Finding an instrument to purge that propensity to invest from the indicator of information is challenging. We want to be able to identify the potential for information to substitute physician services, once the decision to invest has been made. We want to keep the component of information searching purely related to propensity to gain valuable knowledge from search but independent of the need to invest in health (health status and preferences for health). To control for this, Kenkel (1990) and Hsieh and Lin (1997) have used education as an instrument for health information. Education should impact one’s propensity to use information well – their propensity to gain from using information. But preferences for overall good health should theoretically not vary by education alone. Low educated people, holding all observable differences like income and health fixed, should have the same taste for healthy living as high educated people. While health and health insurance status might be correlated with education, thereby driving access to both information and physician services, they are observables
we control for. Education does, however, directly affect access to
information as well as expected gain from using it. After testing,
we find that education has no independent effect on physician
services beyond the observables correlated with it. So education
predicts information but does not predict utilization of physician
services, holding health, health insurance and income fixed. Education
can be an exogenous instrument of information since it
predicts demand for it without being correlated with the error
term of the main equation. Also included as instrumental variables
is a variable indicating proportion of female-headed families
at the county level. As Pauly and Satterthwaite (1981) has noted,
a female-headed family has a tighter time budget and fewer social
networks, thus has a higher opportunity cost for searching health
information. And it is uncorrelated with the residual in the main
equations for health care demand. By using this instrumental variables
approach we hope to capture the component of information
that represents consumption of information independent of those
unobservable preferences when observable health status, health
insurance, income and demographic characteristics are controlled
for.6
6 We also conduct