What has been measured about cocoa farmers' incomes and yields, children's work, forest loss and certification. The design of each study is stated plainly, because most of them can show that two things go together and few can show that one caused the other.
38 studies2005 to 2026Each checked against its published abstract
What these studies add up to
Forest loss is the best measured. Two independent satellite mappings find cocoa inside protected areas on a large scale, and one ties it to over a third of forest loss in Ivorian protected areas.
Income is constrained by land. An analysis across Ghana and Côte d'Ivoire concluded that many smallholders cannot reach a living income from the farms they have.
Yields vary severalfold within a country, with climate zone, and with fertiliser, planting pattern and planting material. Indonesian farms ran at about half their attainable efficiency.
On certification the evidence is now readable and it is modest. A systematic review across crops found higher prices and no higher household income. In cocoa, UTZ outcomes in Ghana and Ivory Coast were mixed and generally modest, Fairtrade raised household spending by 9% in Côte d'Ivoire without improving food security, and in Indonesia certification paid only where the company running it delivered most.
The Living Income Differential has one model, which found anything from nothing to a sizeable gain, and one study of Côte d'Ivoire that found no evidence it had benefited farmers.
Across six data sets, 73 to 90% of cocoa households in Côte d'Ivoire and Ghana earned less than a living income, and doubling the price would leave more than half still below it.
Child labour is hard even to count: parents under-reported it by at least 60% against field data, while children's own reports matched.
Whether higher yields spare forest is disputed between a model that says yes and a household survey that found more clearing where cocoa paid better.
Children harvesting cocoa in Ghana were seen using machetes and pesticides unprotected
What occupational hazards do children face when harvesting cocoa in western Ghana?
How
Observational analysis of children aged 9 through 17 based on personal interviews, focus groups and direct observation of work, with job site analysis. The abstract gives no number of children.
Finding
Children aged 9 through 17 were exposed to strenuous work, sharp tools and pesticides. Reported injuries and illnesses included musculoskeletal disorders, sprains, strains, lacerations to the head, fractures, eye injuries, rashes and coughing.
Does not establish: It gives no prevalence of child labour or of injury.
Limits: Descriptive, with no stated sample size and no rates. It describes conditions two decades ago.
How much deforestation and carbon emission could intensified cocoa production have avoided in the Guinean rain forest?
How
Modelling of land-use change scenarios for Côte d'Ivoire, Ghana, Nigeria and Cameroon, informed by field results on yield responses to seed and fertiliser technology.
Finding
From 1988 to 2007 the area harvested by smallholders of cocoa, cassava and oil palm in the Guinean rain forest increased by 68,000 km². Had intensified cocoa technology been pursued, over 21,000 km² of deforestation and forest degradation could have been avoided, along with nearly 1.4 billion t of carbon dioxide emissions.
Does not establish: It does not demonstrate that intensification reduces deforestation in practice. Ngouhouo-Poufoun and colleagues in this set associate higher cocoa income with more clearing.
Limits: A counterfactual scenario. It assumes that higher yields would have replaced expansion and not encouraged it.
Do certification schemes improve the welfare of farmers and farm workers in low- and middle-income countries?
How
A systematic review of 43 quantitative and 136 qualitative studies of agricultural certification, across crops. Not specific to cocoa.
Finding
Prices were higher for certified producers and so was income from the sale of produce, but overall household income was not, and workers' wages did not benefit. The reviewers found not enough evidence on most outcomes and judged study quality mixed.
Does not establish: That certification does nothing. The finding is that the evidence does not show a gain in household income, on a thin base.
Limits: It covers all certified crops, and the reviewers rate many of the underlying studies as methodologically weak.
Where in the world are commodity crops certified, and do those places match conservation and poverty priorities?
How
Global mapping study synthesising data from over one million certified farms and comparing their locations with environmental, livelihood and physical variables. Organic certification was excluded because spatial data were not available.
Finding
Certification appears to be concentrated in areas important for biodiversity conservation, but not in those areas most in need of poverty alleviation, with exceptions to each pattern.
Does not establish: It does not measure any effect of certification on farmer income or on biodiversity.
Limits: It describes where certification operates, not what it achieves. All commodity crops are pooled.
How do cocoa production, income sources and shade management differ between dry, mid and wet parts of Ghana's cocoa belt?
How
Cross-sectional farm characterisation along a climate gradient in Ghana comparing dry, mid and wet regions. The abstract gives no number of farms.
Finding
Mean cocoa yield was 288 kg per hectare per year in the dry region, significantly lower than 712 and 849 kg in the mid and wet regions. Farmers in the mid and wet regions depended on cocoa for over 80% of annual income, while those in the dry region diversified with non-cocoa crops and off-farm activities.
Does not establish: It does not show that yields will fall by this amount as climate changes, since the gradient is spatial and not a change over time.
Limits: A comparison across places, so soils, farm age and markets differ along with climate.
What have certification and company programmes changed for cocoa farmers in West Africa?
How
Large mixed-method studies comparing UTZ-certified with non-certified farmers at two dates in each country: 2012 and 2015 in Ghana, 2013 and 2017 in Ivory Coast.
Finding
On average outcomes were mixed and generally modest. Certified farmers who received a full package of services had significant gains in productivity and income, but services had since decreased for half the farmers, and the gains were levelling off.
Does not establish: That the certificate caused the gains. They went with the services delivered alongside it.
Limits: A comparison of certified and non-certified farmers who were not assigned at random, in programmes that changed while being studied. The abstract gives no figures.
How efficient are Indonesian cocoa farms and what drove productivity growth?
How
Stochastic frontier analysis of 1,290 panel observations from 722 households in Indonesia.
Finding
Average technical efficiency of the cocoa farmers is 50%. Productivity increased by 75% between 2001 and 2013, mostly from efficiency growth and increased chemicals use supported by government subsidies. Large distortions in input allocation were found.
Does not establish: It does not show that the productivity gain raised household income or reduced poverty.
Limits: Efficiency is relative to the best farms in the sample, so it depends on the model. The abstract does not say which region.
What price premium would compensate Ghanaian cocoa households for giving up child labour?
How
Modelling study: a farm household model developed and calibrated for Ghanaian cocoa, estimating the welfare-neutral price premium.
Finding
Eliminating the worst forms of child labour would require a cocoa price premium of 2.81%. Eliminating regular work over the permitted hours as well would require an 11.81% premium.
Does not establish: It does not show that paying a premium reduces child labour. No premium was paid or tested.
Limits: A calibrated model, not an observed response. It assumes households would withdraw children if compensated and that the rule is enforced.
Is there a gender gap in the technical efficiency of Ghanaian cocoa farms, and what explains it?
How
Cross-sectional analysis using two-stage double bootstrap data envelopment analysis and Blinder-Oaxaca decomposition. The abstract gives no sample size.
Finding
Female plot managers are, on average, less technically efficient than male managers. The gap could be linked to differences in resource endowments, with significant unobservable factors remaining. Education, non-farm activities and farm size may contribute.
Does not establish: It does not show that women are worse farmers, since part of the gap follows from fewer resources.
Limits: No sample size or effect size in the abstract. Efficiency scores depend on the inputs the model includes.
Does security of land tenure decide whether cocoa farmers in Ghana invest in their land?
How
380 surveys covering 796 plots, with key informant interviews, distinguishing legal, practical and perceived security of tenure.
Finding
Farmers planted trees, improved soil and kept naturally occurring trees even where they saw their rights as insecure. The different kinds of security influenced different practices inconsistently, and credit and other factors mattered as well.
Does not establish: That tenure is irrelevant. It shows the link is not the simple one usually assumed.
How many smallholder tree-crop farmers have the potential to earn a living income, and which interventions suit which type of farmer?
How
Analysis of existing datasets on smallholder cocoa and tea farmers in Ghana, Ivory Coast and Kenya combined with literature; an assessment approach rather than an intervention trial. The abstract gives no sample size.
Finding
The authors conclude that a large proportion of these farmers do not have the potential to earn a living income based on their current situation, because they typically cultivate small farms and have low capacity to invest and to diversify. They state there are no silver bullets.
Does not establish: It does not show that higher prices or higher yields alone would close the gap, nor that any named programme has done so.
Limits: Cross-sectional datasets and literature, not a measured effect of any programme. The abstract reports no percentages, so the size of the shortfall cannot be taken from it.
Can cocoa plantations be identified from Sentinel imagery, and how much of the detected cocoa lies in protected areas?
How
Remote sensing study using a multi-temporal stack of Sentinel-1 and Sentinel-2 images and a multi-feature Random Forest classifier across Côte d'Ivoire and Ghana.
Finding
The map achieved 82.89% producer's and 62.22% user's accuracy and detected 3.69 million hectares of cocoa in Côte d'Ivoire and 2.15 million hectares in Ghana. 20% of the detected cocoa plantation area is located in protected areas, and almost 70% of the protected areas in the study area house cocoa plantations.
Does not establish: It does not date the encroachment or show how much forest was cleared for cocoa in any period.
Limits: A user's accuracy of 62.22% means a substantial share of pixels labelled cocoa are not cocoa, so area figures carry wide uncertainty. It is a single-date map, not a measurement of change.
Where and on what crops have sustainability standards been evaluated, and what do the results show?
How
A systematic review of 45 peer-reviewed articles covering 13 agricultural standards.
Finding
Coffee and Fairtrade are over-represented in the literature relative to what is certified; cocoa, and the Ivory Coast, are under-represented. Grouped by case, results were positive in 51%, no different in 41% and negative in 8%.
Does not establish: That standards work half the time. The tally mixes crops, outcomes and study designs.
Limits: A count of results across unlike studies, which the author says should be interpreted carefully.
Does Fairtrade certification change what cocoa farming households in Côte d'Ivoire can spend, and on what?
How
A survey of cocoa farmers analysed by regression with instrumental variables, a method meant to separate the effect of certification from the kind of farmer who becomes certified. The abstract gives no sample size.
Finding
Fairtrade increased household consumption spending by 9% on average and by 14% in poor households, through non-food spending such as housing and clothing. No significant effect was found on food consumption or dietary diversity.
Does not establish: That Fairtrade lifts households to a living income. It measures a rise in spending, not its level.
Limits: One survey at one time, and an instrumental-variable estimate is only as good as its instrument.
Are certified cocoa and oil palm smallholders in Ghana more food secure than uncertified ones?
How
608 household surveys at two sites, compared by propensity score matching on several standard measures of food security.
Finding
Certified farmers scored better on most measures, but the differences were mostly not substantial or statistically significant. 65% of certified cocoa farmers were vulnerable to food insecurity on one standard score.
Does not establish: That certification harms food security. It found a marginal effect, not a negative one.
Limits: Matching compares like with like only on what was measured. Two sites.
How poor are cocoa-producing households in Côte d'Ivoire and Ghana, and what would higher prices or yields change?
How
Analysis of six household data sets, from 385 to 88,896 producers each, with scenarios for doubled prices and higher yields.
Finding
30 to 58% of households had a gross income below the World Bank extreme poverty line and 73 to 90% below a living income. Doubling the cocoa price would still leave 53 to 65% below a living income. Poorer households had lower yields, more members, less land and more reliance on cocoa.
Does not establish: That raising yields is the answer. The scenarios show arithmetic, and the same paper says the poorest are least able to invest.
Limits: Gross income, before costs, for most of the analysis. The yield scenarios assume costs and do not test whether poor households can reach those yields; the authors say they face multiple barriers.
How should a farmer decide between rehabilitating and renovating a declining cocoa orchard?
How
A narrative review of programmes, practices and the evidence on how planting density and yield change with age.
Finding
The decision is described as complex, expensive and often a source of conflict between farmers and advisers. The authors conclude that each case is unique and depends on household, farm, soil, climate and culture.
Does not establish: What replanting costs or returns. The review does not put a figure on either.
Limits: A narrative review. The abstract gives no costs or returns.
How much of traders' sourcing in high-deforestation commodity sectors comes through intermediaries rather than directly from producers?
How
Catalogue of traders' sourcing across four sectors: South American soy, cocoa from Côte d'Ivoire, Indonesian palm oil and Brazilian live cattle exports. The abstract does not describe the data sources or the number of traders.
Finding
Traders often source more than 40% of commodities indirectly via local intermediaries, and the authors describe indirect sourcing as a major blind spot for sustainable sourcing initiatives.
Does not establish: It does not show that indirectly sourced cocoa causes more deforestation than directly sourced cocoa.
Limits: The abstract reports one figure across four sectors and gives no separate figure for cocoa.
Does holding crop insurance change cocoa farmers' income in Ghana?
How
Questionnaire survey of 600 cocoa farmers in Ghana's Ashanti region selected by multi-stage random sampling, analysed with Tobit and propensity score matching.
Finding
Crop insurance had a significant positive impact on cocoa farmers' income in the Ashanti region. Farmer age had a negative effect and savings a positive one.
Does not establish: It does not show that insurance paid out in a drought year or protected income against a specific shock.
Limits: Propensity score matching controls only for measured characteristics. The abstract gives no size of effect.
How much price-setting power do Côte d'Ivoire and Ghana have through their stabilisation systems?
How
Qualitative study based on interviews with actors along the cocoa global value chain. The abstract gives no number of interviews.
Finding
Stabilisation measures address intra-seasonal producer price volatility, and collaboration achieved a living-income differential on top of export prices, but such measures do not shield export and producer prices from inter-seasonal variations in world prices. The two countries largely remain global price-takers.
Does not establish: It does not measure whether the Living Income Differential raised farm incomes.
Limits: Interview-based argument, not a measurement of prices received by farmers before and after the differential.
What might the price surcharge introduced by Côte d'Ivoire and Ghana deliver to farmers?
How
Model simulations of the policy under alternative assumptions about how markets and other producing countries respond. Not a measurement.
Finding
Increases in farmer income ranged from zero to sizeable depending on the assumptions. The policy's success depended strongly on chocolate manufacturers' support unless supply was also managed.
Does not establish: What the differential did. It was written before the outcome could be measured.
How does illegal small-scale gold mining affect cocoa farming where the two coexist?
How
Structured questionnaires with 90 cocoa farmers in two communities of Amansie West District, Ghana, chosen purposively by snowballing.
Finding
Farmers reported labour shortages, polluted water and competition for land. Farms downslope of uncovered pits flooded when it rained, which farmers linked to black pod disease, early loss of young pods and stunted growth.
Does not establish: How much cocoa land or production has been lost to mining anywhere.
Limits: Ninety farmers chosen through each other in one district, reporting their own experience. Nothing was measured on the farms.
Where exactly is cocoa planted in Côte d'Ivoire and Ghana, and how much forest loss in protected areas is associated with it?
How
Remote sensing study. Cocoa plantation data were combined with publicly available satellite imagery in a deep learning framework to produce high-resolution maps for both countries, validated in situ. The abstract gives no count of reference plots.
Finding
The results suggest cocoa cultivation is an underlying driver of over 37% of forest loss in protected areas in Côte d'Ivoire and over 13% in Ghana. Official reports substantially underestimate the planted area, by up to 40% in Ghana.
Does not establish: It does not attribute deforestation to particular companies, certification schemes or farmers, and it does not show that cocoa was the first or only cause of clearing on each plot.
Limits: A map-based association between where cocoa now stands and where forest was lost. Map classification error applies, and the abstract does not give the period over which forest loss was counted.
Does membership of a producer group change the technical efficiency and yield of Ghanaian cocoa smallholders?
How
Cross-sectional comparison of 217 members and 199 non-members of cocoa producer groups using a stochastic frontier model and an endogenous treatment regression to adjust for observed and unobserved differences.
Finding
Producer group membership significantly affected technical efficiency and yield in the endogenous treatment regression model. Farm size, labour and capital had significant positive effects on production.
Does not establish: It does not show an effect on income, and it does not separate certification from group membership.
Limits: Not randomised. The correction for self-selection rests on modelling assumptions, and the abstract gives no effect size.
How is labour organised on smallholder cocoa farms in Ghana, and on what terms?
How
Qualitative interviews, analysed through the idea of bargaining power. The abstract gives no number of interviews.
Finding
Two long-standing informal arrangements are described, hired labour and abusa, in which a caretaker works the landowner's farm, with an imbalance of power between workers and farmers. A newer arrangement of private labour providers is argued to be likely to improve workers' conditions.
Does not establish: How many people work under each arrangement, or what they are paid.
Limits: Qualitative, with the improvement argued and not measured.
How much deforestation in Côte d'Ivoire is driven by cocoa, and how far can exported cocoa be traced back to where it was grown?
How
Publicly available remote-sensing and supply-chain data for Côte d'Ivoire were combined to quantify cocoa-driven deforestation and to trace 2019 cocoa exports, and the deforestation associated with them, from department of origin through trading companies to importing markets.
Finding
The authors find 2.4 Mha of cocoa deforestation and degradation over 2000–2019, 45% of the total over that period. Only 43.6% of exports could be traced to a specific cooperative and department; 23.9% was sourced indirectly through local intermediaries and 32.4% was exported by traders disclosing nothing about their suppliers. Companies in the Cocoa and Forests Initiative had mapped 40% of the farms supplying them, representing 22% of Ivorian exports in 2019.
Does not establish: It does not show which bag of cocoa came from deforested land, and it predates the mapping done since for the European regulation. It says nothing about Ghana or any other origin.
Limits: One country and one export year. Tracing rests on what companies disclose and on modelled links between cooperatives and departments; deforestation is assigned to exports by a model, not observed consignment by consignment.
Do voluntary sustainability standards reduce primary forest loss at country level?
How
Country-level statistical analysis (generalised additive models) of annual certification data for seven standards and five food commodities, including cocoa, against tree cover loss in humid tropical primary forest in 67 tropical countries, plus a spatially explicit analysis in Google Earth Engine.
Finding
Only Fairtrade coincided with lower tree cover loss in primary forests at country level. Higher certification under UTZ, RSPO and Organic coincided with higher tree cover loss. The spatially explicit analysis found no effect of the standards on tree cover loss.
Does not establish: It does not show that certification causes deforestation, and it gives no cocoa-specific estimate.
Limits: Country-level correlations cannot separate the effect of certification from where certification is taken up. Results pool cocoa with coffee, tea, palm oil and banana.
How does price volatility affect producer price share and cocoa supply under Ghana's regulated and Nigeria's liberalised marketing?
How
Time-series econometrics (OLS, GARCH models and a vector error correction model) on annual secondary data from 1970 to 2019.
Finding
Price volatility has no significant relationship with cocoa supply in Ghana and a significantly positive relationship with supply in Nigeria at the 1 % level. In the long run the Ghanaian producer price is negatively affected by international price volatility and inflation, at 5 % and 1 % significance.
Does not establish: It does not show that farmers earn more under either system.
Limits: National aggregates with few observations. The two countries differ in much more than marketing policy.
What drives household deforestation in a conservation landscape, and what role does cocoa income play?
How
Cross-sectional survey of 1035 households in the Tridom landscape in the Congo basin, analysed with a spatial autoregressive model.
Finding
A marginally higher income from cocoa production-based livelihood portfolios is associated with six to seven times higher deforestation compared with other livelihood strategies, with a spillover on neighbouring households. Households imitate the deforestation decisions of their neighbours.
Does not establish: It does not show that cocoa agroforestry itself destroys forest. The authors' point is that promoting it without land use planning may lead to deforestation.
Limits: Self-reported survey data from one landscape with open-access land. Association, not a measured causal effect.
Is it hazardous work specifically, or any cocoa work, that is associated with poorer school progress?
How
Secondary cross-sectional analysis of data on 3,338 children who reported attending school in 2018 across cocoa growing regions of Ghana and Côte d'Ivoire, with mediation analysis for injury.
Finding
Hazardous child labour increases the odds of repeating a class and of work-related injury compared with non-hazardous labour. The effect on academic attainment was mediated by work-related injuries by 14%.
Does not establish: It does not show that non-hazardous farm work is harmless, only that it was not associated with class repetition here.
Limits: Cross-sectional and restricted to children in school, so those who had left school are not counted. The abstract gives no odds ratios.
Can indirect survey methods get accurate reports of child labour from parents?
How
Data from 1741 cocoa producers in Côte d'Ivoire, comparing direct questions with a list experiment and a non-verbal method, with in-depth interviews.
Finding
Indirect methods gave higher rates than direct questions, but lower than surveys of children, which suggest 38%. Parents under-reported for fear of legal consequences and from mistrust.
Does not establish: The true rate. The paper is about how far each method falls short of it.
Limits: One country. The children's figure it compares with is from other surveys.
Whose reports of child labour are accurate, parents' or children's?
How
Comparison of standard surveys of parents and children with data from a cocoa certifier in Côte d'Ivoire that used satellite imagery to select remote communities and visited during harvest, combined with an experiment that randomly assigned a text-message campaign. The abstract gives no sample size.
Finding
Adults under-report child labour relative to the certifier data by a factor of at least 60%, while children's self-reports are statistically identical to the certifier data. Parents' reports can lead to the wrong conclusions about the effects of interventions.
Does not establish: It does not give a prevalence figure for child labour in cocoa.
Limits: The benchmark is itself one organisation's field data, with enumerators imputing work they observe.
Does it matter which company runs a certification scheme for what farmers gain from it?
How
A survey of 458 smallholder cocoa producers under three company-run Rainforest Alliance schemes in South Sulawesi, compared with uncertified farmers by propensity score matching.
Finding
The schemes differed greatly in what they did on farms. Positive effects on production and income appeared in the schemes with the strongest interventions, and not otherwise.
Does not establish: What Rainforest Alliance certification does in general. The point is that the label did not determine the outcome.
Limits: One province, three schemes, and a matched comparison, not an experiment. The abstract gives no size of effect.
Which ways of diversifying bring cocoa households in Côte d'Ivoire more income?
How
A survey of 303 households in five regions, sorted into four types of farming system.
Finding
Households combining cocoa with another income-generating crop on an additional plot earned the most, a system that needs more land. Cocoa intercropped with other crops gave the best cocoa yield. Only 7% of producers had access to finance to invest in diversification.
Does not establish: That diversifying raises income for a household that lacks the land or credit to do it.
Limits: A comparison of household types at one time. Those with more land diversify more and earn more for both reasons.
How did sharecropping come to organise labour in Ghanaian cocoa, and what does it mean for women?
How
Historical analysis with 30 interviews with migrant women working in cocoa production, gathered in Ghana in 2021.
Finding
The article traces sharecropping in Ghanaian cocoa to a shift after the 1920s, and argues that women's exploitation at the base of the chain is rooted in it and in unequal relations within households.
Does not establish: How common sharecropping is now, or how sharecroppers' incomes compare with owners'.
Limits: Thirty interviews and an argument from history; a theoretical piece more than a measurement.
What do voluntary sustainability standards do to trade volumes and export prices of tropical commodities?
How
Gravity models of bilateral trade flows using data on coverage and design of ten standards across seven tropical commodity sectors. The abstract does not list the sectors or years.
Finding
Coverage by standards is associated with increased trade volumes and, in most cases, lower export prices. Price increases occur only for some standards and commodities. Only standards with stringent environmental and ethical requirements and strong compliance control generate price premiums.
Does not establish: It does not measure farmer income, and an export price premium does not establish that farmers received it.
Limits: Country-pair export prices, not prices received by farmers. The abstract does not report cocoa separately.
Which farm, management and biophysical factors explain cocoa yield differences between smallholders in Côte d'Ivoire?
How
Cross-sectional survey of 158 cocoa producers across five sub-prefectures in two climatic zones, analysed with a generalised linear mixed model. Constraints were as perceived by farmers.
Finding
Yields varied from 372 to 812 kg per hectare per year across sub-prefectures, with no statistically significant differences between climatic zones. Mineral fertilisation (+29%), aligned planting (+51%) and identified planting material (+23%) were associated with higher yields, and farmer age (-52%) with lower.
Does not establish: It does not show that giving fertiliser to a farmer raises yield by 29%, and it reports no profitability.
Limits: Associations from one survey. Farmers who fertilise may differ in other ways, and yields appear to be farmer-reported.
How is the cocoa price passed along the supply chain in Côte d'Ivoire, and what is left for farmers?
How
Econometric analysis of price data and a farm household survey, with a qualitative assessment of the institutions. The abstract gives no sample size.
Finding
No inefficiencies were found that would explain persistently low farm-gate prices, and the authors did not find that the Living Income Differential had benefited farmers.
Does not establish: Why farm-gate prices are low. The study rules out the explanations it tested.
Limits: One country, and absence of evidence of a benefit in the data used.
Whether certification itself, apart from the services delivered with it, changes a cocoa farmer's income. No randomised evaluation was found; the comparisons read use matching or instruments.
Whether monitoring and remediation systems reduce child labour.
What it costs to grow cocoa, per hectare and per day of labour, and what farmers have left. Almost every study measures yield, not profit.
How much cocoa land is being lost to illegal gold mining, and how much cocoa crosses borders unrecorded. One small survey of farmers was read on the first and nothing measured on the second.
What share of cocoa can be traced to a farm. The best-known estimate was available only as a preprint and was left out.
These are questions the studies read for this page leave unanswered. They are not a claim that nobody has answered them.
We would like to set two Google Analytics cookies to count page views, and to count them in our own cookie-free measurement as well. They do not identify you and nothing is shared with advertisers. Decline and the site works exactly the same — see cookies for what each one does.