GiveWell assumes that as a result of the pesticide deregistration, agricultural productivity decreases by 30%. The effects of this estimated productivity decrease on the quality of the intended beneficiaries’ lives is not discussed.
This will be huge if true. Fortunately it is probably not true. My guess is that you got the 30% figure from this quoted text:
Bottom line
We apply a sizable downward adjustment (30%) to our cost-effectiveness analysis to account for our uncertainty about potential effects on agricultural production.
My process is that I searched for “30%” in the linked article, here’s the relevant excerpts:
A back-of-the-envelope break-even calculation suggests that a 1% reduction in agricultural revenue from cultivation, or a 1.4% increase in input costs, would be equal in value to (i.e., would cancel out the benefits of) reducing the pesticide suicide burden by 60% in India. This estimate depends on a number of uncertain assumptions, particularly the numbers we use to compare the value of increasing economic welfare to that of preventing deaths.
We reviewed relevant literature and spoke with experts to better understand the extent to which substitutes were available for pesticides commonly used in suicide.
[...]
We found two studies directly addressing the impact of targeted HHP bans on agricultural production and costs.
Manuweera et al. 2008 compared agricultural yields for a number of crops in Sri Lanka with those from other South Asian countries between 1990 and 2003.55 They found that there was no noticeable change in crop yields[...]
Cha et al. 2016 compared agricultural output in South Korea before and after a ban of paraquat in 2011 and concluded the ban had no detectable effect.58
We believe these studies provide some evidence that the impact of selective pesticide bans in these two countries was not large. However, given the many other factors influencing aggregate crop yields, we do not believe these studies provide conclusive evidence to rule out the possibility that pesticide bans could cause a small (1%) reduction in agricultural productivity.59 We therefore considered other relevant data points to weigh this concern[...]
Bottom line
We apply a sizable downward adjustment (30%) to our cost-effectiveness analysis to account for our uncertainty about potential effects on agricultural production. [emphasis mine]
(In context Givewell seems to assume a mean of 0.3% decrease in expected agricultural productivity, and a median of 0%. This is >=100 times lower than 30%)
TLDR: Sure, the 30% seems quite high, although if the price of alternative fertilizer is around double, it could be accurate for many subsistence farmers.
I have the 30% from this cited text and the BOTEC. In the sheet, 30% seems to be subtracted from the overall cost-effectiveness that considers qualitative adjustments (E77 in “Calculations”). “Calculations” E58 specifies 70% adjustment due to −30% due to risk of agricultural harm (“Assumptions” E36). This 70% multiplies other qualitative adjustments (E60), which multiply the cost-effectiveness before qualitative adjustments (E76) to get cost-effectiveness after adjustments (E77).
The number does seem high, though, especially considering that substitutes seem available. However, it may also be accurate, if farmers are able to afford less fertilizer due to its higher price. One Acre Fund (OAF) RCT-based analysis cites about 50% improvement in yield (in a different region) when farmers are given a loan to purchase (and trained to use) fertilizer and improved seed variety (fertilizer:seed cost is about 2:1[1]). Based on anecdotes from The Last Hunger Season, some farmers cannot afford fertilizer.
The price difference between the highly hazardous pesticides and alternatives is not stated, although pesticides constitute only 7.5% of input costs. However, the document (pp. A-12 - A-13 or 58-59 in the pdf) cited by GiveWell that gathers statistics on farm inputs considers relatively high costs for farm labor and land rent which in the case of subsistence farmers can be neglected (thus the cost would be much higher than 7.5%). There is also very high variance among states in India. Some states seem to use much less fertilizer (e. g. 2.5% of seed costs in Mizoram) than others (39% of seed costs in Andhra Pradesh). Thus, it is unclear to what extent any increases in fertilizer price affect yield.
Further, GiveWell cites that
[p]esticides commonly used for suicide may be more convenient, or have a different mechanism of application, in which case agricultural workers will incur some costs in learning how to use replacements.
Farmers in “The Last Hunger Season” were not trained in fertilizer use prior to the OAF program. It can be that farmers who pay attention to using fertilizer correctly will do so even if another type is offered and vice versa. India’s growing network of rural e-centers with agricultural information can provide appropriate fertilizer information. In other countries of CPSP operations, farmers may be less informed. Thus, any decrease in agricultural productivity due to unfamiliar fertilizer use can be limited.
A professor conducted research on the substantiation of sentiments on counterfeiting. It could be possible that when a new type is introduced, farmers will be suspicious. This can be temporary or have limited effect (trust in local retailer not brand).
(More costly) fertilizer can also substitute other items that increase life quality, such as food, education, or health. Thus, even if a higher cost does not lower yields, the −30% (or other) adjustment could still be valid due to the effects of counterfactual spending.
I understand that GiveWell is assuming a 0.3 agricultural productivity decrease high estimate and 0 or 0.01 low estimate. The high estimate is used, while numbers with 0 decrease are cited next to the adjusted ones, possibly due to high uncertainty about the complex effects on agriculture.
So far, I only considered the effects on smallholders. Effects on industrial farms may be much more substantial, even if the price difference is in the order of percent. I assume that in India, most farms are subsistence. That should be 85% (by land holdings?) in Uttar Pradesh. I further assume that industrial productivity is about 5-10x that of subsistence farm (about 1/2-1/3 of land can be used in subsistence compared to commercial and productivity can be about 2-3x lower). This would suggest that commercial farms produce about as much (Fermi estimate) as subsistence farms (15%*5=75%≈85% or (15%*10=150%≈1.8*85%).
In areas where subsistence farmers use little chemical fertilizer, productivity decrease can be negligible (and much lower than that in commercial agriculture). Conversely, in regions where smallholders spend significant proportions on fertilizer, they can be affected disproportionately more than industries. The former suggests that the median would be close to 0 and mean would be the average of the commercial effects and 0 (e. g. 2% if commercial outputs fall by 4%). The latter can suggest a median value of >30% and mean value of the half of that.
The median would be 30% and mean around 0 if few farmers constitute a large majority of output and are relatively unaffected, while the majority of smallholders are affected significantly. This is what makes intuitive sense, upon the assumption that industrial agriculture largely outperforms subsistence in output and can flexibly (with negligible per unit cost) switch to alternative (or is already using it). However, this can be a biased perspective based on the knowledge of US and other developed economies’ agriculture. While the rapidly industrializing India is the largest nation among CPSP partners, other beneficiary countries can be less industrialized.
Secondary effects from forgone commercial agriculture taxation (as well as any decreases in International competitiveness of beneficiary nations) that can support large proportions subsistence farmers could be discussed.
Lower fertilizer use could lead to higher rents accrued to farmers, if their product is sold as organic with a premium.
Another consideration is that CPSP on its previous website cited investigating the possible negative effects on agricultural productivity in Sri Lanka (listing this on the website can suggest a significant concern). This can be considered in conjunction with GiveWell’s cited enthusiasm and great fit of the professor who leads the project/applied for the grant (he could be motivated to gather and interpret evidence in a way that highlights benefits and unhighlights risks).
The effects of highly hazardous pesticides on agricultural productivity (and the impact on populations) will depend on the
Price and effectiveness differences between the currently and newly used fertilizer for smallholders and commercial farmers
Willingness and ability of subsistence farmers to learn any new fertilizer use techniques
Ratio of subsistence and commercial farms
Use of commercial agriculture taxation on smallholder productivity
Price premium for no chemical fertilizer use
Agricultural productivity units (tons, monetary value, % of farmers not experiencing hunger, …)
Guessing these values, measuring productivity in real local currency units and considering effects only on smallholders, based on the above discussion, the decrease can have a mean of 0.04 with SD=0.02 and be normally distributed, with possible other distributions based on country or region.
One Acre Fund provides $75-80 loans for fertilizer and seeds. 10kg of improved corn seeds costs 70,000 UGX. 10-15kg is needed for an acre (used 100,000 UGX or about $25). Based on the book and confirmed by Global Partnerships, the average farm size is about one acre. $25/$75=1/3, so about 1:2.
I should be clearer. Givewell did not assume a 30% decrease in agricultural productivity anywhere. The 30% reduction is to total expected value of the intervention from a 0.3% (by my inference, not directly stated) reduction in agricultural productivity.
This will be huge if true. Fortunately it is probably not true. My guess is that you got the 30% figure from this quoted text:
My process is that I searched for “30%” in the linked article, here’s the relevant excerpts:
(In context Givewell seems to assume a mean of 0.3% decrease in expected agricultural productivity, and a median of 0%. This is >=100 times lower than 30%)
I genuinely appreciate the reasoning transparency and epistemic legibility illustrated in your article that made it easy for me to pinpoint the presumed error.
TLDR: Sure, the 30% seems quite high, although if the price of alternative fertilizer is around double, it could be accurate for many subsistence farmers.
I have the 30% from this cited text and the BOTEC. In the sheet, 30% seems to be subtracted from the overall cost-effectiveness that considers qualitative adjustments (E77 in “Calculations”). “Calculations” E58 specifies 70% adjustment due to −30% due to risk of agricultural harm (“Assumptions” E36). This 70% multiplies other qualitative adjustments (E60), which multiply the cost-effectiveness before qualitative adjustments (E76) to get cost-effectiveness after adjustments (E77).
The number does seem high, though, especially considering that substitutes seem available. However, it may also be accurate, if farmers are able to afford less fertilizer due to its higher price. One Acre Fund (OAF) RCT-based analysis cites about 50% improvement in yield (in a different region) when farmers are given a loan to purchase (and trained to use) fertilizer and improved seed variety (fertilizer:seed cost is about 2:1[1]). Based on anecdotes from The Last Hunger Season, some farmers cannot afford fertilizer.
The price difference between the highly hazardous pesticides and alternatives is not stated, although pesticides constitute only 7.5% of input costs. However, the document (pp. A-12 - A-13 or 58-59 in the pdf) cited by GiveWell that gathers statistics on farm inputs considers relatively high costs for farm labor and land rent which in the case of subsistence farmers can be neglected (thus the cost would be much higher than 7.5%). There is also very high variance among states in India. Some states seem to use much less fertilizer (e. g. 2.5% of seed costs in Mizoram) than others (39% of seed costs in Andhra Pradesh). Thus, it is unclear to what extent any increases in fertilizer price affect yield.
Further, GiveWell cites that
Farmers in “The Last Hunger Season” were not trained in fertilizer use prior to the OAF program. It can be that farmers who pay attention to using fertilizer correctly will do so even if another type is offered and vice versa. India’s growing network of rural e-centers with agricultural information can provide appropriate fertilizer information. In other countries of CPSP operations, farmers may be less informed. Thus, any decrease in agricultural productivity due to unfamiliar fertilizer use can be limited.
A professor conducted research on the substantiation of sentiments on counterfeiting. It could be possible that when a new type is introduced, farmers will be suspicious. This can be temporary or have limited effect (trust in local retailer not brand).
(More costly) fertilizer can also substitute other items that increase life quality, such as food, education, or health. Thus, even if a higher cost does not lower yields, the −30% (or other) adjustment could still be valid due to the effects of counterfactual spending.
I understand that GiveWell is assuming a 0.3 agricultural productivity decrease high estimate and 0 or 0.01 low estimate. The high estimate is used, while numbers with 0 decrease are cited next to the adjusted ones, possibly due to high uncertainty about the complex effects on agriculture.
So far, I only considered the effects on smallholders. Effects on industrial farms may be much more substantial, even if the price difference is in the order of percent. I assume that in India, most farms are subsistence. That should be 85% (by land holdings?) in Uttar Pradesh. I further assume that industrial productivity is about 5-10x that of subsistence farm (about 1/2-1/3 of land can be used in subsistence compared to commercial and productivity can be about 2-3x lower). This would suggest that commercial farms produce about as much (Fermi estimate) as subsistence farms (15%*5=75%≈85% or (15%*10=150%≈1.8*85%).
In areas where subsistence farmers use little chemical fertilizer, productivity decrease can be negligible (and much lower than that in commercial agriculture). Conversely, in regions where smallholders spend significant proportions on fertilizer, they can be affected disproportionately more than industries. The former suggests that the median would be close to 0 and mean would be the average of the commercial effects and 0 (e. g. 2% if commercial outputs fall by 4%). The latter can suggest a median value of >30% and mean value of the half of that.
The median would be 30% and mean around 0 if few farmers constitute a large majority of output and are relatively unaffected, while the majority of smallholders are affected significantly. This is what makes intuitive sense, upon the assumption that industrial agriculture largely outperforms subsistence in output and can flexibly (with negligible per unit cost) switch to alternative (or is already using it). However, this can be a biased perspective based on the knowledge of US and other developed economies’ agriculture. While the rapidly industrializing India is the largest nation among CPSP partners, other beneficiary countries can be less industrialized.
Secondary effects from forgone commercial agriculture taxation (as well as any decreases in International competitiveness of beneficiary nations) that can support large proportions subsistence farmers could be discussed.
Lower fertilizer use could lead to higher rents accrued to farmers, if their product is sold as organic with a premium.
Another consideration is that CPSP on its previous website cited investigating the possible negative effects on agricultural productivity in Sri Lanka (listing this on the website can suggest a significant concern). This can be considered in conjunction with GiveWell’s cited enthusiasm and great fit of the professor who leads the project/applied for the grant (he could be motivated to gather and interpret evidence in a way that highlights benefits and unhighlights risks).
The effects of highly hazardous pesticides on agricultural productivity (and the impact on populations) will depend on the
Price and effectiveness differences between the currently and newly used fertilizer for smallholders and commercial farmers
Willingness and ability of subsistence farmers to learn any new fertilizer use techniques
Ratio of subsistence and commercial farms
Use of commercial agriculture taxation on smallholder productivity
Price premium for no chemical fertilizer use
Agricultural productivity units (tons, monetary value, % of farmers not experiencing hunger, …)
Guessing these values, measuring productivity in real local currency units and considering effects only on smallholders, based on the above discussion, the decrease can have a mean of 0.04 with SD=0.02 and be normally distributed, with possible other distributions based on country or region.
One Acre Fund provides $75-80 loans for fertilizer and seeds. 10kg of improved corn seeds costs 70,000 UGX. 10-15kg is needed for an acre (used 100,000 UGX or about $25). Based on the book and confirmed by Global Partnerships, the average farm size is about one acre. $25/$75=1/3, so about 1:2.
I should be clearer. Givewell did not assume a 30% decrease in agricultural productivity anywhere. The 30% reduction is to total expected value of the intervention from a 0.3% (by my inference, not directly stated) reduction in agricultural productivity.