y i s 2 X 1 n i 1 1. Show that the least squares estimators can be written as n xx i x i 1 ns 2 X i 1 n ` px xqx i x i 1 pδ ij 1 n px i xq x j x

Transcription

1 Question 1 Suppose that we have data Let x 1 n x i px 1, y 1 q,..., px n, y n q. ȳ 1 n y i s 2 X 1 n px i xq 2 Throughout this question, we assume that the simple linear model is correct. We also assume that the noise variables ɛ i are Gaussian. 1. Show that the least squares estimators can be written as pβ 0 β 0 ` pβ 1 β 1 ` ˆ 1 n xx i x ɛ ns 2 i X ˆxi x ɛ i ns 2 X 2. Let mpxq β 0 ` β 1 x and pmpxq p β 0 ` pβ 1 x. Show that pmpxq mpxq ` ˆ 1 n ` px xqx i x ɛ ns 2 i X 3. Define δ ij to be 1 when i j and 0 otherwise. Show that the i-th residual (i.e. the residual at x x i ) can be written as e i ÿ j pδ ij 1 n px i xq x j x qɛ j ns 2 X 4. Given that ˆ Ere 2 i s σ n px i xq 2 ns 2 X show that Erpσ 2 s n 2 n σ2 where pσ 2 p1{nq ř n e2 i. And explain why the estimator s 2 n n 2 pσ2 is most commonly used, rather than pσ 2 p1{nq ř n e2 i. 1

2 Question 2 In a study of the impact of poverty on the standardized test scores students in a US school, data on 22 schools were collected. For each school the data comprised an average standardized math. score MATH, an average standardized reading score READING, and a measure of poverty (POVERTY, percentage of households below the national poverty line). The data were analyzed in R and produced the following output: 1 > head ( Fcat ) # List first six elements 2 MATH READING POVERTY > summary (lm( MATH POVERTY, data = Fcat )) 12 Coefficients : 13 Estimate Std. Error t value Pr ( > t ) 14 ( Intercept ) < 2e -16 *** 15 POVERTY e -06 *** Signif. codes : 0 *** ** 0.01 * Residual standard error : on 20 degrees of freedom 20 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 20 DF, p- value : 2.927e -06 (listing continued on the next page) 2

3 22 > summary (lm( READING POVERTY, data = Fcat )) 23 Coefficients : 24 Estimate Std. Error t value Pr ( > t ) 25 ( Intercept ) < 2e -16 *** 26 POVERTY e -08 *** Signif. codes : 0 *** ** 0.01 * Residual standard error : on 20 degrees of freedom 31 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 20 DF, p- value : 2.088e > summary (lm( MATH READING, data = Fcat )) 35 Coefficients : 36 Estimate Std. Error t value Pr ( > t ) 37 ( Intercept ) READING e -09 *** Signif. codes : 0 *** ** 0.01 * Residual standard error : on 20 degrees of freedom 43 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 20 DF, p- value : 1.763e Three separate analyses have been carried out. Summarize the results of the analyses in terms of the regression relationships between the variables in each analysis, assuming that all the simple linear regression model assumptions are correct. 2. If for MATH, s 2 X 1 n ř n px i xq , from the output, compute the sample correlation coefficient for the relationship between POVERTY and MATH READING and MATH Question 3 The data in the plot below record the results of a survey of 147 US University students who were asked to select the height (in inches) of their ideal spouse or life 3

4 partner. The specified value is recorded as IdealHt. Then the students own height was recorded as Height, as well as their gender, recorded in the study using a binary indicator (Gender) taking the values F -- recorded female -- and M -- recorded male. The data were stored in the R data frame idh and analyzed. Ideal height (in) Female Male Height (in) 45 > summary (lm( IdealHt Height, data = idh )) 46 Coefficients : 47 Estimate Std. Error t value Pr ( > t ) 48 ( Intercept ) XXXXXX < 2e -16 *** 49 Height *** Signif. codes : 0 *** ** 0.01 *

5 53 Residual standard error : on XXX degrees of freedom 54 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 145 DF, p- value : > summary (lm( IdealHt Height, subset =( Gender =='F '), data = idh )) 61 # Females only 62 Coefficients : 63 Estimate Std. Error t value Pr ( > t ) 64 ( Intercept ) e -08 *** 65 Height e -06 *** Signif. codes : 0 *** ** 0.01 * Residual standard error : on 70 degrees of freedom 70 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 70 DF, p- value : 2.474e > summary (lm( IdealHt Height, subset =( Gender =='M '), data = idh )) 77 # Males only 78 Coefficients : 79 Estimate Std. Error t value Pr ( > t ) 80 ( Intercept ) *** 81 Height XXXXXXXX *** Signif. codes : 0 *** ** 0.01 * Residual standard error : on 73 degrees of freedom 86 Multiple R- squared : , Adjusted R- squared : F- statistic : on 1 and 73 DF, p- value : 3.758e

6 91 92 > confint (lm( IdealHt Height, subset =( Gender =='F '), data = idh )) # Compute CIs % 97.5 % 94 ( Intercept ) Height > confint (lm( IdealHt Height, subset =( Gender =='M '), data = idh )) # Compute CIs % 97.5 % 98 ( Intercept ) Height Note that the R command subset=(gender== M ) means perform the analysis for those data for which Gender takes the value M (that is, analyze the males separately). 1. How many males were in the study? 2. There is an entry XXXXXX on line 48 - what is the correct numerical value to enter in that part of that coefficients table? 3. What is the value XXX on line 53? 6

7 4. What is the value of the sum of squares for the residuals, ř n e2 i, in the analysis of the Females only subset? What is the value of pσ What is the missing p-value marked XXXXXXXX on line 81? 6. Explain how the results on lines 49, 65 and 81 should be interpreted. 7. For the dataset subset=(gender== M ), based on the estimated coefficients, can you give an estimate of ErY X 0s? If yes, what is it (and show your work); if not, explain why not. 7

8 8. Note that the confint results reported between lines 92 and 99 report the 95% confidence intervals for the slope and intercept parameters for the relevant data subsets. On the basis of these results, what can an analyst conclude about the relationship between height of subject and selected ideal height of partner in females and males? Justify your answer. 9. For the dataset subset=(gender== M ), give a 95% confidence interval for β 1, (assuming all the model assumptions hold), using one of the following results, qt(0.975,df=70)= qt(0.975,df=73)= qt(0.975,df=145)= On the basis of these results, predict (showing working) the selected ideal height of partner for a female of height 62 inches, a male of height 70 inches. 8