And that’s not a good thing for understanding numbers, says Priceonomics:

Many analysts believe that the unthinking use of the average damages our understanding of quantitative information. This is because when people look at averages, they think it is “the norm”. But in reality, it might be highly impacted by just one huge outlier.

Imagine an analyst who wanted to know the representative value for the cost of real estate on a block with five houses. Four of the houses are worth $100,000 and one is worth $900,000. Given these numbers, the average would be $200,000 and the median $100,000. In this case, and many others, the median gives you a better sense of what is “typical”.

…

The median is also less sensitive to the dirty data that many analysts deal with today. As statisticians and data analysts increasingly collect data from surveys and scraping the web, user error when entering data can have a larger impact on results. When a user accidentally adds an extra zero that changes an answer from 100 to 1,000, it is much more likely to impact the average than the median. More.

One often runs into suspect data proferred by pressure groups armed with averages.

For example, “The average child growing up in Bigtown will have witnessed more than *eight * incidents of violence before the age of fifteen!” [Therefore, you are ethically obligated to do whatever we demand… ]

Okay. But before we sign the bill, what about the “median” child?

Some children live in the most violent neighbourhoods and have probably witnessed or experienced hundreds of such incidents. The median child has maybe witnessed one or two.

Medians can provide a better perspective on typical lived reality. To say nothing of helping us decide how bests to address the situations that are nowhere near the median

Follow UD News at Twitter!

What would be best really would be to educate people regarding the meaning of ‘mean’ and ‘medium’ and to also teach them to critically look at reported statistics. Sometimes ‘mean’ is what you want to use, sometime ‘median’ makes more sense. For a normal distribution they should be just about the same.

I actually agree with ellazimm. I avoid average and use the mean instead.

Using a mean in a highly skewed distribution is misleading. As is the median. Percentiles can be used, or more robust procedures such as the use of algorithm A.