Statistical Methods

Same comments as before. Normal population ? Same gender or bimodal? What is cut off of Old people? How is sample size 7 reflective of greater population? Are all parkinson patients already institutionalized? this fact alone may be explanation for balance.

Difference between means/ medians or standard deviations?

Sample size of 7 is very inauspicious.

What was Ho?

WHat level of significance- remember sample size is only 7.

milo

Yeah, I know. Enough sample size should be considered, caz we want our data to satisfy normality and follow normal distribution etc..

But, it takes a lot effort and time to collect over 30 numbers of samples caz this is sort of human experiment which sometimes needs much of time even in a single experimental trial.

So, I used non-parametric statistical method for small sample size. (I still admit that sample size is quite small, but there are a lot similar results from previous papers, so it seems to be ok in our study.)

We don't care about the gender in our case, and also please just accept each group as a group I mentioned. young is just young. please just focus on my question.

Do not think about the sample size, please.

THEN, Just look at the problem assuming that there are plenty of sample sizes.

My question is.. Do I need additional or different method to be on right way ?

Sorry, I go back to first principles here. I donot turn off thinking because of obstacle.

The point of statistics is to make inference about large population from small sample.

so the nature of selection of the sample is important.

A magazine sampled it s subscribers and predicted landon would win. Roosevelt was a landslide. Issue Sample selection.

Genders do make a difference.

so does fact that parkinsons patients are already in hospital environment or not. You do not seem to have fine tuned what it is that you are describing.

just grabbing 7 available young people and 7 available old people (apparently) parkinsins symptom free, and then grabbing 7 people in the grasp of profound parkinsons and then making claims is statistical malpractice by almost any standard.

To say that these 7 people as a class match the results of another study, and these 7, and these 7, that is a better design, but again what is the null hypothsesis? You have not told us what is being measured, what is threshold, and are we comparing average of the gtroup, or its spread or deviation? what if the seven young people were dancers or gymnasts or athletes? What if one of your convenient non parkinsons old people was avalilable because he was in your hospital for unexplained vertigo and falling issues? That would fix your results, wouldn't it?

THE DETAILS MATTER! The sample selection process matters. We dont turn off thinking because sample acquisition is expensive, difficult or unwieldy. We embrace getting sufficient and representative samples because we want our work to be relevant, and our results significant. And generalizable to the broader population.

How are you testing balance? What is Ho?

milo

I am totally at dark about the sample size. You are saying there are 7 persons with parkinson in a lot (of size)

The numbr of defectives (in my industrial term ) is surprisingly equal for each sample.

I assume the basis of the statistics, the random sampling has been carried out.

To carry out the Mann Whitney test, how did you rank them?

Did you try ANOVA with missing block? But it is all depending on how have you collected sample, why the first block data is missing and so on.

Why are you going for nonparametric at all?

For publishing medical findings in a reputable professional journal one needs to follow an established statistical testing procedure that can be chosen and adopted after reading a number of published papers dealing with similar findings.

For persuading authorities or investors one needs to be persuasive and the targets of the persuation-efforts must be willing or at least willing to listen.

For self-persuasion (or for collegue-group decision making) statistics helps, but is not at all necessary.

I suppose that the heavy arsenal of statistics with its 'null hypotheses', 'alternatives', 'sample sizes', 'significance levels', 'power functions' (and more) is not applicable to a retrospective study evaluating at least 2 factors according to the behaviour of 21 subjects.

Even for self-persuasion a very clear description of how the subject were chosen is extremely important, much more so than any formal statistical method.