Friday 19 November 2010

Do we really need maps for analysis?

This is a response to Barry's QlikFix post -

While interesting that 80% of data may have a spatial component (depending on how you define it), the reality is that the majority of that data does not have a special dependency – it is not really that suitable for display on a map.

A common example for such displays might be something like sales by state. These are often displayed as bubbles geographically positioned over the state in question but do they really help with the analysis? Sure, you may be able to easily see that sales in California are larger than in Oregan, but how do they compare against sales in New York, New Jersey and Rhode Island that may be smudged together on the far side of the map. What might look “cool” on a demo may turn out to be something that is easier to analyse in a bar chart.

One other unfortunate flaw is the level of education that might be encountered. Studies have shown that percentages of schoolchildren in the UK have difficulty picking out their own country versus France or Germany! Can we expect that everyone who needs to analyse European data is actually aware that the boot shaped country is Italy?

Maps can be cool and have some great applications (many of which can be easily done in just Google maps without QlikView). Real world analysis can mostly do without them.

Sunday 14 November 2010

It is often simpler than it looks

Often, as we encounter more and more diverse data, we encounter data structures that would appear to call for more and more complex workings to associate them in QlikView. What we need to do though is take a step back and see if there is a different way. My experience is that once I do this, I find that it is often simpler than it looked.

For example, you have 2 or 3 tables that have several fields in common. Some might start down the route of generating complex keys (for example, with autonumber or autonumberhash256) and one or more key tables. A second look might reveal that if we simply concatenate all the tables - even though it appears that they are unlikely candidates for concatenation - then we will suddenly be left with a simple key table and some dimension tables.

Another way of simplifying things that I use all the time is ApplyMap. This is a really useful function (not unlike vlookup in excel) that enables us to map small tables into larger tables. Always look to see if you can do a mapping load.

It is often simpler than it looks.