According to Gram Research analysis, the way scientists statistically analyze food substitution studies dramatically affects the results—sometimes producing opposite conclusions from the same data. A 2026 study of 1,063 Swedish men found that simple statistical methods gave different mortality risk estimates for regular milk depending on whether food was measured in grams or calories (18% increased risk versus no effect), while more detailed statistical methods produced consistent results regardless of measurement unit. This demonstrates that researchers must be transparent about their statistical approach because different valid methods answer different research questions.
When researchers study how different foods affect our health, the way they analyze the data can dramatically change the results—sometimes even flipping conclusions upside down. A new study looked at 1,063 Swedish men to understand how different statistical methods for studying food substitution (like choosing fermented milk instead of regular milk) can give different answers about mortality risk. The research found that some methods produced very different estimates depending on how they measured food intake, while other methods were more consistent. This matters because it shows scientists need to be clearer about what question they’re actually answering when they study food and health.
Key Statistics
A 2026 methodological study of 1,063 Swedish elderly men found that simple statistical models for food substitution produced opposite conclusions depending on measurement units: regular milk showed an 18% increased mortality risk using mixed units but no effect using calories alone.
According to research reviewed by Gram, detailed statistical models that adjust for other foods produced consistent estimates across different measurement units (grams versus calories), while simple models varied significantly, suggesting more complex approaches are more reliable for nutrition research.
A 2026 analysis of 18 different statistical substitution models found that fermented milk showed similar results across both simple and detailed approaches, while regular milk results varied dramatically by methodology, indicating some food-health relationships are more robust to statistical approach than others.
The Quick Take
- What they studied: How different statistical methods for analyzing food substitution studies produce different results, using milk consumption and death rates as an example.
- Who participated: 1,063 elderly Swedish men tracked over time to see which ones passed away and what they ate.
- Key finding: The way scientists measure and adjust their calculations significantly changes the results. Some methods gave very different answers depending on whether they measured food in grams or calories, while other methods stayed consistent.
- What it means for you: When you read nutrition studies claiming one food is better than another, the method used matters as much as the finding itself. Be cautious about headlines from studies using different statistical approaches, as they may not be directly comparable.
The Research Details
This was a methodological study—basically a study about how to do studies correctly. The researchers took real data from 1,063 Swedish men and tested 18 different statistical approaches to see how each one answered the question: ‘Does eating fermented milk instead of regular milk affect how long people live?’ They compared two main strategies: a simple approach (just adjusting for total calories) and more detailed approaches (adjusting for calories and other foods too). They also tested whether measuring food in grams, calories, or a mix of both changed the answers.
The researchers used a statistical tool called Cox proportional hazard regression, which is a fancy way of analyzing which people died and whether their food choices were connected to that outcome. They got death information from Sweden’s official death registry, so they had accurate data about who passed away during the study period.
The key innovation here is that they deliberately tested multiple methods on the same data to see which ones gave consistent answers and which ones gave wildly different results depending on how you set up the calculation.
This research matters because nutrition science relies heavily on observational studies—watching what people eat and what happens to their health over time. But these studies are tricky because people who eat different foods often have other differences too (like exercise habits or income). The way scientists adjust for these differences mathematically can completely change the answer. If some methods give opposite conclusions from others, we need to know which method is most trustworthy before we change our diets based on the results.
This study has several strengths: it used real data from a large group of people, it had accurate death records from official sources, and it systematically tested multiple approaches. However, it’s important to note this was a methodological study, not a study proving milk is good or bad for you. The findings apply to how scientists should design nutrition studies, not to whether you should drink more milk. The study was limited to elderly Swedish men, so results might differ in other populations.
What the Results Show
When researchers used the simple approach (just adjusting for total calories) to study regular milk, the results changed dramatically depending on whether they measured food in calories or grams. Using mixed units (both calories and grams), regular milk appeared to increase death risk by 18% (hazard ratio 1.18). But using only calories, there was essentially no effect (hazard ratio 1.00). This is a huge difference—one method suggests a risk, the other suggests no risk at all.
Fermented milk (like yogurt) showed a different pattern. Both the simple and detailed approaches gave similar results regardless of how the food was measured. This suggests that fermented milk’s relationship to mortality is more stable across different statistical methods.
When researchers used the more detailed approaches (adjusting for calories and other foods), the results for regular milk became more consistent across different measurement units. This suggests that the detailed method is more reliable because it accounts for what people eat instead of the excluded food.
The study identified 18 different valid statistical models, but each one technically answers a slightly different question about food substitution. The researchers emphasize that scientists need to be explicit about which question they’re answering.
The study found that the choice between measuring food in grams versus kilocalories (calories) matters more in simple statistical models than in detailed ones. The detailed models that adjusted for other foods were more robust to this choice. Additionally, the ’leave-one-out’ approach (removing the substituted food from the analysis) produced more consistent results than the simple approach, suggesting it’s a more reliable method for future studies.
Previous simulation studies (computer models) had shown that different statistical methods could produce opposite results, but this was the first major study to demonstrate this problem using real human data. This research confirms that the theoretical concerns about methodology are actually happening in practice with real nutrition data. It validates the need for researchers to be more careful and transparent about their statistical choices.
The study only included elderly Swedish men, so the results might not apply to women, younger people, or people from other countries with different diets. The study looked at milk consumption and death, but the specific findings about milk don’t prove anything about whether milk is healthy or unhealthy—the study was about methodology, not milk safety. Additionally, observational studies like this can’t prove cause and effect, only associations. The study was also limited to a single food substitution question; results might differ for other foods.
The Bottom Line
For consumers: When reading nutrition headlines, look for studies that clearly explain their statistical methods. Be skeptical of strong claims from single studies, especially if they use simple statistical approaches. For researchers: Use detailed substitution models that adjust for other foods, and be explicit about what research question you’re answering. For journalists: Ask researchers which statistical method they used before reporting results as definitive.
Nutrition researchers and anyone who reads nutrition studies should care about this. If you’re making health decisions based on nutrition research, understanding that methodology matters helps you evaluate claims more critically. Journalists and science communicators should use this to improve how they report nutrition findings.
This is a methodological study, not a treatment study, so there’s no timeline for personal health benefits. However, if researchers adopt better statistical methods going forward, we should see more reliable nutrition research within 2-3 years as new studies using these improved methods are published.
Frequently Asked Questions
Why do different nutrition studies give different answers about the same food?
A 2026 study found that the statistical method researchers use significantly affects results. Simple methods can produce opposite conclusions from the same data depending on how food is measured. Researchers need to be transparent about their methodology so studies can be properly compared.
Does fermented milk affect mortality differently than regular milk?
This study examined methodology rather than milk’s health effects. However, fermented milk showed more consistent results across different statistical approaches than regular milk, suggesting its relationship to health outcomes may be more stable and reliable to study.
What’s the best way for scientists to study food substitution?
According to this 2026 research, detailed statistical models that adjust for total calories and other foods produce more consistent results than simple models. Researchers should also be explicit about what research question they’re answering, since different valid methods answer different questions.
Should I trust nutrition studies that claim one food is better than another?
Be cautious. Ask what statistical method was used—studies using detailed approaches that adjust for multiple factors are generally more reliable than simple approaches. Also check if results have been replicated in multiple studies, since single studies can be affected by methodology choices.
How does measuring food in grams versus calories change nutrition research results?
A 2026 study found that measurement units significantly affected results in simple statistical models but not in detailed ones. Using mixed units (both grams and calories) produced different conclusions than using only calories, showing that consistency in measurement matters for reliable research.
Want to Apply This Research?
- Track your food intake using consistent units (either grams or calories, not mixed) for at least 30 days to see how your measurement choice affects patterns you notice. This helps you understand the real-world impact of measurement methods.
- When logging foods in a nutrition app, choose one consistent measurement unit and stick with it. This mirrors the study’s finding that consistent measurement approaches give more reliable results. If your app switches between grams and calories, note which unit you’re using.
- Over 3 months, compare your nutrition insights when measuring in grams versus calories. You may notice different patterns emerge depending on your measurement choice, which mirrors what researchers found in this study. This personal experiment helps you understand why scientific methodology matters.
This article discusses a methodological study about how to conduct nutrition research—it is not a study proving any food is healthy or unhealthy. The findings apply to research design, not to dietary recommendations. Always consult with a healthcare provider or registered dietitian before making significant dietary changes. This study was conducted in elderly Swedish men and may not apply to other populations. Observational studies show associations, not cause-and-effect relationships.
This research translation is published by Gram Research, the science division of Gram, an AI-powered nutrition tracking app.