# Answer to sample size calculation is n1 = 2 + n2 = 2, can that be correct?

I have computed a (online) sample size calculation for mean difference between two groups. The calculation was done with the following data from literature:

The website used for the sample size calculation is: Sample Size Calculator (clincalc.com). This gave the following calculation with the data provided by literature The results of the calculation was n1 = 2, n2 = 2. Total sample size needed = 4. Im wondering if this can be correct? What should I do different?

Our own dataset consists of only two variables: a) retention measured in Newton for conventional dentures and b) retention measured in Newton for digital dentures. With these variables we would like to conduct a t-test to test if there are significant differences in retention between the two types of dentures. A pilot was conducted with four observations in total, maybe that helps (?)

The purpose of this study is for educational purposes. If there is information missing, please let me know!

• That sounds unlikely. What is the actual analysis model ? Also please tell us about all the variables in the model, and what assumptions you have made in the power/sample size calculation. A summary of your dataset could be useful too. You can edit the question rather than replying to this comment. Nov 18, 2023 at 12:00
• Power should never be computed based on observed differences. Power is based on the difference you would be embarrassed to miss. Nov 18, 2023 at 12:37
• It is not going to take much to show Group II has a higher mean than Group I. But I suspect the interesting question is the absolute or relative change "since insertion" and whether that differs between the two groups Nov 18, 2023 at 15:10

Given this huge difference in means relative to the standard deviations, that does seem possible.

I couldn't find an R power calculator for a t-test with unequal variances, but I ran a program with the smaller of the two SDs:

power.t.test(n = NULL, delta = 15.99-8.77, sd = 1.61, sig.level = 0.05,
power = .8,
type = "two.sample",
alternative = "two.sided",
strict = FALSE)


and got a required N of 2.24 per group. With your larger SD I got a required sample of 3.89 per group.

But, if your data is similar to what you posted (i.e. three time points) then I doubt that this is the analysis you want to run.