statistics question - para. vs non-para.

[nonstoptaxi]

Hey everyone,

I have a controversial question about the use of statistics in our field. The 3 parametric criteria (i.e., interval level of measurement of above, homogeneity of variance, normal distribution) - how strict do people adhere to these? As you might have noticed, there is a much greater number of studies (certainly in I/O psychology) that just stick to parametric stats, and that puzzles me.

 

I've read that parametric tests are quite 'robust' in that there can be violations in the data in terms of the 3 criteria. If this is the case, does it matter if one chooses to use parametric even if the data violates the criteria? If a non-para test produces a significant results, and it's para equivalent does too, should i just report the latter?

 

Obviously, those from the old school would argue for the strict adherence to the 3 criteria, but i'm suspecting so many people don't! I'd prefer to use para as for some implicit reason it seems more impressive (stupid reason I know, but surely i'm not the only one). So how do you feel about this issue? Please advise.

 

Thanks

Jon

[badchad]

I'm no statistician, but I think significance between variables hold more weight when they are parametric. What I think is more important is the type of statistical analysis used. There are strengths and weaknesses involved in any type of statistical analysis. Therefore, it's best to determine these when analyzing the data.

 

I would agree with you that parametric measures seem more impressive. I know it's a vague answer, but it's the best I can do.

[Glider]

More and more people are not bothering with non-parametric analyses. I do, but I think I'm in a minority.

 

I think it comes down to understanding your data. If you understand your data, then you run less risk of abusing them. There are ways of treating non-parametric data to avoid breaching the underlying assumptions of parametric tests. For example, if the data are skewed, a log transformation tends to correct the skew. As long as the same transformation is carried out on all variables, then the underlying relationship between them remains unaffected.

 

Most Parametric tests are quite robust and some more than others, e.g. linear regression is very tolerant of non-parametric data, as long as those data form a meaningful scale. For other tests, there is simply no non-parametric equivalent, e.g. a mixed ANOVA. In this case, the data should be treated to ensure they fulfil the assumptions of the test as far as possible.

 

Basically though, if your data are reasonably sound, then the difference in the results between parametric and non-parametric test is so small as to make no difference. This is why many people are no longer bothering with non-parametric analyses.

 

However, if you are a student, then you will be expected to understand the differences and adhere to the criteria for the tests. It seems to be only postgrads who get lazy.

[nonstoptaxi]

Thanks guys.

 

Glider, yes, that's an interesting response. I'm a postgrad...so yay, I can be lazy!

 

Nonetheless, your advice is useful. I have discussed this with a few peers in my psych. department and it's coming down to ensuring that both a non-para and para equivalent is used on the data, and if both produce the same result in terms of significance, then we'd report the para result.

 

That would hopefully be defendable in a viva situation. Yet, even this to me seems rather 'unscientific', but since most are using parametric tests, I'll go with the tide.

 

Thanks,

J

[Glider]

That's one way of doing it, but it's a lot of work for not much gain. In my Psych dept., many of the researchers tend to default to parametric analyses. However, in some cases I suspect this is due to a tenuous understanding of research methods and the possibility of errors when using parametric tests on ordinal data. However, those that do know their stuff tend either to use an appropriate non-parametric test, or to treat the data to ensure it fulfills the assumptions of the parametric test they want to use.

 

In any event, if you are talking about a viva situation, then using parametric analyses on non-parametric data may be defensible, but you will be expected to do so. However, if you use the appropriate non-parametric test, you won't have to defend your position. I used a Kruskal-Wallace on the data from one of my thesis studies and didn't have to defend it beyond "it was the appropriate test for that level of data".

 

That study was subsequently published, and again, one of the reviewers gave me some grief, saying (as reported by the editor) "why did the author present mean ranks and not means?". Again, my response needed to be no longer than "Because the mean ranks show where the effect is in a Kruskal-Wallace and means are not appropriate for ordinal level data.". The editor found this acceptable, because the point is not debatable.

 

So, for your thesis, I would suggest running non-parametric tests where appropriate. Then you can't go wrong and you are limiting what you have to defend in your viva.