Possibility of errors in parametric and non-parametric tests

Although parametric tests have historically been seen as superior instruments in statistics procedures for evidence-based decision-making, computer simulations have showed that they, like non-parametric calculations, are equally susceptible to type 1 and type 2 errors. The observation is informed by the use of dummy data in the comparative analysis, where employing Monte Carlo techniques under various assumptions that guide conventional calculations reveal that parametric tools such as the t-test can yield incorrect results (Schmider, Ziegler, Danay, Beyer, & Bühner, 2010). It fails to reject the genuine null hypothesis if the data utilized is insufficiently large. The same trend is also reported in in the case of non-parametric tests such as Wilcoxon rank sum test. While parametric tests are more robust in normal distributions when compared non-parametric ones, the robustness that has historically been assigned to t-tests, f-tests, and z-tests was found to be minimal to warrant discriminate use.

Experience with the tests

Being a conventional and the statistical technique of choice in most of the studies surrounding human subjects, I have experience in dealing with parametric tests. Although I do not have vast knowledge in non-parametric tests, comparatively I prefer parametric procedures because of practicability and workability. It does not require voluminous data and tiresome raking to give information that is more applicable to the population. It is also based on pre-existing knowledge, making it easy to detect missteps.

Comfortability in Using Parametric Tests without Professional Help

With Monte Carlo simulations having highlighted those conventional tests can be a source of both type 1 and type 2 errors like other non-parametric tests, I would not be comfortable undertaking the statistical procedures alone. Instead, I would seek professional assistance to be helped on crucial things that make the tests robust as the case of sufficient sample size. The decision to look for an expert is also informed by Hothorn, Bretz & Westfall (2008) argument that while it holds a slight advantage over non-parametric tests, undertaking multiple parametric procedures compromises the ability of the statistical tools in making correct conclusion.

References

Hothorn, T., Bretz, F., & Westfall, P. (2008). Simultaneous inference in general parametric models. Biometrical Journal, 50(3), 346-363.

Schmider, E., Ziegler, M., Danay, E., Beyer, L., & Bühner, M. (2010). Is it really robust?. Methodology.