Doing Statistics for Business Data, Inference, and Decision Making Marilyn K. Pelosi Theresa M. Sandifer Chapter 12 Multiple Regression Models 1 Doing Statistics for Business Chapter 12 Objectives Find the the regression equation for a dependent variable Y as a function of a set of independent variables, X1, X2,…Xk. Determine whether the relationship is significant. Determine which variables contribute to the 2 model and which do not. Doing Statistics for Business Chapter 12 Objectives (con’t) Analyze the results of a regression analysis to determine whether the model is appropriate. 3 Doing Statistics for Business The dependent variable, Y, is often referred to as the output variable , while the set of independent variables, X1, X2, … , Xk, are referred to as the input variables 4 Doing Statistics for Business The true relationship between the independent variable Y and the set of independent variables, X1, X2,…Xk, the multiple regression model, can be described by the equation: y = b0 +b1x1 + b2x2 + ... + bkxk + e 5 Doing Statistics for Business SUMMARY OUTPUT Regression Statistics Multiple R 0.866778185 R Square 0.751304423 Adjusted R Square 0.715776483 Standard Error 37957.7302 Observations 41 ANOVA df Regression Residual Total Intercept Avg. Household Income Total Instructional Expenditures per pupil Mean SAT Score Popultaion Diversity (% non-white) 1993/94 Avg. Violent Crime (per 1000) SS 1.52341E+11 50427624859 2.02768E+11 MS 30468171569 1440789282 F 21.14686162 Significance F 1.08463E-09 Coefficients Standard Error -140351.6689 112957.8302 t Stat -1.242513853 P-value 0.222307657 Lower 95% -369668.5358 5 35 40 Upper 95% Lower 95.0% 88965.198 -369668.5358 1.637671454 0.312118144 5.246960128 7.59702E-06 1.004037162 2.271305746 1.004037162 39.15871129 21.78140875 1.797804345 0.080839566 -5.059953319 83.3773759 -5.059953319 89.63285763 113.8778516 0.787096493 0.436523067 -141.5517541 320.8174694 -141.5517541 -47820.09741 62143.05497 -0.769516359 0.446749366 -173977.3601 78337.1653 -173977.3601 -1132.378569 2146.787097 -0.527475952 0.601191161 -5490.5934 3225.836262 -5490.5934 Figure 12.1 Computer Output from Excel 6 Doing Statistics for Business Regression Analysis Figure 12.1 (con’t) Computer Output from Minitab The regression equation is 1994 Avg. Value _of Home Sold = - 140352 + 1.64 Avg. Household_ Income + 39.2 Total Instructional_ Expenditur + 90 Mean SAT_ Score - 47820 Population Diversity_ (% non-white - 1132 1993/94 Avg. Violent _Crime (per 1000) Predictor Constant Avg. Hou Total In Mean SAT Populati 1993/94 S = 37958 Coef -140352 1.6377 39.16 89.6 -47820 -1132 StDev 112958 0.3121 21.78 113.9 62143 2147 R-Sq = 75.1% T -1.24 5.25 1.80 0.79 -0.77 -0.53 P 0.222 0.000 0.081 0.437 0.447 0.601 R-Sq(adj) = 71.6% Analysis of Variance Source Regression Error Total DF SS MS 5 1.52341E+11 30468171569 35 50427624859 1440789282 40 2.02768E+11 Source Avg. Hou Total In Mean SAT Populati 1993/94 DF Seq SS 1 1.46103E+11 1 3261836377 1 1202680560 1 1372121063 1 400872070 Unusual Observations Obs Avg. Hou 1994 Avg 1 49803 204800 8 30434 58200 21 196697 361100 38 84007 269926 39 87854 269926 Fit 110273 42311 397689 178710 178242 F 21.15 StDev Fit 13150 28127 30411 9319 13518 P 0.000 Residual 94527 15889 -36589 91216 91684 St Resid 2.65R 0.62 X -1.61 X 2.48R 2.58R 7 Doing Statistics for Business TRY IT NOW! Order Filling Finding the Multiple Regression Model A mail-order catalog company is looking at the time it takes to prepare an order for for shipping. In particular, the company is looking for the amount of time that is spent collecting the items ordered and packing them. In this operation, an employee (a checker) is given an order to fill. Items are located in bins in one of six different sections of the warehouse. The checkers move around the warehouse retrieving the items and 8 Doing Statistics for Business TRY IT NOW! Order Filling Finding the Multiple Regression Model (con’t) packing them into the shipping cartons. The company has looked at the operation in some detail and believes that three major variables are involved in the process: the number of items ordered, the number of different locations (sections of the warehouse) in which the items are located, and the experience level (in months) of the checker. Data are collected on 45 orders. A portion of the data is shown next: 9 Doing Statistics for Business TRY IT NOW! Order Filling Finding the Multiple Regression Model (con’t) Time Items Locations Experience (min) 9.3 4.4 4.4 5.6 4.9 8.8 1 1 5 3 11 14 6 3 2 3 1 3 (months) 8 13 3 5 4 5 10 Doing Statistics for Business TRY IT NOW! Order Filling Finding the Multiple Regression Model (con’t) The company wants to know how the time it takes to fill an order is related to the other three variables, so it decides to use a multiple regression model. Write down the equation of the regression model. Interpret the value of each of the coefficients of the model. 11 Doing Statistics for Business TRY IT NOW! Order Filling Finding Predicted Values The mail-order catalog company that is looking at the time it takes to prepare an order for shipping wants to see how well the model it has found predicts the time to fill an order. The data for six of the observations are shown next: 12 Doing Statistics for Business TRY IT NOW! Order Filling Finding Predicted Values (con’t) Time (min) 9.3 4.4 4.4 5.6 4.9 8.8 Items Locations 1 1 5 3 11 14 6 3 2 3 1 3 Experience (months) 8 13 3 5 4 5 13 Doing Statistics for Business TRY IT NOW! Order Filling Finding Predicted Values (con’t) Use the model to find the predicted time to fill an order for each of the six sets of input data you have. Compare the predicted results to the actual data. Do you think that this model does a good job of predicting the time to fill an order? Why or why not? 14 Doing Statistics for Business Source of Variation Regression Degrees of Freedom k Sum of Squares SSR Error n-k-1 SSE Total n-1 SST Mean Square MSR SSR k MSE = SSE n - k -1 F value MSR MSE Figure 12.2 ANOVA Table for Multiple Regression Model 15 Doing Statistics for Business TRY IT NOW! Order Filling Testing the Significance of the Model The mail-order company looking at factors related to the time to fill an order wants to know if its model is significant. Write down the hypotheses that it needs to test. Using the Excel output from your textbook, locate the values for MSR and MSE and their degrees of freedom. 16 Doing Statistics for Business TRY IT NOW! Order Filling Testing the Significance of the Model Use the values of MSR and MSE to calculate the value of the F statistic and compare it to the value in the table. At the 0.05 level of significance, what is the critical value for the test? What can the mail-order company conclude as a result of its test? Verify your answer by using the p value from the printout. 17 Doing Statistics for Business The Coefficient of Multiple Determination, R2, is a measure of the percentage of the variation in the dependent variable, Y, that can be accounted for by the complete set of independent variables, X1, X2,…Xk , in the model. 18 Doing Statistics for Business The Adjusted R2, is the value of the coefficient of multiple determination adjusted to reflect the number of variables in the model. 19 Doing Statistics for Business TRY IT NOW! Order Filling The Coefficient of Multiple Determination The mail-order company looking at factors related to the time to fill an order wants to know if the set of independent variables that it selected does a good job of accounting for the variation in the time SUMMARY OUTPUT to fill an order. Regression Statistics Multiple R 0.915838993 R Square 0.838761061 Adjusted R Square 0.82696309 Standard Error 1.021396967 Observations 45 20 Doing Statistics for Business TRY IT NOW! Order Filling The Coefficient of Multiple Determination (con’t) Using the output, find the value of the coefficient of multiple determination. If you were the manager of the company would you be satisfied with this model? Why or why not? Look at the value of adjusted R2. What do you think this value might be telling the company? 21 Doing Statistics for Business TRY IT NOW! Order Filling Testing Individual Regression Coefficients The mail-order company looking at factors related to the time to fill an order wants to know how the individual variables contribute to the model. It decides to look at the computer analysis again. Look at the coefficients of each of the three variables in the model and perform the appropriate hypothesis tests. 22 Doing Statistics for Business TRY IT NOW! Order Filling Testing Individual Regression Coefficients (con’t) Which variable(s) have nonzero coefficients? Which variable(s) have coefficients that are equal to zero? As a result of these test, what recommendation would you make? 23 Doing Statistics for Business Discovery Exercise 12.1 Find the Best Model The office of Institutional Planning at a university in the Far West is interested in understanding what factors influence the graduation rate, that is, the percentage of entering freshmen who actually graduate from the university. The university planners have collected data on 46 universities with traits similar to theirs and have selected three variables that they think are related to the graduation rate: 25th percentile combined SATs of accepted students, the acceptance rate (percentage of students who apply that are accepted by the university), and educational expenditure per student ($) by the university. 24 Doing Statistics for Business Discovery Exercise 12.1 Find the Best Model (con’t) A sample of the data is shown below: Rank/School Name Indiana University - Bloomington SUNY--Binghamton Allegheny University Univ. of California--Riverside Oregon State University University of Hawaii--Manoa New Jersey Inst. of Technology 1995 Actual Education expenditures Graduation Rate SAT/ACT 25th Acceptance Rate per student 59% NA 80% 9713 74% NA 40% 9080 52% 786 61% 33270 56% 910 78% 13403 53% 949 89% 10182 55% 960 65% 13360 63% 970 67% 14030 25 Doing Statistics for Business Discovery Exercise 12.1 Find the Best Model (con’t) A. Which variable do you think will have the greatest effect on graduation rate? Explain why you chose this variable. B. Find a simple linear model that predicts graduation rate using the variable that you think is most important. What is the value of R2 for your model? Is the model significant? C. Do this again using the other two independent variables. 26 Doing Statistics for Business Discovery Exercise 12.1 Find the Best Model (con’t) D. Which one-variable model do you think is the best? Why? E. How many different models with two independent variables could you find? List them all. F. Find the multiple regression models for all of the two-variable models. G. Which two-variable model do you think is best? Why? Does the model you think is best contain the variable form the best one-variable model? Would you expect it to? 27 Doing Statistics for Business Discovery Exercise 12.1 Find the Best Model (con’t) H. Find the multiple regression model that predicts graduation rate using all three independent variables. What is R2 for this model? Is it significant? I. Now, fill in the table in your textbook for each of the “best” models that you have found. J. If you were going to use a model to predict graduation rate, which of these three models would you choose? Why? 28 Doing Statistics for Business Model Building Techniques are methods used for identifying the best multiple regression model from a set of independent variables. These methods include: Forward Selection Backward Elimination Stepwise Regression All Possible Regressions. 29 Doing Statistics for Business TRY IT NOW! GPA Choosing the Independent Variables Many studies have been done on what factors are related to a college student’s grade point average (GPA). These studies often focus on pre-college factors, such as high school performance, SAT or ACT scores and general socioeconomic factors such as family income and race. Students know that once they are in college, many other factors influence their GPA. 30 Doing Statistics for Business TRY IT NOW! GPA Choosing the Independent Variables (con’t) You are going to try to find a model relating current factors to GPA. Make a list of all variables that you think are related to a college student’s GPA. From this list, identify what you think are the five most important variables. How would you go about gathering the data you need to do your study? 31 Doing Statistics for Business TRY IT NOW! Order Filling Forward Selection The mail-order company wants to use a standard method to find the best model from its set of three variables. It decides on forward selection because this variable is the easiest to understand. The relevant portions of the computer output are shown in your textbook. What variable is added on the first step of the procedure? What is its coefficient? What is the value of R2 for the first model considered? 32 Doing Statistics for Business TRY IT NOW! Order Filling Forward Selection (con’t) What variable is added on the second step? What is its coefficient? What does R2 change to after this step? Does the coefficient of the first variable change when the second variable is added? If so, what is the new value? Write down the final model. 33 Doing Statistics for Business TRY IT NOW! Order Filling Testing Individual Regression Coefficients The mail-order company looking at the model for order filling decides to use stepwise regression to see if it finds a model that is different from the one found using forward selection. From the Minitab output in your textbook, how many iterations did the procedure take? Which variable was added to the model first? What is its coefficient? 34 Doing Statistics for Business TRY IT NOW! Order Filling Testing Individual Regression Coefficients (con’t) Which variable was added second? What is its coefficient? Did the coefficient for the first variable change? If so, what is its coefficient after the second variable is added? What is the is the final model from the stepwise procedure? 35 Doing Statistics for Business TRY IT NOW! Order Filling All Possible Regressions The mail-order company is pretty sure that it has identified the best model, but it decides to use all the possible regressions method to make sure that it is not missing something. Based on the output in your textbook, is there a one-variable model that is best on all criteria? If so, what is it? What is the best two-variable model? 36 Doing Statistics for Business TRY IT NOW! Order Filling All Possible Regressions (con’t) Is there any benefit from moving to a three-variable model? Why or why not? What model do you recommend the company use? Why? 37 Doing Statistics for Business TRY IT NOW! Order Filling All Possible Regressions Now that has a model it likes, the mail-order company would like to make sure that the model does not violate any assumptions of the multiple regression model. The company creates a set of residual plots, which are shown below: Number of Items Residual Plot 3 2 1 0 -1 0 -2 Residuals Residuals Locations Residual Plot 2 4 Locations 6 3 2 1 0 -1 0 -2 5 10 15 20 Number of Items 38 Doing Statistics for Business TRY IT NOW! Order Filling All Possible Regressions (con’t) Look at the plot of residuals versus the locations. Does it appear that there is a problem with the assumption of equal variances? Look at the plot of residuals versus the number of items. Does it appear that there is a problem with the assumption of equal variances? 39 Doing Statistics for Business TRY IT NOW! Order Filling All Possible Regressions (con’t) The company also created a normal probability plot of the residuals: Normal Probability Plot Time 15 10 5 0 0 20 40 60 80 100 Sample Percentile From this plot, what can you say about the assumption of normality? 40 Doing Statistics for Business The Cook’s Distance method compares the values of the regression coefficients with all observations to the values when the ith observation is removed from the model. 41 Doing Statistics for Business A multiple-regression model has Multicollinearity when variables in the set of independent variables are correlated with each other. 42 Doing Statistics for Business TRY IT NOW! Order Filling Checking For Multicollinearity The mail-order company wants to check to see if the model it is thinking about using has any problems with multicollinearity. It calculates the correlation between the two variables that are in the final model and finds that the correlation is – 0.141. Based on this information do you think that multicollinearity is a problem in their model? Why or why not? 43 Doing Statistics for Business Multiple Regression Models in Excel The tools used in Excel for multiple regression models are the same one that are used for the simple linear model. The only difference is that the data range for the X variables will cover more than one column. It is very important, however, that the X variable columns be adjacent to each other. If they are not, you will have to rearrange the worksheet before you do the analysis. 44 Doing Statistics for Business Stepwise Regression with KaddStat From the Kadd menu select Regression and Correlation > Forward Stepwise and the dialog box shown in Figure 12.3 opens. The dialog box is identical to the one for Single/Multiple regression with two additional inputs. 45 Doing Statistics for Business 46 Doing Statistics for Business Just above the where you indicate which graphical output you want is a section labeled Select one:. This allows you to indicate whether you want to have only the final model output or whether you want each step output. In addition, you need to specify the p value to be used for adding and dropping variables from the model. Kaddstat uses the same p value for both. 47 Doing Statistics for Business Chapter 12 Summary In this chapter you have learned: Modeling is an iterative process for which there is no single correct answer. There are several steps to the modeling process: Identifying potential independent variables Collecting data Finding a potential model 48 Doing Statistics for Business Chapter 12 Summary (con’t) Once a potential model is found, the process of Model Building takes place to find a “best” model. The objective is finding a model that does an acceptable job of explaining or predicting the dependent variable with as few independent variables as possible. 49 Doing Statistics for Business Chapter 12 Summary (con’t) Some model-building techniques used are: Forward Selection Backward Elimination Stepwise Regression All Possible Regressions 50 Doing Statistics for Business Chapter 12 Summary (con’t) Once the “best” model is identified, and before it can be used for decision-making purposes, it must be checked for problems such as: Violation of Assumptions Influential Observations Multicollinearity 51