Impact of school gardens in Nepal: a cluster randomised controlled trial

The researchers tested a combined school garden intervention with three components delivered together:

1. School vegetable gardens — Standardised raised beds (10 beds, each 3.5m × 1.5m) where children grew 14 different vegetables under teacher guidance. Each school received ~US$950 worth of seeds, tools, fencing, and a polyhouse for seedling propagation.

2. Teaching curriculum — A 23-week curriculum covering gardening, nutrition, and water/sanitation/hygiene (WASH), delivered in a dedicated 1.5-hour class every Friday plus integration into regular subjects like health and agriculture.

3. Promotional activities — Poster displays, handouts, parent briefings twice per year, seed packets for home gardens, and teacher home visits to observe children's home gardens.

Comparator: Control schools received no intervention (business as usual).

Primary outcome: Fruit and vegetable consumption (measured via 24-hour dietary recall) and nutritional status (measured via anthropometry: height, weight, body mass index).

Secondary outcomes: Awareness about fruits and vegetables, knowledge about sustainable agriculture, knowledge about food/nutrition/health/WASH, and stated preferences for eating fruits and vegetables.

• Sample size: 30 schools (15 intervention, 15 control) in Dolakha and Ramechhap districts, Nepal
• Children: 1,275 children at baseline (2014 school year), 785 children at follow-up (2015 school year)
• Age range: 10–15 years old
• Grade levels: Grades 6 and 7
• Setting: Rural and peri-urban public schools in the mid-hills of Nepal, within one hour walking distance from a road
• Inclusion criteria: Non-boarding public schools teaching at least up to grade 8, minimum 150 students, minimum 300m² available land for gardening, no prior school garden programme involvement
• Exclusion criteria: Boarding schools, urban schools, schools with existing garden programmes

• Fruit and vegetable consumption: 24-hour dietary recall (children reported everything they ate and drank in the previous 24 hours, with quantities estimated using standard portion sizes)
• Nutritional status: Anthropometric measurements — height (cm), weight (kg), body mass index (BMI, kg/m²), and age-adjusted z-scores for height-for-age (stunting indicator) and BMI-for-age (underweight/overweight indicator)
• Awareness, knowledge, and preferences: Structured questionnaire administered face-to-face by trained enumerators, covering:
  • Awareness about fruits and vegetables (recognition of specific items)
  • Knowledge about sustainable agriculture (e.g., composting, crop rotation, pest management)
  • Knowledge about food, nutrition, and health (e.g., vitamin sources, balanced diet concepts)
  • Knowledge about WASH (water, sanitation, hygiene practices)
  • Stated preferences for eating fruits and vegetables (likert-scale questions)
• Data collection timing: Baseline survey conducted before intervention (2014), follow-up survey after one year of intervention (2015)

Study design: Cluster randomised controlled trial (cRCT)

Why cluster randomisation matters: Schools were randomised rather than individual children because the intervention was delivered at the school level (gardens, curriculum, promotional activities). Randomising individual children within the same school would have caused contamination — children in the same school would share the garden and lessons. Cluster randomisation accounts for this by treating the school as the unit of randomisation.

Randomisation procedure: 30 schools were selected from a list of 100 eligible schools. These 30 were then randomly assigned to either the intervention group (15 schools) or the control group (15 schools). The paper does not specify the exact randomisation method (e.g., computer-generated random numbers, lottery), but the design is registered in the Registry for International Development Impact Evaluations (RIDIE).

Blinding: No blinding was possible. Teachers, children, parents, and data collectors all knew which schools received the intervention. This is a major limitation because:

• Children in intervention schools may have reported what they thought researchers wanted to hear (social desirability bias)
• Teachers in intervention schools may have emphasised certain topics more during data collection
• Data collectors may have unconsciously biased responses or measurements

Duration: One school year (approximately 9–10 months of intervention, with data collection at baseline and after one year)

Statistical approach: The analysis used regression models that accounted for the clustered nature of the data (children nested within schools). The paper reports using "multilevel mixed-effects models" to separate variation at the school level from variation at the individual child level. Effect sizes are reported as mean differences between intervention and control groups, with p-values and confidence intervals.

Sample attrition: The sample dropped from 1,275 children at baseline to 785 at follow-up — a 38% attrition rate. This is very high and threatens the validity of the results. The paper does not fully explain why so many children were lost, nor does it provide a detailed attrition analysis comparing dropouts to completers.

What this design can prove:

• Causal relationships between the combined intervention and changes in knowledge/awareness/preferences (because of randomisation)
• Average treatment effects at the school level

What this design cannot prove:

• Which specific component (garden vs. curriculum vs. promotion) caused any observed effects — the three components were bundled together
• Long-term effects beyond one year
• Effects on children who dropped out (attrition bias)
• Generalisability to other contexts (urban areas, other countries, different age groups)
• Effects on actual eating behaviour or nutritional status (the primary outcomes were null)

Major methodological weaknesses:

1. No blinding — high risk of social desirability bias and observer bias
2. High attrition (38%) — threatens internal validity; if dropouts differed systematically from completers, results may be biased
3. Short duration (1 year) — nutritional status changes may require longer exposure
4. Self-reported dietary recall — 24-hour recall is notoriously unreliable, especially with children
5. No process evaluation — the paper does not report how faithfully the intervention was implemented across schools (fidelity)
6. No adjustment for multiple comparisons — testing many outcomes increases the risk of false positives

Primary outcomes (no significant effects):

• Fruit and vegetable consumption: No significant difference between intervention and control groups at follow-up. The paper does not report exact consumption amounts (grams or servings) but states the difference was not statistically significant (p > 0.05).
• Nutritional status (height-for-age, BMI-for-age): No significant difference between groups. The paper does not report exact z-scores or prevalence rates but states no significant improvement.

Secondary outcomes (significant improvements):

• Awareness about fruits and vegetables: Significant increase in intervention group compared to control (p < 0.01). Children in intervention schools could name more fruits and vegetables.
• Knowledge about sustainable agriculture: Significant increase (p < 0.01). Intervention children scored higher on questions about composting, crop rotation, and pest management.
• Knowledge about food, nutrition, and health: Significant increase (p < 0.01). Intervention children knew more about vitamin sources, balanced diets, and nutrition-disease links.
• Knowledge about WASH (water, sanitation, hygiene): Significant increase (p < 0.01). Intervention children knew more about hand washing, safe water, and sanitation practices.
• Stated preferences for eating fruits and vegetables: Significant increase (p < 0.01). Intervention children reported liking fruits and vegetables more than control children.

Effect sizes: The paper reports that improvements in knowledge and preferences were "significant" but does not provide standardised effect sizes (e.g., Cohen's d, Hedges' g) or raw mean differences with confidence intervals. This is a reporting weakness — we know the effects were statistically significant but not how large they were in practical terms.

Summary of findings:

• Knowledge improved ✓
• Preferences improved ✓
• Actual eating behaviour did not change ✗
• Nutritional status did not change ✗

The paper does not provide enough detail to calculate precise effect magnitudes. However, the pattern is clear: the intervention successfully changed what children knew and said they liked, but failed to change what they actually ate or their physical health.

To put this in context:

• The knowledge improvements were statistically significant at p < 0.01, meaning there is less than a 1% chance that these differences occurred by random chance alone.
• The null results for consumption and nutritional status were not even close to significance (p > 0.05), suggesting the intervention genuinely failed to move these outcomes.
• The gap between knowledge and behaviour is a well-documented phenomenon in nutrition research — knowing what is healthy does not automatically translate into eating it.

Acknowledged by authors:

• The intervention was a bundle of three components — cannot attribute effects to any single component
• Implementation varied across schools (altitude, land quality, teacher enthusiasm)
• School gardens were too small to directly supply significant amounts of vegetables to children
• No school meals programme existed to serve garden produce
• The study only measured intermediate outcomes (knowledge, preferences) and short-term nutritional status

Additional critical limitations:

1. No blinding — High risk of social desirability bias. Children who knew they were in the "garden programme" may have exaggerated their knowledge and preferences to please researchers.

2. 38% attrition — Losing more than a third of participants is a serious threat. If the children who dropped out were less engaged, less healthy, or from poorer families, the remaining sample may overestimate the intervention's effects.

3. 24-hour dietary recall with children — Children are notoriously poor at accurately recalling what they ate, especially portion sizes. This method has low validity in this age group.

4. No objective biomarker — The study did not measure any biological marker of fruit and vegetable intake (e.g., blood carotenoid levels, urinary potassium). Anthropometry (height/weight) is a crude measure of nutritional status that may not capture short-term changes.

5. Short follow-up — One year may be insufficient to see changes in nutritional status, especially in children who are already stunted or underweight.

6. No household-level data — The intervention encouraged home gardens, but the study did not measure whether households actually planted gardens or whether household food availability changed.

7. No cost-effectiveness analysis — The intervention cost ~US$950 per school (~US$16 per child), but we don't know whether this is a good investment compared to alternatives (e.g., school feeding programmes, micronutrient supplementation).

8. Generalisability — Results from two districts in the mid-hills of Nepal may not apply to other regions, countries, or age groups.

9. No adjustment for multiple comparisons — Testing multiple outcomes (awareness, knowledge, preferences, consumption, nutritional status) without statistical correction increases the risk of false positives.

10. No process evaluation — The paper does not report how many schools actually implemented the garden, curriculum, and promotion activities as intended. Some schools may have done little.

For someone running their own n=1 experiment (e.g., trying to improve their own or their child's fruit and vegetable intake through gardening):

What to test

• Intervention: Growing vegetables at home (or in a community garden) combined with structured nutrition education (reading about nutrition, watching educational videos, attending workshops)
• Dose: At least 1–2 hours per week of hands-on gardening plus 30 minutes per week of nutrition education
• Comparator: Your baseline behaviour (before starting) or a period without gardening

Minimum meaningful duration

• For knowledge change: 8–12 weeks (this study saw changes in ~9 months, but knowledge can shift faster)
• For behaviour change (actual eating): At least 6 months, possibly 12+ months (this study failed to change behaviour in 9 months)
• For nutritional status (weight, BMI, blood markers): 6–12 months minimum; longer for children

What to measure (specific metrics)

• Primary outcome: Daily fruit and vegetable intake — use a food diary (not recall) for at least 7 consecutive days. Measure in grams or servings. Weigh or photograph portions for accuracy.
• Secondary outcomes:
  • Knowledge: Take a simple quiz about nutrition (vitamin sources, recommended servings, health benefits)
  • Preferences: Rate your liking of 10–20 common fruits and vegetables on a 1–5 scale before and after
  • Gardening engagement: Track hours spent gardening, number of plants harvested, types of vegetables grown
  • Home garden produce: Weigh total harvest per week
• Objective biomarker (optional but valuable): Skin carotenoid levels using a Veggie Meter (non-invasive) or blood carotenoid levels — these reflect fruit and vegetable intake more accurately than self-report

Key confounds to control for

1. Seasonality — Fruit and vegetable availability varies by season. Run your experiment across the same season (e.g., spring to autumn) or control for seasonal effects by measuring at the same time of year.
2. Other dietary changes — Did you also start eating out less, join a CSA, or change your grocery shopping? These could confound results.
3. Social desirability bias — You may report eating more vegetables because you think you should. Use objective measures (photos, weighed food, biomarkers) to reduce this.
4. Maturation effects — Children naturally grow and change over time. Use a control period or a matched comparison (e.g., sibling who does not participate).
5. Hawthorne effect — Just being in an experiment can change behaviour. Consider a run-in period where you measure baseline without any intervention.
6. Home garden yield — If your garden produces a lot, you may eat more simply because vegetables are available, not because your preferences changed. Measure availability separately.
7. Cooking skills — Gardening may increase cooking confidence. Track whether you try new recipes or cooking methods.

What a positive result would look like

• Knowledge: Your nutrition quiz score increases by at least 20% from baseline (e.g., from 60% to 80% correct)
• Preferences: Your average liking rating for fruits and vegetables increases by at least 0.5 points on a 5-point scale
• Intake: Your daily fruit and vegetable intake increases by at least 1 serving (80g) per day, sustained for at least 4 consecutive weeks
• Biomarker (if measured): Skin carotenoid score increases by at least 10% from baseline
• Gardening engagement: You spend at least 2 hours per week in the garden and harvest at least 500g of produce per week during peak season

Important caveat from this study

This Nepal study shows that knowledge and preferences can change without behaviour changing. If you see improvements in what you know and say you like, but not in what you actually eat, you are not alone — this is the most common finding in garden-based nutrition research. To bridge the gap, you may need to:

• Increase the availability of vegetables at home (not just in the garden)
• Improve cooking skills to make vegetables more appealing
• Address barriers like cost, time, and family preferences
• Work with household decision-makers (parents, partners) who control food purchasing and preparation

The authors of this study suggest that "to influence children's food decisions, it may be required to work more intensively with parents and to increase the availability of fruit and vegetables at the household and community level." For your n=1 experiment, this means: don't just grow vegetables — also make them convenient, tasty, and accessible in your daily environment.