Is the question here why the size of a QVD is much smaller than the source data (eg. a CSV of the same data)?
If so, it is due to the way that QlikView handles duplicated values in a column. Rather than re-storing the value each time it occurs (as a CSV must) QlikView stores the value once and then a pointer to that value every other time it is referenced.
This optimisation of size is of course critical for in memory analysis, where every byte counts. The QVD size is as a result of the file being optimised in the same way as the data is stored in memory.
Steve has just told you what you require. QV isn't cruching the data differently each time, but follows a standard set of rules for symbol tables and bit stuffed pointers as mentioned above.
If you're still confused, however, think of it as the difference between a JPEG and a BMP.
Consider a white square, 10px by 10px. A BMP will store colour information for each pixel, 100 lots of RGB(255,255,255) whereas a JPEG will hold 1 lot of colour info RGB(255,255,255) and 99 pointers to that information. This reduces the size of the JPEG file.
QV does the same thing, holding information for unique values and pointers to that information where duplicates exist. This reduces files size, as mentioned in Steve's reply.
