How do mining booms impact the labor market for men and women? Evidence from Indonesia

Abstract

We examine the effects of mining booms in Indonesia on labor market outcomes using exogenous price changes and 452 mines. We do this using labor force surveys between the years 1998 and 2011, and four waves of individual panel data between 1997 and 2014. Surprisingly, female incomes grow during mining booms, not because women work more, but because their work moves from the agricultural to the service sector where paid work is more common. Men experience mixed labor market changes. High average mining incomes attract male labor to mining districts, allowing for some adjustment of labor supply to demand. Suggestive evidence also shows that informal work increases marginally for men, potentially in auxiliary mining jobs. A male dominated industry that supports economic opportunities for women can unexpectedly benefit women as well.

1 Introduction

This paper examines how the growth of a male dominated industry, mining, affects labor market outcomes for men and women. Two factors render the mining industry particularly interesting to study. First, globally women represent between 8 and 17 per cent of the mining labor force (Fernandez-Stark et al., 2019),¹ Second, mining booms depend largely on the availability and value of the natural resource being mined. Variation in mining growth depends on the size of resources available, how difficult they are to obtain, and their value in the global market.

We focus on the mining industry in Indonesia, a country particularly rich in minerals and mining. Mineral rents contributed 1.9 per cent to Indonesia's GDP in 2021, which is twice the global average (World Bank, 2023). While mining comprises 1.5 per cent of overall Indonesian employment, 7 to 10 per cent of men in districts with active mines are employed in mining. Further, the average mining salary is twice that of non-miners (Sakernas, 2021), which makes mining jobs relatively attractive.

We use mining data from the SNL Metals and Mining Database (2018),² which has information on 452 industrial mines across Indonesia. The data are matched with the Indonesian labor force surveys, the Sakernas, between 1998 and 2011, and with four waves of individual panel data of the Indonesian Family Life Survey between 1997 and 2014. These data contain detailed information on migration and allow us to examine how labor moves between districts in response to changes in the mining industry.

To identify the causal effects of mining booms, we exploit exogenous changes in the world price of minerals, as in Berman et al. (2017) and Asher and Novosad (2014). When the world price of a mineral rises, the value of current production increases and existing mines are induced to increase production. About 1.5 per cent of mines also start operation.³ This raises a district's overall mining value, which is increasing in the world price of minerals and in the number of active mines in the district (Berman et al., 2017). The identification relies on world prices being exogenous to other factors within Indonesia, which is plausible because Indonesia does not dominate the global market for most minerals.

Mining booms result in higher monthly incomes for women. While having a greater number of mines is associated with being less economically developed than other districts, a doubling of a district's mining value results in a 7 per cent increase in female monthly incomes, or a 1.5 per cent increase from average female incomes. Income gains are high for low-skilled women, and are reassuringly positively related to the direction of price change. Overall incomes also rise, but there is less evidence that male incomes change.

These results are surprising given that average employment does not strongly improve, though some heterogeneity exists by gender, mine type and the direction of price change. Instead, during mining booms there are larger changes in where individuals work. Employment in agriculture decreases by 0.8 percentage points, and more so for women than for men. Men become more likely to work in services and in mining, and women in services, such as the retail, wholesale, restaurant and trade sectors, household services, and transport and communications.

Simultaneously, women become marginally less likely to work without pay, which is consistent with women in services being three times more likely to be paid than women in agriculture in this context. Most of these shifts occur in rural areas with few other economic opportunities. In these districts, mines are long-term projects. We show evidence that while districts remain mostly agricultural during the early stages of a mine's development, employment in mining increases during a mine's operation and closure stages.

Male labor outcomes may also be reconciled by the fact that men actively move to districts with higher mining values. Using panel data from the Indonesian Family Life Survey (2021), we show that men are 3.3 percentage points more likely to move to a district experiencing a mining boom. Women, on the other hand, only move to urban booming mine districts. Instead, the same woman is 1.9 percentage points less likely to move away from a mining town during a boom if she was already there.

We were not able to test the effects of male dominated industry booms relative to female dominated industry booms due to lack of data. Plausibly, the latter may be more beneficial for women. Instead, the results suggest that local growth accompanying the boom of a male dominated industry such as mining can benefit women in some cases, especially if growth supports opportunities for women in sectors outside of the low paying sectors in which they usually work.

This paper contributes to two branches of literature.

First, it contributes to the literature on the effects of natural resource discovery on economic outcomes, which has yielded mixed results for non-resource tradeable goods (Sachs and Warner, 1995; Aragón and Rud, 2013; Allcott and Keniston, 2018), and female labor force participation (Ross, 2012; Kotsadam and Tolonen, 2016). Most empirical work has relied on repeated cross sections, often across countries, and examined the effects of certain types of mines. We expand on existing research by (i) studying booms of a large number of mines in a single country, and (ii) providing supportive evidence using panel data on how individuals are impacted by booms over time. Our methods allow us to first examine boom effects broadly, and then to narrow our analysis for heterogeneous effects. We provide evidence of structural shifts in the local economy, where mining booms support service industries linked to growth and consumption, instead of to input factors.

Second, a limited recent literature has also explored the effects of mining shocks on local labor markets (Asher and Novosad, 2014; Loayza and Rigolini, 2016; Allcott and Keniston, 2018). We show that mining booms create movements locally which are different for women and men. In fact, for men in particular, open labor markets where labor can move relatively freely may be a reason why equilibrium incomes and quantities adjust and do not change by large magnitudes. Further, given that occupations are often segregated by gender (Goldin, 2014; Blau and Kahn, 2017; Wasserman, 2023), we show that when the mining industry expands, women take specific jobs: they become employed in services and not in mining itself. Overall, our findings imply that even the boom of a male dominated industry can benefit women in some ways, if the growth of sectors that employ women is supported.

The remainder of the paper is structured as follows. We provide contextual background and institutional details in section 2, followed by a discussion of our identification strategy and data in section 3. The results, robustness checks and mechanisms are provided in section 4, and we conclude in section 5.

2 Background

2.0.0.1 The mining sector in Indonesia:

In 2021, Indonesia's mineral rents contributed 1.9 per cent to its GDP, which is twice the global average of 0.8 per cent, and more than the US average of 0.1 per cent (World Bank, 2023). As seen in figure A1 in the online appendix, active mines exist on almost every major island in Indonesia, with a special concentration on the islands of East Kalimantan and Sulawesi. Using data on employment and earnings from the Sakernas (2021), we plotted the average share of a district's labor force in mining and the average income of miners by district, shown in figures A5 and A6, where darker shades represent a higher employment share and average income. Corresponding to figure A1, the average mining employment and income is higher in regions with a greater concentration of mines, such as in East Kalimantan. Miners earn twice as much as non-miners on average, and while mining comprises 1.5 per cent of overall employment, it is double that in mining districts, and averages 7 per cent of male employment in districts with active mines (Sakernas, 2021).

The most important minerals in terms of production are coal and gold, which together comprise 70 per cent of Indonesian mines, followed by copper, tin, bauxite and nickel. The diverse mineral production across geographic regions provides variation in production booms and busts across Indonesia. For example, despite the mining sector performing badly in 2015 overall due to decreased coal production in response to declining prices, Sulawesi experienced a healthy growth in mining that year due to investment in nickel smelters (Panggabean, 2016).

2.0.0.2 The types and stages of mining

While mining often occurs in rural and agricultural areas (Agriorbit, 2023), the type of mining practiced often depends on the mineral and its location. Surface mining (also called open-pit mining) is the most commonly practiced mining for extracting deposits close to the ground. Surface mining is increasingly used for minerals such as coal because it is cheaper, less labor intensive and promotes quick land recoveries (Cashman, 2017). However, surface mines also displace agricultural land to a greater degree, at least temporarily (Jinan Fucheng Hydraulic Equipment Co. Ltd, 2021). On the other hand, underground mining is used to extract deposits located deep in the Earth and is generally more expensive and labor intensive (Ranjith et al., 2017). Underground mines disrupt less surface land, but over time can cause surface subsidence (InTeGrate Carleton College, 2021; Jinan Fucheng Hydraulic Equipment Co. Ltd, 2021).

According to the global mining company (Newmont, 2023), the life of a mine can be as long as 30 years, and sometimes longer. Broadly defined, the key stages of mining are: exploration, development and design, construction, production, closure and post-closure. The first three phases explore the mineral's availability and construct the mine. The production stage itself can last for as long as 10 to 30 years, and is followed by the closure and post-closure stages which involve waste disposal and clearing the mine area.

Lawyers, financial service providers, transport, machine operators and extraction helpers all play a role in the mining industry. Some of the earlier stages of mining involve less labor-intensive and more highly paid jobs, such as those done by geoscientists, physicists and engineers. The production phase tends to be more labor intensive, and generally the miners are paid more than comparable low-skilled jobs to compensate for the additional risks of mining.⁴ The lowest paid jobs are those in occupations that support mining, such as extraction helpers, those who perform hand construction and move freight, and cashiers (Itkin, 2006).

3 Empirical strategy

3.1 Data

3.1.0.3 Mining and price data

The SNL Metals and Mining Database (2018) published by S&P Global Market Intelligence is our primary source of mining data.⁵ Mines are located using their geographic coordinates and the district in which they exist. Due to the presence of illegal and small mines, different sources report different numbers of mines (Siddharta, 2019). For example, the publicly available (US Geological Survey, 2010) identifies 75 Indonesians mines in 2010, while the SNL Metals and Mining Database (2018) contains information on 452 mines that year. In general, the proprietary (SNL Metals and Mining Database, 2018) dataset has a greater coverage of mining locations than publicly available data.

Surface mines comprise 43 per cent and underground mines 6 per cent.⁶ The type of mining also varies by mineral. For example, 91 per cent of coal mines in the sample are surface mines and the remaining are underground mines. Meanwhile, 52 per cent of gold mines are underground mines, 26 per cent are placer and the remaining 22 per cent are surface mines. This difference implies that the type of mineral present naturally leads to different methods of mining and labor demand patterns. Inactive mines are also included in the data. These are often in the pre-production stage of mining.⁷

The SNL Metals and Mining Database (2018) was first merged with price data from the World Bank (2018) commodities’ price data using the name of the mineral and the year. The prices are in nominal US dollars per metric ton for all metals excluding gold, and per troy ounce for gold. The prices are yearly averages to match the frequency of the mining data. The merged dataset includes information on nine minerals.⁸

Next, the data are merged at the district and year level with the Sakernas (2021) data for Indonesian labor market outcomes between 1998 and 2011.⁹ Due to massive administrative reorganization between the late 1990s and mid-2010s that led to the renaming and breaking up of districts and subdistricts (Bazzi and Gudgeon, 2021), this is the most computationally feasible level at which the same geographic regions could be mapped over a period spanning three decades.

3.1.0.4 Labor market data

The Sakernas (2021) are the National Labor Force Surveys of Indonesia, and cover the entire territory of the Republic of Indonesia, except for areas involving active conflict (such as East Timor in the early 2000s).¹⁰ We use repeated cross sections of the Sakernas (2021) between 1998 and 2011. Of the 316 districts that are mapped to the same geographic region over time in the data, there are 30 districts with active mines and 71 districts with active or inactive mines. We define a ‘mining district’ as a district with at least one mine, and an ‘active mining district’ as a district with at least one active mine over the entire study period. Both variables are constant over time. Table A1 provides summary statistics on the number of mines of the most important mineral in a district and year and the log of its price for these mining districts.¹¹ The average number of mines of the most important mineral in a district and year is about 0.5 in both datasets, while the log of price ranges from about 3.5 to 10.6.

Since our main outcomes involve working, we restrict the sample to individuals who are between 22 and 60 years old to allow for completed schooling. The top panel of table 1 shows that half the sample is female, and 80 per cent is married. In this mainly rural population (60 per cent), 90 per cent of individuals have either a primary degree or a secondary degree, which is also the compulsory level of education.¹² The variable Employed is an indicator of having worked for at least one hour the previous week (including as unpaid and/or informal farm or family labor), having a job and temporarily not working, or operating a business. About 70 per cent of the sample engaged in market work five days a week, and half the sample also engaged in some housework.

Table 1. Sample summary statistics

Notes: The table presents summary statistics for the variables used in the paper. The top panel represents a sample of individuals aged 22–60, and data are from repeated cross sections of the Sakernas (2021), 1998–2011. The bottom panel represents a sample of individuals aged between 22 and 60, and data are from the Indonesian Family Life Survey (2021) using the waves from 1997, 2000, 2007 and 2014. M$_{kt}$ is the number of mines of the most important mineral in district k and year t, Log(Price)$_{kt}$ is the logarithm of its price, both shown for mining districts.

Monthly income is created by combining the Sakernas (2021) responses on previous month's income if repondents are casual workers and own-account workers, and a normal month's income if they earn a regular income. To exclude income outliers, the incomes are winsorized so that the top and bottom 2 per cent of monthly income are replaced with their previous and next incomes. The incomes are then converted to logarithm values.¹³ The mean for monthly income is 370,000 IDR, or US $24 by 2024 conversion rates,  which falls in the range of a middle-class individual.¹⁴

3.1.0.5 Individual panel data

Industrial changes in a relatively open local economy are likely to induce migration between districts. The Sakernas (2021) records an individual's place of residence five years before the survey,  which provides some,  but limited information on movement. We therefore supplement labor force survey data with migration data from the Indonesian Family Life Survey (2021) which is a panel survey of individuals and contains detailed migration information. The survey rounds that match the price and mining data are from 1997,  2000,  2007 and 2014. Indonesian Family Life Survey (2021) covers 13 provinces in Indonesia (figure A7). The Sakernas (2021) and Indonesian Family Life Survey (2021) often provide different estimates of labor market outcomes (Dong, 2016). Here,  we use the rich information in the Indonesian Family Life Survey (2021) mainly to add depth to our narrative,  and not necessarily to corroborate the findings from the Sakernas (2021) which is more representative of the population.

Retrospective information on employment,  industry and status of work and migration is used to construct a yearly dataset from 1999 to 2014 to match the frequency of the price data.¹⁵ Meanwhile,  income is constructed by summing monthly net salary and business income for a respondent's main and secondary job. As with the Sakernas (2021),  the top and bottom 2 per cent of incomes are replaced with the next incomes in the distribution. The Indonesian Family Life Survey (2021) data matched with the mining data covers 283 districts,  of which 25 districts have active mines,  and 63 have either active or inactive mines over the study period.¹⁶

The lower panel of table 1 provides summary statistics for individuals aged 22 and 60 in the Indonesian Family Life Survey (2021) sample. The sample is similar in many ways to the Sakernas (2021) sample. About half the sample is male and four-fifths is married. The average respondent has nine years of education,  which corresponds to completing junior secondary school. Interestingly,  a third of the sample lives in a municipality different from their birth municipality,  which indicates a substantial incidence of intra-national migration. Since the Indonesian Family Life Survey (2021) explicitly provides information on business profits,  these are included with net salary to create a total income. Income is winsorized to replace the top and bottom 2 per cent of incomes with the next incomes. The mean income is about 894,000 IDR,  or US $57, monthly, which is at the upper end of middle class incomes.¹⁷

3.2 Identification strategy

Mining districts are often unique, including with regard to their geography, economy and demographics. As seen in table A3, districts with a higher intensity of mining tend to be more male, have greater agricultural employment, and lower average incomes. This is consistent with mining areas being less developed than non-mining areas (Benshaul-Tolonen (2019)). Because of these differences, it is difficult to compare districts with high and low levels of mining to study the effects of mining booms. To identify the causal effects of mining booms on labor market outcomes, we exploit an exogenous source of variation in the mining industry: changes in the value of mining due to world prices. To construct mining value, we first identify the mineral with the highest share of active mines in a district and year (mineral k).¹⁸ Next, we follow Berman et al. (2017)¹⁹ and interact the world price of mineral k with the number of active mines of k to calculate district mining value. We estimate:

(1)\begin{align} Y_{ijt} = \beta_{0}+ \beta_{1}M_{kjt}*\log(Price)_{kt}+ \beta_{2}M_{kjt} + \beta_{3}\log(Price)_{kt} +\alpha_{t}+ \alpha_{j} + X_{ijt} \gamma + u_{ijt}. \end{align}

The outcome Y is defined for individual i in district j and year t. M$_{kjt}$ is the number of active mines of the most important mineral k in district j and year t. Meanwhile, $log(Price)_{kt}$ represents the world price of mineral k in year t. It is positive for districts with active or inactive mines at some point over the study period, and equals 0 for districts that never have a mine, in a similar spirit to Berman et al. (2017). Changes in the world price of minerals (see figure A3) plausibly affect profits from mineral production, and subsequently induce changes in mineral production and the production status of a mine (figure A4).²⁰

Mineral prices are positively correlated with production for most minerals, including for coal, which is the most important mineral by production (figure A8).²¹ Additionally, we provide first-stage regression estimates for the relationship between the the number of active mines and the lagged price of minerals, while controlling for year and district fixed effects (table A2). The correlation between the two is positive and significant for all minerals and is strongest for coal mines, for which a doubling of price correlates with an additional active mine.²²

We use the term ‘district mining value’ for the interaction of M$_{kjt}$ and $log(Price)_{kt}$, and the term ‘mining booms’ for increases in district mining value. The coefficient of interest on the interaction term, $\beta _{1}$, captures the effect of changes in a district's mining value due to the world price. The $\alpha _{ t}$ are year fixed effects which account for factors that are specific to year, such as inflation. The $\alpha _{j}$ are district fixed effects which account for district level factors that are fixed over time such as geography, and the covariate matrix $X_{ijt}$ includes individual demographic characteristics.

In addition to our main analysis, we conduct a number of robustness exercises in section 4.1. The most important of these include creating an alternative measure of district mining value that uses all the minerals mined in a district interacted with their respective world price. In an additional specification in section 4, we control for the total number of mines in a district to account for the total amount of mining in a district. A second concern is that the world price of every mineral may not be exogenously determined, and could be affected by production in Indonesia, especially for coal.²³ In section 4.1 we repeat the analysis for our income and employment outcomes without districts where coal is the most important mineral. Other robustness exercises include: showing that neither (i) placebo prices (constructed using either a random mineral or the price of gold), nor (ii) inactive mines in a district, have the same labor market effects; (iii) using lags and leads in the world price to mitigate the concern that mining booms lead to temporary ‘stop and go’ towns; (iv) displacing mines geographically to a placebo district using randomization inference (section 4.1), and (v) accounting for the spatial correlation between standard errors by estimating (Colella et al., 2019) standard errors in section 4.

While the SNL Metals and Mining Database (2018) does not systematically exclude small mines, artisanal and illegally operated mines are not the focus of the data (Nassar et al., 2022). Artisanal mines employ more women than industrial mines do (Paschal and Kauangal, 2023), and can improve women's financial independence and their transition to the service sector (Buor and Ayim, 2019; Jotham and Mulinya, 2020). These measurement errors could lead to attenuation bias in our estimates, since the true effect of mining on women's employment could be larger if artisanal mines were fully accounted for. However, we believe this concern is mitigated by our identification strategy. If mines are spatially clustered, the effect of artisanal and illegal mines is proxied for by the number of large mines in a district. Further, district fixed effects in our estimations account for the time-invariant presence of smaller mines.

Second, while the SNL Metals and Mining Database (2018) includes the exact latitude and longitude of mines, mines are matched by their district and year with the Sakernas (2021) and Indonesian Family Life Survey (2021), since there are no geographic coordinates in the latter two. This could potentially add imprecision to estimates because all individuals in a district are treated as having the same exposure to mining. This concern is attenuated for small districts, districts with large mines, or districts where mines are spatially clustered.

4 Results

4.0.0.6 Monthly income

When a male dominated industry such as mining expands in a district, it provides opportunities to men but could crowd out industries that employ women, thereby lowering their income. Alternatively, if the mining industry is linked to growth in other sectors such as the service industry, female labor may not be crowded out, and women could experience greater incomes. In table 2 we show that a doubling of mining value increases female incomes by 7 per cent in the full sample of districts (column (3)) and mining districts (column (6)), or a 1.5 per cent increase from the average female income. While districts with more mines have lower average incomes for women, a mining boom partially mitigates this average for women. Within mining districts, overall incomes grow by 5 per cent during mining booms (column (4)). The results for female incomes are robust to accounting for the spatial correlation between errors (Colella et al., 2019) in columns (9) and (10),^{24, 25} controlling for the number of mines of all minerals in a district (table A4), and including zero incomes (table A6). Male incomes do not substantially change, but improve relative to women's incomes by 17 percentage points as shown in a triple difference-in-differences regression where equation (1) is interacted with gender. This result is as expected, since mining is a male-dominated industry.

Table 2. The effect of mining activity on monthly wage

Notes: The table reports the effect of an increase in mining value on the logarithm of monthly income estimated using equation (1). Standard errors are in parentheses, clustered at the district level. Log(monthly income) is the logarithm of last month's income. All regressions include year and district fixed effects, and controls for age, age$^{2}$, indicators for education level and being married. The Sakernas (2021) data sample extends from 1998 to 2011. Columns (1)–(3) examine effects across all districts, columns (4)–(6) across mining districts, and columns (7)–(10) across districts with active mines over the study period. Columns (9) and (10) additionally account for spatial correlation using the method proposed by Colella et al. (2019).

In table A7 we use Indonesian Family Life Survey (2021) data to repeat our analysis. We use data on net salaries, business incomes and the sum of the two for mining districts. In the absence of district and year fixed effects, total incomes increase during mining booms by about 32 percentage points. With year and district fixed effects (columns (4)–(6)) and individual fixed effects (columns (7)–(9)), the effects on total income disappear, suggesting that much of the increase in columns (1)–(3) was due to district-specific factors. Interestingly, net salaries increase (columns (4)–(6)), while business incomes decrease during mining booms, and these opposing effects could drive the overall lack of change in incomes.²⁶ Women who earn a business income in the Indonesian Family Life Survey (2021) typically work by themselves or with one other worker in agriculture, retail and manufacturing. Those who receive a salary usually work in retail, agriculture, social services and manufacturing. A mining boom may support the latter, perhaps because large mining companies establish business in mining towns and out-compete smaller, less profitable businesses, though we cannot rule out alternative explanations.²⁷

We examine the heterogeneity of our income results from the Sakernas (2021) along four dimensions: the direction of price change, an individual's education level, the type of mineral and the type of mine. In table A8, we estimate equation (1) separately for price increases and price decreases, which are defined as $Log (Price_{t,k})-Log (Price_{t-1,k})$ for mineral k and year t. Reassuringly, the world price of minerals and incomes follow the same direction, since a higher value of mining should lead to positive incomes and vice versa. In table A9, we separate effects by the minimum compulsory level of education in Indonesia (completing secondary school). Much of the gains in income accrue to females with less than a college education. To the extent that education is pre-determined, these results suggest that the income gains are experienced by less-skilled women, who usually work in the lowest paying jobs. We discuss heterogeneity by mine type in section 4.

The weak increase in male incomes in the Sakernas (2021) is surprising given the male dominated nature of the mining industry. One explanation is that mining booms result in men moving to mining districts, and this balances labor demand and supply. Another is that the mining sector also includes unpaid and/or informal work for men as discussed in section 4.2, which may lower average wages.

4.0.0.7 Employment

On average, mining booms have mixed to no effects on employment (table 3). Average estimates are mostly positive but small and statistically insignificant when considering all districts (columns (1)–(3)) and mining districts (columns (4)–(6)). There is some evidence that male employment increases by a modest 0.1 percentage point in districts with active mines (column (7)), and when accounting for the spatial correlation between errors (column (9)). However, the effects are largely positive and insignificant for weekly hours worked in the Sakernas (2021) (table A10), focusing on lower-skilled individuals who completed high school (table A9), and using the Indonesian Family Life Survey (2021) (table A11).²⁸

Table 3. The effect of mining activity on employment

Notes: The table reports the effect of an increase in mining value on the likelihood of being employed estimated using equation (1). Standard errors are in parentheses, clustered at the district level. Probability(employed) equals 1 if an individual reported working last week, temporarily did not work but has a job, or started a business last week. All regressions include year and district fixed effects, controls for age, age$^{2}$, indicators for education level and being married. The Sakernas (2021) data sample extends from 1998 to 2011. The first three columns use all districts in the data and the last three columns focus on mining districts.

Notably, these estimates are broad employment averages. When separately examining effects by the direction of price change (table A8), employment for women is higher by 0.9 percentage points during price increases, and is lower by almost the same amount during price decreases, perhaps cancelling out the effect of the former on average. In the next section, we show that employment also varied by the mineral being mined and the structure of mining.

4.0.0.8 Heterogeneity by mine type

In table A12, we present income and employment estimates for booms of gold mines in columns (1)–(2) and coal mines in columns (4)–(5). Gold and coal are the two most important minerals mined in Indonesia. Since both minerals are spread over underground and surface mines, in columns (3) and (6) we also separate gold and coal mines into underground and surface mines respectively for women.²⁹ A gold mining boom leads to a 70 per cent increase in incomes and a 14 percentage point higher employment for women, while these effects are negligible in districts with coal mining booms.³⁰ Incomes and employment are positive and larger in gold mines than in coal mines for men too, but these are more noisily estimated. As gold is valued higher than coal and experienced a steep price increase during this time period, gold mining booms could create more local activity.

While a large proportion of gold is extracted from underground mines in Indonesia, it is actually surface mines that lead to greater income and employment for women. Estimated coefficients for open surface mines are positive and statistically significant, while they are negative for underground mines. Though we cannot rule out alternative explanations, surface mines lead to a greater disruption of cultivable land (InTeGrate Carleton College, 2021). This temporarily decreases agriculture, and creates more activity in the service and mining sectors, as our findings in section 4.2 suggest. Further, since surface mines recover faster (Jinan Fucheng Hydraulic Equipment Co. Ltd, 2021), they might not permanently change the geography, demography and the amenities in a district, which makes a district more livable for its non-mining population. Overall, gold mines and surface mines are more likely to increase employment and earnings for women.

4.1 Robustness

One concern with our identification strategy is that the price of minerals can be spuriously correlated with mining value. To mitigate this concern, we substitute actual prices with two types of placebo prices: a randomly chosen mineral's price, and the price of gold, which is less correlated with the prices of other minerals (figure A3). The zero result in table A15 suggests that our main effects are not driven by spurious correlations. Similarly, we also displace mines to placebo districts by using a randomization inference test, where we randomize the assignment of the number of mines across districts (table A15).³¹ Our results are robust to random assignment of mines to placebo districts. Next, we expect labor market effects to be smaller than those driven by the presence of active mines.³² We substitute active mines with inactive mines in calculating a district's mining value. In table A15, the estimates from using inactive mines are indeed smaller than our main estimates, and some are negative. Urban growth is likely to be slower around inactive than active mines. Our results imply that the combination of active mines and world price results in the labor market effects in section 4.

Third, Indonesia is one of the top ten exporters of the world's coal and Indonesia's coal production may affect the world's price of coal. To mitigate this concern, we estimate our main results without any districts with coal mines. While this reduces the sample size, the results remain largely consistent, and are in fact stronger for incomes. We also use an alternative definition of district mining value that interacts the share of mines of each mineral in a district with the mineral's world price. These individual products are added for all minerals to create the district's mining value. Our main results are hold with this alternative definition.

Further, since a district's mining value may be influenced by lagged prices, in table A17 we examine the temporal robustness of mining value by using lagged ($Price_{t-1,k}$) and leading ($Price_{t+1,k}$) prices of the main mineral in a district.³³ With both prices, the estimates are similar to the main results. Notably, however, estimates in the year before a boom are less strong, and those in the year after are stronger. This suggests that there is a lead-up period to a mining boom, and some persistence in economic activity after year t.

Finally, though our identification strategy relies on exogenous world prices that shift the mining value of districts, in practice, we estimate two-way fixed effects (TWFE) equations. Under heterogeneous effects, TWFE estimations provide a weighted average of comparisons between different units. Some comparisons may lead to negative weights being applied, and can result in an average with the opposite sign to the treatment effect (De Chaisemartin and d'Haultfoeuille, 2020). The bias is mitigated the larger the never treated group, and is negligible when 80 per cent of the sample is never treated (Nguyen, 2022). Since 71 per cent of individuals in the Sakernas (2021) live in non-mining districts, and 88 per cent live in districts without active mines, this concern may be mitigated for our study. Nevertheless, we provide estimates using the TWFE estimator proposed by De Chaisemartin and d'Haultfoeuille (2020) using 500 bootstrap replications. The results, reported in figure A9 and table A16, support the direction of the main estimates but are less statistically significant.

4.2 Mechanisms

4.2.0.9 (1a) Employment industry

Table 4 details the likelihood of working in six industries, including large industries such as agriculture, manufacturing and services, and mining, which is directly relevant. In each column, the outcome is an indicator for working in a specific industry. During mining booms, employment in mining grows by 0.17 percentage points in mining districts, or a 7 per cent increase from the average (table 4, top panel). It is driven by men (middle panel), for whom mining employment increases by 0.23 percentage points. Interestingly, the change in mining is accompanied by changes in two other sectors: agriculture and services. As mentioned earlier, mining tends to take place in rural, agricultural areas. During a mining value boom, the agricultural sector shrinks overall by about 0.8 percentage points, and the retail sector – which is a key service sector – grows by 0.32 percentage points, both of which are substantial changes. The effects are experienced by men and women. Women experience a slightly higher decrease in agricultural employment of 1.1 percentage points during mining booms, which is statistically significant at the 1 per cent level. This decline is accompanied by an increase in employment in the service sector occupations of retail and in household services, including in housekeeping. Together, these findings suggest that mining booms may support other industries related to consumption and urbanization, and result in some structural transformation from the agricultural to the service sector.

Table 4. The effect of mining activity on employment industry

Notes: The table reports the effect of an increase in mining value on the employment industry estimated using equation (1). Standard errors are in parentheses clustered at the district level. The outcome in each column is an indicator for working in a given industry. The sample is of districts with active and inactive mines (‘mining districts’) over the study period. All regressions include year and district fixed effects, and controls for age, age$^{2}$, indicators for education level and being married. The Sakernas (2021) data sample extends from 1998 to 2011.

In table A19, we explore changes in the employment industry for the same individuals using the Indonesian Family Life Survey (2021) and individual fixed effects.³⁴ Our estimates suggest that individuals become more likely to be employed in certain lower-skilled services, such as transport, storage and communications. In the meantime, employment decreases in services that employ higher-skilled individuals, such as social services.³⁵ Employment in agriculture marginally increases with the inclusion of individual fixed effects (table A19). The decrease in the district's share of agricultural employment in table 4 may be at least in part due to individuals from other districts moving in and lowering agriculture's share in employment. In the Indonesian Family Life Survey (2021) sample, casual work in agriculture decreases during mining booms (table A20), perhaps because temporary agricultural workers can easily move to other expanding industries, such as mining.

The different stages of a mine's development also support different types of employment. We provide suggestive evidence of this by regressing district level employment in agriculture, mining and retail on a mine's development stage in table A21.^{36, 37} In columns (1)–(3) the total employment in agriculture in a district and year is regressed separately on indicators for the three main mining stages, and the next columns repeat the regressions for mining and retail employment, respectively. The pre-operation stage is associated with an increase in agricultural employment by 13,000 individuals on average (column (1)), and a decrease in mining and retail employment (columns (4) and (7), respectively), though changes in retail employment are noisily estimated. When a mine is operating or is closing, agricultural employment decreases (column (2)), and mining employment increases (column (5)). Retail employment also increases (column (8)), though these changes are statistically insignificant. Therefore, as a mine develops, a district's employment increasingly shifts from agriculture to mining.

4.2.0.10 (1b) Formal work

Formal work typically involves official contracts and tends to be higher paying.³⁸ In our analysis, we treat being an unpaid and/or informal worker as one category,³⁹ being a paid worker as being formally employed, and combine all self-employed categories into one distinct category.⁴⁰ The outcome in each column of table 5 is an indicator for being in one category of job status.

Table 5. The effect of mining activity on the status of work (occupation)

Notes: The table reports the effect of changes in mining value on the status of work estimated using equation (1). ‘Unpaid/informal’ refers to a combined category of unpaid and/or informal work to create a consistent category of work status across survey rounds. Standard errors are in parentheses clustered at the district level. The outcome in each column is an indicator for a given status of work. All regressions include year and district fixed effects, controls for age, age$^{2}$, indicators for education level and being married. The Sakernas (2021) sample extends from 1998 to 2011.

The likelihood of unpaid and/or informal work decreases marginally by 0.6 percentage points for women (column (3), table 5). Though small, this is important for women, 38 per cent of whom work in unpaid and/or informal jobs. Although the coefficient on self employment is small and statistically insignificant, it is positive for women, suggesting that women could be moving from informal work to self employment. Women in agriculture are three times less likely to be paid than women in services, and movement to the latter could therefore lead to more paid work. Interestingly, men are 0.3 percentage points more likely to do unpaid and/or informal work (column (2), table 5). This could be because some men take on casual work near mines in the hope of eventually doing paid work. This nuance could partly explain why overall earnings do not increase for men as they do for women.

Our main findings are supported in table A20 using Indonesian Family Life Survey (2021) data. During mining booms, the same man is more likely to do unpaid and/or informal work, and women are less likely to do so (though the latter effect is not statistically significant). Interestingly, men and women are less likely to be casual agricultural workers during mining booms.⁴¹ While agricultural employment is not lower in table A19, the likelihood of being a casual worker in agriculture decreases during mining booms, which suggests movement from unpaid and/or informal agriculture work to other sectors.

4.2.0.11 (1c) Do changes occur in urban or in rural settings?

Sakernas (2021) use the definition of urban in the Indonesian population Census.⁴² Twenty-seven per cent of individuals live in an urban village of a mining district, and 23 per cent in an urban village of a district with active mines. Interestingly, almost all changes following mining booms occur in rural areas (table A13). For example, women in rural villages experience income increases of about 9.8 per cent, while these changes are not significant for urban women. Rural women also become less likely to work in agriculture which supports low female incomes. The fact that more men work in unpaid and/or informal jobs is surprising, but is consistent with previous findings, and partially explains the lack of increase in male earnings overall in the Sakernas (2021).

Mining booms also affect the population size and urbanity of a district itself. While there are few notable effects on average, there is heterogeneity in outcomes by the type of mine (table A14). Growth in the production value of open gold mines is associated with population increases and lower agricultural employment. These results tie in well with the findings in table A12, where most labor market improvements are for open gold mines as well. Notably, open surface gold mines lead to a statistically insignificant increase in urbanity, while open surface coal mines to a statistically insignificant decrease in urbanity.

We also examine the role of alternative economic opportunities by computing the Herfindahl-Hirschman Index (HHI) of the concentration of industries in a district.⁴³ Districts with a high HHI are concentrated in one industry, which is usually agriculture, in this context. Columns (5) and (6) of table A13 report results of equation (1) estimated for districts with a higher than average HHI. High HHI districts experience the greatest effects of mining value booms, including increased female incomes, higher employment in retail and less unpaid and/or informal work for women. Together, these findings suggest that much of the labor market and structural changes due to mining booms occur in rural areas where there are scarce opportunities otherwise.

4.2.0.12 (2) Migration

Labor is often mobile between local labor markets, and can respond to shocks such as mining booms. On average, mining districts and those with more mines tend to attract and retain few individuals relative to districts with fewer or no mines (table 6). In the absence of a mining boom, there may be few jobs and amenities in these areas. A mining boom partially mitigates this effect, and attracts and retains more people in the short run. A doubling of mining value results in a 1.8 percentage point decrease in the likelihood of individuals leaving mining districts. This is a decrease of 40 per cent relative to the mean, and is a substantial effect for origin districts. Women who may have otherwise left mining districts are 2.6 percentage points more likely to stay on during boom times. Interestingly, the same woman is also 1.9 percentage points less likely to move away during a mining boom (column (6)). Men become 3.3 percentage points more likely to move to a destination district with active mines during a mining boom (middle panel).⁴⁴ Women are slightly more likely to move to an urban destination that experiences a mining boom, but not otherwise (lower panel).⁴⁵

Table 6. The effect of mining expansions on migration

Notes: The table reports the effect of an increase in mining value on moving away from a mining district using equation (1). Standard errors are in parentheses, clustered at the origin district level. Probability(migrated) equals 1 if the individual reports moving in that year, and 0 otherwise. Columns (1)–(3) include district and year fixed effects and columns (4)–(6) also include individual fixed effects. The sample is from the 1997, 2000, 2007 and 2014 waves of the Indonesian Family Life Survey (2021), which tracks migrations every year between 1994 and 2014.

To further characterize movement to mining towns, in table 7 we present results from a gravity equation, which allows us to analyze movement between every origin and destination district pair. We estimate:

(2)\begin{equation} \begin{aligned} Probability(migrate_{odt}) & = \beta_{0}+ \beta_{1}M_{okt}*log(Price)_{kt}+ \beta_{2}M_{okt} + \beta_{3}log(Price)_{kt} \\ & \quad+ \beta_{4}M_{dit}*log(Price)_{it}+ \beta_{5}M_{dit} + \beta_{6}log(Price)_{it}\\ & \quad +\beta_{t}+ \beta_{od} + u_{odt}, \end{aligned} \end{equation}

where o is an origin district with most important mineral k, and d is a destination district with most important mineral i.⁴⁶

Table 7. The effect of mining activity on migration: gravity equation estimation

Notes: The table reports the effect of mining value changed on migration by estimating a gravity equation corresponding to equation (2). Standard errors are in parentheses, clustered at the origin and destination pair level. Probability(migrated) equals 1 if the individual reports moving in that year from origin district X to destination district Y, and 0 otherwise. All regressions include year fixed effects and origin-destination pair fixed effects. Odd numbered columns depict migrations from any origin district to destination districts with active or inactive mines. Even numbered columns depict migrations from origin district with active or inactive mines to a destination district with active or inactive mines. All regressions exclude origin districts that are the same as destination districts. The sample is from the 1997, 2000, 2007 and 2014 waves of the Indonesian Family Life Survey (2021), which tracks migrations every year between 1994 and 2014.

In table 7, men are about 0.04 percentage points more likely to move to a mining district with a booming mine industry from any origin district (column (3)) and from mining origin districts (column (4)). They are 0.02 percentage points less likely to leave a mining origin district (column (4) of table 7). Together, our results imply that mining booms generate active movements of individuals between districts, and allow mining districts to attract and retain people.

5 Conclusion

We examine how the growth of an industry dominated by one gender, such as mining, affects labor market outcomes for men and women. We exploit changes in district mining value in Indonesia that are driven by the world price of minerals. Surprisingly, we find that a growing mining sector increases average incomes for women and overall, while male incomes are not substantially changed, with a large variation in outcomes by the type of mine. The overall results are not driven by individuals being more likely to work. Instead, we find the largest difference in where individuals work. During mining booms, women become less likely to be employed in unpaid and/or informal and/or informal labor. At the same time, employment in agriculture shrinks and expands in the retail sector and services. Men become more likely to move to a booming mine district, and women to remain in one, potentially in response to greater opportunities.

The implication of our results is that a boom in male dominated industries does not necessarily crowd out jobs for women. Instead, if the mining industry supports growth in the local economy, women may benefit from the industrial growth as well. Our findings are similar to those in other contexts that show that the resource curse in mineral-based economies is mitigated when the mining industry promotes growth in other industries. Our results suggest a policy which creates opportunities for growth in other industries within mining districts which would generate benefits for a wider population.