Alfred DeMaris - Regression with Social Data, Modeling Continuous and Limited Response Variables
Подождите немного. Документ загружается.
ANSWERS TO SELECTED EXERCISES 505
8.21. Sample interpretations (see Instructor’s Solutions Manual for the complete set
of interpretations): EDUCAT: Controlling for other regressors, each additional
year of schooling reduces the odds of “easier” by 14.23%, and the odds of
“more difficult” by 10.631%. MALE: Controlling for other regressors, men’s
odds of “easier” are 28.794% lower than women’s, and their odds of “more
difficult” are 5.578% lower than women’s. On treating DIVLAW as ordinal: no.
(See explanation in Instructor’s Solutions Manual.)
8.23. P(easier) .248. P(more difficult) .699. P(keep as is) .053.
8.25. (a) Sample interpretations (see Instructor’s Solutions Manual for the complete
set of interpretations): SCORE: Controlling for other regressors, a unit
increase in SCORE magnifies the odds of an A by 1.451, and the odds of a B
by 1.304. MALE: Controlling for other regressors, males’ odds of an A are
.573 times females’ odds, whereas their odds of a B are .998 times those of
females. (b) Model χ
2
(12)
97.96, p .00001. (c) College GPA is the only dis-
criminator of an A versus a B grade. (d) χ
2
(7)
31.287, p .0001. The grades
of A and B are not collapsible with respect to the predictors. (e) Test for the
proportional odds assumption: χ
2
(7)
9.7946, p .2005. There is insufficient
evidence to reject the proportional odds assumption. Sample interpretations
(see Instructor’s Solutions Manual for the complete set of interpretations):
SCORE: Controlling for other regressors, a unit increase in SCORE increases
the odds of a better grade by 34%, an effect that is significant at p .0001.
MALE: Controlling for other regressors, men’s odds of a better grade are
38.4% lower than women’s, a nonsignificant effect.
CHAPTER 9
9.1. E( y 冟 y 8) 13.152. SD( y 冟 y 8) 3.172.
9.3. E( y 冟 z 5) 12.278. SD( y 冟 z 5) 3.884.
9.5. 87.365.
9.7. 22.289.
9.9. Self-correcting.
9.11. R
2
p
.254.
9.13. The proportion due to change in the mean is .377. The proportion due to
change in P(uncensored) .623.
9.15. [corr(y,y
ˆ
)]
2
.638.
9.17. (a) y
ˆ
43.819. (b) 83.9% of the effect of child abuse is due to its effect on ele-
vating average PTSD, while 16.1% of its effect is due to raising the risk for
PTSD.
9.19. ρ
eu
β
3
γ
3
V(x
3
), which is 0.
9.21. Bias(b
1
) .697, therefore b
1
will overestimate β
1
.
9.23. There is little evidence of selection bias, since neither LAMBDA (Heckit) nor
ρ
ˆ
(ML) is significant, suggesting that the association between remaining in the
bapp02.qxd 30.8.04 11:59 Page 505
506 ANSWERS TO SELECTED EXERCISES
class and score on exam 2 is pretty much accounted for by the explanatory
variables. Also, the ML and OLS estimates tend to be quite similar.
Interestingly, students with a stronger attitude toward statistics appear to be
less likely to have stayed in the class!
9.25. The ML procedure did not converge in this case. But judging from Heckit,
selection bias does not appear to be a problem. The only significant predic-
tor of inclusion is paying one’s own tuition. Its effect suggests that those
who pay their own way are more likely to be nonmissing on the response.
Otherwise, none of the focus variables are significant predictors of inclu-
sion, and LAMBDA is also nonsignificant. As a result, there is relatively
little difference between OLS and Heckit estimates of the substantive
equation.
CHAPTER 10
10.1.
冱
∞
y 1
y!(
e
1
µ
µ
e
y
µ
)
1
1
e
µ
冱
∞
y 1
e
y
µ
!
µ
y
1
1
e
µ
(1 e
µ
) 1.
10.3. (a) f(0) .041. (b) f(3) .223. (c) f(10) .0013.
10.5. (a) f(0) .0001. (b) f(3) .0015. (c) f(10) .00997.
10.7. (a) p .238. (b) f(0) .238. (c) f(3) .105. (d) f(10) .016.
10.9. L(
ββ
冟y,x)
冱
n
i 1
冤
y
i
冢
冱
K
k0
β
k
X
ik
冣
exp
冢
冱
K
k0
β
k
X
ik
冣
ln y
i
!
ln
冢
1 exp
冢
exp
冢
冱
K
k0
β
k
X
ik
冣冣冣冥
.
10.11. f(0) e
µ
≡ . f(1)
µ
1
e
µ
≡ . f(2)
µ
2
µe
µ
≡ .
f(3)
µ
3
µ
2
e
2
µ
µ
3
6
e
µ
≡
e
3
µ
!
µ
3
. f(4)
µ
4
µ
3
e
6
µ
µ
4
2
e
4
µ
≡
e
4
µ
!
µ
4
.
10.13. f
ˆ
(0 冟 x,
ββ
ˆ
) .575. f
ˆ
(1 冟 x,
ββ
ˆ
) .138. f
ˆ
(2 冟 x,
ββ
ˆ
) .076. f
ˆ
(3 冟 x,
ββ
ˆ
) .050.
f
ˆ
(4 冟x,
ββ
ˆ
) .035.
10.15. (a) Proportionate change is (1 1/N)
d
1. (b) .014.
10.17. (a) Cohabitors’ expected number of children in the household is estimated to
be about 45% lower than that of marrieds. (b) f
ˆ
(0 冟 x,
ββ
ˆ
) .388.
f
ˆ
(1 冟x,
ββ
ˆ
) .367. f
ˆ
(2 冟x,
ββ
ˆ
) .174. (c) χ
2
(4)
129.8468, p .00001. Reject H
0
.
At least one coefficient is nonzero. R
2
L
.105.
10.19. (a) The likelihood of being in the structural-zero group is higher if a couple is
cohabiting, was older at inception of the union, has been together longer, and
has a higher household income. The expected number of children, given that
a couple is at risk for having children, is, correspondingly, lower for couples
with the same characteristics. (b) Since V 10 1.96, the ZIP is to be pre-
ferred over the PRM. (c) R
2
L
.214.
e
µ
2
2!
µ
2
e
2
e
µ
µ
1
1!
e
µ
µ
0
0!
bapp02.qxd 30.8.04 11:59 Page 506
10.21. (a) Selected estimates (significance levels) are: LOGTIME .0472 (.0029);
ETHNIC .3633 (.0000); EDUCCL .0684 (.0005). (See Instructor’s
Solutions Manual for the full set of estimates.) (b) Test 1 [H
1
is
V( Y) µ αµ]: α
ˆ
2.798, p .0000. Test 2 [H
1
is V( Y) µ αµ
2
]:
α
ˆ
2.607, p .0000. Conclusion: the data are overdispersed. (c) r
2
.1139;
R
2
D
.1827; R
2
L
.1437.
10.23. (a) Selected estimates for the log odds of being a structural zero (µ) are:
LOGTIME .0321 (.0453); ETHNIC .5456 (.0677); EDUCCL .0148
(.0546). (See Instructor’s Solutions Manual for the full set of estimates.)
(b) V 27.2561 for ZIP/PRM, so ZIP is preferred. (c) R
2
L
.3516; r
2
.0604.
10.25. See Instructor’s Solutions Manual for the graph.
CHAPTER 11
11.1. Hazard estimates are: h
ˆ
1
.003; h
ˆ
2
.003; h
ˆ
3
. 003; h
ˆ
4
.005. Survival esti-
mates are: S
ˆ
1
1 .000 ; S
ˆ
2
.975; S
ˆ
3
.945; S
ˆ
4
.916; S
ˆ
5
.869.
11.3. Wald: χ
2
.755, n.s. LRχ
2
.811, n.s. Interpretation: the hazard of rearrest for
domestic violence at any given time is 65.8% lower for those who were
employed when first arrested.
11.5. S(t) exp(λt
σ
) ⇒ ln S(t) λt
σ
⇒ ln S(t) λt
σ
⇒ ln[ln
S(t)] ln λ σ lnt.
11.7. 11.29 months.
11.9. Sample interpretations (see Instructor’s Solutions Manual for the complete
set of interpretations): Net of other covariates, women’s hazard of dropping
out is 38.265% higher than men’s at any given time, and the hazard of drop-
ping out is increased by 32.976% for each unit decrement in high-school
grades at any given time.
11.11. (a) ∆χ
2
(2)
13.13, p .0014. There is evidence that at least one of the effects is
nonzero. (b) Marriage: net of other covariates, at any given time being married
raises the hazard of dropping out by a factor of 3.508. Employment: net of other
covariates, at any given time being employed raises the hazard of dropping out
by a factor of 1.669. (c) R
2
G
.211.
11.13. (a) LRχ
2
(5)
43.028, p .0001. The model is significant. (b) Wald chi-
squareds: DURATION: χ
2
6.25; FAGUNION: χ
2
21.778; PARENTAL:
χ
2
2.736; CHDN5: Χ
2
.737; MINORITY: χ
2
2.404. DURATION and
FAGUNION have significant effects. (c) R
2
G
.034.
11.15. (a) Interpretations: each unit increase in the female partner’s relationship hap-
piness reduces the hazard of disruption by 13.4%; each unit increase in the
female partner’s commitment reduces the hazard of disruption by 34.6%. (b)
20.2%. (c) (.480, 3.013). (d) R
2
G
.115.
11.17. 2lnL 6.086.
11.19. Global test: ∆χ
2
(2)
= 3.7116, p .156, n.s. There is insufficient evidence to sug-
gest that there are any gender differences in the effects of income or education
on the hazard of disruption.
ANSWERS TO SELECTED EXERCISES 507
bapp02.qxd 30.8.04 11:59 Page 507
11.21. The effect of DISAGMT at 1 standard deviation above the mean of COM-
STYLE is .02283, a value that is not significant.
CHAPTER 12
12.1. .048.
12.3. ∆χ
2
(2)
= 7.598, p .022. Reject H
0
; conclude that the three events are charac-
terized by different models.
12.5. Equations (a), (b), and (d).
12.7.
Unemployment Duration Censoring
Spell Unemployed Indicator
131
2101
321
4170
12.9. When TIME 1 (i.e., in the first interval), the hazard of reemployment is
50% higher for immigrants residing in Canada. However, each unit increase
in LOGTIME brings a 50% reduction in the effect of Canadian residence.
The coefficient is approximately zero when TIME 1.8.
12.11. The hazards for the first two intervals are: P
ˆ
i1
.191; P
ˆ
i2
.184. The survival
function for the first two intervals is: S
ˆ
i1
.809; S
ˆ
i2
.660.
12.13. The hazards for the first four years are: P
ˆ
i1
.00023; P
ˆ
i2
.0035;
P
ˆ
i3
.0205; P
ˆ
i4
.0573. The survival function for the first four years is:
S
ˆ
i1
.9998; S
ˆ
i2
.996; S
ˆ
i3
.976; S
ˆ
i4
.92.
12.15. (a) and (c) are time invariant, since their values are fixed at the time the
employee enters the given rank (either assistant or associate professor).
(b), although treated as time-invariant in the study, could easily be regarded
as time varying, since the prestige of an institution can change over time.
(d), (e), and (f ) are time varying because their values change with years
in rank.
12.17. Selected estimates for the log-hazard of separation (marriage) are DIS-
AGMT .009 (.073); COMSTYLE .151 (.142); HEHIT 1.467 (2.712).
(See Instructor’s Solutions Manual for the full set of estimates.)
Interpretation: the main effects of covariates not involved in an interaction
with survival time are similar to those in Table 12.1. The block of crossprod-
uct terms is not significant when added to either model, according to ∆χ
2
tests. However, there is a significant nonproportional effect for male violence
in the model for marriage, which suggests that male violence accelerates the
transition to marriage at first, but over time this effect is diminished.
12.19. Selected estimates for the log-hazard of reemployment are PRVJBS .045*;
JOBDUR .052***; FEMALES .314**. (See Instructor’s Solutions
Manual for the full set of estimates.) Tests: ∆χ
2
7.9135, p .005. Wald
508 ANSWERS TO SELECTED EXERCISES
bapp02.qxd 30.8.04 11:59 Page 508
χ
2
8.2846, p .004. By either test, the interaction is significant.
Interpretation: At the time of unemployment, each additional month on the
previous job reduces the hazard of reemployment by about 5%. After being
unemployed for 17.33 months, however, the duration of the previous job no
longer has any impact on the hazard of reemployment.
APPENDIX A
I.Q (1) 3xy 2y 3xz 2z.
(2) 2x
3
x
2
兹y
苶
x
2
z 2x兹y
苶
y 兹y
苶
z 8xw 4兹y
苶
w 4wz.
(3) x 2兹2
苶
4
苶
24.
(4) 15.
(5)
ab(d
c
d
3π)
.
(6)
2x
2
zw
yz
.
(7) x 123.4568.
(8) 15x
5
.
(9) .
(10) 2
7/6
.
(11) 2ln x
1
2
ln z.
(12) f(
1
2
)
4; f
(
1
3
)
3.5; f(2) 18.5
.
(13)
f
(
1
2
)
2.75; f
(
1
3
)
1
9
; f(2) 1
.
(14)
f(g(3))
兹
8
苶
.
(15)
f
(
g
(
1
2
))
.6065
.
(16) f(g(x)) x.
(17) y 9.5 18x.
II.E (1) Self-correcting.
(2) Self-correcting.
(3) Self-correcting.
III.D (1) 5 Cov( X,Y ).
(2) bd Cov( X,Y ).
(3) 65.2.
(4) 10.1.
(5) β
2
V( X) V(ε).
(6) Self-correcting.
(7) Self-correcting.
IV.D (1) 3x
2
.
(2)
5x
4
1/5
.
(3) 12x
3
6x
2
2x.
x
z
2
z
y
z
ANSWERS TO SELECTED EXERCISES 509
bapp02.qxd 30.8.04 11:59 Page 509
(4) f(x)
1
x
3x
2
; f(3) 27
1
3
.
(5) f(2) 4.1353.
(6) e
x
(x
3
3x
2
).
(7) 1 ln x.
(8)
4
2兹
2l
x
苶
n x
.
(9)
2兹
ln
x
苶
x
(l
n
2
x)
2
.
(10) 12x
2
.
(11)
x
1
2
.
(12) 25e
5x
.
(13)
2l
x
n x
.
(14) 2e
x
2
(2x
2
1).
(15)
(1
e
x
e
x
)
2
.
(16) β
(1
e
α
e
α
βx
βx
)
2
.
(17)
2
x
.
(18) 4(x 2)
3
.
(19) 3
冢
1
2
x ln x
冣
2
冢
2
1
1
x
冣
.
(20) 2x 3y.
(21) β
(1
e
α
e
α
βx
β
x
γw
γ
w
λz
λz
)
2
.
V.K (1) 1.
(2) 66.
(3)
冤冥
.
(4)
冤冥
.
(5)
冤冥
.
(6)
冤冥
.
(7)
冤冥
.
(8)
冤冥
.
d
1
x
12
d
2
x
22
d
1
x
11
d
2
x
21
x
12
d
2
x
22
d
2
x
11
d
1
x
21
d
2
3
3
2
2
4
4
1
1
0
1
5
6
7
0
1
1
0
21
45
510 ANSWERS TO SELECTED EXERCISES
bapp02.qxd 30.8.04 11:59 Page 510
(9)
1
8
冤冥
.
(10)
1
2
1
冤冥
.
(11) Self-correcting.
(12) Self-correcting.
(13) System a:
冤冥
=
冤冥
, within rounding error. System b:
冤冥
=
冤冥
, within rounding error.
(14) b
冤冥
,V
ˆ
(b)
冤冥
.
(15) b
冤冥
,V
ˆ
(b)
冤冥
.
.0036
.0014
.0360
.0036
1.1702
.1809
1.075
.43
3.225
1.075
7
1.7
1.5909
.9545
x
l
x
2
2.2857
.1429
x
l
x
2
1
2
1
2
9
2
4
2
8
6
ANSWERS TO SELECTED EXERCISES 511
bapp02.qxd 30.8.04 11:59 Page 511
512
Regression with Social Data: Modeling Continuous and Limited Response Variables,
By Alfred DeMaris
ISBN 0-471-22337-9 Copyright © 2004 John Wiley & Sons, Inc.
References
Aalen, O. O. (1994). Effects of frailty in survival analysis. Statistical Methods in Medical
Research, 3, 227–243.
Agresti, A. (1984). Analysis of ordinal categorical data. New York: Wiley.
Agresti, A. (1989). Tutorial on modeling ordered categorical response data. Psychological
Bulletin, 105, 290–301.
Agresti, A. (2002). Categorical data analysis (2nd ed.). Hoboken, NJ: Wiley.
Agresti, A., and Finlay, B. (1997). Statistical methods for the social sciences (3rd ed.). Upper
Saddle River, NJ: Prentice Hall.
Aiken, L. S., and West, S. G. (1991). Multiple regression: Testing and interpreting interac-
tions. Thousand Oaks, CA: Sage.
Albert, A., and Anderson, J. A. (1984). On the existence of maximum likelihood estimates in
logistic regression models. Biometrika, 71, 1–10.
Aldrich, J. H., and Nelson, F. D. (1984). Linear probability, logit, and probit models. Beverly
Hills, CA: Sage.
Allison, P. D. (1982). Discrete-time methods for the analysis of event histories. In S. Leinhardt
(Ed.), Sociological Methodology 1982 (pp. 61–98). San Francisco, CA: Jossey-Bass.
Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data. Beverly
Hills, CA: Sage.
Allison, P. D. (1995). Survival analysis using the SAS system: A practical guide. Cary, NC:
SAS Institute.
Allison, P. D. (1996). Fixed-effects partial likelihood for repeated events. Sociological
Methods & Research, 25, 207–222.
Allison, P. D. (1999). Comparing logit and probit coefficients across groups. Sociological
Methods & Research, 28, 186–208.
Anton, H. (1984). Calculus with analytic geometry (2nd ed.). New York: Wiley.
Axinn, W. G., and Thornton, A. (1993). Mothers, children, and cohabitation: The intergener-
ational effects of attitudes and behavior. American Sociological Review, 58, 233–246.
bref.qxd 30.8.04 12:00 Page 512
Balzer, W., Boudreau, N. S., Hutchinson, P., Ryan, A. M., Thorsteinson, T., Sullivan, J.,
Yonker, R., and Snavely, D. (1996). Critical modeling principles when testing for gender
equity in faculty salary. Research in Higher Education, 37, 633–658.
Barker, L., and Brown, C. (2001). Logistic regression when binary predictor variables are
highly correlated. Statistics in Medicine, 20, 1431–1442.
Begg, C. B., and Gray, R. (1984). Calculation of polychotomous logistic regression parame-
ters using individualized regressions. Biometrika, 71, 11–18.
Belsley, D. A., Kuh, E., and Welsch, R. E. (1980). Regression diagnostics: Identifying
influential data and sources of collinearity. New York: Wiley.
Bennett, N. G., Blanc, A. K., and Bloom, D. E. (1988). Commitment and the modern union:
Assessing the link between premarital cohabitation and subsequent marital stability.
American Sociological Review, 53, 127–138.
Bentler, P. M., and Chou, C. (1988). Practical issues in structural modeling. In J. S. Long (Ed.),
Common problems/proper solutions: Avoiding error in quantitative research (pp. 161–192).
Newbury Park, CA: Sage.
Berk, R. A. (1983). An introduction to sample selection bias in sociological data. American
Sociological Review, 48, 386–398.
Bickel, P. J., and Doksum, K. A. (1977). Mathematical statistics: Basic ideas and selected
topics. Oakland, CA: Holden-Day.
Blossfeld, H., and Hamerle, A. (1990). Unobserved heterogeneity in hazard rate models: A
test and an illustration from a study of career mobility. In K. U. Mayer and N. B. Tuma
(Eds.), Event history analysis in life course research (pp. 241–252). Madison, WI:
University of Wisconsin Press.
Blossfeld, H., and Rohwer, G. (1995). Techniques of event history modeling. Mahwah, NJ:
Erlbaum.
Blossfeld, H., Hamerle, A., and Mayer, K. U. (1989). Event history analysis. Hillsdale, NJ:
Erlbaum.
Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley.
Booth, A., Johnson, D. R., White, L., and Edwards, J. N. (1984). Women, outside employ-
ment, and marital instability. American Journal of Sociology, 90, 567–583.
Boudreau, N., Sullivan, J., Balzer, W., Ryan, A. M., Yonker, R., Thorsteinson, T., and
Hutchinson, P. (1997). Should faculty rank be included as a predictor variable in studies
of gender equity in university faculty salaries? Research in Higher Education, 38, 297–312.
Bradley, C. S. (2000). Does self-esteem contribute to good grades at the college level? A soci-
ological examination of the factors that impact academic performance in college.
Unpublished master’s thesis, Bowling Green State University.
Breen, R. (1996). Regression models: Censored, sample selected, or truncated data. Thousand
Oaks, CA: Sage.
Breslow, N. E. (1974). Covariance analysis of censored survival data. Biometrics, 30, 89–99.
Breusch, T., and Pagan, A. (1979). A simple test for heteroscedasticity and random coefficient
variation. Econometrica, 47, 1287–1294.
Cameron, A. C., and Trivedi, P. K. (1990). Regression-based tests for overdispersion in the
Poisson model. Journal of Econometrics, 46, 347–364.
Cameron, A. C., and Trivedi, P. K. (1998). Regression analysis of count data. Cambridge:
Cambridge University Press.
REFERENCES 513
bref.qxd 30.8.04 12:00 Page 513
Cameron, A. C., and Windmeijer, F. A. G. (1997). An R-squared measure of goodness of
fit for some common nonlinear regression models. Journal of Econometrics, 77,
329–342.
Caudill, S. B. (1988). An advantage of the linear probability model over probit or logit. Oxford
Bulletin of Economics and Statistics, 50, 425–427.
Chow, G. (1960). Tests of equality between sets of coefficients in two linear regressions.
Econometrica, 28, 591–605.
Chung, K. L. (1974). A course in probability theory (2nd ed.). San Diego, CA: Academic Press.
Clark, M. L. (2001). Factors associated with rule violations in correctional facilities.
Unpublished master’s thesis, Bowling Green State University.
Cleary, P. D., and Angel, R. (1984). The analysis of relationships involving dichotomous
dependent variables. Journal of Health and Social Behavior, 25, 334–348.
Clogg, C. C., and Haritou, A. (1997). The regression method of causal inference and a dilemma
confronting this method. In V. R. McKim and S. P. Turner (Eds.), Causality in crisis?
Statistical methods and the search for causal knowledge in the social sciences (pp. 83–112).
Notre Dame, IN: University of Notre Dame Press.
Clogg, C. C., Petkova, E., and Haritou, A. (1995). Statistical methods for comparing regres-
sion coefficients between models. American Journal of Sociology, 100, 1261–1293.
Collett, D. (1994). Modelling survival data in medical research. London: Chapman & Hall.
Cox, D. R. (1972). Regression models and life tables. Journal of the Royal Statistical Society,
34, 187–220.
Cox, D. R. (1975). Partial likelihood. Biometrika, 62, 269–276.
Cragg, J. G. (1971). Some statistical models for limited dependent variables with application
to the demand for durable goods. Econometrica, 39, 829–844.
Cragg, J. G., and Uhler, R. S. (1970). The demand for automobiles. Canadian Journal of
Economics, 3, 386–406.
DeMaris, A. (1991). A framework for the interpretation of first-order interaction in logit mod-
eling. Psychological Bulletin, 110, 557–570.
DeMaris, A. (1992). Logit modeling: Practical applications. Newbury Park, CA: Sage.
DeMaris, A. (1993). Odds versus probabilities in logit equations: A reply to Roncek. Social
Forces, 71, 1057–1065.
DeMaris, A. (1995). A tutorial in logistic regression. Journal of Marriage and the Family, 57,
956–968.
DeMaris, A. (1997). Elevated sexual activity in violent marriages: Hypersexuality or sexual
extortion? Journal of Sex Research, 34, 361–373.
DeMaris, A. (2000). ’Till discord do us part: The role of physical and verbal conflict in union
disruption. Journal of Marriage and the Family, 62, 683–692.
DeMaris, A. (2001). The influence of intimate violence on transitions out of cohabitation.
Journal of Marriage and Family, 63, 235–246.
DeMaris, A. (2002a). Covariance structure models. In M. Wiederman and B. E. Whitley
(Eds.), Handbook for conducting research on human sexuality (pp. 289–325). Hillsdale,
NJ: Erlbaum.
DeMaris, A. (2002b). Regression models. In M. Wiederman and B. E. Whitley (Eds.), Handbook
for conducting research on human sexuality (pp. 255–287). Hillsdale, NJ: Erlbaum.
514 REFERENCES
bref.qxd 30.8.04 12:00 Page 514