Variance is square of the distance that an item statistically is from the mean. So you calculate it by taking the mean, then taking each of the individual values and subtracting it from the mean then squaring the result. You add this up for each of the individual values, then divide the entire sum by the total number of data points.

In this case Zeke has a mean of 142.5 yards per game, so it would look like this (for the first few games): (122-142.5)^2 + (101-142.5)^2 + (108-142.5)^2.... you sum all of those together for the entire season and you get a really big number. Then you divide that number by the number of games (10) and you get a Variance of 2744.944 (easiest to do this in excel). The standard deviation is the square root of the variance and is much more statistically useful. In this case the standard deviation is 52.4. So it is pretty clear that both of those stats (variance and STDEV) are per rush.

This means that if we assume that Zeke's performances are "Normally distributed" (standard bell curve)... that we can expect Zeke's outcomes to be within one standard deviation of the mean (+ or -1) 68% of the time. So there is a 68% probability that zeke will rush between 90 and 195 yards. There are serious flaws with the assumption, and therefore this would be dumb. One major flaw is the Normal distribution... there are different variables in play (defense of the team he is playing against) so the normal is not necessarily appropriate. Additionally, there is still not a statistically relevant sample size to cause convergence, would like to see more like 30 data points.

The per rush data (STD Dev and Var) are probably a little more relevant as there certainly is enough data to be statistically relevant, but again, the Normal assumption is not great. All we can get out of this data using the Normal Distribution is that 68% of the time Zeke's runs will be between -4 and 18 yards.... I would guess (not going to analyze the data) that more like 95% of his runs fall in that window. If I were analyzing this data officially, I would subtract the outliers from the data set (runs outside of one STDEV or some other parameter) that are badly skewing the STDEV or Var. I would account for those using some new parameter called % explosive plays or something like that, as a statement of the relative "explosive ability of the running backs."

You could also attempt to use some more "exotic" distributions that tend to reduce the effect of head/tail (outlier) probability events like the Cauchy. Standing by to answer any other math questions.