trishrhymes
Senior Member
- Messages
- 2,158
'Presentation of the principles and methods of data description and elementary parametric and nonparametric statistical analysis. Examples are drawn from the biomedical literature, and real data sets are analyzed by the students after a brief introduction to the use of standard statistical computer packages. Statistical techniques covered include description of samples, comparison of two sample means and proportions, simple linear regression and correlation.'
This is a good start, but the kind of statistical tests being quoted in some of the studies we have been shown here are far more sophisticated that this.
What you describe here is at the level I used to teach in A level statistics to 16 to 18 year old school pupils.
And from what I've seen of what is taught in the social science degrees, it's even more basic than this, and mostly non-parametric tests which seem to me pretty unsophisticated.
I think a serious scientific / medical study these days, with all the sophisticated computer stats packages available needs someone with at least a masters degree in statistics and experimental design, not just a module or two. Anyone setting up a research study should have the basics, as described above, but they need experts to help with experimental design as well as analysis and interpretation.
I agree with your general principle, @Barry53 and @CFS_for_19_years , making use of a university department of biostatistics, which will have staff who have PhD's and many years experience should be an essential requirement.
One thing we found with PACE was that they got away with changing outcome definitions (recovery, improvement) and managed to get approval for these, and get them past peer review.
What this tells me is that approval bodies and journal peer reviewers need much more expertise too.
The nonsense misuse of the normal distribution for heavily skewed data should have been a red flag, for example. And in many of the psych studies, there are basic errors of interpretation of correlation as implying causation. This is basic high school statistics.
This is a good start, but the kind of statistical tests being quoted in some of the studies we have been shown here are far more sophisticated that this.
What you describe here is at the level I used to teach in A level statistics to 16 to 18 year old school pupils.
And from what I've seen of what is taught in the social science degrees, it's even more basic than this, and mostly non-parametric tests which seem to me pretty unsophisticated.
I think a serious scientific / medical study these days, with all the sophisticated computer stats packages available needs someone with at least a masters degree in statistics and experimental design, not just a module or two. Anyone setting up a research study should have the basics, as described above, but they need experts to help with experimental design as well as analysis and interpretation.
I agree with your general principle, @Barry53 and @CFS_for_19_years , making use of a university department of biostatistics, which will have staff who have PhD's and many years experience should be an essential requirement.
One thing we found with PACE was that they got away with changing outcome definitions (recovery, improvement) and managed to get approval for these, and get them past peer review.
What this tells me is that approval bodies and journal peer reviewers need much more expertise too.
The nonsense misuse of the normal distribution for heavily skewed data should have been a red flag, for example. And in many of the psych studies, there are basic errors of interpretation of correlation as implying causation. This is basic high school statistics.