Model Testing - 以平衡計分卡與系統動態學觀點探討台灣製藥產業發展之績效

All models are limited, simplified representations of the real world, but we can build confidence that a model is appropriate for the purpose by model testing. Thus, the step 4: Testing the model until satisfied that it is suitable for the purpose will be done.

So far, many kinds of SD modeling testing methods have been advanced by different scholars. For example, Forrester & Senge (1979) in order to increase the confidence in the SD model to propose a wide of variety of tests that include tests of model structure, model behavior, and a model’s policy implication (Forrester & Senge, 1979). Hsieh (1980) recognized that model testing should be begin from the purpose of our model, so he proposed some sample methods such as model stability, pattern of oscillation, time phase and cycle length (Hsieh, 1980; Lin, 2000).

In this section, we will discuss the validity of our research model by three methods, including dimensional-consistency test, extreme-condition test and pattern of oscillation. The first two are advanced by Forrester & Senge and the last one is purposed by Hsieh.

4.3.1 Dimensional-consistency Test

The first and basic task among testing the model structure is checking that an equation is dimensionally consistent. By using the “Unit Check” function of Vensim DSS to test, the units of equations including stocks, flows, auxiliaries and constants are consistent during the process of modeling and simulation.

This result also represents that the model structure of this research is considered robust and no problem in dimensional consistency test.

4.3.2 Extreme-condition Test

The extreme condition test asks whether the model behave appropriately when the inputs take on extreme values such as zero or infinity (Lee, 2002). On the other word, by this test, we can check that the model is robust or not in extreme conditions. In this research, we will assume two situations to test: (1) the ratio of corporate R&D funding is equal to 25% which is the max in global pharmaceutical industry based on Berardi (2003) (see Table 2-3); (2) the ratio of corporate R&D funding is equal to zero.

In the proposed model, the learning and growth dimension absolutely depend on the corporate R&D funding. Once the ratio of corporate R&D funding is equal to 25%, there are more funding to hire more R&D staffs and the R&D capability also will quickly increase. However, in the condition about minimum ratio of corporate R&D funding, there are no funding to pay the salary that no R&D staffs can be hire. If there are no R&D staffs in this industry, the R&D area cannot be developed to accumulate R&D capability.

Figure 4-16 correctly reports the results.

One of the resources about new drug development funding is corporate R&D funding. In max condition about corporate R&D funding, there will be a substantial increase in new drug development funding. And there will be more new drugs appeared on the market (Figure 4-17). More new drugs will bring more domestic demand and more salves value (Figure 4-18). When corporate R&D funding is reduced to zero, the new drug development funding will be decreased, and then the amount of new drugs, domestic market demand and salves value are able to respect to decrease. As the above result of test, the

Figure 4-16: Extreme-condition Test in Learning and Growth Sub-system

Figure 4-17: Extreme-condition Test in Innovation Stage

Figure 4-18: Extreme-condition Test in Customer and Financial Sub-system

Total R&D Staff