Have I calculated my estimated sample size correctly?

I am conducting a study to answer the following question: In women with type 1 or type 2 diabetes in pregnancy, is there a difference in glycated hemoglobin A1C level between women who have a high score on the Diabetes Self-Management Questionnaire and women who have a low score on the Diabetes Self-Management Questionnaire?

I have used G Power to estimate the sample size.

My main question is regarding effect size. As I have learned, effect size is often based on previous studies in the literature, investigators best guesses, or on a value that is clinically significant.

In this case, I have found studies that have suggested that a difference of 0.5% in glycated hemoglobin A1C is usually considered clinically significant. Therefore, I used 0.5% as my effect size in G Power. Is this appropriate or am I totally off-base?

I have attached a screenshot of the G Power results.

Update: I have found a paper that stated the mean A1C and standard deviation among non-diabetic pregnant women (hopefully I can find one for diabetic pregnant women to make it more accurate). The mean is 5.1 and the SD is 0.51. Could I calculate effect size as 5.1 - 4.6 (5.1 - clinically significant difference of 0.5) divided by 0.51? = .9 approximately?

• "Effect size d" is not in percent. Please see p. 4 and p. 49 of the G Power manual.
– whuber
Jun 5, 2019 at 16:25
• Just use the "Determine" button to the left of the effect size. It will open a window where you should easily be able to insert the values that you found in the literature to calculate the effect size in correct terms. Jun 5, 2019 at 18:43
• Yes I did that. I just wanted to make sure that I was entering the right values in. For Mean Group 1 I had 5.1; For Mean Group 2 I had 4.6 (5.1 - .5, the clinically significant difference) and I used the same standard deviation for both 0.51. Does that sound correct? Jun 5, 2019 at 18:58
• Yup, that sounds correct. But be aware of that a difference of 0.5% means that if one group has mean of 5.1, the other group has mean of 5.0745 (since 0.5% of 5.1 is 0.0255). What you most likely meant is difference of 0.5 percentage points which corresponds to a 9.8% reduction if baseline is 5.1. Jun 6, 2019 at 8:16

Effect size has many possible meanings, as the Wikipedia page shows. For G*Power, it means the following unitless ratio:

The difference of the means between the lowest group and the highest group over the common standard deviation is a measure of effect size.

So your value of 0.5 for effect size is not correct; that's the difference of A1C units between the groups that you wish to detect, the numerator in that ratio. You also need to estimate the denominator, the standard deviation of A1C values within a group of individuals.

This paper shows distributions of A1C values in US populations. Its Table 1 indicates a difference of about 0.6 A1C units between the 25th and 75th percentiles, the inter-quartile range. If A1C values follow a normal distribution, then that is about 1.349 times the standard deviation, suggesting a standard deviation of about 0.44 A1C units.

Then the ratio would be: 0.5 A1C units/0.44 A1C units = 1.1 for effect size.

You should of course verify my quick-glance estimates of A1C distributions and would be better off using instead any data you have on A1C distributions within the particular populations you will be studying.

• That is very helpful thank you! I have a further question - I have found a paper that stated the mean A1C and standard deviation among non-diabetic pregnant women (hopefully I can find one for diabetic pregnant women to make it more accurate). The mean is 5.1 and the SD is 0.51. Could I calculate effect size as 5.1 - 4.6 (5.1 - clinically significant difference of 0.5) divided by 0.51? = .9 approximately? Jun 5, 2019 at 18:28
• @KatherineT. Yes, you divide the difference you want to be able to detect by the standard deviation among the members of a population. So if the difference you want to detect is 0.5 A1C units and the SD is 0.51, you have an effect size (as used by G*Power) of a bit under 1.
– EdM
Jun 5, 2019 at 18:41