Internet panels are widely used in market research because they can provide samples of
relatively low incidence subpopulations quickly and at low cost. Critics argue these are non-probability
samples with known and unknown biases. Proponents argue that known demographic biases are
corrected by sample balancing and weighting. This study examines the potential usability of Internet
panels for non-commercial health research.
As part of a study of asthma and its management in the United States, two national surveys were
conducted by telephone in 2009. The first was a national probability sample of 2,500 persons with current
asthma, obtained by telephone screening of 60,000 households sampled by random digit dialing. The
second was a national sample of 1,090 adults sampled by RDD. In both samples, the respondents were
asked whether they participate in any internet panel surveys, which ones, and how often they are
contacted to participate in Internet surveys.
If Internet survey panelists are defined as persons who belong to panels and are contacted as
least once a month to participate in them, then the national RDD surveys suggests approximately five
percent of the adult population of the United States are Internet panelists. These Internet panelists
different from those who do not belong to Internet panels on a number of key demographic characteristics
including age, income, race and ethnicity, marital status, employment status and housing type.
Despite known demographic biases, can Internet panels provide representative samples of
relatively rare disease populations, like asthma? This paper compares the disease and treatment
characteristics of nearly two hundred Internet panelists with more than two thousand non-panelists from
the national RDD asthma patient survey. We find symptom and treatment characteristics are similar for
panel and non-panel members with asthma. Hence, Internet panels should not be dismissed out of hand
as a potential source for health research.