Motivation

In recent decades there has been increased coverage of animal welfare issues, the health risks of high consumption of animal products and the contribution of farming animals to climate change. Multiple high-profile organisations have called for reduced animal consumption through reducetarian, vegetarian and vegan diets [1, 2].

Each of these three issues has been studied extensively and solving these issues has gained broad support. A balanced vegan diet has the opportunity to reduce the negative impacts of all three issues but campaigning for veganism has not been very successful and veganism still has relatively low prevalence of under 10%. For example, in the United Kingdom, where the first vegan society was founded, prevalence of veganism is estimated at 1.16% by the most recent survey [3].

An alternative to veganism called flexitarianism or reducetarianism has been gaining ground. The idea of these diets is to reduce consumption of animal products by partially replacing them with plant-based foods. If one were to adopt a flexitarian diet it would be confusing to decide which species to avoid since consumption of different species have different levels of associated harm on the three scales. For example, poultry has lower health risks and contributes less to climate change than beef but is considered a poor choice from an animal welfare perspective due to the poor conditions of the animals and the high number of animals used.

The goal of this tool is to allow the user to specify their relative concern on two issues: animal welfare and climate change. The tool ranks animal species according to the harm induced by their consumption while taking into accord the user’s values. This ranking could be used to decide which animal products should be replaced with plant-based products.

Rankings based on welfare have been developed previously by various individuals and groups such as Peter Hurford, Brian Tomasik, Charity Entrepreneurship and Dominik Peters. This tool is a minor extension of the work of Dominik Peters that also considers emissions in addition to welfare. I want to thank Dominik for kindly providing the data and methodology that he used.

Animal suffering subscale

To estimate the negative impact on animal welfare, the tool calculates the number of hours animals have spent in a farm in order to produce an amount of food which provides 2000 kcal of energy. For example, Dominik retrieved data on production yields and slaughter age from providers of breed chickens such as Aviagen and Lohmann Tierzucht.

I calculated the amount of produce required for 2000 kcal of energy using data from the FoodData Central website of the US Department of Agriculture. Of course different preparations from the same animal can have varying nutritional value but for the sake of simplicity one value is used per species.

Let $suffering_s$ designate the suffering subscale score of species $s$. Let $lifespan_s,\ production_s,\ refweight_s$ designate the average lifespan of an animal in hours, weight of produce per animal and weight of produce required for 2000 kcal of energy. The basic score is the number of hours suffered on a farm to produce 2000 kcal of produce:

$suffering_s = \frac{lifespan_s}{production_s} \times refweight_s$

There are some issues with this basic approach. For example, our confidence in different species being sentient varies. If we just account for hours lived to produce 2000 kcal then the harm of consuming smaller species will dominate. But the user of the tool might have low confidence in shrimp being sentient and we want to account for that.

Estimates of welfare range were used to account for differences in sentience and capacity for welfare [4]. The welfare range of an animal is the range between the maximum positive welfare and minimum negative welfare the animal can experience. To estimate welfare ranges Rethink Priorities collected data on dozens of empirical traits related to the evolutionary functions of pleasure and pain. Trait data of each species was aggregated into a value between 0 and 1 where 1 corresponds to the welfare range of humans. These estimates were corrected for the probability of sentience. The user can choose to factor in welfare ranges in which case $suffering_s$ is multiplied by the species’ welfare range.

Different animal products might have different price elasticities. Price elasticity of supply shows the responsiveness of production to change in price. Elasticity of demand shows the responsiveness of demand to change in price. Cumulative elasticity is the net effect on supply. If someone spares 10 chickens a year by not eating chickens the actual change could be less than 10. Decreased chicken meat price due to lower demand might motivate someone else to eat more chicken. The user can choose to factor in cumulative elasticity in order to account for this effect. Two sources are provided: the book “Compassion, by the Pound” [5] and the work of the organisation Animal Charity Evaluators [6].

The user can also choose to factor in sleeping time if they assume that animals do not suffer while sleeping and liveability which also factors in animals who die before slaughter.

Climate change subscale

The climate change subscale measures the CO₂ equivalent greenhouse gases produced per kilogram of animal produce. This value is scaled according to the amount of produce required for 2000 kcal. CO₂ equivalent emissions data has been collected from lifecycle analyses [7, 8]. Let $emissions_s$ designate the CO₂ equivalent gases produced per kilogram of produce of species $s$. The climate subscale score of species $s$ is thus:

$climate_s = emissions_s \times refweight_s$

The elasticity parameters apply both to the suffering and climate subscales. That means if elasticity is enabled then $climate_s$ is multiplied by the cumulative elasticity factor.

Note that only the impact on climate change is considered. There are other negative environmental impacts. Saltwater fishing causes marine pollution and fish farms cause eutrophication. But since the risk of climate change outweighs other environmental risks related to the consumption of animal produce I consider these omissions acceptable.

It is sometimes argued that buying local food is more important than reducing meat consumption. In general the climate impact of food is dominated by production [9] so this tool does not make a distinction between where animals were farmed. Imported plant-based food tends to have lower emissions than local animal produce. Life cycle analysis of animal products already includes transportation and this is considered sufficient.

The tool also does not consider which plant-based foods are substituted for animal products. Plant-based food production in general causes significantly lower emissions [10, 11] but from the perspective of the environment it might make sense to prefer whole foods because these do not require additional energy-intensive processing.

An issue with using CO₂ equivalent greenhouse gas emissions as a measure of warming is that farming different species puts different types of greenhouse gases in the environment. The high impact of ruminants is caused by their methane emissions. While methane warms the atmosphere more than CO₂ it is also removed from the atmosphere significantly faster. Climate scientists sometimes use a metric called CO₂-equivalent with Global Warming Potential 100 which considers methane to cause 25x as much warming as an equivalent amount of CO₂ over a century. Some physicists disagree with this approach [12]. If one is concerned about the effects of warming over thousands of years as opposed to a hundred years this approach understates the impact of CO₂ compared to methane.

Combined model

A weighted product model is used to combine the subscales. Weighted product models are dimensionless and are used for ranking options when making decisions. Because the scores are dimensionless they are normalised to the range $[0, 100]$.

Let $w_{suffering}$ and $w_{climate}$ designate the suffering and climate weights. The combined score of species $s$ is calculated using:

$harm_s = suffering_s^{w_{suffering}} \cdot climate_s^{w_{climate}}$

A product model ensures that the subscales affect the combined score equivalently. A 1% increase in CO₂ emissions changes the combined score by the same amount that a 1% increase in the animal suffering subscale would. Adding weights to the model allows us to change the relative contribution of each subscale to the combined score.

Why is there no health impacts subscale?

I considered designing a health subscale using data from the Global Burden of Disease (GBD) study but eventually opted against it.

Understanding nutrition is notoriously difficult. It is impossible to conduct trials that assess long-term health impacts of diets due to costs and ethical concerns with assigning people to diets with unknown health effects. Due to this dietary decisions must be made based mostly on observational data which is not as reliable as randomised controlled trials. Even well studied questions such as the impact of saturated fat consumption have not been fully resolved [13, 14].

GBD data has been aggregated from a large number of sources by a large team. It might be close to the consensus of nutritional science but even if we have reasons to trust the data it would be difficult to use due to non-linear effects. If two people decided to eat one less chicken this year their climate and animal welfare impact would be similar even if one of them previously ate 10 chickens per year and the other ate a single chicken each year. Positive health effects of reduced meat consumption on the other hand have diminishing returns and the reduction of their health risks would depend on the current composition of their diets.

Adequately planned reducetarian, vegetarian and vegan diets are believed to be healthy based on existing evidence [15] but positive effects over conventional diets are not considered in this tool due the uncertainty of the effects and difficulties of modeling.

Limitations

Multiple issues with the method used to estimate suffering are outlined in [16]. The tool could be helpful to make decisions in the face of uncertainty but is not a true measure of harm.

The model does not consider the suffering of wild animals which could significantly exceed that of farm animals [17]. There is also no consideration of which plant-based products are substituted for animal products. Farming of some plants causes less emissions [10] or wild animal suffering [18].

I consider the greatest limitation of the tool the fact that setting subscale priorities based on intuitions can be misleading. It would make sense to compare emissions and harm based on the underlying values which cause us to be concerned about the issues in the first place. If for example I am motivated by increased welfare, it would be helpful to estimate the welfare impacts of climate change and factory farming on a common scale.

Sources

Food energy

References