes of sickness of different sorts is a key variable that influences the household's actions (Jeuland et al., Reference Jeuland, Pattanayak and Bluffstone2015b).

Total exposure is a weighted sum of exposure in different micro-environments, which in turn, are equal to the product of time spent in these micro-environments ($t$) and the concentrations of air pollution in these micro-environments ($e$). We consider four micro-environments on which we have data: outdoors, indexed by $o$; cooking, indexed by $c$; work, indexed by $w$; and indoors, indexed by $i$,

$E^{i}=t_{o}^{i}e_{o}+t_{c}^{i}e_{c}+t_{w}^{i}e_{w}+t_{in}^{i}e_{in}.$

While the time spent in different micro-environments is person specific, the concentrations are not. To simplify, we assume that the time spent by the child in the four different micro-environments is the same as that of the adult female.

In our sample, almost all households cook with LPG or biomass. We take $t_{c}^{i}$, the time in the cooking micro-environment, to be the sum of $t_{c}^{lpg}$ and $t_{c}^{b}$, the time cooking with LPG and biomass, respectively. This is an approximation, since it is possible that LPG may be used with biomass at the same time. The key point though is that greater use of LPG is likely to reduce the amount of biomass burnt.

In our sample, cooking is mainly done by women, and so we assume that the adult female does the cooking. The concentration of air pollution in the cooking environment $(e_{c})$ is a function of the concentration outdoors (or ambient concentration) and the length and type of cooking, so

\[e_{c}=e_{c}(t_{c}^{lpg},\,t_{c}^{b},\,e_{o}).\]

We note that the concentration outdoors will be influenced by the total cooking pattern in a village; most notably the contrast will be between a village where every household uses LPG only and a village where every household uses biomass only.

Similar to the concentration of air pollution in the cooking environment, the concentration indoors will depend on time cooking and the concentration outdoors, such that

\[e_{in}=e_{in}(t_{c}^{lpg},\,t_{c}^{b},\,e_{o}).\]

The total amount of food cooked in the household is a function of the time spent cooking:

(1)\begin{equation} C_{F}=C_{F}(t_{c}^{lpg},\,t_{c}^{b}).\end{equation}

Equation (1) may give the impression that more cooked food requires more cooking time irrespective of fuel, but LPG cooking can reduce cooking time compared to biomass cooking.

$C_{F}^{i}$, the amount of food consumed by each family member, is assumed to be some norm-based share $(\theta ^{i}\in \lbrack 0,\,1])$ of the total amount of food cooked in the household. The amount of raw food consumed $(R_{F})$ is assumed to be a linear function of the food cooked, $R_{F}=\eta _{1}C_{F},$ where $\eta _{1}$ is a constant; this is an approximation. With LPG we can quickly vary the intensity, from off to medium and high, but with biomass burning, it is more like a batch process. Similarly, the amount of fuel used ($q$) is assumed to be a linear function of the time spent cooking:

\[q^{LPG}=\eta _{2}t_{c}^{lpg}, \quad \textrm{and} \quad q^{B}=\eta _{3}t_{c}^{b}.\]

We also assume that a certain proportion $(\eta _{4}\in \lbrack 0,\,1])$ of the biomass fuel is gathered and we assume that it is the adult female who gathers biomass fuel, $q^{BG}=\eta _{4}\eta _{3}t_{c}^{b}.$

The time spent in gathering this fuel $(t_{c}^{g})$ is proportional to the quantity to be gathered. This is an approximation; for example, the same person may gather the same amount of fuel from different locations at different times, taking different time to gather the same amount of fuel, because the gathering of fuel may be combined with some other activity, $(t_{c}^{g})=\eta _{5}\eta _{4}\eta _{3}t_{c}^{b}.$

We assume, based on examining our data (see table 6 and associated discussion) that the amount an individual works is predetermined by the occupation of the person. In other words, the amount an individual works is not influenced by marginal cost and benefit considerations, and for this model, is predetermined. We assume that after cooking, working and gathering biomass, the adult female divides her remaining time in some given proportion $(\alpha _{AF})$ between the indoor $(t_{in}^{AF})$ and outdoor micro-environments. Total time outside $(t_{o}^{AF})$ is equal to remaining time spent outside and time gathering biomass,

\begin{align*}t_{in}^{AF}& =\alpha _{AF}(T^{AF}-t_{c}^{AF}-t_{w}^{AF}-t_{g}^{AF}),\\ t_{o}^{AF}& =T^{AF}-t_{in}^{AF}.\end{align*}

Since the adult male does not cook or gather biomass, the expressions for time indoors and time outdoors are different in the case of the adult male,

\begin{align*}t_{in}^{AM}& =\alpha _{AM}(T^{AM}-t_{w}^{AM}),\\ t_{o}^{AM}& =T^{AM}-t_{in}^{AM}.\end{align*}

The household maximizes utility subject to the following budget constraint:

\[t_{w}^{AM}P_{w}+t_{w}^{AF}P_{w}=P_{NF}C_{NF}+P_{r}R_{F}+P_{D}\sum D^{i}+q_{P}^{B}P_{B}+q_{LPG}P_{LPG},\]

by choosing $C_{NF},\,t_{c}^{lpg},\,t_{c}^{b}$ and $D^{i}.$

The first-order conditions are (denoting the Lagrange by L):

(2)\begin{equation} \frac{\partial L}{\partial C_{NF}}=\frac{\partial U}{\partial C_{NF}} +\lambda \lbrack -P_{NF}]=0.\end{equation}

Equation (2) is the usual consumer theory condition for consumption and says that the marginal utility from an additional unit of consumption should equal the marginal cost in utility terms, which is the product of the multiplier and the price.

(3)\begin{equation} \frac{\partial L}{\partial D^{i}}=\frac{\partial U}{\partial S^{i}}\frac{ \partial S^{i}}{\partial D^{i}}+\lambda \lbrack -P_{D}]=0.\end{equation}

In equation (3) the marginal benefit of spending a unit of money on doctor visits of the $i^{\text {th}}$ person in the household is the marginal utility of lower sickness of the $i^{\text {th}}$ person times the marginal product (in terms of lower sickness) from an additional doctor visit. The marginal cost is the price of a doctor visit multiplied by the multiplier, so that

\[\frac{\partial L}{\partial t_{c}^{lpg}}=\sum \frac{\partial U}{\partial S^{i}}\left[\frac{\partial S^{i}}{\partial E^{i}}\frac{\partial E^{i}}{\partial t_{c}^{lpg}}+\frac{\partial S^{i}}{\partial C_{F}^{i}}\theta ^{i}\frac{ \partial C_{F}}{\partial t_{c}^{lpg}}\right]-\lambda \left[ \eta _{2}P_{LPG}+\eta _{1}\frac{\partial C_{F}}{\partial t_{c}^{lpg}}P_{r}\right].\]

A change in time spent cooking with LPG or biomass is associated with higher emissions and therefore higher exposure and sickness (of all members), and higher cooked food and therefore lower sickness. It will also entail greater cost concerning the gathering of biomass or expenditure on purchase of fuel and raw food. The household will have imperfect information about the effects of cooking on exposure. Moreover, cooking affects women and children more than adult males since they stay in the cooking micro-environment:

\[\frac{\partial L}{\partial t_{c}^{b}}=\sum \frac{\partial U}{\partial S^{i}} \left[\frac{\partial S^{i}}{\partial E^{i}}\frac{\partial E^{i}}{\partial t_{c}^{b}}+\frac{\partial S^{i}}{\partial C_{F}^{i}}\theta ^{i}\frac{ \partial C_{F}}{\partial t_{c}^{b}}\right]-\lambda \left[ (1-\eta _{4})\eta _{3}P_{B}+\eta _{1}\frac{\partial C_{F}}{\partial t_{c}^{b}}P_{r}\right].\]

A change in time cooking has several effects on exposure, since it affects the time spent in different micro-environments (in the case of the adult female and the child) and the concentration in the indoor and cooking micro-environment. So, for example,

\[\frac{\partial E^{AF}}{\partial t_{c}^{lpg}}={-}e_{o}+e_{c}+\frac{\partial e_{c}}{\partial t_{c}^{lpg}}t_{c}-e_{in}+\frac{\partial e_{in}}{\partial t_{c}^{lpg}}t_{in}^{AF}.\]

Our model is static for simplicity. However, in reality, what we witness today is the outcome of the past. Mining tends to follow a life-cycle, with the initial expansion of mining and economic activity in an area finally leading to a slowing down of mining as new areas are found and exploited. During this mining life-cycle, the economic context and the environment (of which air pollution is one indicator) of the households change. Moreover, human health is affected by cumulative exposure, especially in the case of chronic air pollution-related ailments. In our main estimations, we study the association between cumulative exposure and health.

4.1. Time in micro-environments

We begin by estimating the associations of time spent in micro-environments (reported in table 7), where we include biomass fuel usage and interact its usage with the female dummy along with the individual and household characteristics, as we discussed in theory (where we assumed that only females gathered biomass). The individual level attributes are age, age-squared, gender and never-smoker (dummy), and the household characteristics include the number of adults and children by gender and whether or not the house was pucca (constructed with solid materials as a permanent dwelling).

Table 7.

Time spent in micro-environments

OLS estimations at individual level; $^{\ast }$ $p<0.10$, $^{\ast \ast }$ $p<0.05$, $^{\ast \ast \ast }$ $p<0.01$.

Standard errors in parentheses and clustered at household level.

Other controls: number of adults and children (by gender).

Table 7 presents the results of regressions on the dependent variables of time spent: indoors, outdoors, in the kitchen and at work. We control for cluster level differences by including cluster dummies in our estimation. For time spent in the kitchen, we examine mean time spent in the kitchen by adults in the household. The time adults spend in the kitchen is expected to depend on the composition of adults and children, since one person can cook for several members. We therefore include the number of adults and children in the household by gender in these regressions.

Table 7 shows that age and gender are statistically significant regressors. Age is related negatively to time spent indoors, shown in column (1), but positively with time in the kitchen, shown in column (3). Males spent less time indoors and in the kitchen and more time outside the house or working. Age and education have a positive relationship with time spent working (column (4)). We call the reader's attention to the positive relationship between biomass fuel usage and time spent in the kitchen (column (3)). Also noteworthy is the positive relationship between biomass usage $\times$ female (dummy) on time spent outdoors (which is consistent with the assumption in our model that females spent time gathering biomass fuels).

4.2. Choice of fuel

In table 8, we present the relationship between household characteristics and their choice of fuel, biomass or LPG. The unit of observation here is the household (N = 308) and we estimate a linear probability model.Footnote ⁶ As a robustness check, we estimated the models with a binary dependent variable using a maximum likelihood method and find similar results (see columns (2) and (4) in table 8).

Table 8.

Household fuel choice: biomass and LPG

Estimations at household level; probit marginal effects reported in columns (2) & (4).

Standard errors in parentheses; $^{\ast }$ $p<0.10$, $^{\ast \ast }$ $p<0.05$, $^{\ast \ast \ast }$ $p<0.01$.

Other controls: number of adults and children in the household (by gender).

The dependent variables in table 8 are households who used biomass or LPG for cooking. We include the cluster dummies in the specifications. The regressor pucca house (dummy) is negatively associated with biomass only used for cooking – columns (1) and (2) – and positively for LPG only – columns (3) and (4), as pucca house proxies for higher income households. As expected, all four (mining-related) clusters are negatively associated with biomass only used for cooking compared to the control cluster. Except for Cluster 1, the other three mining clusters are more likely to be using LPG for cooking.

4.3. Health indicators

We now examine the association between cumulative exposure and health. In table 9, we present the results for both reported health indicators (respiratory and chronic respiratory sick days) and clinically-diagnosed respiratory health (expert's diagnosis of the X-ray and lung function test (PFT)). The respiratory sick days (e.g., laryngitis, sinusitis, pharyngitis) and chronic respiratory sick days (e.g., asthma, bronchitis, wheezing, emphysema) were self-reported by the participants. According to clinical experts, respiratory health (measured by X-ray reports) is a function of cumulative exposure rather than immediate exposure (Cooper et al., Reference Cooper, Sridharan, Uma, D'Souza, Murugesan and Dayal2006). As our key variable of interest is cumulative exposure to air pollution, the cardio-respiratory expert's diagnosis of X-rays provides the best measure of respiratory health for our purposes.Footnote ⁷ The pulmonary function test (as clinically measured with the peak flow meter instrument) indicates age-specific lung capacity and can be influenced by immediate 24-hr exposure.

Table 9.

Cumulative exposure and health indicators

Standard errors (clustered at household level) in parentheses.

*$p<0.10$, **$p<0.05$, ***$p<0.01$. Other controls: male and pucca (dummies).

Estimated only for adults (so numbers lower than table 2).

Table 9 presents the main results of the paper, the association of health measures and cumulative exposure to air pollution for adults. In addition to the detailed survey questionnaire administered by local enumerators, sampled households were provided with individual health diaries to record the type of ailment, date, number of days sick, number of visits to the doctor, doctor's fees and any additional comments. We chose the self-reported sick days for respiratory and chronic respiratory illness as the dependent variables for the results presented in columns (1) and (2). The key variable of interest, cumulative exposure, is statistically significant and positive, indicating a positive association between exposure and respiratory sick days. In all the estimates in table 9, we control for individual (age, age-squared, education, male dummy), household level pucca dummy and cluster dummies to account for fixed effects at the regional level. A one-unit change in cumulative exposure is associated with a 0.0529-unit change in respiratory sick days, shown in column (1), and a 0.0378-unit change in chronic respiratory sick days, shown in column (2). In terms of elasticity, a 1 per cent change in cumulative exposure (at the mean) is associated with a 0.79 per cent increase in respiratory sick days and a 0.86 per cent increase in chronic respiratory sick days.

Lastly, a crucial concern in the literature when using self-reported measures of health as the dependent variable are issues of under- (or over-) reporting (Short et al., Reference Short, Goetzel, Pei, Tabrizi, Ozminkowski and Gibson2009; Vaillant and Wolff, Reference Vaillant and Wolff2012). The use of health diaries may mitigate the concerns with self-reported health based on recall methods but does not completely address issues of the heterogeneity problem in reporting, since different populations may use different threshold levels when asked about their health (Shmueli, Reference Shmueli2003; Lindeboom and Van Doorslaer, Reference Lindeboom and Van Doorslaer2004). Studies find correlations between attributes such as education and self-reported health which may arise from measurement errors in self-assessment. We ameliorate some of these concerns by controlling for education and income. Furthermore, the results in columns (3) and (4) in table 9, where the dependent variables are diagnosed clinical measures of health, are consistent with our findings with self-reported health.

As noted, we were advised by the cardio-respiratory specialist that respiratory health (indicated by X-ray reports) is a function of cumulative exposure. Therefore we argue that a diagnosed respiratory issue with the X-ray reports is the key health indicator in our study (column (3) in table 9). The pulmonary function test (as clinically measured with the peak flow meter instrument) indicates age-specific lung capacity and can be influenced by immediate 24-hr exposure, so it will be more responsive to 24-hr average exposure as a determinant.

Columns (3) and (4) in table 9 present the results for the relationship between cumulative exposure and clinically-diagnosed respiratory health status. A sub-sample (about 50 per cent of the total) of individuals volunteered for these medical tests that were offered for free and these observations are therefore lower than previous individual level regressions. In column (3) we use the X-ray diagnosis by the respiratory health expert and find a positive relationship between cumulative exposure to air pollution and an X-ray report diagnosed with respiratory problems. In terms of elasticity, a 1 per cent change in cumulative exposure is associated with a 0.90 per cent change in the likelihood of an X-ray report diagnosing a respiratory issue. Column (4) reports a positive association between cumulative exposure and the lung function test (i.e., ‘PFT not okay’) although not statistically significant. A 1 per cent change in exposure is associated with a 0.75 per cent change in the PFT measure recording an abnormality. As we noted, PFT is responsive to recent exposure and therefore noisily captures long-run effects.

The finding in our study that the relationship between cumulative exposure and health indicators is similar (in terms of sign and significance for X-ray measure) for both self-reported measures as well as clinical assessments is useful to related studies. The detailed clinical assessment and medical expert diagnosis, as in our study, may be infeasible to collect or the data and resources may not be available. The positive correlation we find between clinical and self-reported health illustrates the value of other field studies even if they only use self-reported health.

4.4. Study limitations

Our study has some limitations. Firstly, part of the study uses survey data and the recall method for self-reported health, which is open to measurement errors and biases. Self-reports are amenable to social desirability biases when responding to questions about health (Ezzati et al., Reference Ezzati, Martin, Skjold, Hoorn and Murray2006), for example, when responding to questions about smoking habits in our survey. Our provision of health diaries at the start of the study to all sampled households could have potentially improved participant's recording and recall during the health survey. We find consistent results with the clinical measures of health. Participants’ reports of time spent in micro-environments can be affected by such biases as well, but the presence of field assistants in the households during the indoor and kitchen concentration measurements and their independent verification of time spent should constrain the bias.

Although we cannot make causal claims in the paper about the effect of mining or traditional fuel usage on health, our elaborate data collection allows us to make careful inferences about source apportionment for pollution. We treat the assignment of mining activity as exogenous to households in our computation of cumulative exposure, but selective in- or out-migration could bias our estimates. Even if we do not deal with the out-migration issue, the fact that 77 per cent of our sampled households were originally from the cluster (the main results are qualitatively similar when restricting the analysis to this sub-sample) allows us to have confidence in the results.

Despite our attempts to tie the theory closely to the data collection process that allowed us to apportion exposure to pollution sources, we were still limited in our empirical strategy by the data. Our measure of cumulative exposure assumes that the current 24-hr exposure is indicative of exposure across the years for the individuals living in the location. But pollution levels could have varied considerably over the years in the locations which we do not account for in our exposure construction. Similarly, we do not measure cumulative smoking years of individuals. Moreover, we measured $PM_{10}$ rather than $PM_{2.5}$, which could arguably be a better indicator of respirable pollutants. We only measured particulate matter concentrations while health is also impacted by other noxious matter (e.g., sulphur oxides).

Sophisticated treatments of costs and benefits have been published since the data collection for this study (Jeuland et al., Reference Jeuland, Bhojvaid, Kar, Lewis, Patange and Pattanayak2015a, Reference Jeuland, Soo J-S and Shindell2018). Our particular contribution is the incorporation of total exposure and micro-environments. Air pollution valuation studies may to some extent abstract from that or, at times, simply ignore HAP.