If the data is not following Normal Distribution then the researcher has to proceed with Non Parametric Tests.
Hello, I strongly recommend to you to enter in the site www.edx.org, where you find a lot of interesting courses, for free and imparted for some of the most prestigious universities, there you can find one or two courses of statistics and epidemiology.
The thing is, that if your sample is not following a normal distribution, you still can use some of the parametric test given that you have a large number and some other assumptions, you will find the answer in this courses, I hope.
"If the data is not following Normal Distribution then the researcher has to proceed with Non Parametric Tests." - This is simply not true in several respects.
A log-transformation is appropriate if you assume a multiplicative error model.
http://rds.epi-ucsf.org/ticr/syllabus/courses/25/2009/04/21/Lecture/readings/log.pdf
Hello, I strongly recommend to you to enter in the site www.edx.org, where you find a lot of interesting courses, for free and imparted for some of the most prestigious universities, there you can find one or two courses of statistics and epidemiology.
The thing is, that if your sample is not following a normal distribution, you still can use some of the parametric test given that you have a large number and some other assumptions, you will find the answer in this courses, I hope.
This may seem a terrible heresy, but I find that Wikipedia often offers very well-written descriptions of simple statistical tests.
As far I know, the "transformations" are extremely problematic. I don't dare to call them non-parametric, because they allow you to use parametric methods. In non-normal distributions, non-parametric tests are more useful, because they have long tails with extreme values: parametric tests compare means of the distributions and non-parametric tests don't. The mean is too sensitive to extreme values than even transformed data may suffer, since they don't reflect what is happening in the original data.
Hello Mr Sathian, if you go to my website and click on favourite links there is an electronic textbook. http://sites.google.com/site/deborahhilton/ there is a good chapter in there on non parametric tests. Cheers, Debbie
Consider that modern statistics includes generalized linear modelsfor non-normal data such as Poisson, binomial, and other exponential families, so no guarantee that nonparametric is really the right direction.
Finally, consider that bootstrap. and randomization methods are also now strong competitors to classical nonparametric methods. Particularly for problems in which inference is needed for complex functions of model parameters.
For classical nonparametrics, I would recommend Conover's book: Practical Nonparametric Statistics.
http://www.amazon.com/Practical-Nonparametric-Statistics-Wiley-Probability/dp/0471160687
Hi Brijesh, I agree with John, in that many modern techniques address problems that were solved with nonparametrics. Still some nonparametric tests are handy and are still in pretty current usage, but I would suggest you consider the modern approaches first. Also, in small samples exact tests are often used because the asymptotic assumptions of common nonparametics often given faulty p-values (exact tests would be used more, but many people don't have the software to run them). As far as resources, with similar apologies, Googling often gives good leads. I haven't checked spefically for resources on nonparametrics, but UCLA often has good statistical infromation, so look for that when you Google. The Survey Research Center at the University of Michigan used to have a nice guide to choosing the right statistical test. I understand it is now incorporated into free statistical software called MicrOsiris you can dowload from there. Although the guide is probably out of date, the nonparametrics themselves haven't really changed much. Bob
if data are not approximately normal behavior, the first is to analyze these as they often present data outliers, therefore, explore about the quality of the data, find variables that explain the behavior. If everything is clear, you can still transform before applying nonparametric tests
Hi Brijesh, as Conover points out in his book "Practical Nonparametric Statistics" the simplest way to employ nonparametric statistics is to transform the data to ranks and then perform ordinary parametric tests on the rank-transformed data. This method immediately tames the outliers and provides conservative inference, as long as there are not too many ties.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics