Statistics 350 Lecture 24 Today • Last Day: Exam • Today: Start Chapter 9 Model Building Process • Read 9.1-9.2…important ideas • Four Steps to Model Building: 1. Data collection and preparation: • What sort of data is being collect…source of data? What sort of study has been conducted? • Response and explanatory variables? • Much of the work Masters statisticians do involves data preparation…look for proablems Model Building Process 2. Reduction of number of explanatory variables: • Researchers are wise to gather all they can, since any unmeasured phenomena get lumped into random error. • Key things are: • Need number of variables in final model to be large enough so that your model provides an adequate approximation to reality, but small enough to be practical for use • Do not omit anything that the researcher considers vital just because it is highly correlated with another variable Model Building Process 3. Model refinement and selection: • Once step 2 is done, then it is time to use techniques learned in Chapters 2-7 (transformations, interactions, plots, tests, …) • Try to develop one or more plausible models that could potentially be considered as final answers Model Building Process 4. Model validation: • Not always possible, but it's best not to claim to have discovered a new scientific principle before you try it out on an independent data set Example • Investigators studied physical characteristics and ability in 13 football punters • Each volunteer punted a football ten times • The investigators recorded the average distance for the ten punts, in feet • In addition, the investigators recorded five measures of strength and flexibility for each punter: right leg strength (pounds), left leg strength (pounds), right hamstring muscle flexibility (degrees), left hamstring muscle flexibility (degrees), and overall leg strength (foot-pounds) • From the study "The relationship between selected physical performance variables and football punting ability" by the Department of Health, Physical Education and Recreation at the Virginia Polytechnic Institute and State University, 1983 Example • Variables: • • • • • • Y: Distance traveled in feet X1: Right leg strength in pounds X2: Left leg strength in pounds X3: Right leg flexibility in degrees X4: Left leg flexibility in degrees X5: Overall leg strength in pounds Example Example Model Summary Model 1 R .902a R Square .814 Adjusted R Square .682 • When we looked at this example in lecture 22, we saw that the regression equation was significant, but that the individual t-tests identified none of the variables as important • Why? • We then went through a process of selecting variable to include in the model…seemed a bit ad-hoc • Chapter 9 deals with the problem of model selection Std. Error of the Estimate 14.64982 a. Predictors: (Constant), X5, X4, X2, X1, X3 ANOVAb Model 1 Regres sion Residual Total Sum of Squares 6590.987 1502.321 8093.308 df 5 7 12 Mean Square 1318.197 214.617 F 6.142 Sig. .017a a. Predic tors: (Constant), X5, X4, X2, X1, X3 b. Dependent Variable: Y a Coeffi cients Model 1 (Const ant) X1 X2 X3 X4 X5 Unstandardized Coeffic ients B St d. Error -29.580 65.700 .279 .456 .070 .484 1.241 1.449 -.395 .745 .224 .131 a. Dependent Variable: Y St andardiz ed Coeffic ients Beta .245 .062 .373 -.131 .412 t -.450 .611 .144 .857 -.531 1.714 Sig. .666 .561 .890 .420 .612 .130 Criteria for Model Selection • So how do we deal with Step 2 of our procedure? • Would like tests that identify some variables as unimportant • Suppose have P-1 potential explanatory variables and fir all possible sub-sets of variables • There are such models • So, in the football example, there are possible subsets Criteria for Model Selection • Goal is to choose a “good” set from these variables that explain the data well • So, may wish to choose a few possible models that appear “good” • What is good? • There exist criteria that assess the relative goodness of models Criteria for Model Selection • Criteria: R2p • • • • Is the coefficient of determination (SSR/SSTO) p represents Good models have What happens if you add more variables? • How to use? Criteria for Model Selection • Criteria: R2a Criteria for Model Selection • Mallows Cp Criteria for Model Selection • AIC Criteria for Model Selection • PRESS: