Skip to content

Base selection

Analysis selection as a subset of the base selection

You may be just interested in a sub-set of your customer base. For example, you may just be interested in customers who live in Scotland, and in understanding which of these people went to Sweden. In this case you could set the base selection to “Region = Scotland”. You would then be identifying characteristics that distinguish:

  • Customers who live in Scotland and took a holiday to Sweden.
  • From all customers who live in Scotland.

The analysis customers must be a subset of the base customers – people cannot be in the analysis set who are not in the base. In this example, the base is just people living in Scotland, so the analysis et cannot contain people living outside of Scotland. You can achieve this in two ways:

  1. Make this explicit in your selections, for example:

    Base: Region = Scotland

    Analysis: Region = Scotland, Destination = Sweden

  2. Let the system enforce the condition for you. It does this by first applying the base selection and then applying the analysis selection on top of this. For example, specifying the following would achieve the same result as above.

    Base: Region = Scotland

    Analysis: Destination = Sweden

Be careful if you are relying on this second method as it can be complicated interpreting the overlap between the selections when variables are taken from different levels. For example, an analysis selection of “destination = Sweden” applied on top of a base selection of “Booking Year = 2005”, would not enforce the fact that the holiday to Sweden was in 2005 – the holiday to Sweden could have been any year, and the person will be in the base selection provided that they have had at least one holiday in 2005. irrespective of the destination.

Using a random base selection

Using a random base selection provides a useful way of restricting the volume of data used in your analysis. This can be particularly useful to either:

The easiest way to set up a random base selection is to use the Apply Limits option and select a random percentage or fixed total, as shown below. There is no need to modify the analysis selection, since this will be applied on top of the base selection (as explained above).

Using a random base selection to identify main effects

If you use say a sample of 5000 people, then only the major factors will be visible. For example, using all the data could lead to very detailed rules based on Year of Birth (e.g. people born in 1970 behave differently to those born in 1972) – these rules could be too specific and actually modelling small quirks in your data. If you used a small sample, then only the big effects will be visible and these quirks will be hidden. You may end up with just a few rules based on age, perhaps dividing your customers in to a handful of age bands, but these would be robust and transferable rules.