GOODNESS OF FIT TEST using F test may be helpful. In the two group. Let n1 > be. Find the variance for both groups, var1 and var2, then the Fvtest is given by:

F = var1 / var2

See the F* value by reading in the F table by degree of freedom of both groups. F* > F not significant and F* < F significant difference.

I do not see a problema as long as groups are large enough, as you compare summary statistics. In purely experimental settings, particularly, in Medicine, nowadays, it is recommended to have similar sizes in control and treatment groups, but even there, we can see trials using a different number of people in placebo and control groups.

Hi Tewodros! This is a very general question for a particular problem. The first step is to explore your data. If the group variances are equal and the sample size is unequal, there is no problem. But one must consider how unbalanced the groups are. Another important element is that statistical software assumes that the sample size is balanced; therefore, the estimators may not be reliable.

The most reliable methodology, and good practice research, is to specify an unbalanced design, i.e., power and sample-size analysis for hypothesis tests in an unbalanced design.

1. Specifies the sample-size ratio of the experimental group relative to the control group, Ncontrol/Ntreatment, where Ncontrol=sample size of the control group and Ntreatment=sample size of the experimental group. The sample-size ratio is necessary for power or effect-size determination for two-sample tests. You can see https://www.stata.com/manuals/pss.pdf for consult how to analyze the power and sample-size.

2. The recommended methods to estimate the treatment effect with an unbalanced group size are inverse-probability-weighted (AIPW) and inverse-probability-weighted regression-adjustment (IPWRA). Although these methods are not specific for unbalanced samples, can correct the missing data; therefore, can face differences between sample size in groups. You can see https://www.stata.com/manuals/te.pdf (p.1 onwards) for details.

I hope this information helps you. Regards!

Let's connect! www.linkedin.com/in/maría-josé-ibáñez

The term "unequal" ranges from 51/49 to 99/1. So, it is a very general term. Anyway, equal proportion is more preferable as the models assume that. Using 60/40 is fair, but using 90/10 will limit your sample size to the level below the rule of thumb.

Tidak ada persoalan selama, secara statistik, tidak ada perbedaan yang signifikan antara kelompok treatment dan kelompok control pada pra-estimasi. Yudhi Dharma Nauly