Can any researcher calculate the level of significance through percentage scores? How much is this authentic? What is the technique to calculate it?
Hello,
I`m afraid you will have to provide more information for me (or anyone else) to help you. But no, the answer to your question is a sound no, from the information you provided. I`ll be happy to help if you provide more details.
Best,
The "p-value" from null hypothesis tests *is* a percentage. p*100% is the expected percentage of experiments showing an effect of the observed size (or a stronger) in repeated experiments under the assumption that "true" effect is in fact zero.
Hello, what do you want to compare? Mean or variance between both groups. For testing mean you can use a student´s t-test (one or two side) and for variance a F-test. I think it does not matter the form of the data you have in percentage or fraction, they are both numbers.
Roland, when the values are close to 0% or 100% then the distributions are not symmetric. It would be more appropriate to compare the parameters of beta-distributions rather than those of normal distributions (see http://www.ime.usp.br/~sferrari/beta.pdf).
Jochen, I agree with your last statement. I think the question need to be more defined to be able to give a clear and correct answer
You can try to fix the problem of distribution (see Jochen Wilhelm response) by a classical mathematical transormation used with accuracy scores converting data in the arcsine of the root square (Zubin, 1935).
It is very obvious as pointed out by scholars, your question is not clear to give a correct answer. However, supposing from the main query you want to compare to sets of data (again supposing giving information on a single variable in percentages) either you may use Analysis of variance or preferably that of means to determine whether the two sets (samples) of data are taken from the same population or not. If you want to know how much they vary differently , you may calculate simple statistics of coefficient of variation for the two sets (sample) and may decide which one has more variable values (percentages). These answers are based only on guess, if you state your problem very cleanly, answer may be unequivocal.
Both variance and % should be used. Not use % alone. Depending on your sample size and groups Duncan multiple comparison or Tukey tests are helpfull tools for comparision of the means at 0.05 probability level. When you use either of tests, if possible you can also give another statistics underlying of the frequency (%) of your tests. Otherwise, the term, namely "significantly different "cannot be figured out. I guess. See the attachment and the Tables in it. Since your question is not clear, I've understood this way only.
As previously stated, more details are needed in order to properly answer you question. With this in mind, and as stated by Luis, I think you should be careful when evaluating statistical significance in data set expressed as %. My feeling is that in your case a G-test or a Chi2 test would answer your question in a safe way. Buhara just touched a bit on those (comment above). Let me know if you need help with those test.
God luck!
As mentioned, significance level is a percentage. To compare percentage data, you can transform it (e.g. using arcsine or square root transformation) and use non-parametric statistics if the data does not fit a normal distribution :)
P value is a percentage. So, to compare data, you should transform it (using square root transformation or another methods).
If P-values were percentages, there was no need to call something "probability".
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics