Animal Welfare and Policy Risk Index (AWPRI)

The AWPRI compares the volume of animal use a country carries against the governance capacity standing over that use.

What This Page Is For

The question this page answers. Vietnam scores 0.663 on the AWPRI in 2022, the second highest risk score of the 24 countries in the panel. A reader who sees that number asks what is holding it up, and what would have to be different for it to fall. This page answers that question, for any country in the panel. Choose a country, choose one or more of the eight changes listed in the table below, and the page reports what the country's score would read, how far the score moved, which of the three layers the movement came out of, where the country would then rank, and how often any country in the panel has managed a change of that size since 2010.

What a change on this page is, and what it is not. All 24 countries are scored on the same 12 variables, in every year, before and after. No variable is removed from the index and no variable is put into it, and no variable is swapped for another variable. What a change does is take the selected country's own figure on the variables named below and set that figure to a level a country in this panel actually reached in 2022. The index is then rebuilt from the same 12 variables it is always built from. Everything the page reports is the difference between the index computed on the country's observed figures and the index computed on that one altered figure.

Which variables each change moves. The left-hand column names the change, the middle column names the variables that change moves, and every variable not named in the middle column keeps the country's observed value.

The change	Variables it moves	Where it moves them to
Welfare legislation stock at the panel's upper quartile	Animal welfare legislation, stock	better placed than 75 per cent of the panel on that variable in 2022, and no further
Five-year legislative flow at the panel's upper quartile	Animal welfare legislation, five-year flow	better placed than 75 per cent of the panel on that variable in 2022, and no further
Welfare legislation stock at the panel maximum	Animal welfare legislation, stock	the best value any country in the panel recorded in 2022
Governance conditions at the regional best	Rule of law Civic space Civil liberties	the best value recorded in 2022 within the country's own World Bank region
Whole governance layer at the panel best	Animal welfare legislation, stock Rule of law Animal welfare legislation, five-year flow Civic space Civil liberties	the best value any country in the panel recorded in 2022
Animal welfare research base at the panel's upper quartile	Animal welfare research base	better placed than 75 per cent of the panel on that variable in 2022, and no further
Meat supply and animal protein share down to the panel's 75th percentile	Meat supply per capita Animal share of protein supply	better placed than 25 per cent of the panel on that variable in 2022, and no further
Whole exposure layer down to the panel's 75th percentile	Farmed animals slaughtered per capita Aquatic animal share of flesh supply Meat supply per capita Animal share of protein supply	better placed than 25 per cent of the panel on that variable in 2022, and no further

Every level in the right-hand column is one that a country in this panel actually recorded in 2022, so nothing put into the index is invented. The full rule behind each change, and the raw units it is stated in, are in What Each Benchmark Is below.

A worked example, which is the state this page opens in. Vietnam holds 0.839 binding animal welfare instruments per 1,000 indexed FAOLEX records in 2022, the second smallest stock of welfare law in the panel. Three-quarters of the panel stands at 3.521 or below. Raising Vietnam's own figure on that one variable to 3.521, and rebuilding 2022 around it, takes Vietnam's Layer 2 score from 0.852 to 0.729 and its overall score from 0.663 to 0.622. The stock of welfare law is therefore carrying about 0.041 of Vietnam's score, on a scale that runs from 0 to 1. Run the same exercise across the eight changes and the page shows which part of a country's risk score is the largest, and which part is closest to a level other countries have already reached.

Why a country you did not select can still move. Each variable is placed on the 0 to 1 scale by the lowest and the highest value present in the panel that year, so changing one country's figure can move that scale for everyone. The whole 2022 panel is therefore rebuilt, through the same code that produces the published index. All computation runs in the reader's own browser and no data is sent anywhere.

This page has three limits, and each limit is stated here so that no figure below is read for more than the figure is worth. The page does not estimate what a policy would achieve, because nothing in this study identifies the effect of a law or a programme on any of the 12 variables. The page does not forecast, because the arithmetic is contained entirely within 2022. The page does not claim that a change is achievable. What the page reports is how often a change of that size has been observed between 2010 and 2022, which is a weaker statement than achievability.

What the 12 variables are. The AWPRI scores 24 countries from 2010 to 2022 on three layers. Layer 1 records the number of farmed animals slaughtered per person, the quantity of meat supplied per person, the share of dietary protein drawn from animals, and the aquatic share of flesh supply. Layer 2 records the cumulative stock and the five-year flow of binding animal welfare instruments, together with rule of law, civic space and civil liberties, which are the institutional conditions under which animal welfare law is enforced and contested. Layer 3 records how a country's research output divides between artificial intelligence and precision agriculture on one side and animal welfare science on the other. Scores run from 0 to 1 and a higher score means greater risk. The AWPRI does not measure the welfare of animals. The AWPRI measures animal-use exposure against governance capacity, and the methodology page sets out that distinction in full before any number is given.

Choose a Country and a Change

Country

Select a change to apply it, and select the same change again to remove it. More than one change can be active at once, and every figure below is recomputed on each selection. What each active change does to the selected country is printed underneath, and all eight changes are defined in What Each Benchmark Is.

The Full Panel After the Substitution

All 24 countries in 2022, ordered by their observed score. The selected country is the only country whose inputs were changed. Any other country that moves does so because the 0 to 1 scale is set by the lowest and the highest value present in the panel, so changing one country can change the scale for every country.

Effect on the Layer Scores and on Each Variable Moved

The composite score is the mean of the three layer scores. A substitution that touches Layer 2 alone therefore moves the composite by one third of the change in Layer 2.

The three figures on each row are the observed 2022 score, the score after the substitution, and the difference between the two. A negative difference means the composite score falls, which on this scale means lower risk.

How the Number Is Worked Out

Step 1. Start from the raw values

The panel holds 12 variables in their own units for 24 countries, such as head of animal slaughtered per person, kilograms of meat supplied per person, binding instruments per 1,000 indexed FAOLEX records, and publications per capita. The calculator starts from those raw values for 2022 and not from the normalised scores, which is what makes Step 3 possible.

Step 2. Substitute the benchmark

A benchmark is a value that some country in the panel actually recorded. Each benchmark is derived on the normalised risk scale, because the normalised scale is the one on which a lower value means lower risk for every variable, and the benchmark is then converted back into the variable's own units so that the substituted figure is a real quantity. Where the selected country already stands at or beyond the benchmark, the variable is left unchanged and reported as already met. The calculator never moves a country backwards in order to meet a target.

Step 3. Renormalise the whole year

Every variable is rescaled to the interval [0, 1] by min-max normalisation within 2022 across all 24 countries, and the six protective variables are then inverted so that a higher value always means greater risk. This step carries the substantive consequence. Because the bounds are the panel's own minimum and maximum, changing one country can change the scale for every country, and a calculator that edited a normalised score in place would conceal that consequence. The panel is therefore rebuilt in full, and the number of other countries whose score moved is reported on every run.

Step 4. Average the layers and check against the published index

Each layer score is the mean of the variables belonging to that layer, and the composite score is the mean of the three layer scores, so a Layer 1 variable carries 0.0833 of the composite, a Layer 2 variable 0.0667, and a Layer 3 variable 0.1111. Before any figure is displayed, the baseline that this page computes is compared against the published panel. Where the two disagree by more than 0.0005, the page states the discrepancy at the top instead of showing figures that differ from the rankings table without warning.

How unusual a move of that size would be

The final column of the variable table compares the requested move against every within-country change on that variable observed between 2010 and 2022, taken over all ordered pairs of years so that a 13-year change is counted alongside a one-year change. A reading of 0.0 per cent means that no country in this panel has ever moved that far on that variable. That share is the closest statement about feasibility the data supports. The share does not report whether a policy would work. The share reports whether a change of that size has ever occurred.

The substitution, the renormalisation and the averaging are the same three operations, performed in the same order, that panel_assemble.py performs when building the published index. The benchmarks are computed by policy_sim.py in the pipeline and are sent to this page as raw target values. The full construction, the codebook for all 12 variables, the weights and the 312 observed country-years are set out on the methodology page.

What Each Benchmark Is

There are eight benchmarks, and each benchmark is a level that exists in the observed data. Three benchmarks are panel percentiles, two are the lowest-risk value that any country in the panel reaches, one is the lowest-risk value reached within the country's own World Bank region, and two move an entire layer at once in order to show the arithmetic limit of that layer. No benchmark is a policy proposal.

Welfare legislation stock at the panel's upper quartile

The country's cumulative stock of binding animal welfare instruments, per 1,000 indexed FAOLEX records, reaches the level that three-quarters of the panel already stands at or below.

The target value is the 25th percentile of the risk scale in 2022, which leaves the country better placed than 75% of the 24 countries on that variable and no further.
The variables moved are Animal welfare legislation, stock (L2).
All remaining variables are left unchanged.

Five-year legislative flow at the panel's upper quartile

The country enacts new binding instruments over five years at the rate the panel's upper quartile already manages. The flow variable is a rate and the stock variable is an accumulation, so this benchmark states a pace of legislating and does not state a quantity of accumulated law.

The target value is the 25th percentile of the risk scale in 2022, which leaves the country better placed than 75% of the 24 countries on that variable and no further.
The variables moved are Animal welfare legislation, five-year flow (L2).
All remaining variables are left unchanged.

Welfare legislation stock at the panel maximum

The country matches whichever country holds the densest stock of binding instruments in the panel that year. This benchmark is the upper bound on what the legislation variable can contribute.

The target value is the lowest risk value any of the 24 countries recorded in 2022.
The variables moved are Animal welfare legislation, stock (L2).
All remaining variables are left unchanged.

Governance conditions at the regional best

Rule of law, civic space and civil liberties reach the strongest level observed in the country's own World Bank region that year. Rule of law, civic space and civil liberties are conditions under which welfare law is enforced and contested, and are not welfare measures in themselves. The codebook records that pathway as indirect.

The target value is the lowest risk value recorded in 2022 by another country in the same World Bank region.
The variables moved are Rule of law, Civic space, Civil liberties (L2).
All remaining variables are left unchanged.

Whole governance layer at the panel best

All five governance variables reach the strongest value any country in the panel reaches. That combination is the arithmetic ceiling of the layer, and is included to show how much of the index the governance layer can account for. The combination is not offered as a policy proposal.

The target value is the lowest risk value any of the 24 countries recorded in 2022.
The variables moved are Animal welfare legislation, stock, Rule of law, Animal welfare legislation, five-year flow, Civic space, Civil liberties (L2).
All remaining variables are left unchanged.

Animal welfare research base at the panel's upper quartile

Animal welfare publications per capita reach the panel's upper quartile. The codebook records this pathway as assumed, because the index treats a larger welfare research base as offsetting agricultural automation and this study does not establish that the research base has that effect.

The target value is the 25th percentile of the risk scale in 2022, which leaves the country better placed than 75% of the 24 countries on that variable and no further.
The variables moved are Animal welfare research base (L3).
All remaining variables are left unchanged.

Meat supply and animal protein share down to the panel's 75th percentile

Meat supply per person and the animal share of protein supply fall to the level that three-quarters of the panel already stands at or below, which leaves the country better placed than a quarter of the panel and no further. This benchmark is the smallest of the exposure benchmarks, and it applies only to countries currently above that level. FAO Food Balance Sheets measure supply and not consumption, so the benchmark states what enters the food system.

The target value is the 75th percentile of the risk scale in 2022, which leaves the country better placed than 25% of the 24 countries on that variable and no further.
The variables moved are Meat supply per capita, Animal share of protein supply (L1).
All remaining variables are left unchanged.

Whole exposure layer down to the panel's 75th percentile

All four exposure variables fall to the level that three-quarters of the panel already stands at or below, meaning fewer animals slaughtered per person, less meat and less animal protein in supply, and a smaller aquatic share of flesh supply. The country is then better placed than a quarter of the panel on exposure, and is not in the lower quartile of the panel.

The target value is the 75th percentile of the risk scale in 2022, which leaves the country better placed than 25% of the 24 countries on that variable and no further.
The variables moved are Farmed animals slaughtered per capita, Aquatic animal share of flesh supply, Meat supply per capita, Animal share of protein supply (L1).
All remaining variables are left unchanged.

The word best always means the lowest risk on the normalised scale. For a protective variable such as rule of law the lowest-risk position is that variable's highest raw value, and for an exposure variable such as meat supply the lowest-risk position is that variable's lowest raw value. The conversion is performed in the pipeline so that the substituted figure shown in the table stands in the variable's own units.

What These Numbers Cannot Tell You

No causal claim is made. This page performs arithmetic on a published formula. The page reports what the AWPRI would read if an input value were different, and reports nothing about what would bring that value about or what would follow for animals.
The score is relative and not absolute. Min-max normalisation is performed within year and within this panel, so a value of 1 means that no country in the panel scored higher in that year, and does not mean that the underlying quantity reached a ceiling. The same substitution performed in a panel of different countries would return a different figure.
A layer score moving is not a country changing. When other countries shift in the chart, nothing has occurred in those countries. The scale beneath those countries has moved.
Layer 3 rests on an assumed pathway. The codebook grades all three Layer 3 variables as assumed, which is to say that the AWPRI treats research effort in artificial intelligence and precision agriculture as bearing on welfare harm, and this study does not establish that the effort does bear on welfare harm. Benchmarks that touch Layer 3 should be read on that footing.
The AWPRI is not validated against welfare outcomes. No harmonised cross-national series of on-farm welfare outcomes exists for these countries and years. Tested against World Animal Protection's Animal Protection Index, the composite score shows no significant association (Spearman ρ = −0.264, p = 0.213), and the methodology page reports that divergence as a result about what the two instruments measure.