For everyone, the health aspect of interest is measured and the groups with and without the specific diet are compared. An alternative is to conduct an experiment: a number of people are asked if they want to participate.

Then the experimenter decides who should the diet use during the following month and who not, often using “randomization” (for example by tossing a coin). After the month “treatment” with diet or without, the health aspect of interest is measured and the groups with and without the specific diet are compared.

Makes it easier to draw conclusions

If randomization is used in an experiment, it is easier to draw causal conclusions compared to an observational study. We expect that the randomization helps us to ensure that the group of people using diet and the group not using it are comparable (as everyone has the same chance to receive the diet).

So if we can prove a difference between the groups with regard to the health aspect, we can then conclude that the difference is most likely due to the diet.

In contrast, if we let everyone decide themselves to use the diet or not (in the observational study), then it might be that people who are in better health choose to use the diet.

In this case, if we see a difference between diet group and the non-diet group with regard to the health aspect, this could have other reasons than the use of the diet.