Alfred DeMaris - Regression with Social Data, Modeling Continuous and Limited Response Variables
Подождите немного. Документ загружается.
MATHEMATICS TUTORIALS 495
(14) For the data y 6, 2, 3, 0 and x 1, 2, 3, 4, find the vector b of least-
squares estimates for the regression of y on x using matrix algebra. Also, if
SSE y y bXy, give the estimated variance–covariance matrix of
regression parameter estimates.
(15) For the data y 1, 1.5, 1, 3 and x .5, 0, .5, 10, find the vector b of least
squares estimates for the regression of y on x using matrix algebra. Also,
give the estimated variance–covariance matrix of regression parameter esti-
mates.
bapp01.qxd 30.8.04 11:46 Page 495
496
Regression with Social Data: Modeling Continuous and Limited Response Variables,
By Alfred DeMaris
ISBN 0-471-22337-9 Copyright © 2004 John Wiley & Sons, Inc.
APPENDIX B
Answers to Selected Exercises
This appendix contains short answers to the odd-numbered chapter exercises, plus
short answers to all exercises in Appendix A. Note that the term self-correcting
means that it is obvious when the correct solution has been reached.
CHAPTER 2
2.1. (a) b
0
: the estimated average score on the first exam for those with a college
GPA of zero is 25.092; b
1
: each unit increase in college GPA is expected to
result in a 16.783-point increase in the first exam score. (b) σ
ˆ
b
0
6.347;
σ
ˆ
b
1
2.027. (c) F 68.579; p .0001. (d) r
2
.244; r
2
adj
.241. (e)
σ
ˆ
2
219.717. (f ) 75.441. (g) (12.810, 20.756).
2.3. The orthogonality assumption is not very reasonable. For example, health pos-
itively affects coital activity and is negatively correlated with age. Therefore,
the true model for coital frequency is more likely Y β
0
β
1
male age β
2
male health ε. When we estimate Y β
0
β
1
male age ε, ε β
2
male
health ε. So even if Cov(male age, ε) 0, Cov(male age, ε) Cov(male
age, β
2
male health ε) β
2
Cov(male age, male health) 0.
2.5. b
0
y
b
1
x
苶
implies that y
苶
b
0
b
1
x
苶
, which means that the point (x
苶
, y
苶
) is on
the regression line.
2.7. Self-correcting.
2.9. (a) corr(SAT,GPA) .834. (b) t
(8)
4.28; p .01. Yes. (c) E
ˆ
(GPA) 1.049
.337SAT. (d) 2.67. (e) .27.
2.11. (a) b
0
13.052; b
1
.7681. (b) r .364. (c) b
0
: the estimated average
homicide rate for cities with a zero reading quotient is 13.052; b
1
: being a unit
higher on the reading quotient is associated, on average, with a homicide rate
bapp02.qxd 30.8.04 11:59 Page 496
that is lower by .7681. (d) t
(52)
2.818; p .01. Yes, there is a significant
linear relationship. (e) .1325. (f ) 7.58.
2.13. Via SAS: The Shapiro–Wilk test produces a p-value of .0003, suggesting that
the null hypothesis of normally distributed errors should be rejected.
2.15. (a) yˆ 2.7575 .1542x. (b) σ
ˆ
2
3.0811. (c) r
2
.1039. It has modest dis-
criminatory power, at best. (d) F
(2,6)
.9703/3.7847 .2564; p .7. We can-
not reject the null hypothesis that the model is empirically consistent with the
data. (e) r
2
adj
.0081.
2.17. Self-correcting.
2.19. r
2
adj
1
n
n
2
1
S
T
S
S
E
S
1
n
n
2
1
r
2
n 2
n 1
r
2
冢
1
n
n
1
2
冣
, and this is r
2
, since
冢
1
n
n
1
2
冣
0 for n 3.
2.21. Self-correcting.
2.23. corr(error,STATMOOD) 0. This is an artifact of OLS estimates;
corr(error,COLGPA) .4603. This suggests that there is some other system-
atic factor (namely, COLGPA) in the error term that should be entered as a
regressor in the model; corr(error,SCORE) .4738. This suggests that there is
some other systematic factor (namely, SCORE) in the error term that should
be entered as a regressor in the model; corr(error,EXAM1) .95908. This
suggests that error accounts for a substantial part of the variation in EXAM1;
corr(yˆ, STATMOOD) 1.0. This is due to the fact that yˆ is an exact linear
function of STATMOOD; corr(yˆ, error) 0. This follows from the fact that the
error term is uncorrelated with STATMOOD; corr(yˆ, EXAM1) .2832. This
is equal to the correlation between STATMOOD and EXAM1, since yˆ is just
a linear “translation” of STATMOOD.
2.25. s
2
y
b
2
1
s
2
x
s
2
e
.
2.27. Self-correcting.
CHAPTER 3
3.1. X
2
3 ⇒ yˆ (b
0
3b
2
) b
1
X
1
; X
2
6 ⇒ yˆ (b
0
6b
2
) b
1
X
1
;
X
2
9 ⇒ yˆ (b
0
9b
2
) b
1
X
1
.
3.3. (a) .454. (b) .454.
3.5. F
(5,408)
3.802, p .0022. The addition of these variables appears to produce
a significant improvement in the model.
3.7. (a) F
(4,209)
13.899, p .00001. Reject H
0
. At least one coefficient is not zero.
(b) .2101. (c) 24.586. (d) COLGPA: t 1.597/.604 2.644; significant at
p .009; SCORE: t .296/.110 2.691; significant at p .008; HOURS:
SSE
TSS
SSE
TSS
SSE
TSS
SSE
TSS
SSE
TSS
SSE
TSS
SSE
TSS
ANSWERS TO SELECTED EXERCISES 497
bapp02.qxd 30.8.04 11:59 Page 497
t .490/.145 3.379; significant at p .0009; PREVMATH:
t 1.516/.344 4.407; significant at p .000017. (e) 7.275. (f ) Not really,
since zero is outside of the observed, and in some cases, logical range of most
of the regressor values. No one has zero college GPA, math diagnostic score,
or hours in the current semester.
3.9. The plot of standardized residuals against yˆ from the regression reveals no
major irregularities, and all z
e
are under 4 in absolute value. Similarly, the par-
tial regression leverage plots reveal no noticeable nonlinear trends. In conclu-
sion, the model appears, at least, to demonstrate empirical consistency.
3.11. (a) .2052. (b) F
(6,228)
11.07, p .00001. Yes, the model appears to have
some utility for predicting STATMOOD. (c) .475. (d) 23.9679.
3.13. t for MCHATT versus FCHATT effects is .2599 and is not significant; t for
MALEAGE versus FEMAGE effects is .6164 and is not significant; t for
MEDUC versus FEDUC effects is .2934 and is not significant.
3.15. The test of equality of male versus female parent effects is F
(4,347)
.0949,
p .98. There is not enough evidence to conclude that there is a difference in
male versus female parent effects on offspring’s sexual adventurism.
3.17. (a) If r
x
1
x
2
0 then b
1
r
yx
1
(s
y
/s
x
1
), which is the b for the SLR of Y on X
1
. (b)
As r
2
x
1
x
2
approaches 1.0, the denominator of b
1
becomes ever smaller, mean-
ing that the magnitude of b
1
grows ever larger.
3.19. E(Y 冟x
k
1, x
k
) E(Y 冟x
k
, x
k
) (note that X
j
is one of the other X
1
, X
2
,...,X
K
)
β
0
β
1
X
1
β
2
X
2
...
β
k
( x
k
1)
...
β
K
X
K
γ (x
k
1)X
j
β
0
β
1
X
1
β
2
X
2
...
β
k
x
k
...
β
K
X
K
γ x
k
X
j
β
k
(x
k
1) β
k
x
k
γ (x
k
1)X
j
γx
k
X
j
β
k
(x
k
1 x
k
) γ [(x
k
1)X
j
x
k
X
j
]
β
k
(1) γ (x
k
X
j
X
j
x
k
X
j
) β
k
γ X
j
.
3.21. At 2 years below mean years in rank the estimated partial slope is
1008.267 15.481(2) 10039.229; at 2 years above mean years in rank it
is 1008.267 15.481(2) 977.305; at 9 years above mean years in rank it is
1008.267 15.481(9) 868.938. Estimated mean salaries: 57444.749,
57410.961, 57351.832, respectively.
3.23. (a) Net of other factors, for kids with an average level of parental monitoring,
a unit increase in their sexual permissiveness is worth, on average, a .1625-
unit increase in their sexual adventurism. (b) Partial slope: .1625 .024
MONITOR. Partial slope at 1 standard deviation above mean parental moni-
toring: .2441. (c) At 1 SD above mean parental monitoring: t
(352)
4.253,
p .001; significant. At 2 SD below mean parental monitoring:
t
(352)
.0079; not significant.
3.25. F
(7,228)
3.8805, p .001. The models for STATMOOD do appear to be
significantly different for males and females.
498 ANSWERS TO SELECTED EXERCISES
bapp02.qxd 30.8.04 11:59 Page 498
3.27. Nested F 2.627, which implies a t of 1.6208, p .1058. t for
difference 1.624, p .1051. t for θ 1.624, p .1051.
CHAPTER 4
4.1. (a) yˆ 1.5658 .4101 PRESCHDN. (b) r .341, r
2
.116. (c) Two-sample t
test: t 7.36, p .001; test for r: t 7.38, p .001; test for d: t 7.35, p .001.
4.3. (a)
SOPH JUN POST
Senior 1 1 1
Sophomore 1 0 0
Junior 0 1 0
Postgrad/grad 0 0 1
(b) yˆ 78.806 2.789 SOPH4.125 JUN 5.258 POST.
4.5. (a) F
(4,1255)
33.135, p .00001. Net of gender, marital status has a
significant effect on cohabitation attitude. (b) F
(4,1251)
1.045, p .38. There
is not enough evidence to conclude that there is an interaction effect in the pop-
ulation. (c) Cell means: married female 5.603; married male 5.688; wid-
owed female 4.802; widowed male 5.474; divorced female 6.522;
divorced male 6.662; separated female 6.909; separated male 5.95;
never married female 7.013; never married male 7.396.
4.7. (a) Married couples: 8.171; cohabiting couples: 11.485. (b) Married couples:
8.23. cohabiting couples: 10.55. The adjusted means are closer together, sug-
gesting that part of the gap in sexual frequency between marrieds and cohab-
itors is accounted for by differences in age and relationship duration between
these types of couples.
4.9. (a) Interpretations:
• Intercept: Males in couples where both are working full time do, on aver-
age, 23.766% of the housework, when MINCOME and FINCOME are both
zero (this is, of course, nonsensical, but is nevertheless the interpretation!).
• MPARTIME: Controlling for partners’ incomes and females’ employment
status, males’ average percent contribution to housework is 2.104 more
when males work part time, compared to when they work full time.
• MUNEMP: Controlling for partners’ incomes and females’ employment sta-
tus, unemployed males do, on average, .177% less housework, compared to
those who work full time.
• FPARTIME: Controlling for partners’ incomes and females’ employment sta-
tus, the average percent contribution to housework of male partners of
women who work part time is 6.043 less, compared to males whose partners
work full time.
• FUNEMP: Controlling for partners’ incomes and females’ employment status,
the average percent contribution to housework of male partners of unemployed
women is 4.738 less, compared to males whose partners work full time.
ANSWERS TO SELECTED EXERCISES 499
bapp02.qxd 30.8.04 11:59 Page 499
(b) Adjusted mean percent of housework done by males when: both work full
time: 25.881; male works part time, female full time: 27.985; male unem-
ployed, female full time: 25.704; male works full time, female part time:
19.838; male works full time, female unemployed: 21.143; both unemployed:
20.966; both work part time: 21.942; male works part time, female unem-
ployed: 23.247; male unemployed, female part time: 19.661.
4.11. (a) F
(2,226)
3.176, p .05. Yes, it is. (b) Controlling for couple conflict,
women with natural children are, on average, 5.0846 units lower on depres-
sive symptomatology, compared to those without children. (c) Since both
types of kids are associated with greater conflict, but less depression with
conflict controlled, and conflict is strongly positively related to depression,
omitting conflict from the model reverses the impact of kids on depression.
4.13. Seniors: 75.892; sophomores: 79.522; juniors: 76.366; postgrad/grads: 79.398.
The means are the same, within rounding error, as in Exercise 4.12(b).
4.15. (a) Interpretation of rank effects: Full professor: net of the covariates, full pro-
fessors’ average salary is $17,500 higher than for assistant professors. Associate
professor: net of the covariates, associate professors’ average salary is $6213.53
higher than for assistant professors. Instructor/lecturer: net of the covariates,
instructor/lecturers make, on average, $7870.20 less than assistant professors.
(b) Adjusted means: full professors 56243; associate professors 44956.526;
assistant professors 38743; instructor/lecturers 30872.797.
4.17. (a) F
(171,215)
1.2459, p .064. There is not enough evidence to reject homo-
geneity of error variance. (b) Chow test: F
(15,386)
.895, p .57. The models do
not appear to differ for males versus females. (c) Effect of COITFREQ:
.0183 1.2076 MALEHIT .2841 FEMAHIT. Interpretation: coital fre-
quency has virtually no effect on the female’s depressive symptomatology if
neither partner has been violent; it increases depressive symptomatology if the
male has been violent; and it lowers depressive symptomatology if the female
has been violent.
4.19. Intercept constrained: F
(10,1853)
2.381. Intercept unconstrained: F
(8,1853)
2.336.
4.21. Constrained intercept: F
(12,701)
1.422. Unconstrained intercept:
F
(11,701)
1.361.
4.23. The interaction is disordinal in Z, but ordinal in X.
4.25. Self-correcting.
CHAPTER 5
5.1. (a) ps, husbands: 12.343 2(1.944) power. ps; wives: 11.567 2(2.537)
power. (b) Husbands: ps at power 1 is 8.455. ps at power 4 is 3.209;
wives: ps at power 1 is 6.493. ps at power 4 is 8.729. (c) Husband’s
minimum occurs at power 3.175; wife’s minimum occurs at power 2.28.
500 ANSWERS TO SELECTED EXERCISES
bapp02.qxd 30.8.04 11:59 Page 500
ANSWERS TO SELECTED EXERCISES 501
5.3. With γ
0
1.5, γ
1
.25, ε
1
.5, ε
2
.5, we have:
X 12 3 45
Model (5.13):
Y
1
3.176 4.077 5.236 6.723 8.632
Y
2
1.168 1.500 1.926 2.473 3.176
Model (5.12):
Y
1
2.426 2.973 3.676 4.577 5.736
Y
2
1.426 1.973 2.676 3.577 4.736
Variances of Y:
S
2
5.13
2.016 3.320 5.478 9.031 14.884
S
2
5.12
.5 .5 .5 .5 .5
5.5. (a) F
(2,7)
.151, n.s. A linear model appears adequate for the data. (b) A quiz
score of 60 implies yˆ 49.152; a quiz score of 92 implies: yˆ 91.903. (c) ps
at mean 1 SD(Quiz): .929. ps at mean(Quiz): 1.376. ps at mean 1
SD(Quiz): 1.823.
5.7. (a) log yˆ 2.773 .024 MALEAGE. (b) Each additional year older the male
partner is results in a 100(e
.024
1) 2.37% reduction in coital frequency.
For a 10-year increase in age, the reduction is 21.34%. (c) Male age 25:
yˆ 12.043. Male age 35: yˆ 9.474. Male age 55: yˆ 5.862. (d) .127. (e)
When the residuals are plotted against log yˆ, they appear to have zero mean,
but their variance appears to increase with log yˆ. According to a formal test,
the normality of the errors would be rejected. So the assumptions on the errors
may not be warranted.
5.9. (a) yˆ 1.693 .508x
1/3
.896z .365zx
1/3
. (b)
∂
∂
y
x
ˆ
.50
3
8
x
2/3
.365z
;
∂
∂
y
z
ˆ
.896 .365x
1/3
. (c) Effect of X at z
苶
1s
z
:
3
.
x
2
2
0
/3
4
; at z
苶
:
1
3
.
x
2
2
5
/3
2
; at z
苶
1s
z
:
2
3
.
x
7
2
0
/3
7
.
(d) Effect of Z at x
苶
1s
x
: .471; at x
苶
: .293; at x
苶
1s
x
: .184. (e) See
Instructor’s Solutions Manual.
5.11. (a) See Instructor’s Solutions Manual. The trend resembles a segment I
curve. (b) F
(8,406)
2.444, p .014. It appears that the data trend is
significantly nonlinear. (c) F
(7,406)
.521, n.s. It appears that a quadratic
model adequately captures the nonlinearity in the relationship. (d) yˆ 1.824
.013 DURYRS .0002 DURYRS
2
. Slope at mean 1 SD: .007; at
mean: .013; at mean 1 SD: .019. A segment I curve is indicated.
5.13. (a) At x 0:
5
2
2(0)
.4. At x 2.5:
5 2
2
(2.5)
.2. At x 5:
5
2
2(5)
.133. (b) At x 0: 1.609. At x 2.5: 2.303. At x 5: 2.708.
(c) See Instructor’s Solutions Manual.
5.15. (a) U-shaped curve. (b) U-shaped curve. (c) Segment I curve. (d) Inverted
U-shaped curve.
5.17. At z
苶
1s
z
: 3.263 .5x. At z
苶
: 3 .5x. At z
苶
1s
z
: 2.738 .5x.
bapp02.qxd 30.8.04 11:59 Page 501
5.19. At z
苶
1s
z
: 2.378 .375x. At z
苶
: 3 .5x. At z
苶
1s
z
: 3.263 1.375x.
5.21. (a) Linear model. (c) Nonlinear interaction in X. Linear interaction in Z.
5.23. (a) Nonlinear model. (c) Nonlinear interaction in X and in Z.
5.25. (a) Nonlinear model. (c) Nonlinear interaction in X and in Z.
CHAPTER 6
6.1. (a) Self-correcting. (b) R
xx
冤冥
.
6.3. Self-correcting.
6.5.
R
xx
冤冥
⇒ R
1
xx
1 .
1
675
2
冤冥
冤冥
6.7. λuuλ
冤
冥
u
冤
冥
, and each λu
j
u is a scalar multiple of the same
vector, u, so each λu
j
u is an exact linear function of every other vector (e.g.,
λu
2
u
u
u
2
1
λu
1
u, and so on). Therefore there is only one linearly independent
row in the matrix, hence the matrix has rank 1.
6.9. (a) h
ii
for cases 1 through 9, respectively: .475, .1646, .1646, .1112, .1112, .1112,
.1572, .1572, .5479. (b) t
9
2.172; dffits
9
2.392; dfbetas
9
1.521;
D
9
1.867.
6.11. (a) E(b
s
1
)
冤冥
. (b) The cross-product term, and therefore the interaction
effect, are completely suppressed when the intervening variables are omitted.
6.13. b
s
pc
冤冥
.
6.15. (a) V(b
s
4
) σ
2
*
冢
2
.0
.9
1
3
3
0
7
5
2
.
.
9
9
9
9
9
9
9
6
2
.
.
0
0
0
6
1
8
7
9
2
.00
.0
0
0
9
1
1
4
8
2
06
冣
σ
2
*
(1.00248).
(b) p
13
.000408; p
23
.00000024; p
33
.02198; p
43
.9776.
6.17. Self-correcting.
6.19. WLS estimate of the equation is: 10.8721 .1993 ADVENTRE .0852
FSTYLE1 .2041 MSEXATT .1988 FSEXATT. R
2
WLS
.1049.
6.21. Gender difference in exam scores at: college GPA 2.5: d 1.429, p .68, n.s.;
at college GPA 3.0: d 2.149, p .34, n.s.; at college GPA 3.5:
d 5.727, p .07, n.s.
.35
.35
.25
.15
.00
λu
1
u
λu
2
u
λu
n
u
u
1
u
2
u
n
1 .6
1
75
2
1
.
.
6
6
7
7
5
5
2
1
.
.
6
6
7
7
5
5
2
1 .
1
675
2
1 .675
.675 1
1 .675
.675 1
1 .876
.876 1
502 ANSWERS TO SELECTED EXERCISES
bapp02.qxd 30.8.04 11:59 Page 502
6.23. One serious near linear dependency involving YRDG, YRBG, and PRIOREXP.
The VIF’s for their coefficients (52.836, 49.587, and 14.719, respectively) sug-
gest that the coefficient variances are substantially affected.
6.25. Partial principal component results in the form of VARIABLE (b
pc
) are: INTER-
CEPT (7873.500), YRDG (13.897), YRBG (80.556), PRIOREXP
(179.726). (See Instructor’s Solutions Manual for the full set of estimates.) The
ridge estimates are the most intuitively appealing, since one would expect that,
all else equal, YRDG, YRBG, and PRIOREXP would each have a positive effect
on salary.
CHAPTER 7
7.1. 3.0089.
7.3. R
2
L
.1058. R
2
G
.1328. R
2
GSC
.1795.
7.5. .0176.
7.7. .718.
7.9. Logit: .0093; probit: .0082.
7.11. Baseline π
ˆ
.1062. Change in probability: ∆π
ˆ
.0042.
7.13. (a) The odds of violence for those with no alcohol or drug problems at aver-
age economic disadvantage is .143. Having alcohol or drug problems
magnifies the odds by a factor of 2.876, net of economic disadvantage. Each
unit increase in economic disadvantage magnifies the odds by a factor of
1.031, controlling for substance abuse problems. (b) For those without alco-
hol/drug problems with mean economic disadvantage, the change in the prob-
ability of violence for a unit increase in ECNDISAD is ∆π
ˆ
.0034. For those
with alcohol/drug problems, on the other hand, the change is ∆π
ˆ
.0064.
7.15. The model χ
2
’s are: logit model 1 128.634; logit model 2 275.859; probit
model 1 126.631; probit model 2 273.353. The test for the additional
conflict block is: logit: χ
2
(3)
147.225; probit: χ
2
(3)
146.722. Both tests are
significant at p .0001.
7.17. (a) .638. (b) .886. (c) .725.
7.19. Logit: ∆πˆ .044; probit: ∆πˆ .04.
7.21. Partial results are as follows. Logit equation: ln O
ˆ
1.08251.1050 lived
apart .2936 number of children
...
; model χ
2
(5)
62.4470. Probit equa-
tion: Φ
1
(π
ˆ
) .8325 .6055 lived apart .1550 number of children
...
;
model χ
2
(5)
63.1250. (See Instructor’s Solutions Manual for the full set
of estimates.) Interpretations: Logit: Each additional year the couple has
been together lowers the odds of female aggression by
100[exp(.1219) 1] 11.5%. Probit: Each additional year the couple has
been together reduces the latent female-aggression scale by .0614, on average.
7.23. Probit: ∆π
ˆ
.0512; logit: ∆π
ˆ
.0362.
7.25. Probit: R
2
L
.258; R
2
MZ
.53; logit: R
2
L
.255; R
2
MZ
.564.
ANSWERS TO SELECTED EXERCISES 503
bapp02.qxd 30.8.04 11:59 Page 503
504 ANSWERS TO SELECTED EXERCISES
7.27. Interpretations: Each unit increase in GPA adds .464 to the probability of an
A. Those taught with the PSI method have .379 greater probability of getting
an A. Each unit increase in TUCE adds .01 to the probability of an A.
Estimated probability: π
ˆ
A,11
.078.
7.29. Interpretations: Each unit increase in GPA raises the odds of getting an A by
a factor of 16.88. Those taught with the PSI method have 10.791 times higher
odds of getting an A. Each unit increase in TUCE raises the odds of getting an
A by a factor of 1.1. Estimated probability: π
ˆ
A,11
.024. R
2
L
.374.
CHAPTER 8
8.1. Self-correcting.
8.3. At 1 SD economic disadvantage: ψ 1.848. At .5 SD economic disadvan-
tage: ψ 3.142. At 1.5 SD economic disadvantage: ψ 4.476. At 1 SD
economic disadvantage, the impact of alcohol/drug problem is reduced by about
30%. At .5 SD, it’s increased by 19.3%, and at 1.5 SD, it’s increased by 70%.
8.5. At .5 SD: .041. At 1.5 SD: .016. At 2 SD: .003.
8.7. Self-correcting.
8.9. Self-correcting.
8.11. With 0 “intense male violence,” 1 “physical aggression,” and 2 “nonvi-
olence,” the estimated probabilities are: P(Y 0) .085; P(Y 1) .225;
P(Y 2) .69.
8.13. (a) χ
2
(12)
11.5317, p .484. There is not enough evidence to reject the pro-
portional odds hypothesis. (b) Interpretations: Controlling for other regres-
sors, each unit increase in COLGPA magnifies the odds of a higher grade by
a factor of 8.929. Controlling for other regressors, each unit increase in math
diagnostic score magnifies the odds of a higher grade by a factor of 1.305.
Controlling for other regressors, each unit increase in STATMOOD magnifies
the odds of a higher grade by 1.064. Controlling for other regressors, each unit
increase in the ratio of study hours to TV hours diminishes the odds of a higher
grade by about 2%.
8.15. Model χ
2
(10)
182.2603, p .00001. The model is significant as a whole.
R
2
L
.126.
8.17. Sample interpretations (see Instructor’s Solutions Manual for the complete set
of interpretations): AGE: Controlling for other regressors, each additional year
of age increases the odds of “no prohibition” by a factor of 1.022, and
increases the odds of “prohibition for all” by a factor of 1.029. MALE:
Controlling for other regressors, men’s odds of “no prohibition” are 26.125%
lower than women’s, and their odds of “prohibition for all” are 38.097% lower
than women’s. On treating PORN18 as ordinal: no. (See explanation in
Instructor’s Solutions Manual.)
8.19. P(no prohibition) .026. P(prohibition for all) .178. P(prohibition if
18) .796.
bapp02.qxd 30.8.04 11:59 Page 504