DISCLAIMER: I have limited knowledge of G*Power. But AFAICT, it uses Cohen's t-shirt size (S, M, L) effect size estimates as input. In this short document, Russ Lenth explains why he does not recommend doing that:

• https://homepage.stat.uiowa.edu/~rlenth/Power/2badHabits.pdf

See also this opinion piece describing some other issues with G*Power:

• https://pubmed.ncbi.nlm.nih.gov/31954629/

Regarding your request for alternatives to G*Power, are you looking for free software only, or are you open to commercial products?

What statistical software do you use currently?

Finally, have you thought about using simulation for power analysis?

Thanks for clarifying.

Thank you for your quick reply and the references!

I am looking for free software as I would like to extend gained knowledge to university students that could benefit from proper sample size estimations.

I have thought about bootstrapping/simulation but was hoping for a less time consuming method.

I am mainly using JASP, MATLAB and SPSS if JASP doesn't do it. I am not very familiar with R but open to learning it.

Something I am currently looking into is 'Superpower"

https://aaroncaldwell.us/SuperpowerBook/appendix-3-comparison-to-brysbaert.html#repeated-measures

I just came across this article as well. It should be noted that my lab moved away from GPower now. :D

Article Simulation-Based Power Analysis for Factorial Analysis of Va...

I am not sure I understand your comparison with R. Do you mean base R (which only has a few functions for power analysis) or any of the packages that exist (or even what you could produce with the simulation packages)?. Some R packages (e.g., pwr) are designed in a similar way to GPower, so if you set the options you get the same values, and can do a few loops or apply statements and get similar plots (though with more flexibility that comes with R). Other packages do simulation, which is more flexible, and if you have specific power questions and a good idea of what the data will be like you can write your own. Others allow different things. So GPower is good if you are willing to make those assumptions, but you may still want to use the similar R packages if you want to use the output, for example, to produce plots the ones it produces may not look like what you would want to publish. But if not willing to make those assumptions, or needing power for tests not in GPower, then searching through R packages would likely be prefered.

ps. On reliable, with the non-simulation methods you should get the exact same thing each time you press the buttons, but with the simulation approaches (assuming different seeds) you'll get different answers. Is this what you mean by "reliability"? This obviously doesn't mean one is better than the other.

Bruce Weaver I believe it should be mentioned that while it is true that G*Power has certain default values, it does not force you to use those defaults or any of Cohen's effect size guidelines. You can enter whichever effect size makes the most sense for your power analysis.

Hi Christian Geiser. I could have been clearer. Using an unpaired t-test as the context, I believe Russ Lenth's point was that just plugging in a Small, Medium, or Large value of Cohen's d without giving any thought to what the expected pattern of means and SDs might be is ill-advised. I doubt very much that he would object to using G*Power if one first thought carefully about what pattern of means and SDs is expected (and reasonable) if some minimally important effect were to occur, and then converted that pattern into a value of d. But IMO, that is very different than mindlessly plugging in a "medium" value of d.

More generally, anything that absolves researchers of the responsibility of thinking carefully about things and understanding the variables under study is suspect.

For the benefit of those who have not had time to look at the Correll et al. article, another "feature" of G*Power is that the estimated sample size for two statistically identical analyses can be quite different. Here are a couple of paragraphs discussing that problem in the context of comparing the means of two independent groups.

--- Start of excerpt ---

Definitions Are Inconsistent

As we begin this section, is important to note that t-tests, simple regression/correlation, ANOVA, ANCOVA, and multiple regression are special cases of the general linear model. The results of an analysis do not differ if we conduct an ‘independent samples t-test’ (assuming equal variance) to compare group A with group B, or a ‘regression’ in which group A is coded –1 and group B is coded +1. The tests differ in name only: they are fundamentally equivalent because they employ the same underlying model comparisons [12,13]. Similarly, if we want to examine differences between groups A, B, and C, while controlling for some continuous predictor, we can use ‘ANCOVA’ or ‘multiple regression’. The analyses are interchangeable.

Despite this equivalence, when we consider the question of power analyses for t-test, ANOVA, regression, and multiple regression, strange inconsistencies emerge. These inconsistencies can be seen both in Cohen’s writings [3,4] and in software programs, like G*Power [14]. For example, imagine two researchers who want to test whether 8-year-olds and 10-year-olds differ in emotional intelligence. Before collecting data, each consults Cohen’s ‘power primer’ [4]. The first researcher conceptualizes the study in terms of an independent samples t-test; the second conceptualizes it in terms of a regression in which emotional intelligence is predicted from contrast-coded age group. Otherwise, the two researchers make identical choices: alpha = 0.05, two tailed, a ‘medium’ effect (d = 0.50, r = 0.30, respectively), and required power = 0.80. The first researcher discovers that the total required sample size is n = 128. The second researcher finds that the required n = 82.

--- End of excerpt ---

Table 1 in the article shows other examples (see attached).

I suggest Stata and PASS

Seyyed Amir Yasin Ahmadi if I had the means to purchase these products, I would definitely want to go that route! Thank you for the recommendations!

Martina Speck If you want an open source, you can use R. You can find each specific package in google. The R packages for power analysis are easy and you can reach your purpose with a one line syntax!

https://stats.stackexchange.com/questions/600397/power-analysis-for-two-way-repeated-measures-anova

Yes, there are alternatives to GPower. Here are some of the best alternatives and competitors to GPower according to G2 community reviews: IBM SPSS Statistics, Posit, Minitab Statistical Software, Stata, JMP, eviews, Grapher and OriginPro.

For repeated measures, there is also the freeware MorePower https://link.springer.com/article/10.3758/s13428-012-0186-0