Correcting non-response bias in United States national survey estimates

A photo of US Census

We are extending the application of developed non-response methodology to the United States’ National Health Interview Survey (NHIS) and National Health and Nutrition Examination Survey (NHANES) which have experienced the typical declines in response levels over time. Specifically, in the US context, there are substantial gender and racial/ethnic disparities in survey response as well as gender and race-specific health disparities. For example, men and Black and Hispanic individuals are less likely to respond, and also appear to have different smoking and alcohol consumption patterns than other groups. National surveys indicate that women as well as some minority race/ethnic groups are less likely be heavy drinkers and less likely to smoke, yet among drinkers and smokers, women and racial/ethnic minorities have higher alcohol- and tobacco-related mortality rates. Thus the impact of differential non-response patterns on estimation of health behaviours by gender/race is unknown but could bias assessment of inequalities. By addressing non-response bias through linked mortality records in the United States, a more accurate assessment of disparities can be determined, and efforts to reduce inequalities can be evaluated more rigorously using national survey data.

In work conducted so far, we have found survey respondents to have lower death rates than the general US population. This suggests that they are a systematically healthier source population. Incorporating non-household samples and revised weighting strategies to account for sample frame exclusion and non-response may allow for more rigorous estimation of the US population’s health.

This project is funded by the MRC and CSO and Columbia University in the City of New York Calderone award.


Columbia University in the City of New York

Katherine Keyes

Silvia Martins

Caroline Rutherford

MRC/CSO Social and Public Health Sciences Unit logo 800 wide