The Problem with Changing Five Things at Once

A familiar story: someone starts a new way of eating, and within a few weeks their weight, energy or blood test results have changed. They credit the one feature of the plan they find most interesting, such as meal timing, the absence of a particular ingredient, or a certain balance of nutrients. But look closer and they changed five things at once. The result is real. The explanation is a guess.

The question and why it matters

Many diet programmes and personal accounts are bundles of changes. That is not necessarily a criticism of the bundle; it may be a sensible package to try. The problem is explanatory. If theories are built on bundled results, they can claim support they have not earned.

A hypothetical bundle

This example is invented for illustration. Imagine someone who, in the same week:

1. Stops eating packaged snacks.
2. Eats all meals within an eight-hour window.
3. Starts cooking most dinners at home.
4. Walks to work instead of driving.
5. Goes to bed an hour earlier, because they are no longer snacking late.

They also start weighing themselves daily, and the change coincides with a new job and a quieter period at home. Suppose their weight falls noticeably over the following months. What caused it?

Competing explanations for the same result

Each of these could contribute, alone or together, and several have direct evidence behind them.

• Food composition: removing packaged snacks changed what they ate.
• Total intake: the plan may have reduced how much they ate, through appetite, convenience or fewer eating occasions.
• Activity: walking added movement.
• Sleep: this is not a neutral background change. In a randomised trial of 80 adults with overweight who habitually slept under 6.5 hours, a single counselling session to extend sleep added about 1.2 hours a night and reduced energy intake by about 270 kcal a day compared with controls, in ordinary home life [1].
• Attention: a review of 22 studies found self-monitoring consistently associated with weight loss, though the evidence was weak because of design limitations [2]. More broadly, a systematic review of research-participation effects concluded that being observed does change behaviour, though little is known about how much or when [3].
• Regression to the mean: unusually high or low measurements tend to be followed by ones closer to average. People often start a plan at a high point, so some change might have happened anyway [4].
• Early water shifts: glycogen is stored with roughly three to four parts water, so early weight change on a restrictive diet partly reflects glycogen and water rather than fat [5].

A theory about meal timing, one about processed food and one about energy balance can all point to this person’s result as support. When the same evidence confirms every theory, it cannot tell them apart.

When a bundle is taken apart: meal timing

Time-restricted eating is a useful case because it is often promoted on the strength of bundled results. What happens when timing is isolated?

• In the TREAT trial, 116 adults either ate freely between noon and 8 pm or ate three structured meals a day, for 12 weeks. The difference in weight loss between groups was −0.26 kg (95% CI −1.30 to 0.78), not significant, and estimated energy intake did not differ [6].
• In a 12-month trial of 139 adults with obesity, everyone followed a calorie-restricted diet, and half also ate only between 8 am and 4 pm. Weight fell by 8.0 kg with the window and 6.3 kg without; the difference of −1.8 kg (95% CI −4.0 to 0.4) was not significant [7].

The fair reading: in these trials, the timing component added little measurable benefit for weight once other things were held similar. The confidence intervals still allow a modest effect, so this is weak evidence against a large benefit, not proof of none.

A matched comparison can work the other way too. In a ward study of ultra-processed versus unprocessed diets, the researchers matched the menus for presented calories, energy density, macronutrients, sugar, sodium and fibre, so that processing was the main difference. People still ate more and gained weight on the ultra-processed diet [8]. Isolating a variable does not always shrink the effect; it tells you whether the effect belongs to that variable.

How studies untangle bundles

• Factorial designs randomise people to every combination of components switched on or off. The multiphase optimisation strategy uses them to screen components before assembling a final programme [9]. In the Opt-IN study, 562 adults were randomised across five components, including the number of coaching calls, text messaging and meal replacements. Doubling coaching calls from 12 to 24, the most expensive component, did not increase weight loss and was dropped [10].
• Matched comparisons hold one factor constant, such as presented calories, to isolate another, such as processing or timing.
• Dismantling studies remove one component at a time from a package to see which pieces matter.
• Measuring the in-between steps means tracking intake, activity and sleep, so you can see which of them actually changed.

None of these is perfect, but each turns “something in the bundle worked” into a more specific claim.

Why anecdotes add another layer

Personal accounts pass through a filter. People whose bundle worked are more likely to share their story than people whose bundle did not. Across many anecdotes, that selection can make an approach look more reliable than it is, even if each story is true.

If you are running a personal experiment

• Take several baseline measurements over a couple of weeks before changing anything, to blunt regression to the mean.
• Decide in advance what you will change and what you will measure.
• Change one thing at a time where you can, and give it long enough to show an effect before adding another. Expect early weight changes to include water.
• Keep brief notes on other changes in your life, such as sleep, stress, travel or illness.
• Accept the limits: one person’s results describe one person under particular conditions. They can generate a hypothesis; they cannot establish a general treatment.

Confidence assessment

Using the labels defined in what would count as evidence for a metabolic theory:

• A bundle of simultaneous changes cannot, by itself, show which component caused an effect: a point of logic, supported by well-documented artefacts such as regression to the mean [4] and early glycogen shifts [5].
• An eating window adds meaningful weight loss beyond what people eat: weakened in the two trials examined here [6] [7].
• Extending sleep in short sleepers reduces intake: plausible but unproven as a durable effect; one well-measured randomised trial supports it [1], and longer replication is needed.

Speculation, clearly labelled: some components may matter mainly in combination. More sleep might make cooking at home easier, which in turn changes what someone eats. There is now some trial evidence for this kind of interaction: in a factorial experiment with 384 adults, none of five added components improved weight loss on its own, but a combination of three appeared to [11]. That is one trial, and interaction findings need replication. If interactions like this are common, testing one change at a time could miss effects that only appear together, which is exactly why factorial designs exist. Whether any particular popular bundle works this way is an open question.

Before you change several things at once. If you take medicine for diabetes or blood pressure, are pregnant, have a chronic illness, or have a history of disordered eating, talk to your clinician before making significant changes to how or when you eat. Changes to when and how much you eat can affect blood sugar, particularly with glucose-lowering medicines, and a clinician can help you decide what to change first.

For a different kind of translation problem, see what a mouse study can tell us about human weight regulation.