Let's say we have a box that contains a mysterious and infinite universe. It is impossible to examine every item in this universe, just as it is impossible to examine every child in the universe. Research involves sampling the reality inside the box to learn about it. If we see a pattern in the samples drawn from the box, then we can begin to predict the contents of the box. The goal of scientific activity is to accurately predict what we'll draw from the box.

Let's say the first item a researcher draws randomly from the box is a drinking glass. It is impossible to predict the nature of the other items in the box from one item-a case study. A case study out of the context of other research is not very informative.

Let's say the next item a researcher draws from the box is another drinking glass. We might start to predict the contents of the box to be drinking glasses.

In selecting the next item, a researcher who has formed a hypothesis about the contents of the box might be tempted to look in the box and select an item that shows his or her hypothesis is right. Or the researcher might throw out an item if it doesn't confirm the hypothesis. Throwing out an unwanted item or choosing only items that confirm a hypothesis is not good research. A good researcher is careful to continue to select randomly from the box to test and develop the hypothesis. A good researcher accounts for all the samples drawn from the box.

Let's say the next item drawn is a cup. Our original hypothesis didn't predict accurately enough, so we could revise our prediction now from "drinking glasses" to "drinking containers."

One more item drawn from the box is a canteen. The object confirms our hypothesis. Now our samples include two drinking glasses, a cup, and a canteen. The box seems to contain all types of drinking vessels.

Let's say now that the researcher draws a rock from the box. A rock does not fit the pattern of drinking vessels at all. Must we now throw out our entire knowledge base regarding the contents of the box because we have run across one item that does not fit? No, we shouldn't. Based on the samples taken from the box, we can still predict with some certainty what will come out of the box next. Our hypothesis is now that 4 out of 5 objects are drinking vessels. If we select another item randomly from the box, then we most likely will get some kind of drinking vessel next. To toss our knowledge away and say we have no knowledge because it is not perfect would be silly.