I'm testing whether sugar consumption is making my stomach upset, with an unknown lag.

My basic plan is:

  • Get 7 sugarfree and 7 regular, or 'sugarfull' Grape Powerades
  • Label 14 paper bags with consecutive dates, and have someone else secretly put them into the bags in consecutive pairs
    • So day 1 and day 2 will be the same kind of drink, then day 3 and day 4 might be the same
  • Drink each on its date, and record how I feel throughout each day
  • Join the data to have the following columns: date, sugarfull_1day_ago (binary), sugarfull_2day_ago (binary), sugarfull_1day_ago*sugarfull_2day_ago (binary), outcome (avg of ~5 0-5 ratings)

Then I'd do a standard linear regression. Here are the potential issues that I see:

  • Outcome is subjective, and a '2' doesn't necessarily mean 2x a '1' (ordinal variable)
  • I'm not sure that I'm accounting for 'spillover effects' properly, particularly if the effect lasts 3 days, or is cumulative
  • 14 samples is not very many

What do people think, are there any tweaks that could make this experimental plan more rigorous? I have not started yet, so I'm open to any suggestions.


1 Answer 1


One point of random allocation is to minimise the chances of guessing which is coming next. I do not think your design of randomising in pairs accomplishes that since if you guess one on an odd numbered day then you know the next one on the even day. It would be better to do a simple randomisation.

I say you are worried about cross-over (as it is called in the clinical trials literature) so I would suggest having a longer period between trials. This will make your study last longer admittedly but it will give cleaner results.

You may need to search this site for ordered categorical regression which looks more appropriate for your ratings.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.