Tham khảo tài liệu 'engineering statistics handbook episode 5 part 13', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | . Accounting for Non-Constant Variation Across the Data Example of WLS Using the Power Function Model The power function model for the weights mentioned above is often especially convenient when there is only one predictor variable. In this situation the general model given above can usually be simplified to the power function t - i- t which does not require the use of iterative fitting methods. This model will be used with the modified version of the Pressure Temperature data plotted below to illustrate the steps needed to carry out a weighted least squares fit. Modified Pressure Temperature Data 300 250 200 150 100 50 30 60 70 2 0 40 50 Temperature Defining Sets of Approximate Replicate Measurements From the data plotted above it is clear that there are not many true replicates in this data set. As a result sets of approximate replicate measurements need to be defined in order to use the power function model to estimate the weights. In this case this was done by rounding a multiple of the temperature to the nearest degree and then converting the rounded data back to the original scale. Temperaturerep 3 roundiTemperature 3 This is an easy way to identify sets of measurements that have temperatures that are relatively close together. If this process had produced too few sets of replicates a smaller factor than three could have been used to spread the data out further before rounding. If fewer replicate sets were needed then a larger factor could have been used. The appropriate value to use is a matter of judgment. An ideal value is one that doesn t combine values that are too different and that yields sets of replicates that aren t too different in size. A table showing the original data the rounded temperatures that define the approximate replicates and the replicate standard deviations is listed below. http div898 handbook pmd section4 10 of 14 5 1 2006 10 22 20 AM . Accounting for Non-Constant Variation Across the Data .
