Confounding of a Public School Health Intervention in a Nonrandomized Experiment

• Activity Number: 40
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Terrence Murphy*+
• Keywords: instrumental variable ; non-randomized ; school intervention ; healthy food

Abstract

In a nonrandomized experiment it is challenging to find the unconfounded association between a public school health intervention and an outcome. The Healthy Food Certification (HFC) was implemented in Connecticut (CT) in 2006 as a voluntary program that subsidized CT school districts participating in the National School Lunch Program (NSLP). HFC implemented state nutritional standards on all competitive foods sold at schools in participating districts. Roughly half the eligible districts signed up for HFC in the first year. We test for the association between initial HFC participation and the rate of paid middle school NSLP meals served in 2007 in HFC and non-HFC districts. Several modeling approaches are considered, including a spatial distribution among neighboring districts and a naively (not adjusted for confounding) adjusted random effects model. The naively adjusted model shows a significant positive association. We repeat the random effects analysis with an instrumental variable (IV) and see that the naively significant association with HFC becomes nonsignificant.