Poster #70 - Sensitive Parenting Mediates the Association Between Maternal Representations and Child Behavior: A Demonstration of Bayesian Mediation

• In Event: Poster Session 17

Abstract

Maternal representations, or mothers’ internalized schemas of their children, begin to form during pregnancy. Longitudinal research indicates that prenatal representations impact postnatal parenting behaviors, such that representations that are emotionally disengaged and distorted (i.e. nonbalanced) predict poor postnatal parenting. In turn, parenting is one of the most salient influences on children’s development. However, due to little extant research, it is unclear whether prenatal representations have a direct or indirect effect through parenting on child development. To address this gap, the present study examined the impact of prenatal representations on children’s externalizing behavior mediated by sensitive parenting. We hypothesized that non-balanced prenatal representations would be associated with less sensitive parenting which, in turn, would be associated with greater externalizing behavior to result in a positive indirect effect of nonbalanced representations on externalizing behavior. We tested this hypothesis using 3 commonly recommended methods to illustrate the advantages of the Bayesian mediation approach.

Mother-child dyads (N= 206) were enrolled in a longitudinal study. Maternal representations were assessed with semi-structured interviews during pregnancy (Zeanah et al., 1994), parenting sensitivity was coded from free-play when infants were one year old (Ainsworth et al., 1971), and externalizing behavior was assessed via maternal report when children were four years old (Achenbach, 1991). Because participants were oversampled for intimate partner violence, IPV exposure was covaried in all analyses.

First, the indirect effect was tested with Bayesian estimation, an alternative to frequentist approaches, which has advantages in developmental research including high precision, low sample size requirements, and clear interpretability yet is less commonly used that frequentist methods (Van de Shoot et al., 2014). Bayesian mediation results were compared to two frequentist resampling methods, bias-corrected and percentile bootstrapping in order to illustrate similarities and differences in the three approaches. Frequentists disagree regarding the most ideal bootstrapping method: some recommend the bias-corrected method, arguing that the percentile method is overly conservative, while others recommend the percentile method, asserting that the bias-corrected method is overly liberal (Falk, 2018; Fritz & MacKinnon, 2007). These two methods often produce contradictory results leaving uncertainty about conclusions. Bayesian mediation is an alternative that can clarify effects. Thus, we illustrate the utility of Bayesian mediation.

The Bayesian approach supported the hypothesis; the 95% credibility interval for was estimated to lie between 0.03 and 2.56, indicating a positive indirect effect. The bias-corrected mediation effect was also significant, 95% CI [0.05 to 2.5], however the percentile was not, 95% CI [-0.04, 2.38]. Simulation studies in the methodology literature (Fritz & MacKinnon, 2007) suggest the sample was underpowered using the percentile method. Thus, the nonsignificant percentile result may represent a Type II error. Alternatively, given arguments that the bias-corrected approach is too liberal, the significant bias-corrected result may be a Type I error. However, the Bayesian method, which is the most appropriate, provided additional evidence of a significant indirect effect. Thus, the Bayesian approach clarified this indirect effect, demonstrating its utility for detecting mediation in underpowered samples.

Authors

• Alexandra Lauren Ballinger, Michigan State University
  Presenting Author

• Amy K. Nuttall, Michigan State University

• Alytia Akiko Levendosky, Michigan State University

• G. Anne Bogat, Michigan State University