Nonparametric statistics v.s Parametric statistics

[katlego]

Why nonparametric statistics is less preferred compared to parametric statistics even if it requires fewer assumptions?

 

[Dason]

Typically the corresponding parametric test has higher power than the non-parametric test (as long as the assumptions for the parametric test are met).

 

[Jake]

The answer is a little different depending on exactly which parametric vs. nonparametric approach we're talking about, but to make some broad generalizations, nonparametric approaches can often have drawbacks due to (a) reduced power as a consequence of information loss (e.g., nonparametric statistics that are based on rank transforming the data), and/or (b) they are sometimes computationally expensive (e.g., nonparametric bootstrap). They can be useful, but they are emphatically not a cure-all, and they must be deployed wisely. See my signature...

 

[Dason]

Jake said it better than I did. There's also the fact that parametric statistics is what is typically taught in intro stats courses so it's what most people end up using...

 

[Jake]

Heh, yes, let's definitely not forget that factor...

 

[katlego]

Hi, do you mind if you explain your signature to me?

 

[Jake]

That was just a slight joke. My signature just means that we need to think hard about what analyses would be appropriate or inappropriate to solve the specific problem that we have at hand--before we sit down to conduct the actual analysis. It may seem like a trivial point, but you will find that people often skip the first part and just sit down to mindlessly run their ANOVA program or whatever, without ever thinking too hard about what it all means.

 

[katlego]

Tragedy of the commons! I presume they just examine whether something has a perception of being right, then the rest is history.