• Home
  • Textbooks
  • Statistical Methods for the Social Sciences
  • Model Building with Multiple Regression

Statistical Methods for the Social Sciences

Alan Agresti, Barbara Finlay

Chapter 14

Model Building with Multiple Regression - all with Video Answers

Educators


Chapter Questions

02:37

Problem 1

Refer to Table 9.1, deleting the observation for DC With $Y=$ violent crime ratc and the five predictors as explanatory tariables (all except inurder 1 ate) and using $\alpha=10$ in tests. select a model using (a) backurard elimination. (b) fol ward selection. Jiterpret the model selected.

Raymond Matshanda
Raymond Matshanda
Numerade Educator

Problem 2

Refer to Table 9.1, excluding the obscrvation for D C L.et $Y=$ muder rate. For the five predictors in that table (excluding tiolent crime ratc), the $r$ test ol independence for the bivariare model has $P$-valuc below 05
a) Fil the multiple regression mode,, using all five predictors. Are the $\mu_{\text {-ralucs for the }}$ partial lests all significant? Explain why rccults of these tests nay differ from those in the separate tests of independence
b) Use bachu ard elimination to select a model. deleling a variable if it does not have $P$. value below 10 Interpres.
c) Use forward selection to select a model. udding a vanaahle it it has $P$-talue below 10 . Interprct.
d) Use stepwise regression with the . 10 level fot significance. Interpiet.
e) Compare results of the thee selection procedures How is it possible that a varable (percent with a high school education) can be the first variable dropped in (b) yel the secoud added in (c)?

Check back soon!

Problem 3

Refer to the previous problem Now include the DC observation
a) Use backw ard elimination. and cumparc ıesults to part (b) above.
b) Use forward selection, and compare results to part (c) above.
c) What does this exercise suggest about how influential outlicrs can be on the 1esults of autonatic selection piocedures?

Check back soon!

Problem 4

Refer to Problem 14 2(h) Use buckward climination again with the variables chosen in (b) and theil interactions Does the resulting model make sense?

Check back soon!

Problem 5

Refer to Example 11.2 on $Y=$ mental impairment, $X_1=$ life events. and $X_2=$ SES. a) Slow that forw ard selection with $X_1, X_2, X_3=X_1 X_2, X_4=X_1^2$, and $X_4=X_1^2$ and the $\alpha=.10$ level for inchusion selects only $X_1$ dnd $X_2$ for the model.
b) Lise backu aid elimination. What is the final model? Interpret.
c) Use the $C_p$ index to describe the fit provided by each model consideled in the process in (b).

Check back soon!
01:15

Problem 6

Use bach wrard elimination with the hone salec data of Table 94 , using as candidales the four explanatory variables and all their interaction and quadratic (square) terms What model do you end up with? Is this a reasonable model? Explain

Raymond Matshanda
Raymond Matshanda
Numerade Educator
02:37

Problem 7

Problem 13.6 showed thal for the home sales data. a single ohcervation has a large impact on whether an inleraction term seems needed in the model. Lel's check whether that observation affects results of selection procedures Ising Icgression software after deleting that observ ation from the data set. conduct cithes bach u und elimination or forward selection with the vanables in Example 14.1. Compare results

Raymond Matshanda
Raymond Matshanda
Numerade Educator
00:34

Problem 8

Figure 14.14 is a SAS plot of the residuals versus the piedicted values for the analysis of covariance model discussed in Example 13.1 ıelating income to education and racialethnic goup. What does this plot suggest?

Emily Himsel
Emily Himsel
Numerade Educator

Problem 9

Reter to the nodel for housing price selected in Example 14.1.
a) Study the studentized residuals. and show that only one is unusually large. What docs this reflect?
b) Siudy the hat \alues. Which three observations have the greatest leverage. and hence the predictor values with the greatest potential to affect the fil of the regression model?
c) Study the DFFTTS values. W'hich three observations secm to have much more influence on the fitted values than any others?
d) Study the DFBEIAS for the $S$ predictor. Which three observations have the greatest influence on the predicted partial effect or size of honce?
e) Study the DFBETAS for the $B A$ and $N$ predictors. Are the three observations highlighted above as influential for these predictors? Explain.
f) Refit the model without the three highly influential observations. Compare the prediction equation, standard errors. and $R^2$ to the fit for the complete data set. Summarize the influence of this set of influentaal observations

Check back soon!

Problem 10

Refer to Problen 917 and Table 9 13. Table 14.8 chows a SAS computer printoul of various diagnostics Irom filling the muluple regression model relating birth rate to literacy and women's economic activity (delelung Germany. South Africa. and Vjetnam). Figure 1415 plots the residuals against the predicted values
a) Construct a histograin or stem and leaf plot of the residuals Are there any appareit oulliers?
b) Study the plot of the residuals in Figure 14.15. Does it suggest any lack of fit or unusual observations?
c) Study the studentized residuals. Are therc any apparent outliers?
d) Study the hat values. Which, it any, observations seem to have noticeable leverage for affecting results?

Check back soon!

Problem 11

Based on thc answers in the previous exercise. remove an obsen ation that seems polentially influential to you, and re-fit the model. Is the new fit different in any sabstantive way?

Check back soon!

Problem 12

Refer to Table 9.1. and fit the lincar regrcssion model relating $Y=$ violent crime rate to $X=$ percent melrupolitan for all 51 observations
a) Find the piediction equation. Using a stem and leaf plot or a histogram. plot the residuals. Interpret.
b) Plot the residuals againsl the predictor Interpret.
c) Refit the line without the outlier, and compare resuls.

Check back soon!

Problem 13

Refer to the previous exercise, and use software to obtain ıegression diagnostics
a) Swdy the studentized residuals. Arc there any clear outliers?
b) Study the lat values. Are there any observations with noticcable leverage?
c) Based on the answers in (a) and (b). docs it seem as if any observ ations may bc particularly inffuential? Explain.
d) Siudy the DFFITS values. Which. if any. observations have a strong influence on the fitted values?
e) Study the DFBETAS values. For each term. which if any observations have a strong influence on tle parameler cstimate?
f) Renove the observation thal seems most influential to you. and refit the model is the fit different in an! substantive way?

Check back soon!
02:27

Problem 14

Refer to Example 111. Fit the multiple regression model discussed there. for the data from Problent 9.24 Plot the residuals against eacl predictor and / or against the predicted values Do the plots show any irregulanties?

Jameson Kuper
Jameson Kuper
Numerade Educator

Problem 15

Refer to Table 9.1. Let $Y=$ violent crime rate. Fit the model to the 51 observations with percentage in poverty and percentage of siogle-parent families as predictors.
a) Repori the prediction equation.
b) Constuct a stem and ledf plot or a histogram of the residuals. Interpret.
c) Plot the residuals agrainst the predicted valucs. Interpret.
d) Plot the residuals against percentage of single-parent families Interprel
e) Refit the model without dic D.C. observation. and discuss the changes in the model fit and residual patterns

Check back soon!

Problem 16

Refer to the previous exercise. and use software to obtain regression diagnostics
a) Based on hat values and studentized residuals, does it seem as if any observations may be influential? Explain.
b) Sludy the DFFITS values Which, if any, observations have a strong influence on the fitted values?
c) Study the DFBETAS values. For each predictot, which if any observations have a strong influence on the parameter estimates?
d) Remove the observation that seems nost influential to you. and refit the model. Is the fit different in any substantive way?

Check back soon!

Problem 17

This problem shows that multicollinearity also affects precision of estimation of partial correlations.
a) Suppose the true correlations are $\rho_{X_1 X_2}=.85, \rho_{Y X_1}=.65$. and $\rho_1 X_z=.65$. Show that $\rho_{Y x_1, x-}=\rho_\gamma x_2 \delta=.244$.
b) In a sample. $r_{X_1 \lambda_1}=.9$. $r_{Y X_1}=.7$ and $r_1 x_2=.6$. Unless the sample is very large, these are well within the linits of sampling enor for the true valucs Slow that $r_r x_1 x$, $=$ .46 and $r_{Y X_2 \lambda_1}=-.10$ Note how small differences in $r_{Y X_1}$ and $r_{Y X_2}$ yield large differences in partial correlations when multicollineanity exists. (This illustrates that partial correlations have large standard errors when multicollinearity exists. For these values, an unwary observer mught conclude that the partial effects of $X_t$ and $X_2$ have opposite signs and that the partial effect of $X_t$ is much stronger. when in fact they are identical in the population of interest.)
c) For companson. compute the purtial correlations in (b) when $r_{Y X_1}=.7$ and $/ y X=.6$, but (i) $r_{X_1 x_2}=0$. (ii) $r_{X_1 X_2}=.6$.

Check back soon!
06:08

Problem 18

$Y=$ height. $X_1=$ length of left leg, and $X_2=$ length of right leg are measured for a sample of 100 adults The model $E(Y)=\alpha+\beta_1 X_1+\beta_2 X_2$ is fitled to the data. and neither $H_0: \beta_1=0$ nor $H_0$. $\beta_2=0$ is rejected.
a) Does this imply that length of leg is not a good predictor of height? Why?
b) Does this imply that $H_0: \beta_1=\beta_2=0$ would not have a small $P$-value? Why?
c) Suppose $r_{Y x_1}=.901, r_{Y x_2}=.902$. and $r_{X_3 x_2}=.999$. What model would you expect to sclect, using forward selection and the predictors $X_1$ and $X_2$ ? Why?

Lucas Finney
Lucas Finney
Numerade Educator
02:45

Problem 19

Refer to Tables 11.5 and 11.7. Note that $X_1$ and $X_2$ lose ther significance after entering the interaction term, even though that term is not significant
a) Explain why this happens
b) Rerun the interaction model, after centering the predicior scoles about their mean. that is. take $X_1^{\prime \prime}=X_1-44.425$ and take $X_2^*=X_2-56.60$. Note that now the estimates and their standard errors resemble those foi the no-interaction model. This is a useful way of dealing with nulticollineanıy for models that have intelaction or quadratic terms

Khoobchandra Agrawal
Khoobchandra Agrawal
Numerade Educator
00:23

Problem 20

Shecch the follou ing mathematical functions on the samc set of axes, for values of $X$ between 0 and 4 .
a) $\bar{Y}=10+4 X$
b) $\hat{Y}=10+4 X+X^2$
c) $\hat{Y}=10+4 X-X^2$
d) $\hat{Y}=10-4 X$
e) $\hat{Y}=10-4 X+X^2$
f) $\hat{Y}=10-4 X-X^2$
g) $\hat{Y}=10(1.50)^x$
h) $\hat{Y}=10(.50)^X$
For the quadratic models, use thesc curves to descnbe how the coefficients of $X$ and $X^2$ affect theil shape.

Allison Knapp
Allison Knapp
Numerade Educator
06:24

Problem 21

Refer to the housing dada in Table 9.4 The quadratic model relating selling price to size of house has fit $\hat{Y}=-2.04+49.34 S+6.74 S^2$.
a) Interpret the coefficients of this cquation. What shape does it have?
b) Find the predicted selling price for homes with (i) $S=1$, (ii) $S=2$, tiii) $S=3$. Explain why the effect of a one-unit increase in $S$ increases as $S$ increases.
c) Find the $S$ value for which thus curve takes its manimum. Note that the curve increases oser the entire runge of possible house sizes.
d) Using size as a straight-line predictor in a bivariate model, $r^2=.808$, whereas $R^2=$ 815 for the quadratic model. Do you think thai the degree of nonlineanty is major, or minor? Do you think that the degree of linear association is strong, or weak?

Robin Corrigan
Robin Corrigan
Numerade Educator

Problem 22

Table 14.9 shows results of fitting models to the housing data in Table 9 4. using number of bedrooms as a predictor of selling price.

Check back soon!
01:32

Problem 23

Refer to the previous exercise. Fit the exponenual regression nodel, and interpret. Find the predicted selling price when the number of bedroonss equals (i) 2 , (ii) 3 , (iii) 4 , and compare to the predictions with the other models.

Shu Naito
Shu Naito
Numerade Educator

Problem 24

. Refer to Table 9.13 l.et $Y=$ binh rate and $X=$ women's economic activity
a) Fit the straighl-line tegression model. Interpret coefhcients.
b) Fit the quadratic regression model. and interpret paranete1s.
c) Using the quadiatic fil. find the $X$-value at which predicted hirth rate takes its minimum valuc Is the prediction equation decreasing over the entirc range of observed $X$-values?
d) Does the quadratic model provide a much improv ed fit over the linear model? Answer (i) descnptively, by compaing $R^2$-1 alues, (ii) inferentially, by conducting a I test for the quadıatic effec. (iii) graphically. by plorting the two fits thıough the scatter diagram.

Check back soon!
01:21

Problem 25

Refer to the previous example. Fit the exponential regiession model, and interpret the parameter estimates Describe the estimated effect of a 10 unit increasc in economic activity.

James Kiss
James Kiss
Numerade Educator
02:36

Problem 26

Refer to Example 91 . Table 91. and Figure 9.4. Though inference may not be relevant for these data, we use them to illustrale how a single observation can be highll) influential in detennining whether to assume a nonlincal relationship
a) Using all 51 observalions. fi the quaduatic model between murder rate and percentage in povery Interpuer the estimales.
b) Refer to (a). Test whether the quadratic term is needed in the snodel Report the $P$. valuc, and interpret.
c) Now refit the quadratic inodel deleting the observation fon D.C. Compare the estumates to (a), and interpret.
d) Refer to (c). Test whethel the quadratic term is needed in the model. Report the $P$. ialuc. and interpret.
e) Compare results of the tesis in (b) and (d), and note how a single observation can have a large impact on the fit of a quadratic model Show how you would be wamed of this by infinence diagnostics for the fit in (a).
f) Show the resull of the $I$ test for the coefficient of $X$ in the model in (c) Do the results of this test and the one in (d) imply that poverty does not affect murder rale? Explain.

Tyler Moulton
Tyler Moulton
Numerade Educator
01:24

Problem 27

Refer to the previous example Fit the exponential regression model to the full data set, and interpıct the parameter estimales.

AG
Ankit Gupta
Numerade Educator

Problem 28

Table 14 I0 shous the results of filling two models to 54 observarions on $Y=$ mental health score. $X_1=$ degree of social interaction. and $X_2=$ SES The vanables $X_1$ and $X_2$ are measured on scales of $0-100$. and large! $Y$-scores represent betcer mental health. The saliable syinbol $\mathrm{X1}^{\top}{ }^r 2$ represents $X_1^2$, and $\mathrm{X} 1{ }^* \mathrm{X} 2$ repiesents $X_1 X_2$.

Check back soon!

Problem 29

Refer to Problem 14.12.
a) What does the plot of the residuals against the predictor suggest?
b) Using a gamma GLM, fit the linear model Compare the parameler estimates and the standard error of the slope to results obtained using least squares. Interpret.
c) Compare the two fits to the least squares fit of the model with the outlier deleted. Which fit seems more robust to the effect of the outlier?

Check back soon!

Problem 30

Refer to Example 14.4 on bird rates. To allow for greater variation at higher values of the mean. use the gamma GLM with identity link.
a) Fit the gamma linear model. Compare estimates to those obtained using least squares (i.e., the GLM with normal random component)
b) Fit the gamma quadratic model. Compare estimates to those obtained using least squares. Find the GNP value at which predicted birth rate takes its minimum value.

Check back soon!
01:22

Problem 31

Draw rough sketches of the following matliematical functions on the same set of aixes, for values of $X$ berween 0 and 35 .
a) $\hat{Y}=5(1.02)^X$. $(\hat{Y}=$ wolld population size in billions $X$ ycars from now, if there is a $2 \%$ rate of growth every year.)
b) $\hat{Y}=5(1.04)^x$. (What docs this represent?)

Carson Merrill
Carson Merrill
Numerade Educator
01:14

Problem 32

Consider the formula $\hat{Y}=4(2)^x$.
a) Calculate the $\hat{Y}$-values for integer values of $X$ between 0 and 5 , and graph the function.
b) Plot log $\hat{Y}$ against $X$. What is the intercept and what is the slope of this line?

James Kiss
James Kiss
Numerade Educator
13:44

Problem 33

Table 14.11 presents, for whte men $m$ the United Stares. the number of deaths per thousand individuals of a fixed age within a period of a year Lel $X$ denole agc and $Y$ denote death rate
a) Plot $X$ against $Y$ and indicate whether a linear model seems reasonable.
b) Plot $X$ against $\log Y$ What does this plot suggest about the , elationship between death rate and age?
c) Using generalzed linear models, find the prediction equation for the model $\log [E(Y)]=$ $\alpha+\beta X$.
d) Find the prediction equation for $Y$. Interprel the parameter estimates. Obtain the six predicted values for $Y$ and plot this prediction equation on the graph from (a).

Victoria Dollar
Victoria Dollar
Numerade Educator
02:25

Problem 34

Consider the birth rate data in Table 14.5 .
a) To check whether an exponential regression model seems appropnate. plot the log hirth rate values against GNP and report the correlation between them Compare to the correlation between birth rate and GNP.
b) Using GLM software. fit the exponential regression model. Interpret the effect of GNP on birth rate.
c) What advantages does the exponential model have over the quadratic model?

AG
Ankit Gupta
Numerade Educator

Problem 35

Refer to the WW'W data set (Problem 1.7).
a) Using software, conduct and interprel a regression analysis using $Y=$ poitical ideology, selecting predictors from the lariables in that filc Prepare a report describing the analyses and diagnostic checks that you conducted, and indicate how you selecied a final model literprel results.
b) Repear the analysis, using $Y=$ college GPA.

Check back soon!

Problem 36

Refer to the data filc crcated in Problem 17 . For the models you fitted in Problems 1125 and 13.12, use methods of this chapter to check model adequacy Interpret and summarize your findings

Check back soon!
View

Problem 37

Refer to the model used to predict income in Example 13.1. Conduct an analysis of residuals and of influence diagnostics, and summanze your findings.

Shu Naito
Shu Naito
Numerade Educator
01:15

Problem 38

Refcr to Table 111 and Example 11.2 Allowing for potential nonlinear relationships and interaction. and checking residuals and influence diagnostics. use the methods of this chapter to find a suitable prediction equation for $y=$ incntal impairment.

Raymond Matshanda
Raymond Matshanda
Numerade Educator
04:31

Problem 39

Table 14.12 shows population size of Florida, by decade. from 1830 to 1990. Analyze these data.

Carson Merrill
Carson Merrill
Numerade Educator
03:12

Problem 40

Refer to Table 913 Úsing nethods of this chapter. find a good prediction equation relating $X=$ per capita GNP $10 Y^{\prime}=$ life expectancy (Hinr Try a transform of $X$. such as $\log X$ or $1 / X$.)

Jameson Kuper
Jameson Kuper
Numerade Educator
01:15

Problem 41

Refer to Table 9.13. L sing nethods of this chapter, analy $\angle$ these data, finding a good prediction equation for birth 1 ate Explain heu yuu selected a anabies for the model and how you handled the missing datd. (Most software uses lismise deletion, deleting an observation from the list if it is mussing data on any of the varialles. New and better methods have been developed recently: see R Littlc and D. Rubin, Sociological Methods and Research, Vol. 18. 1989. pp 292-326.)

Raymond Matshanda
Raymond Matshanda
Numerade Educator

Problem 42

Refer to Problem 9 24. Using methods of this chapter, find a good model for predicung crime rale. In your report. show how you checked for ponlinearity, considered poten-tial multicollinearity, checked the results of stcpwise procedures, analyzed residuals, and considered influential observations Interpret your final model.

Check back soon!

Problem 43

Refer to Problem 14.26. Now use violent cnme rate as the response vanable Use influence diagnostics to analyze the fil of the quadratic model to the 51 obset vations on violent crime rate and poverty rate. Is the simpler. straight-line, inodel adequate?

Check back soon!
15:09

Problem 44

In multiple reguession, the standard error of the estimator $b_j$ of the coefficient $\beta_j$ of $X_j$ equals $\hat{\sigma} / s, \sqrt{(n-1)\left(1-R_i^2\right)}$. Where $\hat{\sigma}$ is the estimated conditional standard doviation. $s$, is the sample standard deviation of $X_j$ and $R_j$ is the inultiple conclation from the regression of $X_j$ on the other pecdictors The quantity $1 /\left(1-R_2^2\right.$ ) is called a variance inflation factor, since it replesents the multiplicative increase in the vanance due to $X_j$ being correlated with the other predictors. Using this formula, explain how precision of cstumated regression cocfficients is affected by:
a) Multicollineanty
b) The conditional vanability of the response variable.
c) The variability of the explanatory variables.
d) The sample size

Oluwadamilola Ameobi
Oluwadamilola Ameobi
Numerade Educator

Problem 45

Give an example of a Iesponse variable and a pair of explanatory variables for wiuch an automated variable selection procedue would probably produce a model with only one predictor. Explain.

Check back soon!

Problem 46

A sociologist \& first reaction upon studying automated variable selection routines was that they had the danger of leading to "crass cmpiricism" in theory building. Frons a theoretical perspective. describe the dangers with such methods. Whal guidelines would you suggest for avoiding these problems?

Check back soon!

Problem 47

Give an example of two variables you expect to have a nonlinear relationship Describe the pattern you expect lor the relationship and explain how 10 model that patiem.

Check back soon!
02:01

Problem 48

You invest $\$ 1000$ in a savings account with interest compounded annually at $10 \%$.
a) How much money do you have after $X$ jears?
b) Hou many years does it take your savings to double in size?

Sheryl Ezze
Sheryl Ezze
Numerade Educator
00:53

Problem 49

A recent newpaper article quoted a planner in a Florida city as saying. "This city has been growing at the rate of $4.2 \%$ per year. That's not slow growth by any means. It corresponds to $42 \%$ growth per decade." Explain what is incorrect about this statement If. in fact. the current population size of the city is 100.000 and in each of the next ten jears the city increases in size by $4.2 \%$ relative to the previous jcar, then
a) What is the population size after a decade?
b) What percent growth occurs for the decade?

Allison Knapp
Allison Knapp
Numerade Educator
03:42

Problem 50

Example 14.7 showed a predicted growth rate of $13.77 \%$ per decade
a) Show that this is equivalent to a $1.30 \%$ predicted growth per Year. $\left[H i n f .(1.0130)^{10}=\right.$ 1.1377.]
b) Explain why the predicted U.S. population size (in millions) $X$ vears after 1890 is $70.46(1.0130)^X$.

Ryan Mcalister
Ryan Mcalister
Numerade Educator
View

Problem 51

Sketch the relationshup between $E\left(Y^{\prime}\right)$ and $X$ if
a) $\log [E(Y)]$ is linearly related to $X$.
b) $E(Y)=\alpha+\beta_1 X+\beta_2 X^2$ with $\beta_1<0$ and $\beta_2>0$.
c) $E(Y)=\alpha+\beta_1 X+\beta_2 X^2$ with $\beta_1>0$ and $\beta_2<0$.
For multiple choice problems $14.52-1455$, select the correct response(s)

Lainey Roebuck
Lainey Roebuck
Numerade Educator

Problem 52

In the model $E(Y)=\alpha+\beta_1 X+\beta_2 X^2$, the coefficjent $\beta_2$
a) Is the mean change in $Y$ as $X^2$ is increased one unit with $X$ held constant.
b) Is a curvature coefficient that describes whether the regression equation is convex or concave.
c) Equals 0 if the relationship between $Y$ and $X$ is linear
d) Equals 0 if the population value of $R^2$ for this model equals $\rho_{Y X}^2$.

Check back soon!
View

Problem 53

The log transformation of the mean ıesponse in regression is useful when
a) $E(Y)$ is approximately a logarthmic function of $X$.
b) $E\left(Y^{\prime}\right)$ is approximately an exponential function of $X$.
c) $\log E(Y)$ is approximatel) a linear function of $X$
d) Unit chanyes in $X$ have a multiplicative, rather than addiuve, effect on the mean of $Y$.

Rashmi Sinha
Rashmi Sinha
Numerade Educator

Problem 54

Foru ard selection and stepwise regression are similar in the sense that. if they have the same $\alpha$-level for entry.
a) They always select the same final regression model.
b) They always select the same initial regression model (when the) enter the first explanatory variable).
c) Any variable not in the final model does not have a significant partial associatuon with $Y$, controlling for the variables in the final model.
d) It is impossible that all the variables listed for potential inclusion are in the final model.
5

Check back soon!

Problem 55

Evidence of multicollinearity exists in a multiple regression fit when
a) Strong intercorrelations occur among explanatory variables.
b) The $R^2$-ialue is very small
c) The partal correlation between $Y$ and $X_{\mathrm{J}}$ is the same as the bivariate correlation betw'een $Y$ and $X_1$.
d) The $F$ test of $H_0: \beta_1=\cdots=\beta_h=0$ has a small $P-1$ alue, but the individual tests of $H_0: \beta_1=0 \ldots, H_4: \beta_k=0$ do not.

Check back soon!
01:00

Problem 56

True or False?
a) Possible effects of an influential observation include changing a correlation from positive to negative, a $P$-value from .01 to .99 , and $R^2$ from .99 to .01 .

Bryan Luo
Bryan Luo
Numerade Educator
View

Problem 57

Sclect the besi response for each of the following temns (not cvery 1 esponse is used).
Heleroscedasticity
Multicollinearity
Forward selection
Interaction
Exponential inodel
Stepuise regression
Studentuzed iesidual
Gencralized linear model
a) The mean of $Y$ multiplics by $\beta$ for each unit increase in $X$
b) The $\log$ of $E(Y)$ is lineally related to the $\log$ of $X$.
c) A residual plot indicates that the residuals are inucls more spread out at high $X$ than at low $X$.
d) The bivariate association between $Y$ and $X_1$ is different from the partial dssociation between $Y$ and $X_1$ controlling for $X_2$.
e) Suong intercorrelations anıong explanatory variables.
f) When there is multicollinearity. inay provide regiession parameter estimates with smaller standard errors
g) At each stage the variable considered lor entry into the model cxplains the greatest portion of the remaining unexplained varnability in $y^z$.
h) The ıesponse vanable need not be uormal. and we can model a function of the mcan as a linear function of the predictors.
i) At each stage after entering a new vanable. all variablec in the model aure retested in see if they still have a significant partial effect on $Y$
j) The slope between $Y$ and $X_1$ has different values depending on the value of $X_1$.
k) Measurcs the number of standard errors that an observation falls fioun its predicted value

Rashmi Sinha
Rashmi Sinha
Numerade Educator
View

Problem 58

Plot the prediction equation $\hat{Y}=10-4 X_1-X_1^2-2 X_2$ in the following cases:
a) Between $\hat{Y}$ and $X_1$ for values of $X_1$ between 0 and $S$. when (i) $X_2=2$. (ii) $X_2=5$. lise the same axes. Interpret.
b) Between $\hat{Y}$ and $X_2$ for values of $X_2$ belween 0 and 5 when (i) $X_1=2$, (ii) $X_1=5$ I Isc the same axes. Interpret

Shu Naito
Shu Naito
Numerade Educator
View

Problem 59

Show that using a cioss-product tem to model inleraction assunes that the slope of the relationship between $Y$ and $X_1$ changes lincarly as $X_2$ changes. How would you suggest modeling interaction if, instead, the slope of the linear relarionship between $Y$ and $X_1$ first increases as $X_2$ changes from low to moderate values and then decreases as $X_2$ changes from moderate to high $\$ alues?
6

Shu Naito
Shu Naito
Numerade Educator
00:56

Problem 60

Foru ard sclection is used with ten potential predictors Ior $Y$ In reality, Done ane truly correlated with F or with each other For a landom sample, show that the probability equals 40 that at least one is entered into the regression inodel. when the criterion for adinission is a $P$-value below . 05 for the $t$ test (Hin: Use the binomial distribution to first find the probubility that none are selected.)

Hast Aggarwal
Hast Aggarwal
Numerade Educator