Does the sampling method significant determinant of whether a study requires parametric test or non-parametric test?
Dear Sir. Concerning your issue about the Sampling Method for Parametric Tests. All statistical tests assume that the data captured in the sample are randomly chosen from the population as a whole. Selection bias will obviously affect the validity of the outcome of the analysis.I think the following below links may help you in your analysis:
http://math.tut.fi/~ruohonen/S_1.pdf
http://rcompanion.org/handbook/I_01.html
http://www.creative-wisdom.com/teaching/WBI/parametric_test.shtml
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4667138/
Thanks
Does the sampling method significant determinant of whether a study requires parametric test or non-parametric test?
Beside sampling method, other factors that influence the decision to use parametric or non-parametric tests include sample size, dependent variable data types e.g. ordinal, interval etc. You can refer the following links for further reference:
http://blog.minitab.com/blog/adventures-in-statistics-2/choosing-between-a-nonparametric-test-and-a-parametric-test
http://www.statisticssolutions.com/selecting-between-parametric-and-non-parametric-analyses/
http://users.monash.edu/~smarkham/resources/param.htm
https://www.slideshare.net/gonencdalgic/statistics-40363156
Han, could you specify why and how precisely the sampling method would i/should influence the decision between using parametric/non-parametric tests?
I just peeked into the first link you gave and I got upset reading the same old misconceptions promoted in too many bad stats books... Just one example: "When your distribution is skewed enough, [...] the median continues to more closely reflect the center of the distribution. " -- what the heck?! In skewed distributions there exists not the one definition of a center. And if one looks for a typical value (as the authors also state somewehere), it would be much more sensible, at least in unimodal distributions, to go for the most probable value (aka the mode). That's just one example, the mess goes on.
In the second link, you finde "Fortunately, the most frequently used parametric analyses have non-parametric counterparts." - another shot against helping to understand statistics. There "counterparts" do address entirely different questions. Again, just one example.
All links agree that "If your measurement scale is nominal or ordinal then you use non-parametric statistics". How this? The binomial test assumes a binomial distribution, the parameter is p - so it's a parametric test; the chi²-test is based on the chi²-distribution of the test statistic, that is based on the assumption of the normal distribution of the cell frequencies (why should one assertain a minimum cell count of 5, otherwise?).
However, I didn't find anything relevant to the OP's question...
I raised this question because recently I heard that multiple linear regression can’t be performed with convenient sampling. However, there are well reputed papers employing convenient sampling for multiple linear regression by addressing the limitation of generalizability.
Convenience sampling limits the generalizability of the data. If this is clearly stated and taken care of, there is no big problem. This is not related to what kind of analysis I do with the data. The data has a possibly large risk of not being representative for what I want actually want to investigate. This is the problem. It has nothing to do with parametric or non-parametric tests, linear regressions or rank-correlations. Garbage in -> garbage out.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size