Các dữ liệu thu thập phải được tuân theo phân tích theo cách như vậy để làm sáng tỏ mô hình. Đối với vấn đề khó khăn, việc xác định các thông số phù hợp nhất là một thủ tục mang lại lợi ích cho rất nhiều từ kinh nghiệm, trực giác, sự kiên trì, hoài nghi, và lý luận khoa học. Một câu trả lời tốt đòi hỏi phải có các ước tính tốt ban đầu. | COMPLEX REACTION PATHWAYS 35 properties of the model. The data gathered must be amenable to analysis in such a way as to shed light on the model. For difficult problems the determination of best-fit parameters is a procedure that benefits greatly from experience intuition perseverance skepticism and scientific reasoning. A good answer requires good initial estimates. Start the minimization procedure with the best possible initial estimates for parameters and if the parameters have physical limits specify constraints on their value. For complicated models begin model fitting by floating a single parameter and using a subset of the data that may be most sensitive to changes in the value of the particular parameter. Subsequently add parameters and data until it is possible to fit the full model to the complete data set. After the minimization is accomplished test the answers by carrying out sensitivity analysis. Perhaps run a simplex minimization procedure to determine if there are other minima nearby and whether or not the minimization wanders off in another direction. Finally plot the data and calculated values and check visually for goodness of fit the human eye is a powerful tool. Above all care should be exercised if curve fitting is approached blindly without understanding its inherent limitations and nuances erroneous results will be obtained. The F-test is the most common statistical tool used to judge whether a model fits the data better than another. The models to be compared are fitted to data and reduced X2 values X2 obtained. The ratio of the Xv2 values obtained is the F-statistic X 2 a - b where df stands for degrees of freedom which are determined from df n - p - 1 where n and p correspond respectively to the total number of data points and the number of parameters in the model. Using standard statistical tables it is possible to determine if the fits of the models to the data are significantly different from each other at a certain level of
