How to calculate the error bar (e.g. stdev.s) for a slope based on independent samples? In my experiment, I prepared 6 identical samples in the beginning. At Time 1, 3 samples were sacrificed and analysed; at Time 2, the rest 3 samples were sacrificed and analysed. So I got the data as below. The slope (-0.5243) is obtained as the rate. Can I ask how to caculate the error bar of the slope? Do I need to do this experiment e.g. three times, and get three slopes to draw the error bar?
I do not think I can simply get three slopes from these 6 data points. Because I can get a slope either from "Time2_repeat1 - Time1_repeat1" or "Time2_repeat2 - Time1_repeat1". Is there a particular definition for the slope, e.g. "average slope", "individual slope"?
Time    repeat1 repeat2 repeat3 average STDEV.S
1   0.8628  0.8675  0.8661  0.8655  0.0024 
2   0.3445  0.3343  0.3448  0.3412  0.0060 


Thank you Glen_b. I have attached it as below according to your words.
Can I ask, 
1) I use "VAR.S" for the VAR, is that correct?
2) When you say "standard error", do you mean "standard error of the mean"? If I can calculate the Standard Error of the Mean (SEM) by:
SEM = STDEV.S/sqrt(count(n))
Thank you.

 A: The usual assumption of constant variance are not usually reasonable for percentage data (they might be okay in some situations). [Linearity might also matter if you're trying to infer anything but the means at those two particular times, or the change in mean there.]
In any case, if you're prepared to assume the population standard deviations are about the same at the two time periods then one easy way to get a standard error of the slope would be to fit a least squares regression line to the entire six values (i.e. don't average them first).
[i.e. Set up a y column by stacking the six y-values, and an x-column by stacking the times (three 1 values on top of three 2 values).]
There are a variety of equivalent available formulas for the standard error of the slope, but the following approach will probably be the easiest for you to implement. In simple linear regression,
$$\widehat{\text{Var}}(\hat{\beta})=\frac{(1-R^2)}{n-2}\frac{\text{Var}(y)}{\text{Var}(x)}$$
The standard error of the slope is the square root of that.
[In Excel, you can get the slope using the SLOPE function, and the $R^2$ value using the RSQ function between the y and x values. The ordinary VAR functions are applied to the y and x columns and then COUNT to (say) the y-values (here assuming there are no missing values in the x-column) to get the $n$.]
In the simple case you have here the slope and its standard error can be calculated a bit more simply, but the approach here is more general. 
