Effect of a packer on speed of checkout operators

The 8 operators on the morning shift were provided with a packer to fill customer bags.

Use this diagram to explain how poor allocation of treatments to experimental units can result in the treatment effect being over- or under-estimated. Click Allocate treatments then Conduct experiment to see typical experimental results.

The checkout operators getting the packer were generally older, so they would have taken less time even without getting a packer (as explained in the diagram on the previous page). As a result, their average time is 0.15 less than those without a packer even though the effect of the packer is only 0.1 second per dollar.

The experiment has been badly designed.

Repeat a few times. To help explain, change the true effect of the packer to zero with the pop-up menu and repeat, noting that the experiment estimates that the packer still improves the time taken.