7. Testing of Data

This is all about confidence.

Are you confident that the data provided is reliable and accurate?

Testing

To provide a reasonable level of confidence in the data, it is often a good idea to test the system that provides it.

Every car in the UK has to pass an MOT test. It is extremely inconvenient if your car fails the MOT!

 

An important MOT test is to measure the exhaust to make sure it is within legal limits. The machine that carries out this test has to be regularly calibrated to make sure it is providing good data

 

Purpose of testing

There are three things you want to know about the system providing data:-

• Does work under normal conditions?
• Does it work at the extreme limits of operation?
• Is it capable of warning that bad data is being created?

Suppose that this machine is providing car weight data. In order to test the system you would look at three situations:

Normal

Test that the system working under normal conditions, is providing the correct data

Extreme

Put the heaviest car it is designed to weigh and check that the data is correct. Then put the lightest car on the system and check that data is good as well.  You test both possible extremes to ensure that the system accepts them.

Error - out of calibration.

It is normal to use test data that you know will fail.  You need to know that the system will reject values outside of the acceptable limits. 

With our car example, the machine checks itself against calibrated weights and sounds a warning if it fails the test.

 

Importance of test plans

It is a real skill to put together a test plan for a complicated system because there are so many 'paths' to check.  Even a test plan for a 'simple' system will require many data checks.

In fact some systems are so complicated, they need other computers to work out a  test plan that can be carried out in the time available.

 

A good test plan will have the following features:

• To ensure all aspects of the data are covered, including normal, extreme and erroneous values
• To document the data used - if anything goes wrong with the system later on, the original data tested can be rechecked.
• To list the actions taken
• To enable all tests to be reproduced
• To list expected results
• To be able to tie expected results to actual results