Marginal Structural Models: An Application to Incarceration and Marriage During Young Adulthood

• Author: Edward H. Kennedy,Valerio Bacak
• DOI: http://doi.org/10.1111/jomf.12159
• Published date: 01 February 2015
• Date: 01 February 2015

V B University of Pennsylvania

E H. K University of Pennsylvania∗

Marginal Structural Models: An Application to

Incarceration and Marriage During Young Adulthood

Advanced methods for panel data analysis

are commonly used in research on family

life and relationships, but the fundamental

issue of simultaneous time-dependent con-

founding and mediation has received little

attention. In this article the authors intro-

duce inverse-probability-weighted estimation

of marginal structural models, an approach

to causal analysis that (unlike conventional

regression modeling) appropriately adjusts for

confounding variables on the causal pathway

linking the treatment with the outcome. They

discuss the need for marginal structural models

in social science research and describe their

estimation in detail. Substantively, the authors

contribute to the ongoing debate on the effects

of incarceration on marriage by applying a

marginal structural model approach to panel

data from the National Longitudinal Survey of

Youth1997 (N=4,781). In line with the increas-

ing evidence on the collateral consequences

of contact with the criminal justice system, the

authors nd that incarceration is associated

with reduced chances of entering marriage.

Recovering causal effects from observational

data is complicated by the possible biases from

confounding and nonrandom selection of units

Department of Sociology, Universityof Pennsylvania, 3718

Locust Walk, McNeil Building, Ste. 113, Philadelphia, PA

19104 (vbacak@sas.upenn.edu).

∗Department of Biostatistics and Epidemiology, University

of Pennsylvania, Room 507, Blockley Hall, 423 Guardian

Dr., Philadelphia, PA19104.

Key Words: incarceration, marginalstructural models, mar-

riage, quantitative methodology.

into treatment and control conditions. This

is especially the case in the social sciences,

where few research questions allow for a fea-

sible and ethical experimental assignment of

study participants. Even though researchers are

increasingly relying on methods that exploit

natural experiments to estimate causal effects,

these approaches are hardly a silver bullet for

identifying causality (Rosenzweig & Wolpin,

2000; Sekhon & Titiunik, 2012). Although

imperfect, the most popular resource for causal

analysis in social science research is survey

data, in particular data from an increasing num-

ber of nationally representative panel studies

(Elder, Kirkpatrick, & Crosnoe, 2003; Hal-

aby, 2004). Such panel data are a substantial

improvement over cross-sectional data when

testing causal hypotheses, but they present

researchers with unique challenges. In the

presence of time-dependent covariates that

have both a confounding and a mediating role

on the causal pathway between the treatment

and the outcome, estimates from conventional

regression approaches will be misleading. This

critical issue has received little attention in most

social science applications, including research

on family life and relationships, despite an

effective and practical solution grounded in

the increasingly popular potential outcomes

framework (Morgan & Winship, 2007).

In this article we introduce an approach

to causal analysis designed to appropriately

address simultaneous time-dependent con-

founding and mediation in longitudinal settings

(Robins, Hernán, & Brumback, 2000). Briey,

this occurs when the treatment predicts one

or multiple variables that subsequently predict

112 Journal of Marriage and Family 77 (February 2015): 112–125

DOI:10.1111/jomf.12159