I used to think that if the sample size is small and we assume nonnormality of the outcome variable, then we use a nonparametric test instead of an ANOVA. But non-normality also seems to lead to the necessity of using generalised linear models in other cases. Could you explain to me the difference in situations in which either would become necessary to use ?
Is a generalised linear model simply the equivalent of the nonparametric test, but for regression instead of ANOVA ?
There are two ways of doing a "nonparametric" analysis: analyzing the ranks, and using resampling or bootstrap. Both can have serious disadvantages: With a rank-based analysis you do not test hypotheses about expectations ("population means", if you like) and do not allow to address interactions or non-linear relationships. Resampling/bootstrap methods are not always feasable, due to small sample size or model structure.
You are usually assuming a distribution with a GLM, just often not a normal distribution. See any review of GLMs.
Forget non-normality problems, theuse of non parametric test is mandatory if (and only if) you have an outlirt and/ot you do not trust so much on the actual values of your measurment and you prefer to downsize the detail to greater/lower. See attached files for these points...
Generalized linear models (GLMs) are used when the outcome variable is not normally distributed or when the relationship between the outcome variable and the predictor variables is not linear. Nonparametric tests are used when the assumptions of parametric tests are not met. GLMs are more flexible than nonparametric tests because they can handle a wider range of data types and distributions. GLMs can also be used to model relationships between variables that are not linear.
- Badges
- Science topic
- Similar topics
- Agricultural Science
- Agronomy