Fitting and interpreting multilevel models in maternal and child health research

Background: Multilevel models allow researchers the ability to examine individual and contextual predictors of health and simultaneously investigate what predicts variance within and across contexts. However, the potential of multilevel models has gone under used in maternal and child health research, partly because few applied examples demonstrate the application of multilevel models in maternal and child health research. To address this, I take a non-mathematical approach and introduce the key interpretative concepts in multilevel modeling with a maternal and child health relevant example.

Methods: I used multilevel modeling and data from the 2005-2006 National Survey of Children with Special Health Care Needs (NS-CSHCN: n = 40,723) to investigate how an individual level variable (family income) and a contextual level variable (the proportion of families in poverty within in a state) relate to the length of time CSHCN go uninsured.

Results: Both family income and the proportion of families in poverty in a state predicted the length of time CSHCN go uninsured. Additionally, low family income had an especially strong negative effect on CSHCN living in states with larger proportions of the population in poverty. However, substantial variance across contexts (states) remained unexplained.

Conclusions: Results show how simultaneously investigating individual and contextual level predictors of children's health outcomes and partitioning the outcome variance into individual and contextual level components leads to more fully informed pictures of maternal and child health. In my discussion, I provide practical advice describing how maternal and child health researchers should fit and interpret multilevel models.