Does Excel STDEV calculate in 3 stdev (standard deviation)?

[T Dutton]

Does Excel STDEV calculate in 3 stdev?

[Scott Catron]

If I'm interpreting the questions correctly - NO.

STDEV is one standard deviation - you'll have to multiply it by 3 in a formula somewhere.

[Howard Lee]

In Excel STDEV yeilds one sample standard deviation. To get 3 sigma you need to multiply it by 3. Also, if you need the standard deviation of a population, you should use STDEVP instead.

[Statistical Steven]

Originally Posted by Howard Lee

In Excel STDEV yeilds one sample standard deviation. To get 3 sigma you need to multiply it by 3. Also, if you need the standard deviation of a population, you should use STDEVP instead.

I wish Microsoft would do away with the STDEVP function. It has no use. The difference is that STDEV divides by n-1 and STDEVP divides by n. Can anyone tell me when you would use STDEVP?

[Wesley Richardson]

Originally Posted by Statistical Steven

Can anyone tell me when you would use STDEVP?

Hi Statistical Steven,

It would be used if you had the entire data set. Although this is less common, there are a few cases: 1. The standard deviation of test scores for the entire number of students in a class. 2. The standard deviation of a part characteristic when you have measured all of the parts in the production run. Both cases would typically have a fairly small number in the population, such as less than 100 items.

I accept that if you have measured all of the population values, then you could graph the values and see how they are distributed.

Wes R.

[Steve Prevette]

Originally Posted by Statistical Steven

I wish Microsoft would do away with the STDEVP function. It has no use. The difference is that STDEV divides by n-1 and STDEVP divides by n. Can anyone tell me when you would use STDEVP?

I think I've also answered this on the ASQ board from a question there. But here goes:

The formula for sigma (STDEVP) is the maximum likelihood estimator for the population standard deviation. However, it is a biased estimator. If you only have a sample of the population (and consider - if the "future" is part of your population, you still only have a sample), STDEVP will have a bias in it. If you repeatedly sample from a population, you will notice that the average STDEVP values will not tend to match the actual population standard deviation. However, if you use STDEV (dividing by n-1), the average of the STDEV's will converge on the population standard deviation.

[Tim Folkerts]

I like what Steve said, but I want to make two minor points.

1. sigma (STDEVP) would be the population standard deviation, not an estimate. When you have the full set, you don't need to estimate!
2. I believe it is actually (STDEV)^2 that averages to (STDEVP)^2, not STDEV that average to STDEVP. That is, you take the variances calculated from the sample standard deviations and average them to get an unbiased estimate of the true variance.

Of course, all this is all WAY to esoteric for most people!

[Statistical Steven]

Actually S^2 is the MLE for sigma squared...but that is not the point. If as Wes stated, you have the entire population, then why would you need to know the standard deviation. That is to say, any inferences to be made can be made from the data in the population. There are no confidence intervals that need to be calculated. I am stumped, regardless of the theoretical differences between sigma and S, why there is an application that someone would be using Excel for that requires the population stdev.