# What is the practical application of variance?

I'm teaching myself probability theory, and I'm not sure I understand any use for variance, as opposed to standard deviation. In the practice situations I'm looking at, the variance is larger than the range, so it doesn't seem intuitively useful.

-
Take a look at an ANOVA table. –  whuber Apr 28 '11 at 14:36
The SD is more intuitive because it is on the same scale as the data. However, when working with the normal distribution, the variance is the parameter not the SD. Thus, variances can be more useful when working with distributions mathematically. Eg, variances add, but SDs don't. –  gung Oct 3 at 16:35

In practice, you calculate the SD through calculating the variance (as abutcher indicated). I believe the variance is used more often (apart from interpretation, as you indicated yourself) because it has a lot of statistically interesting properties: it has unbiased estimators in a lot of cases, leads to known distributions for hypothesis testing etc.

As to the variance being bigger: if the variance were 1/4, the SD would be 1/2. As soon as your variance/SD are smaller than 1, this order reverses.

-
Would you think that one should use arbitrarily use units that prevent variance being less than one? I would even go so far as to suggest that the units used should be such that the measure having its variance assessed should not have decimal places. Take for example, measurements of the same length in metres and its various multiples and subdivisions. –  Robert Jones Jun 29 at 23:51

In portfolio theory, variance is additive. In other words, just as the return of a portfolio is the weighted average of the returns of its members, so to is the portfolio variance the weighted average of the securities' variances. However, this property does not hold true for standard deviation.

-
although it's been a while, but your answer helped me understand a totally different question that I had about portfolio theory :) –  PhD Nov 8 '11 at 2:33
Variance is additive outside of portfolio theory, too. –  gung Oct 3 at 16:31

Variance is the most basic of the two measures... stddev = sqrt(variance). While exaggerated, it's good enough for a comparison and grows very large when there is mixed-up-ness in the distribution.

variance(22, 25, 29, 30, 37) = 32.3
variance(22, 25, 29, 30, 900) = 152611.0


Standard deviation is used way more often because the result has the same units as the data, making standard deviation more appropriate for any sort of visual analysis.

-
"No practical use" is a bit overly-strong. $\beta$, for example, is calculated using variance and covariance and variance shows up in many, many other calculations as well. People often prefer to report standard deviation instead because the units match the mean (and it's often closer in magnitude to it too), but I'd argue that reporting the raw means and variances/standard is hardly the only thing one can do with business-related data! –  Matt Krause Jun 29 at 14:18