Nonresponse error in injury-risk surveys

Area of Science:

• Epidemiology
• Public Health
• Survey Methodology

Background:

• Nonresponse in epidemiologic surveys can introduce significant error, particularly in injury control and risk assessment.
• Investigating nonresponse bias is crucial for accurate public health data.
• Random-digit-dialed (RDD) surveys are susceptible to nonresponse bias.

Purpose of the Study:

• To assess the impact of survey nonresponse on injury indicators using a records-matching approach.
• To determine if nonresponse bias affects estimates of violence-related and unintentional injuries.
• To evaluate the utility of a ZIP code-based linkage for understanding nonresponse error.

Main Methods:

• Merged RDD survey data (Illinois, 2003) with 2000 Census ZIP code-level data.
• Employed hierarchical linear models to predict individual survey response propensity using ZIP code indicators.
• Utilized ZIP code measures to predict injury risk indicators, including partner violence and DUI.

Main Results:

• ZIP code characteristics predicted both survey response propensity and partner violence reporting.
• Higher income ZIP codes showed lower response rates and less reporting of partner violence, suggesting overestimation of this violence.
• Smaller housing ZIP codes had lower response rates but higher reporting of partner isolation, suggesting underestimation.

Conclusions:

• Survey response propensity is linked to partner violence indicators but not driving-under-the-influence (DUI) indicators.
• The records-matching approach using ZIP codes is an effective and cost-efficient method for assessing nonresponse error.
• This methodology can be applied to surveys of injury prevention and other health-related outcomes.