I have a data series of results from an experiment ranging from a minimum negative value to a maximum positive value. The t-test done with the data as it is doesn't correspond to the trends I expected.
The t-test does not assume that the data is non-negative. However, if the t-test is a good choice for your data cannot be answered without understanding what the data is and how it looks like. Given this is ok, the unepected result is a rather good thing, because it tries to tell you something new! Btw - what "trend" did you expect and what is the result of the test?
The T-test has assumption for normal distributed data, so you need to test your data for normality, in case of non-normal distributed data you can use transformation or test your data with nonparametric test.
Good Luck
Khalid: "The T-test has assumption for normal distributed data, so you need to test your data for normality" - from what, please, does this follow?
There is no need to *test* the data for normality (apart from that, a test does not provide any useful information here). The assumption shoudl be reasonable, for sure. But real data will never follow the ideal theoretical distribution model. So fainig to "see" that the data is non-normal is only a sign that the sample size wasn't large enough.
The t-test assumes normal distributed data. This is right. And *if* the data *would be* normal distributed, then the t-test *would be* an exact test. For any real data, the test is neccesarily approximate. This is ok, as long the approximation is good enough. If one does not know if the assumption is reasonable, one must only check if the approximation is good enough, and this depends also on the sample size. Note that the assumption is less imprtants with increasing sample size - but at the same time the test for normality becomes more sensitive. I wonder why so many people seem to ignore or to miss this.
After all, "statistics" is not using tests. Statistics is using the brain.
Dear Jochen: Thank you for your comments, you wrote a valuable points deals with flexibility of decisions in statistics and that is the role of brain, I think most textbooks for different authors used by students are mention to assumptions for parametric tests and that nonparametric tests can be use instead of parametric tests in absence of normal distribution.
Are textbooks need to be change?
I think there is no idea in the live will be acceptable for all people, and that is the diversity of thoughts among scientists.
Best Regards
Khalid, yes, I think ist great that we do have different ideas and different opinions. And it is up to us to provide evidence that our ideas and opinions are good and useful. As all evidence must be weighted, there is always plenty of room for controversies. Done in a right way, this can all be a very constructive process.
My opinion is that we really should start to change the textbooks. My evidence is that we see generations of scientists bein utterly confused in statistics. Even many statisticians are. So I think history has demonstrated that the current practice of teaching stats is "not ideal". I don't know if it is still the "best we can do", or if there is a better way. I think there must be a better way, but I have not (yet?) found it, unfortunately.
It's interesting how 6 different people have an opinion but none of them understands the question or knows what he is talking about.
- Badges
- Science method