Experiment on friction coefficient Here you can see the results of the experiment about a friction coefficient:

The mean of the friction coefficient becomes 0.262 but when I do a linear regression in the form of y=mx the slope is 0.31. Shouldn't it be the same? I used $F_N$ as x values and $F_D$ (friction force) as y values.
regression: https://www.desmos.com/calculator/njj4utvsdk 
 A: 
The mean of the friction coefficient becomes 0.262 but when I do a linear regression in the form of y=mx the slope is 0,31. Shouldn't it be the same? 

No.
Linear regression and arithmetic mean are not the same thing.
Linear regression is trying to fit a linear plot to the points you gave it with the best $R^2$ value.
Arithmetic mean is just the sum of all the values divided by the total number of values.  Statistically they measure different things, so you can't expect the same value from them.
A: You have a $F_D$ measurement issue because from the numbers it appears the linear regression gives you a negative y-offset. This means you are missing some force that is not measured.

Go with the linear regression slope, removing this error as any constant value in the measurements is taken out. This would be the 0.386 value according to Excel.
I suggest looking for ways to remove any other sources of error, like stickiness or friction in the areas that you are not interested in. Also, add the trailing zeros to $F_D$ because it is misleading they way presented here. A value of 0.3 can be really anything between 0.25 and 0.35, but if you specify the value of 0.3000, the range of possible values is 0.2995 to 0.3005 which is much more well defined than the values posted.
A: Expounding on sammy's comment, when you divide $F_D$ by $F_N$, since $F_D$ only is measured to one significant figure you can only report $\mu_S$ to one significant figure.
Value of $\mu_S$ from calculating the mean of all experiments: 0.262 = 0.3 
Value of $\mu_S$ from linear regression: 0.311 = 0.3 
So for your case you do get the same answer either way.
If you would take a ton more data points with more significant figures on $F_D$ you should still get the same answer both ways. It would probably be a crapshoot to guess which way will give you a more accurate answer with four data points. Plotting the data points and looking at the regression will at least give you a heads up if the experiment shows non-linearity (which would be an indication of some limitation of the experimental setup to reflect the model).
