Page 51 - IJPS-2-2

Mark Lyons-Amos

effect of national level policies on life course events, such as fertility or partnership transitions, or
the effect of some other exogenous factor (e.g. Neels, Theunynck, and Wood, 2013; Billingsley and
Farrini, 2014). The major advantage of this approach is that it allows the integration of policy indi-
cators as contextual variables with more general clustering parameters to capture unobserved or un-
specified country level characteristics (e.g., cultural variation). Country level variation is also of in-
terest in many other research areas, such as politics or sociology (Hox, van de Schoot, and Mat-
thijsse, 2012).
Unfortunately, standard multilevel or random effects models, and the aim of examining mi-
cro-macro interactions are not coherent when the higher level clusters are countries. Specifically, the
model can complicate the interpretation of country specific variation, small numbers of coun-
tries can complicate model estimation, and the fundamental assumptions of the multilevel model are
not compatible with national level data. Random effects multilevel modelling is therefore not an ap-
propriate solution when examining the effect of national level characteristics on individual demo-
graphic behaviour. Unfortunately, fixed effects models provide little by way of alternative, due to
their limited ability to provide inference and statistical inefficiency. Fixed effects also exhibit other
issues such as potentially inefficient estimation for large numbers of countries (since each country
within the dataset requires a new parameter within a regression model) and, critically where country
level policies are of research interest, the inability to include covariate information at the cluster lev-
el (since the country specific parameter now confounds the national level policy information).
Some authors have attempted to overcome the limitations of a purely data driven approach by us-
ing a priori specifications of country typologies. Esping-Andersen (1990 and 1999) provided an at-
tempt at these classifying countries as belonging to different social welfare regimes within Western
Europe; countries belonging to either liberal, corporatist-statist or social democratic welfare regimes,
extended by Blossfeld and Drobnic (2001) to incorporate former socialist countries. Further attempts
at linking welfare regimes to demographic behaviour have been made by Blossfeld (2006) characte-
rising countries as being conservative (Germany and the Netherlands), southern (Italy and Spain),
liberal (UK and US), social democratic (Sweden, Norway, Denmark) and post- socialist (Czech Re-
public, Estonia, Hungary), as well as other demographic examples (Korpi, Ferrarini, and Englund,
2013; Korpi, 2000; Kalwij, 2010). There are considerable advantages to the typology driven ap-
proach. The typologies derived show few of the disadvantages of a purely empirical based approach,
since typologies can be derived from a small sample of countries and by their nature are interpretable.
Additionally, the grouping derived will be conceptually valid and consistent with existing theoretical
understandings; this potentially may not occur in a purely empirical approach. That said, the major
drawback of the typology based approach is that typologies have to be specified a priori by the re-
searcher, and the ability to validate these groupings can often be neglected when linking typologies
to the variable of interest.
In this paper an alternative means for the analysis of individual and national interactions is pro-
posed, through the use of two level latent class growth models. These models provide the ability to
generate clusters at the national level, but based on observed characteristics, rather than the distribu-
tional characterisation of random effects models. The observed characteristics used to define
the class provide a substantive interpretation. This has the additional advantage of being a means by
which theoretically derived country level typologies (Esping-Andersen, 1990; Esping-Andersen,
1999; Blossfeld, 2006) can be validated empirically. I apply this model to data from the Harmonized
Histories dataset and GGP contextual database, which captures individual level demographic
and country level welfare data in the European context. Individual level demographic behaviour is
measured through the three processes of the timing of first marriage, the timing of first cohabitation
and the timing of first birth. I classify countries based on relevant socio-economic (family allowance,
social support) and legal (recognition of cohabitation within the legal framework) characteristics,
and allow the timing of demographic behaviour to vary by class. This provides results which expli-

International Journal of Population Studies | 2016, Volume 2, Issue 2 45