I would like to know if there is an existing statistical test, like for instance the Kaiser-Meyer-Olkin test for sampling adequacy used for the PCA, to test if a sample size is numerically big enough to perform statistical tests (parametric or non-parametric based on data distribution) or if they can "just" be used for an Exploratory Data Analysis.
Just to be clear I'm not referring to sample size estimation a priori for experimental design, but I'm talking about opportunistic data collected randomly (i.e., number of road-killed animals) for which I want to assess the effect of specific predictors over an explanatory variable.
Furthermore, can data analysed using an Exploratory Data Analysis approach be published in peer-review journals (Q2, Q3, Q4) even if they "only" provide a preliminary assessment, or they best fit for preliminary reports?
Many thanks in advance.
You should say what you mean by "adequacy", but googling "statistical power" might help.
Daniel Wright thanks for your reply. I changed a bit the answer in order to be more specific.
I searched deeply on the net but I found only partial satisfactory answers. Therefore, I preferred to ask an opinon to those that are more experts than me.
So if you mean will the statistical procedure produce estimates (good or bad), it depends on the procedure. For example, to calculate a regression, 2 points (call these (x1,y1) and (x2,y2) ) will provide enough data to allow you to estimate a straight line providing x1 != x2 and y1 != y2. (!= is not equals). I don't think this is what you are actually asking though, so you might want to edit your question again.
On your final paragraph, about EDA being publishable, this question does not make sense to me, but maybe it will to others.
Daniel Wright first, I want to apology because English is not my native language so, sometimes, I have some difficulties to explain what I want to mean. However, I'll try again: I have 36 animals and I would like to test the effect of different environmental predictors over a specific explanatory variable. What I desire to know is if with such a small sample I can use regression models or GLMs. I read that a general rule of thumb is that to obtain reliable results you need for at least 10 data for predictor, so in my case I need to use 3 or 4 predictors, at the very most. However, I don't know if a EDA approach is more adequate in this case and if studies realized using such method (based on a small sample size and no robust statistical approach is used) can be published even in peer-review journals or if I cannot have such ambition. Is that more clear? Thanks again.
Article Power Analysis, Sample Size, and Assessment of Statistical A...
- Badges
- Science topic