8.6 Presenting a General Linear Model

Optimal Retail Design. Imagine your client is a major retailer. They are trying to determine the best checkout counter configuration, best being fastest one for your cashiers to complete transactions with customers. They also want an understanding of the differences between the styles. The client has given you a large sample of 20,000 records, 5,000 records per configuration. The first few records are given in Figure ??. The information in the file is as follows:

• configuration=configuration of the checkout counter and cash register. There are 4 types. Two different counter shapes and two different cash register types.
• nitems=number of items purchased during transaction
• tender=0 if cash is used,1 if credit is used
• massist=1 if manager assist 0 otherwise
• timer1=time first item is scanned
• timer2=time last item is scanned
• timer3=time transaction completed

The dependent variable was not given directly. The dependent variable used was timer3 minus timer1. The following presentation is one example of presenting the findings from a GLM in a manner that most can understand. After the presentation some issues will be discussed concerning the way in which the data were analyzed.

What was shown on the last few slides was the complicated statistics. The same findings were presented with simple graphs and means. Your job is to enlighten, not impress people with complicated statistics. Why overwhelm and confuse the client with complicated statistics when the statistical findings can be presented in a simpler and thus more informative/enlightening manner.

Many of you may work with statisticians at one point in the future. These statisticians may use advanced statistical techniques, but the statistical findings can almost always be presented in a simple manner, as shown here. Try to get the statistics into a simpler form for presentation. Do not have anything in the presentation you do not understand. You are expected to know what you are presenting. You may think it is ok if you do not understand but the statistician does understand. This is often not ok. If the statistician can not explain it you, there is a good chance he cannot explain it to others. People do not buy/trust what they do not understand and they often do not even bother to listen after awhile. If you are presenting then you are responsible for avoiding this situation when delivering the presentation.

The general linear model shown in the SPSS output was not the optimal model for answering the problem at hand. Some of the issues with the model are:

1. Register type and counter shape should have been separated into two indicator variables as opposed to a single categorical with four levels.
2. The configurations were not tested under busy times but under general conditions. What if people work faster when the store is busy? Which configuration is best under busy times was the real question that needed answering, not performance under general conditions.

How could a person determine if the store was busy or not? One possibility for attempting to answer this question concerns the time between the first item scanned for transaction i, timer1, and the end of transaction i - 1, timer3. Very little time between these two timers might indicate the existence of a que and that the store was busy, or at least the specific cashier was busy.