- July 5, 2020
Unlikeparametric tests, non-parametric statistics do not work under theassumption that a data follow a particular distribution (Siegal &Castellan, 2011). Therefore, these tests can be used when the datadoes not follow the assumption made in parametric of normaldistribution i.e. when distribution of one variable is not normal.Non-parametric statistics can also be used when there is an analysisof dichotomous response. This is used to test whether the proportionof replies in a data are different from the reference data.Additionally, Non-parametric statistics is applied when analyzingfrequencies (Rosner, 2012). This kind of analysis involvesexamination of the differences between the expected number ofoccurrence of an outcome and the observed number of the outcome. Forsubtle samples, it is advisable to use non-parametric tests in placeof parametric test. This is primarily due to the assumptions centrallimit theorem that is majorly applicable to parametric tests
Non-parametricstatistics have various demerits over parametric tests. They includeone the non-parametric statistics are less powerful as compared toparametric when the data is normal (Weiss 2016). The researcher willrequire an enormous amount of data so as to get the same result as aparametric test. Non-parametric tests usually use ranks. Thus if youhave data like 40, 50, 60, and 70, this would be replaced by ranks 1,2, 3 and 4. This positioning means that the original information onmagnitude is lost in the process. Additionally, non-parametricprocedures only focus on hypothesis testing and do not give valuableinformation about the data set such as mean, variances, and thestandard errors.
Rosner,B. (2012). Fundamentalsof biostatistics.California Duxbury Press.
Siegal,S. & Castellan, N. J. (2011).Nonparametric statistics for the behavior sciences.New York: McGraw Hill
Weiss,N. A. (2016). Introductory Statistics (10th Ed.). New York, NY:Pearson Education.