I did some lab experiments and because of huge variance I needed to log and square root transform the data to be able to proceed.
I'm now planning to submit for publication but I can't recall seeing anyone presenting log transformed results. This may be because they haven't transformed the data, but I don't know for certain.
I'd convert (un-transform) as it provides more sensible numbers, but is that what others would do?
Gareth -
If you collected data for which you found a high variance, then you can reduce the standard error of a mean (but not the population standard deviation) by increasing the sample size. But if you transform the data as it seems to me that you indicated, you are just looking at a different data distribution, which does not sound like it has anything to do with your study.
Transformations often just make it harder to interpret your results. So I suggest that you be sure any transformations you might use serve a subject matter related purpose. If the original data distribution is what is of interest, I don't think that a transformation helps anything, if I understand your question.
Cheers -
Jim
Dear Garret,
In the first instance, it is not clear why you applied the transformation using square roots or logarithms, I suppose you wanted to apply an ANOVA and because of the high dispersion the data did not reach the assumptions of that method; If that is so, personally when I see the need to make some transformation on the data, I usually present in the publication the transformed values and next to parentheses the values without transforming. To illustrate what I mean, I attach a figure from one of my published works in which the means of the transformed data are presented in logarithms and next to, in parentheses, the means of the untransformed data. Please see Fig 1, page 747. The article is written in Spanish, I hope that is not a problem.
if you dont use transformation, then you dont need to think about this issue
Article TO DETERMINE SKEWNESS, MEAN AND DEVIATION WITH A NEW APPROAC...
I agree with Luis. There is nothing abnormal about including both the untransformed and transformed data. For example a dependent variable may not have linear relationship with an independent variable, but may be linearly related to the logarithm of it. This has to clearly stated in the report. This is because the original relationship may polynomial and not linear in nature.
Transformations are usually for purposes of clarity on what exactly is related to what. So they are not purposes.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics