You are on page 1of 4
Deparrwenr oF Sraristics BT April 1999 ‘Total marks = 30 Notes: Short answers are preferable to long answers (answers in “point form" are acceptable) -B 2 “A . . > : . = “Cc o ° 5 19 15 1, Consider a linear regression model that involves just one explanatory variable Y= fot AX te e~NOe) and suppose that the data contains a single unusual observation. The points labelled A, B and Cin the above figure above illustrate three possible ways an observation could be ‘unusual. Now consider three eases: for each ease the data set consists of the unlabelled points and one of the labelled points. For each of these three cases (a) Describe how the labelled observation is unusual (is it an outlier? a high leverage point? an influential point? all ree?) [B marks] (b) Explain what effect deleting the labelled point would have on the fitted model (How would fo, 81, and & be affcted?) [4 marks] 475.390 ‘Terms ‘Test Page 1 of 4 2. Example 3 in your course nots contains some tyze abrasion data. ‘The idea was to relate the amount of abrasion in a standard abrasion test to the hardness and tensile strength of the subber. Two trellis pots are given below for this dataset (a) Use these plots to brielly explain how abrasion loss (abloss) is related vo hardness and tensile strength, [marks] (b) Leis desirable to have low abrasion loss. Cleaely, indicate what combination of values for hardness and tensile strength result in low values of abloss, [2 marks] thes 475.380 ‘Terms Test 3. Creatinine clearance (clearance difficult to obtain because it requires a 24 hour urine election. Data {s an important measure of kidney finetion but is was collected by a kidney specialist to determine whether creatinine clearance can be predicted using ‘measurements of creatinine concentration in mg per declitre (cone), age in years (age). ‘and weight in ky (keSght) since these measurements are much easier to collect. Summary statisties were generated for the data using Splus > summary (kidney dt) clearance Min. $30.00 Min, tet gu Median genes ‘The following model was used. kidney. fit3¢-In(clearance > unary (kidney fi¢3) cone + I(conc"2) + age + weight, data coefficients: Value Std. Brror t value Pr(ltl) (Imercept) 151.2033 21-1118 7.1620 0.0000 cone -24.9245 73.2016 -2.6434 0.0011 Teonc*2) 15.465 7.7885 «1.9807 0.0575, ‘age -0.7924 0.1948 © 6.4941 0.0000, weight 0.7491 0.16K7 4.5125 0.0001 Residual standard error: 11.97 on 28 doge Multiple R-Squared: 0.8727 of freedom kidney dt) Festatietic: 47.97 on 4 and 28 degrecs of freedom, the p-value ie 3.806e-12 (a) Write down the theoretical fonn (ie, clearance = + ... ) for this model [2 marks] (b) What docs each of the following lines from the output indicate about the fitted ‘model? i, Y(come"2) 18.4265 7.7885 1.9807 0.0875, fi, Multiple R-Squared: 0.9727 [2 marks each] (6) Deseribe how the fitted model relates age to creatinine clearance (be more speviic than “creatinine clearance increases /decreases a8 age increases” ) [2 marks] (a) How much dilference in creatinine concentration does the fitted model predict be- tween a patient for has cone = 1 and cone = 2 (assuming age and weight are the same)? 475.380 ‘Terms Test [B marks] Page 3 of 4 4 (a) What docs an unusual value of the variance ratio statistic for an observation indi- cate? Approximately, what value of covariance ratiodo you expect for an observation that does not have a large effect on the regression analysis? [B marks] (b) What problem are variance inflation factors (VIFs) used to detect? Explain how VIP's are used diagnose this problem (What do the VIFs actually measure? What range of values can the VIPs take? What valués indicate that a probless is present?) [3 marks} 475.380 ‘Terms Test Page 4 of 4

You might also like