Typical Customer Order Value I am analysing some sales data and looking at the customer orders for 550 customers and want to answer the question "What is the typical customer order value", however, the orders are skewed as follows;
Order Value - No of Orders
£0 to £100 - 400
£101 to £200 - 78
£201 to £300 - 35
£301 to £400 - 25
£400 and over - 12
The average is £27.5 with almost 50% of the orders under £28 and 72% under £100, the mode is 7. What method can I use to answer the question of what is the typical customer order?
 A: The median is usually a decent summary statistic for skewed data, as it isn't affected very much by a small number of outliers. The mean is a reasonable choice in some instances, but it could be thrown off if you had a single order for £1M, for example. In this case, your mean and median are almost the same, so either is a decent choice.
The mode may not be very useful for this kind of continuous data where you have a few hundred samples covering a range of a few hundred pounds. Without knowing more about the data, I'd imagine that you have relatively few repeated values, so seeing 5 orders of £7 isn't really much more meaningful that 4 orders of £100. The mode can be highly affected by exactly what samples appear in your data, but a distributional summary statistic like mean or median should be more stable. You can get around that continuous issue somewhat by binning data, but then you run into issues in how to choose the bin boundaries. The mode is usually most meaningful in well-sampled data that has some kind of central tendency or distributional peak, but would not be as meaningful in uniformly distributed data, for example - in a uniform distribution, the mode will just pick up on random sampling fluctuations rather than tell you something meaningful about the underlying distribution.
Without knowing anything about the underlying distribution, I usually find the median to be most informative, followed by the mean, and lastly the mode. That said, the best choice will depend on what you want to capture, and what the underlying distribution you're trying to summarize is. You could imagine a bimodal distribution with an empty gulf between them, in which the mean/median could actually be values that never appear in the data, so neither would be very good summary statistics themselves. You'll need to have a look at the distribution of values, and define exactly what you want to capture with the "typical customer order" - that could feasibly be the most frequent order, or the mean/median value of customer orders.
A: +1 to Nuclear Hoagie's answer. However, I would disagree.
The "typical customer order" to me strongly suggests what the "most common" order would be. And the most frequent = most common order is by definition the mode. So I would say the straightforward answer is the mode.
As Nuclear Hoagie writes, the mode can be sensitive. Just one order more of a certain value can make the mode jump wildly. Assume you have 10 orders of 7 GBP, and 9 of 31 GBP, and then two more orders for 31 BP come in - suddenly your mode jumps from 7 to 31 GBP.
I would recommend you discuss what the result of your analysis will be used for, and what "typical" is intended to mean.
