As analysts we are often interested in averages. But an average is a tricky thing because the word average actually covers several related concepts. In this blog post I am going to discuss two different concepts of average – the Mean and the Median – and why it’s crucial to understand their differences.

The Mean is what we typically mean when we use the term average. The Mean is calculated by adding everything up and dividing by the number of items. The resulting value is supposed to represent the centre of all the values. People intuitively understand the Mean because a lifetime of being exposed to the concept of the bell curve (or the normal distribution as it is called in math circles) has taught them that the big peak in the middle of the curve is the average.

The Median is less well known but just as easy to conceptualize. Consider all of the items in your collection and sort them from smallest to largest. The Median is then the value of the item that sits in the middle of the sorted list. In other words, it’s the value of the item that splits the collection in half with an equal number of items above and below it.

When dealing with data that follows the bell curve, that is data that produces a symmetric and unimodal (single peak) distribution, the Mean and the Median have the same value and we don’t have any problems. The issue is that a lot of data we’re likely to encounter in our day-to-day work— such as call response times, time spent on patrol, man hours spent on an occurrence—are not accurately represented by a symmetric bell curve but instead follow a skewed distribution that is weighted more to one side than the other.

But what does this mean for averages? It means that for data that is positively skewed (the bulk of the data is on the left side of the chart) that the Mean is going to be larger than the Median and vice versa for negatively skewed data. The problem arises when the difference between the Mean and Median gets large enough to meaningfully change the stat that you are reporting.

For example, the number of man-hours spent on an occurrence is the kind of data that is going to be positively skewed. This is because of the nature of the data: 0 is the lowest value that can exist but the upper limit is not bounded. If you create a histogram of the data you’re likely to find a whole lot of occurrences with a small to medium amount of time being spent (the peak on the left) but also a few occurrences with a lot of time being spent (the long tail on the right). I’ve created the following simplified data set below to illustrate the scenario.

You can see in the screen shot that the occurrence that took 24 hours is clearly an extreme example but it has a significant impact on the Mean. If a Commander came to you and asked for the average number of man-hours spent on occurrences do you feel that the Mean of 8.8 would be representative, considering that only one occurrence actually took longer than that? Clearly the Median value of 5 hours is more representative of man-hours and is a more accurate average.

This result comes about because the Mean is sensitive to the inclusion of extreme values because of the way it is calculated while the Median, which is relatively indifferent to the inclusion of extreme values because it only concerns itself with the middle value for a data set, is more likely to provide an appropriate number that is stable in the face of outliers.

The screen shot also conveniently illustrates the function calls used to calculate both the Mean and Median in Excel. Frustratingly, Microsoft has opted to use the function name AVERAGE for the Mean further erroneously cementing them as synonyms. Thankfully the Median is a straightforward call to the MEDIAN function.

You’ll notice a third function in the list: SKEW. I discuss above about the positive and negative skewness of data and the SKEW function is useful for coming to grips with the direction and magnitude of a data set’s skewness. In the occurrence man-hours example I state that the data is positively skewed and that’s reflected by the positive (greater than 0) value returned by the SKEW function and that indicates that the Mean likely overestimates the Median. Contrarily, if SKEW returns a value that is less than 0 you’ll know that the data is negatively skewed and that the Mean likely underestimates the Median. Finally, if the value is close to 0 your data set likely follows the bell curve or normal distribution and the Mean and Median will be nearly the same.

The key take away from this post is that it is important to not just blindly take the average (by which I mean the Mean) of a stat without first considering how the data is distributed. So much law enforcement data is skewed one way or the other that it makes sense to take a minute and run the MEDIAN and SKEW functions in Excel to make sure you’re reporting the most appropriate number to your superiors.