The controls you see in the picture can be divided into six different groups:
|
On this page we describe the elements of the G*Power Analysis Window shown in the picture below. |
|
|
The controls you see in the picture can be divided into six different groups:
|
|
You can choose the type of analysis you want to perform by clicking onto one of the buttons you see in the picture. |
|
|
For more information concerning the different types of power analyses supported by G*Power, check:
|
|
|
Use these buttons to choose which type of test you want to perform. The tests supported by G*Power are divided in three categories,
These are the tests used most frequently in behavioral research. |
|
G*Power knows two ways to calculate your power analysis. You select the one you prefer by clicking onto the corresponding button as shown in the picture below. |
|
|
|
Speed mode is fast but inaccurate, accuracy mode is very accurate (up to five significant digits at least) but the price is that you will usually have to wait a little longer until you see a result. More detailed information about the accuracy of G*Power calculations is available in the description of the algorithms used in G*Power. However, we generally recommend that you use the accuracy mode whenever possible. This is so because the speed mode can be quite inaccurate, particularly with extreme values of the parameters. The only reasonable application for a speed mode calculation is when you want a quick overview, for instance when you want plot a graph. But even here you may find the speed mode to be too inaccurate when you get jagged edges in what you know should be a smooth curve. The speed mode calculations are based on approximations which were also used for most of the tables published in Cohen (1977). The accuracy mode calculations use the 'exact' noncentral t, F, and chi-square distributions. However, be aware that even the accuracy mode calculations have their limits. For instance, when you conduct a post-hoc power analysis and the power in the program output is 1.0000, then this may simply reflect a rounding effect (i.e., the 'true' power could be something like .9999999999999) and you should check the total sample size with an a priori power analysis using a power of .9999. If the sample sizes under these two conditions differ, then the critical value as it was specified in the post hoc power analysis must be treated with extreme caution! (Note also that in a priori power analyses the sample size is usually rounded to the next multiple of the number of groups in your design.) |
|
In this part of the Analysis Window you can specify the input values for your test. Type the numbers into the editable text fields and start the calculation. |
|
|
We provide further information about these parameters here:
|
|
|
G*Power always displays Cohen's (1977) effect size conventions for what he calls 'small', 'medium', and 'large effects. The example display shows the conventions for parameter d.
|
|
For your convenience, these effect size conventions appear synchronously with the test you select. The distinction between 'large', 'medium' and 'small' effect sizes is designed to help you specify the kind of effect you expect when you have no other source but your 'intuition'. Cohen (1977, 1988) justifies these levels of effect sizes.
|
|
|
|
Index |
|
|
|
|
|
|
|
d |
|
|
| |
|
|
r |
|
|
| |
|
|
f |
|
|
| |
|
|
f2 |
|
|
| |
|
|
w |
|
|
|
|
In order to prepare the appropriate G*Power input, it may sometimes be necessary to know the relation between the total sample size and the effect size measure on the one hand and the noncentrality parameter of the noncentral distributions on the other hand. We have provided the relation between the sample size, the effect size measure, and the noncentrality parameter on a separate page. |
|
|
The bottom right section of the Analysis Window contains the two buttons that you use to start calculations and to draw graphs that display one of the parameters of your power analysis as a function of another parameter. |
|
Hitting the "Calculate" button starts a power analysis with your input values (provided no necessary information is missing). Hitting the "Draw Graph" brings up the dialog field in which you can specify details of the graph you want to plot. The results of your power analysis will be displayed in the section of the Analysis Window shown below. |
|
|
You may wonder what "actual power" means. Here is the answer: If you perform an a priori power analysis, G*Power calculates the 'exact' sample size for you. Assume that this exact sample size for a t-test is 60.70. Of course, you cannot recruit 60.70 subjects. Therefore, G*Power rounds to the next reasonable integer for your t-test, which would be 62 (two groups of 31 subjects each). |
|
However, 62 is larger than 60.70, and one way to express what this means is to say that with 62 subjects your t-test has more power to detect an effect than it would have given the 'exact' number of 60.70 subjects. This larger power value is displayed as "Actual power". Note that in this way G*Power guarantees that with the sample size computed for an a priori power analysis, the power of your test is always at least the power you specified.
|
|
|
|
|
|
|
|
|
|
|
|
Please report suggestions for improvements to Axel Buchner, Franz Faul, or Edgar Erdfelder. |
