Answer

You should not adjust the funnel but leave it positioned directly over the target. This is an abbreviated version of Deming's 'funnel' experiment, the full version is described below.

It is essentially a 'thought' experiment. It is hard to make it work properly in practice and in any case the design is flawed. The marble will never come to rest directly under the funnel. A more practical version involves dropping playing cards.

The Deming Funnel Experiment

Mark the target point on a horizontal surface. Then position the funnel directly above the target and drop a marble through the funnel.

The rest position of the marble, and the distance from the target, are recorded. This is repeated about 50 times for each of the following four strategies for positioning the funnel:

Strategy	Description
1	Do not move the funnel, keep it aimed at the target
2	Move the funnel from its previous position a distance equal and opposite to the last error
3	Move the funnel from the target a distance equal and opposite to the last error
4	Move the funnel over the last marble

The results of each of the four strategies are:

Strategy	Description
1	Do not move the funnel, keep it aimed at the target
	This is the best strategy, it is the one that will minimize the scatter
2	Move the funnel from its previous position a distance equal and opposite to the last error
	This is akin to making a part, measuring it and then adjusting the machine to compensate. In the salinity example the sample mean was 52 and the machine would be adjusted to decrease the salinity by two units. This strategy will increase the variation by a factor of two
3	Move the funnel from the target a distance equal and opposite to the last error
	The logic here is that the machine is badly calibrated and so the salinity level setting is reduced from 50 to 48. On the first drop this is no different to strategy 2, but subsequent adjustments will be larger than for strategy 2, and lead to increasingly large swings
4	Move the funnel over the last marble
	A young man in his first job was told to cut pieces of rubber tube to a particular length. As he cut each piece he used it to measure for the next cut. After cutting a few dozen he realized that the tube lengths had gradually decreased and were now about half what they should be, and useless (the tube was taken from a coil and the error was caused by the curvature). Another common example is when workers are given instructions that they pass on to others who join later and pass them on in turn