Calculating sample size for a 4-arm study

[Marianne Lionny]

Hello!
I was hoping that someone could try to help me understand how I could calculate a sample size for a 4 arm study.
This would entail testing the effectiveness of two devices and have the following groups:
Group 1 - both devices;
Group 2 - device A,
Group 3 - device B,
Group 4 - no devices.
My assumptions would be that the effect would range from 40% of success in group 4, through 50% success in groups 2 and 3, and 55% in group 4.
Please could you help me understand how I can calculate the sample size for this.
Thank you so much,
M

[Rich Goldstein]

within Stata, see "help power twoway"; note, however, that it also sounds like a factorial design so you might want to see a specialist text such as Ryan, TP (2013), Sample size determination and power, Wiley, which has a special chapter on traditional experimental designs

[Joseph Coveney]

. . . the effect would range from 40% of success in group 4, through 50% success in groups 2 and 3, and 55% in group [1].

within Stata, see "help power twoway" . . .

Rich, wouldn't that be for a linear model (ANOVA)?

The way I read Marianne's description, it sounds more like she's comparing proportions, which would call for power twoproportions if she intends to do a set of pairwise comparisons using, for example, Pearson's chi-squared test.

Otherwise, if her choice of analysis method is, for example, fitting a logistic regression model, then it could be more a job for simulation.

She hasn't stated what her primary comparison of interest is, but in the absence of that, she could plan to power the study for the worst-case comparisons, which would be Groups 2 and 3 versus Group 1, making adjustments for multiple comparisons as desired.

[Rich Goldstein]

Joseph Coveney - thanks, yes, I may have mis-read #1, but it is not clearly written anyway and I should have said that in my original response