Background Recent research in several countries has demonstrated that small-holder tobacco farming is typically not a profitable enterprise. Many farming households report losing money in this economic endeavour, even without incorporating the value of their household labour. Losses are typically considerably worse when household labour is considered. We take advantage of panel data that include information about both current and former tobacco farming households’ characteristics and economic decisions to be the first to rigorously estimate the effects of both tobacco and non-tobacco farming on income.
Methods We designed and implemented a two-wave economic survey of current and former tobacco farming households in Indonesia’s two largest tobacco-growing regions. We use regression analysis to estimate the effects of tobacco farming on household income per farming area in both survey waves.
Results We find that former tobacco farming households are typically generating profits from their non-tobacco farming, while current tobacco farming households experience greater variability, including experiencing economic losses. Former tobacco farming households’ income were comparable to current tobacco farming households’ even in the period in which tobacco leaf production and prices of tobacco leaf were relatively high. We find a negative and significant effect of tobacco farming on household income.
Conclusions One of the main arguments from those opposing tobacco control policies—especially increasing cigarette excise taxes—is their alleged effect on tobacco farming households’ livelihoods through a lower demand for tobacco leaves. Our finding that there is a negative effect of tobacco farming on household income shows that the narrative is grossly inaccurate. Shifting to non-tobacco farming would allow farming households to reallocate their resources to other more lucrative economic opportunities.
- low/middle income country
- priority/special populations
Data availability statement
Data are available on reasonable request. Researchers who provide a methodologically sound proposal may request individual participant data that underlie the results reported in this article, after deidentification (text, tables, figures and supplementary materials). Proposals should be directed to JD at firstname.lastname@example.org.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Tobacco control policies are often stymied by arguments—typically put forward by tobacco companies and their allies—that these policies would impose detrimental effects on the livelihoods of relevant economic actors, especially tobacco farmers. These arguments assume that tobacco farming provides prosperous livelihoods for farmers. Tobacco control measures, such as increasing cigarette excise taxes, are argued to reduce cigarette consumption, the broader demand of tobacco leaves, and thus the livelihoods of tobacco farmers.1 2 This argument often puts significant pressure on the government to delay, stop or dilute significant improvements to tobacco control measures. However, to date, the argument that tobacco farming provides a viable economic livelihood is not supported by rigorous empirical evidence3–7 and the global demand for leaf has decreased only incrementally in the last decade.8
This research rigorously examines the economic livelihoods of small-holder farming households in Indonesia, a major tobacco producer. This research is part of a larger study to analyse economic livelihoods of workers in the Indonesian tobacco sector including clove farmers and kretek rollers.9–12 We build on an established line of research that investigates the profitability of tobacco farming, while systematically considering household labour costs.3 6 7
We designed and implemented a longitudinal survey of current and former tobacco farming households representative of the largest tobacco-growing regions, comprising more than 90% of tobacco farming households in Indonesia. We use the survey to estimate, among other dynamics, current and former tobacco farming households’ income and profits from their economic endeavours. Analyses of the rich panel data facilitate significant contributions to the literature on the livelihoods of small-holder farming households particularly because we speak directly to conditions in multiple time frames and change over time. First, we investigate differences in economic livelihoods of current and former tobacco farming households across two time periods, the 2016 and 2017 tobacco-growing seasons. Second, we estimate the effects of tobacco farming on household income.
The context of tobacco farming in Indonesia
The focus on Indonesia is important for at least three reasons. The contribution of tobacco leaf production to the broader economy is quite small, estimated at only 0.03% of Indonesian Gross Domestic Product (GDP).9 12 Indonesia is the second-largest cigarette market by volume after China with a retail volume of 307.1 billion sticks in 2018.13 It is also the sixth largest producer of tobacco leaf in the world.14 Given the large size of consumption, tobacco companies have been using most of the domestically produced tobacco leaf and importing large amounts of tobacco leaf from other countries. In fact, Indonesia has been a net importer of tobacco leaves for many years to satisfy domestic tobacco product manufacturers’ demand.5 Despite this situation, tobacco companies continue to create a somewhat counterintuitive narrative that tobacco control policies that reduce cigarette consumption would adversely affect tobacco employment, particularly tobacco farmers.
Second, although the share of tobacco farmers to total farmers is quite small, the absolute number is quite large at more than 500 000. Moreover, they are concentrated in just several politically important regions in Indonesia, mainly East and Central Java provinces, and as a result are consequential to policies that relate to tobacco and tobacco control. The concentration of tobacco farmers in these regions also parallels the concentration of tobacco manufacturing in Indonesia.9 Moreover, identifying the factors that affect tobacco farmers’ income is important to comprehensively understand alternative livelihoods, which as a goal has been enshrined internationally in Articles 17 and 18 of the World Health Organization Framework Convention on Tobacco Control (WHO FCTC).
Lastly, the government still has no comprehensive agricultural policy in place for tobacco-growing regions and tobacco farmers in particular. Understanding the effects of tobacco farming and alternative livelihoods available for tobacco farmers is essential for government to identify and formulate a comprehensive policy.
Data: Tobacco Farmer Survey
The data in this study come from a longitudinal survey of current and former tobacco-farming households. The first wave of the Tobacco Farmer Survey (TFS) was conducted at the end of 2016 in seven top tobacco-producing districts in Indonesia comprising 84% of production: Magelang and Temanggung in Central Java Province, Lumajang, Jember and Bojonegoro in East Java Province, and Lombok Tengah and Lombok Timur in West Nusa Tenggara Province. Within the selected districts, we chose the top producing and one second tier producing subdistricts from which two or three top-producing villages were selected as the study areas. Within each selected village (18 in total), we randomly selected 20 current and 5 former tobacco farming households (a ratio of 4:1). In Lumajang, the ratio of current and former tobacco farming households was 1:1 because of difficulties finding farming households. We obtained a total sample of 1350 households. We explain the sampling method, sample size determination and survey instruments for the wave 1 TFS in online supplementary section A.5
Due to cost considerations, for the wave 2 TFS conducted at the end of 2017, we revisited a subset of the baseline sample. Specifically, we did not revisit the sample from West Nusa Tenggara because the preponderance of farmers live in the Javanese districts, representing 90% of the country’s tobacco farmers.9 We reinterviewed 15 of the 25 baseline households for a total sample of 660 households. The 15 households were selected randomly with an aim towards maintaining the ratio of 4:1 of the baseline sample. A strict rule was in place to replace the target households with the households from the 10 remaining households when a household could not be reinterviewed for various reasons. In total, only eight households were replaced.15 We explain the detailed sampling method for the wave 2 TFS and the replacement protocol in online supplementary section A.
We summarise the final composition of the second wave survey sample in table 1. We observe a non-negligible number of switchers. For example, in Lumajang, the current to former tobacco farming households ratio is nearly 2:1 in the second wave, which is quite different from the 1:1 ratio in the first wave. All in all, however, the share of tobacco farming households in the second wave is relatively similar to the share of tobacco farming households in the first wave of the tobacco survey.
We implemented a sample selection procedure to obtain a final sample for the descriptive and regression analyses. We summarize the sample selection procedure in a consort diagram and present it in online supplementary figure S1. Specifically, we dropped 35 non-tobacco and 39 tobacco farming households’ observation points in the first wave whose profit per hectare was below the 1st percentile or above the 99th percentile. We also dropped 21 non-tobacco and 32 tobacco farming households’ observations in the second wave whose profit per hectare was below the 1st percentile or above the 99th percentile. Lastly, we kept households who are observed in both survey waves. Our final analytical sample is a balanced panel consisting of 563 farming households and a total of 1126 observations. We also calculated similar descriptive statistics and regression analyses using all households sampled in wave 1 and wave 2 TFS. We report the results in online supplementary section E, and the results are qualitatively consistent with the balanced panel presented here.
The data were analysed using the STATA (V.15.1) statistical package.
Calculation of household income
We calculated household income by incorporating household labour costs, which are a significant opportunity cost for the household.3 6 7 Specifically, household income is the sum of tobacco farming profit, non-tobacco farming profit, household enterprise profit, wage income and other income minus rent and household labour costs. Tobacco farming proves to be considerably more labour intensive than other types of farming. At the same time, research demonstrates that non-tobacco farming households dedicate some of these hours to other economically viable activities. For example, a household member could have spent his or her time working for other non-tobacco farms, enterprises or entrepreneurial activities.
We apply the method used in a recent study in Indonesia to estimate household labour costs.5 This method was developed based on established methodologies in the literature.3 6 7 Individuals within a farming household could have spent their hours on other economic endeavours and earned an income. Therefore, to calculate the opportunity costs of household labour, we multiply regional agricultural minimum hourly wages by the number of household labour hours reported.
The ultimate objective of this study is to estimate the effect of tobacco and non-tobacco farming on household income. The base specification to estimate the effect of tobacco farming on household income is:
where i indicates household, t indicates time, income indicates total household income per hectare of farming land and tobacco indicates the share of a household’s land used for tobacco farming. The vector X includes household characteristics such as log of total cultivation area, log of labour hours, log of assets, log of agriculture wage, log of non-agriculture wage, an indicator of whether a farming household enter a contract with a leaf-buying entity, household size, head of household age, head of household educational attainment and a time fixed effect (FE). We also include dummy variables to indicate households with no labour hours, assets, agricultural wage and non-agricultural wage. The term a indicates farming-household-specific unobserved characteristics while the term u indicates idiosyncratic shocks.
We estimate the model in specification 1 using the random-effects (RE) model to take advantage of the rich panel dataset. The main assumption for the model to yield a consistent estimate of the tobacco farming effect is that there is no correlation between unobserved farming-household-specific characteristics and the farming portfolio, This assumption suggests that the choice of tobacco farming is not related to farming households’ unobserved characteristics such as knowledge, information and farming skills.
This assumption is rather difficult to fulfil because unobserved farming households’ characteristics can be correlated with the farming portfolio in many different ways. We confirm this by implementing an endogeneity test of tobacco in the context of a RE model.16 The result of the test suggests that tobacco is indeed endogenous (p value of 0.015). The result implies that an RE model would produce a biased estimate of the tobacco farming effect.
To mediate the endogeneity issue of tobacco, we use an instrumental variable approach.16–18 We use rainfall shocks in the previous tobacco season—April to September—as the instrumental variable for the farming portfolio. Let rain_deviation be the log of squared deviation from long-run rainfall. The first-stage equation is:
The vector X includes all variables used for the estimation of the structural model shown in equation (1). As in the structural model specification, the term a indicates farming-household-specific unobserved characteristics while the term u indicates idiosyncratic shocks.
For the validity of the instrument, it must fulfil both the relevance and exogeneity criteria. We first discuss the relevance criteria which requires that deviation from long-run rainfall is correlated with the share of land dedicated for tobacco farming, . From the survey and our qualitative study, weather—particularly rainfall—was cited as one of the factors that determines farmers’ decisions to farm tobacco in the next farming season. A study finds that relatively dry conditions in the preharvest period led to better tobacco farming productivity and leaf quality.19 Their results demonstrate that tobacco farming generally produces higher yields and quality leaf under drier conditions.
We show in online supplementary table S2 that rainfall shock in the previous tobacco season was indeed a significant predictor of the share of land dedicated for tobacco farming. A 1% increase in squared deviation of rainfall to the long-run average was associated with a higher share of land by 0.07% point. The weak identification test also rejects the null hypothesis of a weak instrument with an F-test of 50.26 and a Stock-Yogo critical value of 16.38.20 21 This is quite reasonable because higher-than-usual rainfalls rarely extends to two consecutive tobacco farming seasons. Tobacco farmers predicted favourable rainfall in 2017 tobacco farming season after the relatively high rainfall in the 2016 tobacco farming season. Given their prediction, they dedicated more land for tobacco farming in the current season.
We now discuss the exogeneity criterion which requires that the instrument is not correlated with household-specific unobserved characteristics and any idiosyncratic shock, . Previous studies have exploited exogenous variations in rainfall shocks to estimate causal relationships or to address endogeneity issues.22–26 The amount of rainfall in the previous tobacco season does not directly correlate with current household income. For example, higher-than-usual rainfalls in the previous season should not directly affect leaf production, sales and costs in the current season. The amount of rainfall in the previous tobacco season should only affect household income through the share of land dedicated for tobacco farming in the current season.
We run a robust Hausman test to test the null hypothesis that both FE and RE are consistent estimates. The robust Hausman test is an extension of the Hausman test that accounts for serial correlation and heteroscedasticity, which are common in panel data.16 27 28 In addition, for every specification, we use heteroscedasticity-robust or cluster-robust SEs to accommodate a general form of heteroscedasticity.29 30
The analytical sample consists of 563 farming households. In the first wave of the survey, there were 129 former tobacco farming households and 434 current tobacco farming households. In the second wave, there were 115 former tobacco farming households and 448 current tobacco farming households. Using the second wave data, we find that most of the heads of the tobacco farming households in the sample were male (96.80%), were married (94.32%) and attended or completed primary education (75.67%). They were predominantly in the working age of 15–64 years old (90.05%) and indicated farming as their primary source of income (85.61%). Some households (20.78%) were eligible for the social protection programmes provided by the government of Indonesia.
Current tobacco farming and former tobacco farming households have relatively similar land sizes. In the second wave of the survey, the median current tobacco farming household had 0.90 hectare of farming land, while the median former tobacco farming household had 0.75 hectare of land (median test, p=0.55). We observe a significant increase in farming land among current tobacco farming household across the two waves. The median current tobacco farming household had 0.90 hectare of farming land, larger than the land size of the median current tobacco farming household of 0.60 in the first wave (median test, p=0.00). In a particular year, the median current tobacco farming household dedicated about a third of their farming land for tobacco farming (33.30%). The rest of their farming land is dedicated for non-tobacco farming (66.70%).
Profit and income analysis
Table 2 presents the production, costs, profit and income of former and current small-holder tobacco farming households in both survey waves. The median tobacco leaf production per household was significantly higher in the second wave, increasing from 200kg to 461kg (or 130.50%), and the difference was significant at p<0.01.
The production increase resulted in higher tobacco profits. We calculate tobacco profit by subtracting costs of inputs, marketing (costs related to selling the crop) and hired labour from tobacco sales. The median tobacco farming household experienced a loss of US$17.47 purchasing power parity (PPP) in the first wave of the survey. However, the median tobacco farming household gained a profit of US$967.92 PPP in the second. The difference in the median profit between the two survey waves is significant at p<0.01.
We hypothesise that the large difference in the median tobacco profit is due to, among other reasons, relatively favourable amounts of rainfall during the tobacco farming season in 2017. According to the survey respondents almost uniformly, weather conditions during the 2017 dry season were quite favourable to tobacco growing, particularly in terms of nearly ideal rainfall. This contrasted the 2016 tobacco-growing ‘dry’ season, which farmers reported widely as unseasonably wet. Tobacco does not fare well under conditions with significant moisture, particularly as the leaves mature.
We construct rainfall data for each of the 56 enumeration areas in the TFS study to investigate the amount of rainfalls in both survey waves. The weather data comes from the University of Delaware data set, with a spatial coverage of 0.5×0.5 degree.31 We calculate the long-run average rainfall during the tobacco seasons, and we present the deviation of rainfall from the long-run average in online supplementary figure S2. In 2016, the average rainfall during the tobacco farming season was indeed much higher than the long-run average in all study areas. On the contrary, in 2017, the average rainfall during the tobacco farming season was close to the long-run average.
We explain two mechanisms such that the favourable rainfall led to better tobacco farming outcomes. First, current tobacco farming household allocated more land—0.20 ha vs 0.25 ha (median test, p<0.01)—to tobacco cultivation in the second wave. Second, the unseasonably wet tobacco farming season in 2016 resulted in significantly lower tobacco leaf production and quality, and largely because of the lower quality, a low leaf prices. As discussed earlier, tobacco leaf production increased significantly by 130.50% in the wave 2 TFS. In online supplementary section F, we also show that prices of Virginia and Burley leaves—accounting for 76.81% of total leaves grown in our survey sample—were significantly higher in the second wave of the survey. Higher production and prices led to higher sales and ultimately income.
Current small-holder tobacco farming household also planted non-tobacco crops such as potatoes, chilis, cassava, corn and green vegetables. The median tobacco farming household dedicated 66.70% of their land for non-tobacco farming in each time period. Production of these crops is also dependent on weather conditions. We hypothesise that due largely to the favourable weather conditions in 2017 farming households also experienced higher non-tobacco farming profits in the second survey wave compared with the first. The median current tobacco farming household generated US$281 PPP crop farming profit in the second wave of the survey, a significant increase from the median crop farming profit in wave one (median test, p<0.01). On the other hand, the median former tobacco farming household was relatively stable across the two waves. The median former tobacco farming household generated US$151 PPP of non-tobacco profits in the second wave of the survey, comparable to the median of US$158 PPP in the first wave.
The data show that current tobacco households dedicated significantly longer farming hours than former tobacco households. In the first wave of the survey, the median current tobacco farming household spent 600 hours per year in the field almost entirely for tobacco farming. In contrast, the median former tobacco farming household spent 360 hours per year in the field (median test, p<0.01). The median current tobacco farming household spent less time tobacco farming in the second wave, but the total hours they spent were still higher than the hours spent by the median former tobacco farming household (median test, p<0.01). The significant difference in household labour hours between the two groups reflects higher opportunity costs—time they could have spent in other economic activities—that current tobacco households incur from tobacco farming. In the second wave of the survey, the household labour costs incurred by the median current tobacco-farming household for tobacco farming was US$590 PPP compared with US$173 PPP for the median former tobacco-farming household for non-tobacco farming (median test, p<0.01).
To further understand the heterogeneity of the tobacco farming population, we calculated household income per hectare for each wave using the estimated farming profits, non-farming income, other income and household labour costs divided by the total cultivated area. We stratified the results in four groups. The first group includes farming households who only farmed non-tobacco crops in both survey waves, while the second group includes farming households who farmed tobacco crops in both survey waves. The third group includes those who switched from farming non-tobacco to tobacco crops, whereas the fourth group includes those who switched from farming tobacco to non-tobacco crops. We compare the median income per hectare across survey waves for each group in figure 1.
In columns A and B of figure 1, we observe that the median always former tobacco farming households made a slight profit in the first wave of the survey. On the other hand, the median always current tobacco farming household was making a loss (US$15 PPP) although the difference in medians between the two groups is not significant. The household income of both median always former and always current farming household increased quite significantly in the second wave of the survey. The p values from Wilcoxon matched-pairs signed-rank tests were lower than 0.01 for both groups.
In column C of figure 1, we observe that there was a slight decrease in the median household income among farming household who switched from non-tobacco back to tobacco farming although the difference was not statistically significant. On the other hand, farming households who switched to farming only non-tobacco crops in the second wave of the survey experienced higher income. As shown in column D of figure 1, the median household income of these farming household increased from—US$566 PPP per hectare to US$1816 PPP per hectare (Wilcoxon matched-pairs signed-rank test, p<0.01).
The key takeaways from this figure are twofold. Farming households in general are facing variability in income across periods. More importantly, switching away from tobacco farming resulted in a significantly higher income. This finding is important evidence in support of Article 17 of the FCTC because it provides empirical support for the notion that farming households can shift away from tobacco and improve their livelihoods.
The effects of tobacco and non-tobacco farming on household income
We estimate the effects of tobacco and non-tobacco farming on household income using specifications discussed in the previous section. We present estimation results in table 3, omitting several control variables from the table for brevity. We focus on the estimated coefficients for the share of land for tobacco farming which represent the effects of tobacco farming on income.
We first discuss the results of ordinary least square (OLS) regression for the first and the second waves. In column A and B of table 3, we observe that the estimated correlations between the share of land allocated to tobacco cultivation and household income are negative although the correlation is only significant in the first wave. The smaller estimated tobacco effect in the second wave is consistent with the descriptive finding above that current tobacco farming households were doing much better in the second wave of the survey.
In column C of table 3, the estimated coefficient of tobacco farming is negative and statistically significant. However, as discussed in the estimation strategy section, an estimation using RE model may yield a biased estimate owing to the endogeneity of the share of land for tobacco farming. We conduct an endogeneity test and we find that the results reject the null hypothesis that the share of land for tobacco farming is exogenous. Accordingly, we estimate the specification using an RE-IV model and we present the results in column D of table 3. The result from the first-stage regression confirms that deviation from the long-run average rainfall is a significant predictor of the share of tobacco land. The Klaibergen-Paap F-statistic of 50.261 also rejects the null hypothesis of a weak instrument.32 The estimated first-stage coefficient suggests that a 1% increase in the squared deviation of rainfall is associated with a higher share of land by 0.067% point.
The results of the RE-IV estimation confirm an endogeneity issue, but the signs are still consistent with those derived from the RE estimation. The estimated coefficients from the RE-IV estimation suggest that a one percentage point increase in the share of land for tobacco farming reduces profit by US$333 per hectare. For a median tobacco farming households with 0.25 hectare of land, a 1% point in the share of land for tobacco farming would reduce profit by US$83.25.
We conduct a robust Hausman test to determine if the differences in the RE-IV and FE-IV estimates are not systematic. The result of the test implies that we do not reject the null hypothesis of no systematic difference between the RE-IV and FE-IV estimates. The result also implies that the RE-IV strategies yield consistent and efficient estimates, suggesting that indeed tobacco farming has negative effects on household income. This result is quite the contrary to the industry’s narrative that tobacco farming is a more prosperous agricultural endeavour.
We can also observe that total assets are positively and significantly correlated with household income. This result is reasonable because households with larger assets presumably have more capital, and have and/or were able to rent larger farming land. We also find that both agricultural and non-agricultural wages are positively and significantly correlated with household income. This finding implies that engaging in alternative activities other than tobacco farming may improve households’ income. This result suggests that a relatively higher engagement of household members to tobacco farming imposes higher opportunity costs.
The results of the regression analyses are important evidence on discussions to identify factors that affect tobacco farming households’ income. Identification of these factors is important to comprehensively understand alternative livelihoods, as enshrined in Articles 17 and 18 of the FCTC.
The descriptive statistics across the two waves show a variability of rewards of both tobacco and non-tobacco farming. However, tobacco farming households faced a greater variability of rewards across the two periods. The median tobacco farming household incurred an economic loss when conditions were more difficult, as in the relatively wet 2016 dry season. During this season, the harvested tobacco leaves were priced quite low owing to the relatively wet growing season and the resulting poorer leaf quality. In contrast, tobacco farming was more lucrative in 2017 and generated economic profits for tobacco farming households. The profit was primarily driven by both higher leaf production and price of tobacco. The average and median leaf production was 501 kg and 200 kg, respectively, in the first wave of the survey. Tobacco leaf production doubled in the second wave of the survey, with an average of 981 kg and a median of 461 kg.
The price of tobacco leaves or processed tobacco leaves increased in 2017. The average price of grade-A-sliced-and-dried Virginia tobacco leaves was US$5.96/kg PPP in the first wave of the survey, compared with US$8.19/kg PPP in the second wave of the survey. The average price of grade-D-slice-and-dried Virginia tobacco leaves increased from US$5.51/kg PPP in the survey’s first wave to US$10.55/kg PPP in the second. Given that leaf production is significantly higher as shown in table 2, it is not surprising that the average tobacco household earned much higher income in the second wave. But, as outlined above, better growing conditions for tobacco are likely to have positive effects on non-tobacco crops, too. Despite the significant increase in household income, the average per-hectare household income of current tobacco farming households (US$7808) was still slightly lower than the average household income of former tobacco farming households (US$12 661) in the second wave of the survey (a difference of US$4852, p<0.05).
The variability of farming outcomes, particularly from tobacco farming, can be explained by, among others, the amount of rainfall during the tobacco season. Our analysis of long-run rainfall data confirms a high amount of rainfall during the 2016 tobacco farming season. In contrast, the amount of rainfall during the 2017 tobacco farming season is relatively close to the long-run average. Farmers who experienced higher-than-average rainfall in 2016 could have predicted a more desirable weather in 2017. We find evidence for our hypothesis that farmers predicted more desirable weather in 2017 for farming tobacco. The median and mean total tobacco farming land size was higher in 2017 than in the previous year. Specifically, the median total tobacco farming land was 0.20 hectare in 2016, while the median rose to 0.25 in 2017. Among farming households who farmed tobacco in both survey waves, the average farming area increased significantly by 0.085 (Wilcoxon matched-pairs signed rank test, p<0.01) in the second wave of the survey.
The negative and significant correlations between the share of land for tobacco farming and household income are expected given the findings that tobacco farming profits are lower and that household labour costs are higher than those of non-tobacco farming. The negative and significant correlations between the size of farming land and household income suggest no economies of scale in their farming activities. These results are contrary to the tobacco industry’s narrative that tobacco farming is a generally profitable economic activity.
The results from the estimations also show positive and significant effects of agricultural and non-agricultural wages on household income. This finding suggests that small-holder farming households have alternative livelihoods that they can use to raise their income.
We purposively choose top-producing districts in Central and East Java provinces for our study. In each district, we randomly selected one top-producing and one second tier producing subdistrict. We acknowledge that the purposive sampling method may lead to an upward bias of tobacco farming outcomes including production, sales and income. However, statistics that we obtain from the survey would serve as an upper bound on tobacco farming households’ outcomes and ultimately livelihoods.
The balanced panel data allow us to control for unobserved household characters, but more time periods would help us to disentangle the relationships better. Multiple follow-up surveys would allow us to better account for variations in weather, tobacco leaf prices, and more importantly other unobserved factors that affect income. Multiple follow-up surveys would also allow us to investigate factors that would affect farming households’ livelihoods such as decisions on how much area to allocate to tobacco farming, allocation of household members’ time to tobacco farming and whether to pursue off-farm employment.
The analyses suggest that households who switched away from tobacco experienced higher income, while those who switched back into tobacco farming did not experience higher income after the switch. Multiple follow-up surveys would allow us to investigate factors that determine farming households’ switching decisions, and the consequence of their switching decisions on income and livelihoods. We aim to explore these inquiries in future studies.
When measuring household labour hours and inputs, tobacco farming was measured by itself, while all other crops were measured collectively to limit the survey’s length. This may lead to underestimation of the labour and input cost of certain other crops depending on what farming households are growing. Future studies with each crop measured individually and systematically will resolve this challenge.
These results shed significant light on the dynamics of farming decisions for current and former tobacco farming households in much of Indonesia. In general, relying significantly on tobacco farming was not an optimal economic decision for farming households. First, while tobacco and crop farming both depended on the weather, crop farming generated higher household income than tobacco farming regardless of the weather. Second, farming households incurred significantly higher household labour costs from tobacco farming compared with non-tobacco farming, implying that non-tobacco households could allocate their labour resources for other fruitful economic opportunities including those off-farm. These dynamics suggest strongly that farming non-tobacco crops is a viable alternative livelihood for tobacco farming households.
Our findings also have important policy implications given the public debate about the effects of a policy to increase cigarette excise tax rates. One of the main arguments against the policy is the possible effects on tobacco farming households’ livelihoods. Specifically, higher cigarette tax rates would reduce the demand for tobacco leaves and reduce tobacco farming households’ income. Our results show quite the opposite: decreasing the share of tobacco crops in a farming household’s farming portfolio would likely raise their income. The clear policy recommendation would be to encourage farming households to strengthen non-tobacco crops in their farming portfolio. Finally, the findings suggest that decreases in tobacco in a farming household’s crop portfolio generate greater overall household income, suggesting strongly that FCTC Article 17 is likely a prosperity-enhancing provision.
What this paper adds
There is a growing literature that demonstrates through household surveys that small-holder tobacco farming households across many countries generally have challenging economic circumstances.
There is little to no analysis of the economic livelihoods of tobacco farming households who switch to other livelihoods even though the WHO Framework Convention on Tobacco Control explicitly counsels Parties to help farming households switch.
Using panel data of current and former tobacco farming households, we show that there is a negative effect of tobacco farming on household income.
Small-holder farming households can be economically better off by making a switch to non-tobacco farming, potentially providing useful guidance to other countries facing similar challenges.
Data availability statement
Data are available on reasonable request. Researchers who provide a methodologically sound proposal may request individual participant data that underlie the results reported in this article, after deidentification (text, tables, figures and supplementary materials). Proposals should be directed to JD at email@example.com.
All activities for this research were approved by the Institutional Review Board (IRB) of the Morehouse School of Medicine, the IRB of record for the American Cancer Society, and the IRB of SurveyMeter, the implementation survey organisation in Indonesia.
We thank Edson Correia Araujo, Pandu Harimurti, and Josefine Durazo for their oustanding inputs for the Tobacco Farmer Survey. We also thank Bondan Sikoki and SurveyMeter for collecting the survey data. Kandrika Fadhlan Pitularga provided outstanding research assistance. This paper has benefitted from comments by participants of the 2019 Economics of Tobacco Farming Annual Research Meeting.
Contributors GAS, JD, QL, NN and FW contributed to the study design. FW, JD, QL and NN drafted the survey tool and contributed to subsequent refinements of the survey tool in the second wave. FW collected the survey data with SurveyMeter. GAS completed the statistical analysis. GAS and JD wrote the first draft of the manuscript. All team members contributed to the writing and revision of the manuscript. GAS submitted the manuscript on behalf of the team.
Funding This research was supported by the Office of the Director, National Institutes of Health (OD) and the National Cancer Institute (NCI) under Award Number R01TW010898; NCI through a CRDF Global grant; the World Bank; and the American Cancer Society.
Disclaimer Its contents are solely the responsibility of the authors and do not necessarily represent the official views of these funders.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.