Interactive charts enhance the values below when scripts are available. The data tables, explanation and contact links remain available.
Investigate what shapes nutrient density
Vitagri Vitals connects farm records, food measurements and scientific evidence to investigate nutrient density and farming practices. Explore published studies, then follow our wheat example.
Barn Door Farm · harvest 2025 · mg/kg dry matter
Causal inference in farming and food
12 studies of farm productivity, food choices and nutrition. Choose a theme, then a region or country.
Positive findings from independent research, curated with the Vitals library. Results are specific to each study.
1. Choose a theme
2. Choose a region or country
Start with these three
United Kingdom · Food retailHealthier checkout choices17.3%fewer checkout snack purchases →Greece · Cotton farmingBetter sowing advice, higher yields12–17%higher cotton yield →Australia · Beef farmingSeaweed feed, lower methaneUp to 98%less methane per kg dry feed →Healthier checkout choices17.3%fewer checkout snack purchasesRead case study
- Sector
- Food retail
- Objective
- Estimate how removing sweets, chocolate and crisps from checkouts changes household purchases.
- How it was estimated
- Controlled interrupted time series: compare purchase trends at supermarkets with and without checkout policies.
- Outcome
- Purchases of the targeted foods fell an estimated 17.3% immediately; the 15.5% reduction at one year was less robust.
- Lesson
- Product placement can influence buying. Test longer-term effects before assuming a lasting change.
- Limits
- Purchases are not consumption; other retailer changes could affect the estimate.
Read the original study ↗ (opens in a new tab)Ejlerskov et al. · PLOS Medicine · 2018
Better sowing advice, higher yields12–17%higher cotton yieldRead case study
- Sector
- Cotton farming
- Objective
- Estimate the yield effect of sowing on recommended days rather than unfavourable days.
- How it was estimated
- A causal graph, matching and weighting compare 171 fields from a cooperative’s 2021 season.
- Outcome
- Estimated yield gains ranged from 12% to 17% across methods.
- Lesson
- Judge farm advice by its effect on production, alongside the accuracy of its forecasts.
- Limits
- One season; observational assumptions and unmeasured farm differences remain. Workshop paper.
Read the original study ↗ (opens in a new tab)Tsoumas et al. · NeurIPS workshop / arXiv · 2022
Target fertiliser for better yields0.9–2.8t/ha median yield responseRead case study
- Sector
- Maize farming
- Objective
- Estimate how fertiliser changes maize yield across local growing conditions.
- How it was estimated
- Causal forests use 2,854 yield observations from randomised trials; further modelling explores variation.
- Outcome
- Median yield responses were 2.8 t/ha in Sudan Savannah and 0.9 t/ha in Forest–Savannah Transition.
- Lesson
- Tailor fertiliser advice to local conditions. A national average can hide large differences.
- Limits
- Regional medians, not guaranteed field gains; explanatory soil and weather patterns are not separately randomised.
Read the original study ↗ (opens in a new tab)Kouame et al. · Field Crops Research · 2026
Clear labels, lower-carbon meals4.3%lower emissions per mealRead case study
- Sector
- Food service
- Objective
- Estimate how carbon labels change meal choices and emissions compared with no labels.
- How it was estimated
- A non-randomised field experiment compares labelled and unlabelled cafeterias, covering over 80,000 meal choices.
- Outcome
- Estimated emissions per meal fell 4.3%; high-carbon meal choices fell 2.7 percentage points.
- Lesson
- Clear information can change food choices at low cost. Measure the actual sales mix.
- Limits
- University cafeterias; results may differ in other settings. Emissions are calculated, not directly measured.
Read the original study ↗ (opens in a new tab)Lohmann et al. · Journal of Environmental Economics and Management · 2022
Use less nitrogen, earn more34%less sidedress nitrogenRead case study
- Sector
- Precision maize farming
- Objective
- Estimate how field-specific nitrogen advice changes fertiliser use, yield and profit.
- How it was estimated
- 113 replicated strip trials in Iowa and New York compare model-recommended rates with grower-selected rates.
- Outcome
- Sidedress nitrogen fell 34% on average; estimated partial profit rose US$65/ha, with no significant yield difference.
- Lesson
- Testing advice in real fields can reveal savings that a yield forecast alone misses.
- Limits
- Volunteer farms, four seasons and historical prices. Environmental-loss reductions were simulated.
Read the original study ↗ (opens in a new tab)Sela et al. · Agronomy Journal · 2016
Clearer packs, lower-sugar choices14.2%less added sugar selectedRead case study
- Sector
- Beverage packaging
- Objective
- Estimate how sugar warnings and teaspoon disclosures change parents’ drink choices.
- How it was estimated
- A randomised online trial assigns 5,005 parents to seven packaging conditions.
- Outcome
- Warnings plus teaspoon disclosures reduced added sugar in selected drinks by 14.2% and calories by 6.5%.
- Lesson
- Test packaging with customers to find which information helps them choose.
- Limits
- A one-off simulated choice, not actual purchases or measured consumption.
Read the original study ↗ (opens in a new tab)Musicus et al. · JAMA Network Open · 2022
More efficient organic olive farms~10%higher production efficiencyRead case study
- Sector
- Olive and olive-oil production
- Objective
- Estimate how organic certification changes farm production efficiency.
- How it was estimated
- A production model and propensity-score matching compare similar organic and conventional olive farms.
- Outcome
- Estimated technical efficiency was about 10% higher on organic farms, with the largest gains on small farms.
- Lesson
- Matching similar farms helps separate the effect of production practices from farm size.
- Limits
- Efficiency is not a 10% yield gain. Cross-sectional matching cannot remove unmeasured differences.
Read the original study ↗ (opens in a new tab)Raimondo et al. · Agriculture · 2021
Connect farm support with better nutrition+0.11child height-for-age score unitsRead case study
- Sector
- Farm-linked food and nutrition
- Objective
- Estimate whether adding food, nutrition and hygiene support to an agricultural programme improves child growth.
- How it was estimated
- A cluster-randomised trial compares farming families receiving the combined package with agricultural support alone.
- Outcome
- Child height-for-age improved by 0.11 standard-score units over two years; dietary diversity and iron-rich food intake also improved.
- Lesson
- Link farm programmes to the foods people eat. More production alone may not improve nutrition.
- Limits
- A modest growth effect. The combined package cannot identify which individual component caused it.
Read the original study ↗ (opens in a new tab)Wegmüller et al. · American Journal of Clinical Nutrition · 2022
Seaweed feed, lower methaneUp to 98%less methane per kg dry feedRead case study
- Sector
- Beef farming
- Objective
- Estimate how adding Asparagopsis red seaweed to a feedlot diet changes methane emissions.
- How it was estimated
- Twenty Brangus steers in a randomised incomplete-block trial: four diets, five animals each. Respiration chambers measured methane fortnightly over 90 days; final doses were held for the last 60 days.
- Outcome
- At 0.20% of feed organic matter, methane yield (g/kg dry-matter intake) fell 98% against controls. At 0.10%, it fell 38%.
- Lesson
- A controlled feeding trial with direct emissions measurements can show whether a feed change works.
- Limits
- 20 animals on a specific feedlot ration; doses were adjusted during the first 30 days. This is an enteric methane result, not a whole-farm footprint or a proven grazing-cattle effect. Weight-gain findings need confirmation.
Read the original study ↗ (opens in a new tab)MLA project report ↗ (opens in a new tab)Kinley et al. · Journal of Cleaner Production · 2020
Farmer groups spread better practice1.81×odds of nutrient-management adoptionRead case study
- Sector
- Dairy farming
- Objective
- Estimate how farmer-group participation changes adoption of sustainable management practices.
- How it was estimated
- Spatial propensity-score matching accounts for observed farm differences and neighbouring farmers’ choices.
- Outcome
- Estimated odds of adopting nutrient management were 1.81 times those of comparable non-participants.
- Lesson
- Peer learning can help practices spread. Account for local connections when assessing its effect.
- Limits
- Odds are not probabilities or yield gains; unmeasured farmer differences can still influence results.
Read the original study ↗ (opens in a new tab)Yang & Wang · Annals of Public and Cooperative Economics · 2023
Less sugar, steady sales25%less added sugar sold per personRead case study
- Sector
- Packaged food and drink manufacturing
- Objective
- Estimate how warning-label policy changes the nutrients sold in packaged foods and drinks.
- How it was estimated
- Interrupted time series compares monthly sales and product nutrition with projected pre-policy trends.
- Outcome
- Added sugar sold per person fell an estimated 25%, calories 15%, saturated fat 12% and sodium 7%; overall sales volume stayed stable.
- Lesson
- Track nutrition alongside sales volume to assess changes in the food supply. Reformulation may contribute to the gains.
- Limits
- No untreated comparison group; pandemic effects remain possible. Reformulation is inferred, not separately proven.
Read the original study ↗ (opens in a new tab)Garduño-Alanis et al. · Social Science & Medicine · 2026
Better seed access, better returns>20%higher paddy yieldRead case study
- Sector
- Smallholder crop farming
- Objective
- Estimate how formal rather than informal seed sources affect yield and net crop income.
- How it was estimated
- Propensity-score matching compares similar smallholders in India’s National Sample Survey.
- Outcome
- Estimated yields rose over 20% for paddy and nearly 50% for arhar; paddy net income rose 23%.
- Lesson
- Seed access can improve returns, but benefits differ by crop. Include seed costs in the decision.
- Limits
- Survey-based estimates; matching cannot remove unmeasured differences or isolate every feature of formal seeds.
Read the original study ↗ (opens in a new tab)Nandi · Indian Journal of Agricultural Economics · 2024
Five samples.
A pattern worth investigating.
The grain samples differ in nutritional composition. Select a nutrient, reorder the samples or follow one sample across the charts.
Grain zinc across five samples
MeasuredZinc and fibre
MeasuredMeasured grain only · Pearson r = 0.978 · R² = 0.957 · n = 5.
Read measured values
| Sample | Zinc (mg/kg DM) | Iron (mg/kg DM) | Protein (% DM) | Manganese (mg/kg DM) | Phosphorus (mg/kg DM) | Fibre (% DM) |
|---|---|---|---|---|---|---|
| S1 | 31.690 | 35.666 | 13.158 | 20.698 | 2549.228 | 8.560 |
| S2 | 37.569 | 42.420 | 11.844 | 18.615 | 2989.742 | 11.102 |
| S3 | 29.188 | 31.726 | 12.647 | 25.727 | 2930.318 | 8.929 |
| S4 | 24.266 | 28.043 | 10.755 | 19.001 | 2689.886 | 6.112 |
| S5 | 21.198 | 28.186 | 12.227 | 14.675 | 2469.190 | 4.717 |
What this shows
Across five measured samples, higher fibre goes with higher zinc. That's a lead worth a closer look.
What it can't show
One season and one sample per field can't establish a treatment effect. S4 and S5 have provisional field links; their lab results don't change if those links change.
Bring the soil-to-food
question to life.
Explore how soil indicators could be compared with grain nutrition. Change the soil indicator and test the two possible field assignments to see which patterns move.
Select a point to see its illustrative soil value and measured grain result.
Read the plotted data and its status
| Indicator and timing | Unit | S1 | S2 | S3 | S4 | S5 |
|---|---|---|---|---|---|---|
| Solvita CO₂ burst · before crop | ppm CO₂-C | 75.000 | 75.783 | 65.467 | 64.200 | 62.167 |
| Active carbon (POXC) · after harvest | mg/kg | 498.817 | 545.933 | 438.600 | 466.883 | 417.950 |
| Fungi:bacteria ratio · before crop | ratio | 0.710 | 0.832 | 0.713 | 0.502 | 0.588 |
| Earthworm biomass · before crop | g per spade | 4.583 | 5.850 | 3.700 | 3.250 | 2.100 |
| Microbial biomass carbon · before crop | mg C/kg | 408.700 | 363.117 | 468.817 | 339.767 | 338.867 |
| Mycorrhizal colonisation · at flowering | % root length | 19.617 | 21.917 | 19.000 | 24.250 | 18.167 |
The next evidence to collect
Pair measured
soil biology and grain samples at matched locations, record
variety and yield, and confirm the field identities. The case
proposes six sampling points per field.
Understand the causes
of the outcome.
Agriculture connects biological, physical, social and economic systems. Predictive models help anticipate outcomes. Causal reasoning adds the intervention question: what might change if we farm differently?
Vitagri’s causal engine brings scientific evidence, farm data and causal reasoning together. Its map makes possible pathways and competing explanations visible, helping assess whether the available evidence can support an estimate of an intervention’s effect.
The case asks whether foliar zinc could change grain zinc. Baseline soil may influence both treatment decisions and outcomes, so it needs measuring before any effect can be estimated.
Read the selected relationships
- Baseline soil → soil chemistry (case-proposed assumption)
- Soil chemistry → grain composition (existing wheat-graph assumption)
- Baseline soil → foliar zinc (case-proposed assumption)
- Foliar zinc → grain composition (case-proposed assumption)
- Grain composition → laboratory result (existing wheat-graph assumption)
A selected extract, not the complete causal model or adjustment set. Graph relationships are assumptions to examine. The full Vitals workspace remains private.
Choose the next step
the evidence needs.
A promising pattern doesn't always call for a field trial. Vitals weighs what the question asks, what the records and published evidence can already support, and what is practical on the farm. It then proposes the most proportionate next step, or says plainly that the evidence isn't there yet.
Proposed Worked example: an observational sampling study for this case
No treatment is assigned and nothing about the farming changes. Instead, soil, crop and grain are measured at six matched points in each of the five fields through one season, so the soil-to-grain question can be examined with real soil data, not illustrative values.
What “paired” means: one sampling point, followed through the season
- Before the crop Soil core: chemistry, respiration, fungi:bacteria, earthworms, microbial biomass
- At flowering Roots: mycorrhizal colonisation
- At harvest Grain: composition, bran fraction and yield
- One data point Soil, roots and grain all from the same spot. Repeated 30 times, that's a real soil-to-grain comparison.
Field records (confirmed field identity, variety, sowing date, nitrogen and programme) are kept for every point, so other explanations can be checked too.
What it could show
Whether measured soil biology and grain minerals move together across 30 matched points, and whether yield or bran fraction explains part of the zinc–fibre pattern.
What it can't show
That a practice or soil property causes the difference. Other factors could still explain it, so any link it finds is a lead for further work, not a recommendation.
An illustrative study outline, not an agreed protocol. No results are shown.
Evidence today.
Better models through testing.
Our predictive-modelling research asks how farming conditions relate to food composition, whether food measurements can help test claims about how it was grown, and which interventions deserve field testing. These capabilities need independent validation across farms and seasons.
Pulse · scientific evidence
Frame the question and assess possible explanations.
Vitals · your farm and food data
Inspect patterns, trace pathways and prioritise measurements.
Predictive modelling · research
Test and refine forecasts and verification methods against independent field evidence.
Nutrient density and farm intelligence: questions
What is nutrient density?
Nutrient density describes the nutrients a food provides relative to its energy content. Compare food measurements on the same basis, such as per 100 g or per 100 kcal.
How can causal inference help investigate nutrient density?
Causal inference can estimate how a farming practice changes a measured nutrient when the study design and data support that comparison. Vitals helps identify the evidence and measurements needed to test it.
What is causal inference in agriculture?
Causal inference in agriculture investigates how an outcome would change under a different farming practice, rather than simply identifying which measurements move together. It can use controlled trials or suitable observational data, with explicit assumptions about confounding, timing and comparable groups.
How does Vitagri Vitals use causal inference?
Vitagri Vitals connects farm records, food measurements and research to investigate possible causes. Causal maps show assumptions and competing explanations, helping you choose what to measure or test next.
What is Vitagri Vitals?
Vitagri Vitals is a farm intelligence workspace for investigating nutrient density, farming practices and food quality. This page offers public examples; full access is arranged privately.
What does the Barn Door Farm wheat case show?
The case compares five measured grain samples from one farm in the 2025 harvest. Grain zinc and fibre have a descriptive Pearson correlation of 0.978. Five observations from one season cannot establish a treatment effect or show that the pattern generalises to other farms.
Are the soil biology values measured on the farm?
No. The six soil indicators shown here are illustrative values paired with measured grain nutrition. They are not farm soil measurements and cannot validate a soil-to-food relationship. S4 and S5 have provisional field assignments; switching them tests how sensitive the illustrated correlations are to those assignments.
How does causal reasoning differ from prediction?
Prediction asks what outcome is likely. Causal reasoning asks what might change after an intervention, such as a different nitrogen rate or foliar-zinc treatment. A directed acyclic graph (DAG) makes assumptions and competing explanations explicit. Estimating effects still requires suitable data, defensible assumptions and validation.
Can I explore the full Vitals workspace?
Full Vitals access is arranged privately. Request a demonstration to discuss your farm, laboratory or supply-chain question. The public charts and causal-map extract do not provide access to the full workspace.
Published by Vitagri · Updated . Source: selected Barn Door Farm wheat case, harvest 2025; case snapshot reviewed 23 September 2026. Pearson correlations use five paired values without adjustment for confounding. The full case report is private. For wider context, explore soil and nutrient-density research and the Growing Health report.
What question could
your data answer?
Discuss your farm, laboratory or supply-chain data with Vitagri. Explore the evidence you have and a practical route to stronger answers.
Tell us the question you would like to investigate. Full Vitals workspace access is arranged privately.