Caspi et al. (1998) [75]
Predictors of youth unemployment. Data from a longitudinal study of a
birth cohort from New Zealand, followed up for 21 years. Tobit models
for months unemployed between ages 15-21. $N=954$. BIC mentioned
in a footnote, used to reject a model with gender-by-predictor
interactions.
D'Unger et al. (1998) [117]
Poisson regression models to describe delinquent / criminal careers,
with the number of convictions or police contacts as the response. Data
from three longitudinal studies, $N$ approx. 400-13000.
Semiparametric mixture models where the baseline age trajectory in rate
is different for groups with different patterns of career, while
dependence on covariates is the same for all groups. [Thus also an
attempt to deal with overdispersion.] BIC used to select the number of
patterns (`latent classes'). BIC explained in an Appendix.
Manza and Brooks (1998) [257]
Gender gap in U.§. presidential elections, adjusting for other
factors. Logistic regression, pooled data from election surveys
1952-92, $N$ approx. 11000-13000. BIC used for selection of the
baseline (gender only) model, less when adding covariates.
Macintosh (1998) [251]
Testing the unidimensionality of an international attitude
(`postmaterialism') scale. Log-linear Rasch models, $N=50 177$. Model
with one latent trait is comprehensively rejected by both BIC and
$L^{2}$. Extensions of the model identify cross-national variation in
traits (item bias).
Smits et al. (1998) [343]
Educational homogamy [how similar are the education levels of husbands
and wives] in 65 countries. Loglinear models, true $N=487000$ but
$N=65000$ (scaled so that 10000 for each country) used in some
analyses. Differences in homogamy modelled using country-level
explanatory variables (one at a time). BIC used unquestioningly,
$L^{2}$ and $R^{2}_{L}$ also reported.
Breen and Goldthorpe (1998a) [56]
Analysis of the meritocracy of the British society, especially a
critique of Saunders [1996, 1997 etc.] Data from NCDS, $N=5090$ (2511
men, 2579 women); variables on origin, destination, ability, effort and
educational qualifications. Multinomial logit models examining whether
merit variables eliminate the effect of origin on destination. Education
most important mediating variable, effect stronger on men than women,
effect of origin remains strong.
Breen and Goldthorpe (1998b) [57] Extending the analysis of Breen and Goldthorpe (1998a) [56]. Comparing two cohorts (58 and 70), with data from two panel studies, to examine whether the British has become more meritocratic over time. No such evidence, partially even the opposite. The relative but not the absolute strength of education as a mediating merit variable increased.