AGENDA Experimental Design ROLE OF EXPERIMENTAL DESIGN Validate base model of actual system Design experiment Identifying different alternatives to examine Replication analysis How many replications are needed for comparison Output analysis Confidence intervals ANOVA Duncan multiple ranges test Recommendations EXPERIMENTAL DESIGN Factors and Levels Two Alternative Experimental Designs One Factor Experimental Designs Two Factor Experimental Designs Multifactor Experimental Designs 2^k Experimental Designs Interactions Refining the Experimental Alternatives FACTORS AND LEVELS Factors are different variables thought to have an effect on the performance of the system Levels are the values that the different factors may take EXAMPLES OF FACTORS Workers performing specific functions Machines which perform specific operations Machine capacities Priority Sequencing Policies Worker Schedules Stocking Levels EXAMPLES OF LEVELS Four versus five versus six workers performing a specific function An old versus a new machine performing specific operations A Five ton versus a two and a half ton capacity truck. First-in-first-out versus last-in-first-out priority sequence policies Five eight hour shifts versus four ten hour shifts. Restocking order levels between 10 and 25 percent TWO ALTERNATIVE EXPERIMENTAL DESIGNS Most simple experimental design A base system exists… No base system exists… WHEN A BASE SYSTEM EXISTS First alternative is the base model Second alternative is: An alternate operating policy An alternate resource policy WHEN A BASE SYSTEM DOES NOT EXIST Both alternatives are for proposed models Examples Equipment for a new process from two different manufacturers Facility layout for a service facility of two different ONE FACTOR EXPERIMENTAL DESIGNS Next level of sophistication One specific factor that we are going to examine at three or more levels Same level of resources but different operating policies ONE FACTOR EXPERIMENTAL DESIGN RESOURCE EXAMPLE Three clerks Four clerks Five clerks ONE FACTOR EXPERIMENTAL DESIGN OPERATING POLICY EXAMPLE Three individual parallel clerk queues One single snake queue feeding two clerks and one queue feeding one clerk One single snake queue feeding into all three clerks TWO FACTOR EXPERIMENTAL DESIGNS Two factors Examine each of these factors at a number of different levels Number of alternatives is equal to: Number of Levels in Factor A x Number of levels in Factor B TWO FACTOR EXPERIMENTAL DESIGN RESOURCE EXAMPLE Three regular clerks with parallel queues Four regular clerks with parallel queues Three novice clerks with parallel queues Four novice clerks with parallel queues TWO FACTOR EXPERIMENTAL DESIGN OPERATING POLICY EXAMPLE Three clerks with regular parallel queues Three clerks with a single queue Two clerks with regular parallel queues and one clerk with a single express queue Two clerks with a single regular queue and one clerk with a single express queue MULTIFACTOR EXPERIMENTAL DESIGNS Much more complicated type of experiment Number of alternatives can quickly explode into an unmanageable level Number of alternatives is equal to: 2^K EXPERIMENTAL DESIGNS Reduces the number of levels in each factor to two Low level and high level High and low levels can be different between factors EXAMPLE A security check point at an airport has people checking for the presence of tickets, two metal detectors and two xray machines. A 2^3 experiment is set up with the three factors being ticket checkers (2 or 3), metal detectors (1 or 2), and xray machines (1 or 2) FACTOR COMBINATIONS INTERACTIONS When two particular factors have some sort of synergistic effect The effect of both of the factors may be larger than the sum of the effect of each of the individual factors Interactions can be examined through sophisticated statistical analysis techniques Practitioner will probably want to assume there are no special interactions REFINING THE EXPERIMENTAL ALTERNATIVES An initial experiment is conducted to test the factors Follow-up experiments are conducted which focus on the significant factors May need to bracket the point at which the levels in the factors become significant