Help - problem

#1
Hi

I have a variable VA and want to know if there are significant differences in this variable between group A and group B.

N = 93 samples

Situation 1:
In Group A the n = 53 in Group B n = 40 (these groups were defined considering that the group A have the treatment X and the group B don’t have the treatment DX).



Situation 2:
In Group A the n = 76 in Group B n = 17 (these groups were defined considering that the group A was submitted to surgery S and the group B don’t was submitted to surgery DS).




N = 17 samples (subgroups of situation 2)

Situation 3:
In Group C the n = 11 in Group D n = 6 (these groups were defined considering that the group C is (X + S) and the group D (DX + S).


Situation 4:
In Group C the n = 42 in Group D n = 34 (these groups were defined considering that the group C is (X + DS) and the group D (DX + DS).


Questions?

- How to verify that the number of individuals is sufficient to verify the statistical significance?

- Is it possible to justify that these are solid, that is, allowing a sound statistical interpretation / reasonable despite the n presented?

I cannot change the n because the samples are difficult to obtain and very time consuming. Furthermore many of the studies present an overall sample less than 93.

best regards
 

CB

Super Moderator
#2
Try this statistical power calculator:

http://www.dssresearch.com/toolkit/spcalc/power.asp

You will need to provide estimates of the group differences and population standard deviation. If you already have the data for the sample you can use the figures you've found to calculate the "observed power".

You'll probably find that in situations 1 and 4 you will have borderline-sufficient or insufficient power (depending on the differences and SD), while in situations 2 and 3 it is very unlikely you will have sufficient power to detect statistically significant differences.

This is coherent with the practical situation, because for some of these comparisons you just don't have enough participants to be making inferences about population parameters (the point of "statistical significance", really).

One way to increase your power would be to use one-tailed testing - i.e. pre-specify the direction of the expected difference (e.g. hypothesising that treatment group will have have a lower score than the non-treatment group, rather than just a different score). This may be looked upon sceptically by reviewers though.

Is it possible to justify that these are solid, that is, allowing a sound statistical interpretation / reasonable despite the n presented?
Your descriptive statistics will be "solid" for the sample itself - an accurate reflection of the variables for this particular group of people (as long as your measurements are reliable). They probably WON'T be a solid basis for making inferences about the general population though - you just don't have enough people (at least in situations 2 and 3). Descriptive statistics can be useful though - they can provide suggestions for future research, and your results could be incorporated into important meta-analyses by future reviewers (where the "statistical significance" of your results won't matter).