In vivo, in vitro, in silico
Some people get unit testing and some people don't. The reasons vary, usually based on a mixture of previous experience, lack of experience, fear of the unknown or joy at a safer quicker way of developing. One specific doubt crops up from time to time. It comes in the form of "If I test small bits, i.e. units, whatever *that* means, it proves nothing. I need to test the whole thing or small parts of the whole thing live."
My PhD was in toxicity prediction, which involves testing if something will be toxic or not. You can test a chemical "in vivo" - administer it to a several animals in varying doses. You sit back and wait til half of them die, or show toxicity symptoms and record the doses. This gives you the Lx50 - for example the LD50 is the lethal dose that kills 50% of the animals. Notice I said you can do this. You can also test the chemical on a set of cells in a test tube or petri dish - "in vitro" (in glass). Again you can find the dose which affects 50% of the specimens. I personally find this less upsetting, but I want to focus on parallels with testing code here. Finally, given all this data the previous tests have generated, you can analyse the data, probably on a computer, perhaps finding chemical structure to activity relationships - SAR, or quantitative SARS i.e. QSARs. These are referred to as "in silico" - for obvious reasons. Some in silico experiments will just find clusters of similar chemical, which can either alert you to groups that might need more detailed toxicity testing, or even guide drug discovery by steering clear of molecules, say containing benzene rings which can be carcinogenic, saving time and money if you are trying to invent a drug that cures cancer. The value of testing on a computer outside a live organism should be clear. It can save time, money and even lives.
If we keep this in mind while considering testing a software system, rather than a biological system, we should be able to see some parallels. It is possible to test a live system - maybe on beta rather than "TIP" (Test in production). This can be a good thing. However, it might save time and money, and though maybe not lives, certainly headaches, to test parts of the live system in a sandboxed environment, analogous to in vitro. Running an end to end test against a test database instance with data in a specific state might count. Pushing the analogy further, you could even test small parts of the system, say units, whatever they are, in silico. Just try this small part away from the live system in a computer. This is worthwhile. It will be quicker, as toxicity in silico experiments are quicker - they tend to take hours rather than days. This is a good thing. Of course, you won't know exactly what will happen in a full live system, but you can catch problems earlier, before killing something. This is a Good Thing.
Other industries also test things in units - I could put together a car or a computer hit the on switch and see if it works. However, I am given to believe that the components are tested thoroughly *before* the full system is built. If I build a PC and it doesn't work I will then have to go through one part at a time and check. If someone tests the parts first, this will ensure I haven't put a dodgy power block in the whole thing. Testing small parts, preferably before testing the whole system, is a Good Thing.
I don't believe this short observation will change anyone's minds. But I hope it will give pause for thought to those who think only testing from end to end matters, and testing "in silico" is a waste of time.