An introduction to power and sample size estimation (2003). Jones, Carley, Harrison

R Jones, S Carley, M Harrison. An introduction to power and sample size estimation. Emerg Med J2003;20:453–458

The importance of power and sample size estimation for study design and analysis.

OBJECTIVES
1. Understand power and sample size estimation.
2 Understand why power is an important part of both study design and analysis.
3 Understand the differences between sample size calculations in comparative and diagnostic studies.
4 Learn how to perform a sample size calculation.
– (a) For continuous data
– (b) For non-continuous data
– (c) For diagnostic tests

Free full text (link to pdf): https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1726174/

 

One thing I note here as I read though, is an assumption that seems common in clinical trials.

This excerpt I fully understand ...

[my bold]


[my bold]

Why is it always presumed that any difference will be positive, i.e beneficial? To me this is has massive potential for bias, if negative differences get ignored.

Unless of course I am misinterpreting something here, and that 'benefit' is deemed itself to be a signed parameter, and can adopt both positive and negative values? A negative benefit thereby being a deterioration?

 

In my efforts to better understand, also found this:

https://stats.idre.ucla.edu/other/m...nces-between-one-tailed-and-two-tailed-tests/

Still trying to get an intuitive grasp of why, for a one-tailed test, it is OK to 'steal' the other tail's 2.5% to add to the power on your favoured side, when the values are still occurring on the other side regardless of whether you are interested in them or not. Feels like having your cake and eating it.

ETA: Answering my own question. I guess if it really is valid (for your particular experiment) to disregard values one side of the mean, then you genuinely do only need consider the probability of chance values the other side of the mean. What would be an issue is if you did a one sided analysis for an experiment that requires two-sided, as the article emphasises. I note PACE did do two sided.

ETA2: I should make clear that stats is a very weak subject for me, and is why I'm trying to make the effort to understand a bit better. So please do not take my words as gospel on this, as they most certainly are not! There are many people here in S4ME who are very competent statisticians, but I'm not one of them .

 

My browser doesn't like that link, but thanks for explaining what you read

 

Very odd! It's actually well worth reading.

Can you get to https://stats.idre.ucla.edu/ and then browse through ...

Resources -> Frequently Asked Questions -> General FAQs -> Other statistical questions -> What are the differences between one-tailed and two-tailed tests?

 

There are cases where mortality was ignored. For example, antiarrythmia drugs. These were touted as successful as they treated the symptom well. Long after, someone did mortality stats ... overall mortality, that is death, was increased. They worked, but you died more. This was one of the case examples used to promote evidence based medicine by some. Total mortality should be considered in every study. In the case of ME, objective functional capacity must also be considered. We might be less functional with treatments that "work".

 

unfortunately not.

Could be an issue with certificate standards?