TY - JOUR
T1 - Impact of missing data strategies in studies of parental employment and health
T2 - Missing items, missing waves, and missing mothers
AU - Nguyen, Cattram D.
AU - Strazdins, Lyndall
AU - Nicholson, Jan M.
AU - Cooklin, Amanda R.
N1 - Publisher Copyright:
© 2018
PY - 2018/7
Y1 - 2018/7
N2 - Background: Understanding the long-term health effects of employment – a major social determinant – on population health is best understood via longitudinal cohort studies, yet missing data (attrition, item non-response) remain a ubiquitous challenge. Additionally, and unique to the work-family context, is the intermittent participation of parents, particularly mothers, in employment, yielding ‘incomplete’ data. Missing data are patterned by gender and social circumstances, and the extent and nature of resulting biases are unknown. Method: This study investigates how estimates of the association between work-family conflict and mental health depend on the use of four different approaches to missing data treatment, each of which allows for progressive inclusion of more cases in the analyses. We used 5 waves of data from 4983 mothers participating in the Longitudinal Study of Australian Children. Results: Only 23% had completely observed work-family conflict data across all waves. Participants with and without missing data differed such that complete cases were the most advantaged group. Comparison of the missing data treatments indicate the expected narrowing of confidence intervals when more sample were included. However, impact on the estimated strength of association varied by level of exposure: At the lower levels of work-family conflict, estimates strengthened (were larger); at higher levels they weakened (were smaller). Conclusions: Our results suggest that inadequate handling of missing data in extant longitudinal studies of work-family conflict and mental health may have misestimated the adverse effects of work-family conflict, particularly for mothers. Considerable caution should be exercised in interpreting analyses that fail to explore and account for biases arising from missing data.
AB - Background: Understanding the long-term health effects of employment – a major social determinant – on population health is best understood via longitudinal cohort studies, yet missing data (attrition, item non-response) remain a ubiquitous challenge. Additionally, and unique to the work-family context, is the intermittent participation of parents, particularly mothers, in employment, yielding ‘incomplete’ data. Missing data are patterned by gender and social circumstances, and the extent and nature of resulting biases are unknown. Method: This study investigates how estimates of the association between work-family conflict and mental health depend on the use of four different approaches to missing data treatment, each of which allows for progressive inclusion of more cases in the analyses. We used 5 waves of data from 4983 mothers participating in the Longitudinal Study of Australian Children. Results: Only 23% had completely observed work-family conflict data across all waves. Participants with and without missing data differed such that complete cases were the most advantaged group. Comparison of the missing data treatments indicate the expected narrowing of confidence intervals when more sample were included. However, impact on the estimated strength of association varied by level of exposure: At the lower levels of work-family conflict, estimates strengthened (were larger); at higher levels they weakened (were smaller). Conclusions: Our results suggest that inadequate handling of missing data in extant longitudinal studies of work-family conflict and mental health may have misestimated the adverse effects of work-family conflict, particularly for mothers. Considerable caution should be exercised in interpreting analyses that fail to explore and account for biases arising from missing data.
KW - Maternal mental health
KW - Missing data
KW - Multiple imputation
KW - Non-response
KW - Work-family conflict
UR - http://www.scopus.com/inward/record.url?scp=85044132937&partnerID=8YFLogxK
U2 - 10.1016/j.socscimed.2018.03.009
DO - 10.1016/j.socscimed.2018.03.009
M3 - Article
SN - 0277-9536
VL - 209
SP - 160
EP - 168
JO - Social Science and Medicine
JF - Social Science and Medicine
ER -