Abstract:
|
The use of non-probability samples in survey research has increased in recent years, but the validity of estimates varies, and it is not clear under which circumstances weighted non-probability samples yield valid inferences of population-level estimates. This study assesses the validity of data from non-probability panels in the context of health-related outcomes with items from the Heath Information National Trends Survey (HINTS), a national probability-based mail survey. We review differences in population-level estimates between HINTS and online surveys administered by four vendors, using a range of sampling methods including probability, quota-based, and convenience. First, we compare weighted estimates from the different methodologies to eight national benchmark estimates, and examine whether differences from benchmarks are associated with demographic characteristics. Second, we reproduce associations from the health literature and determine whether data from online and non-probability samples yield similar inferences. These analyses will help clarify conditions under which online panel data can be used to make valid inferences about health-related knowledge and behaviors.
|