Poster #100 - Classifying Parents' Behavioral Application of Information About Infant and Toddler Feeding Using Decision Tree Modeling

• In Event: 3-206 - Poster Session 13
  In Poster Session: PS 13 Section - Health, Growth, Injury

Integrative Statement

Increasing the reach and impact of evidence-based information about childhood obesity prevention remains an important public health challenge (Ludwig, 2018). Parents of infants and toddlers are an important target audience for childhood obesity prevention (Paul et al., 2009). However, little is known about how parents naturally find and apply information regarding infants’ growth and nutrition needs (Bentley et al., 2014). Such research could inform the content and delivery characteristics of childhood obesity prevention programming and promote effectiveness. This study explored features of parents’ behavioral application of information about infant and toddler feeding that parents intentionally obtained during a naturally occurring information seeking occasion.

Methods. First-time parents (N = 423; 89% Female; 75% White/Caucasian) of children less than 24 months old completed an online survey on their most recent information seeking occasion about infant and toddler feeding. Parents reported on potential features of behavioral application of information drawn from a novel conceptual model of information seeking and use: Demographic and socioeconomic characteristics; Information seeking aptitude (i.e. topic interest; information urgency); Behavioral capacity and intent (i.e. prior and current topic knowledge; strength of behavioral intent); and Information integration (i.e. cognitive and affective responses; search status; information satisfaction). Parents were coded as having applied information if they endorsed: Making a purchase; Preparing a recipe; Using a new feeding approach; and/or Using a new parenting approach.

Analyses. Decision tree modeling is a machine-learning approach to identifying how independent variables and their interactions discriminate the value of a dependent variable (Breiman et al., 1984). A pruned decision tree model was generated using a two-step growth and pruning approach (Breiman et al., 1984; Hastie et al., 2009). Training data (80%) were balanced using the random over-sampling examples (ROSE) approach (Lunardon et al., 2014). The pruned model was determined by the lowest cross-validated error rate (Hastie et al., 2009).

Results. Information integration variables, specifically motivation, information satisfaction, and search status, were among the most salient features of behavioral application. Behavioral capacity and intent variables also emerged as salient features, specifically current and prior topic knowledge, and information urgency (see Table 1). The sub-tree with 9 terminal nodes (8 splits) had the lowest cross-validated error value (see Figure 1). This pruned model had an accuracy of 62.4% and an area under the receiver operating curve (AUC) value of 0.740 (SE = 0.059), which suggests the features that collectively contributed to the model had a medium to large effect size in classifying behavioral application (Rice & Harris, 2005). Including demographic and socioeconomic characteristics did not significantly improve classification performance.

Conclusion. The results of this cross-sectional study indicated parents were more likely to behaviorally apply information about infant and toddler feeding when they felt motivated and moderately-to-highly satisfied with information. Parents were less likely to apply information when they did not feel motivated and perceived their search as ongoing. To inform childhood obesity prevention programming, future research should examine longitudinal associations between the salient features identified in this study and parents’ behavioral application of information about infant and toddler feeding.

Authors

• Allison Hepworth, University of Maryland School of Social Work
  Presenting Author

• Meg Small, Penn State University
  Non-Presenting Author

• Timothy R Brick, The Pennsylvania State University
  Non-Presenting Author