Pearson correlation minimum number of pairs I have only 7 pairs of data, is this enough for me to do a correlation using pearson r? Or should i use spearman's r instead because of this small sample size?
How many pairs is needed in utilizing pearson r?
 A: If $n=2$ the possible values of $r$ are -1, 0, 1. If $n=3$ then other values are clearly possible but when you come to compute a confidence interval using Fisher's $z$ transformation you find that the standard error is undefined. For $n>3$ you can compute both $r$ and its confidence interval. Suppose in the example quoted $r=0.5$ with $n=7$ the a 95% confidence interval for $r$ is from -0.41 to 0.91. Since before you started you know that the 100% confidence interval is from -1 to +1 you have not really gained much. So my conclusion is that you can compute $r$ for any sample size but it may not tell you much.
A: Yes, seven paired observations are enough to estimate a correlation if you think of it in terms of a phase 0, pharmaceutical clinical trial where 5, 6 or a few more human subjects are the preliminary focus. It is in contexts like these that Fisher's exact test is employed but that test is reserved for contingency tables. Finding its analogue for continuously distributed information is useful. 
To @mdewey's point about confidence intervals, it is precisely for that reason that resampling tools such as Efron's bootstrap work well since it sheds light on the magnitude of fluctuations around a parameter such as a correlation. In fact, some of Efron's original papers about the bootstrap focused on resampling correlations with fifteen observations. As a method, the bootstrap involves repeated sampling with replacement (of the same n) from a set of data some large number of times. As such, it is an approximating tool subject to the inherent limitations of all finite data samples. Hastie and Efron in their book Computer Age Statistical Inference (2016) state that, in most cases, ten bootstrap draws is sufficient to establish basic behaviors. In the theoretical, asymptotic limit, larger and larger numbers of bootstrap replications will more tightly approximate the parameter of interest, again, given the limitations of finite data samples.
Here's an example for toy data:

Here are summary statistics about this data (that the SDs are the same is an accident):

Here are the correlations based on these seven original observations, neither shows significant association:

Here is an example of one bootstrap replication:

You can see that, in the bootstrap replicate, pairs 1, 4 and 6 are sampled once, pairs two and three repeat twice while pairs five and seven are not included. Other replications show similar resampling properties. 
Here are summary statistics for this replicate. You can see that the original means and std devs change slightly due to the resampling and the limitations of a finite data sample:

The correlations for this replicate are somewhat higher than for the original data but, again, neither correlation is significant:

For this toy example, across ten replications (with seven resampled observations each) changes in these summary values are to be observed insofar as the std devs are slightly tighter. Unless a fixed seed is used, ten different draws would have differing summary statistics:

And while the correlations show larger association, they both remain nonsignificant but with smaller p-values:

In other words, bootstrapping reduced the significance levels and, thereby, the confidence intervals, but was not able to turn a sow's ear into a silk purse. Other data, perhaps yours, might show better performance.
Regarding the choice between Pearson and Spearman metrics, Pearson is a parametric test of symmetric linear association where Spearman is a nonparametric test of monotonicity. Clearly, the Pearson is the more stringent test. Choosing between the two is the analyst's subjective choice.
Other measures of pairwise dependence are possible including nonlinear metrics such as distance correlations, maximum distance correlations, mutual information correlation, and so on. That's a different topic and literature. A Columbia workshop on such measures occurred a few years ago and is probably the best single source for information about this literature (http://dependence2013.wikischolars.columbia.edu/Nonparametric+measures+of+dependence+workshop).
