Are samples independent or dependent when measuring the same subject? Hey I read into dependent vs. independent samples but can't figure out what they are in my case. So I'll describe briefly what my samples are, and how I derived them.
I have measured 4 muscles on  20 subject. For each muscle I have one parameter:
            muscle1   muscle2   muscle3   muscle4
subject1       1.2      2.4       2.3        1.3
subject2       2.2      3.3       3.3        2.6
subject3       2.4      ...
...

Now I want to compare the groups of different muscles. My understanding of dependence is that if I measure subjects twice, the samples are dependent. When we predict a model from these values, the outcome and errors of the outcome are not independent. 
The trouble I have to understand is if I measure the same subject, but a different muscle, are the samples still dependent?  My guess is, they are somehow dependent, since the measures are derived from the same subject and the muscles belong to the same region. However, I really would like your opinion on this.
Edit: I don't know if it is important, but the muscles are not measured at the same time.  
 A: You have to ask yourself whether there is some kind of natural correlation between the measures of the muscles in each subject. This is not exactly a statistical question, but it depends very much on the specific research subject. 
Assume that I analyze your data and I come up with the conclusion: "hey Nelson, I found out that if a person has a relatively high measurement in muscle1, that also means they have a relatively high measurement in muscle2". 
If your answer is: "phew, well, that was kind of obvious", you have an obvious dependent structure. In statistical literature, this is also called a multilevel structure, since you have two level, where the second level is nested within the first level:
first level: the subjects (represented e.g. by the mean on the measurements across all muscles)
second level: the different muscles within each subject
Of course, there are ways to quantify the amount of dependency of your data, e.g. the intraclass coefficient (ICC) which expresses how much variance in your sample can be explained by the underlying dependency. There are no strict rules when exactly your samples are, so to say, independent enough to be analyzed while ignoring the multilevel structure. However, if you falsely ignore this structure, you may get distorted coefficients and p-values.
The concept of multilevel structures is very nicely explained in this short article by Aarts et al. (2015): https://bmcneurosci.biomedcentral.com/articles/10.1186/s12868-015-0228-5
