The following is a summary of a research project by the authors. For more information please see the published working paper:
Author: Nathan Kettlewell1
- Economics Discipline Group, University of Technology Sydney, and the Institute of Labor Economics (IZA).
Summary of Research
Most decisions involve an element of uncertainty. In these situations, people form expectations about the likelihood of an outcome occurring, and then use these expectations to inform their choices.
In a current project, Nathan Kettlewell address the question: are people’s expectations about their likelihood of using certain types of health care (e.g. being hospitalised) accurate? This question has important policy implications. If people have biased beliefs, they may be making sub-optimal health and lifestyle decisions. On the other hand, if beliefs are accurate, this information could be leveraged to improve decision making. For example, private beliefs could be used together with plan selection software to better match people to insurance policies.
Kettlewell’s research suggests that people’s expectations about health care utilisation are quite accurate over a 12 months horizon.
To evaluate people’s expectations, he surveyed 1,528 people aged 25-64 years and asked them about their expected probability (0-100%) of visiting various health care providers in the next 12 months. He also asked about actual visits in the previous 12 months. If expectations are accurate on average, then the means of expected use and past use should be roughly equal. The figure below compares these means for childless singles (who are less likely to conflate own health care use with that of a spouse or child). People somewhat underestimate hospitalisation risk, and overestimate visits to a naturopath, but otherwise their expectations are remarkably close to actual behaviour.
To further test the accuracy of people’s expectations, he estimated the correlations between predictors of expectations and past behaviour (e.g. age, self-assessed health). Across all health services, the magnitudes and signs of the correlations were similar between expectations and past behaviour. Finally, he looked at the correlations between ‘objective risk’ – obtained with a prediction model estimated out-of-sample – and expectations, and found these were meaningfully positively correlated.
Kettlewell’s paper also demonstrates how subjective expectations data could be used in empirical research to answer important policy questions. The appeal of these questions is that they are simple and quick – adding a couple of questions on expected hospitalisation risk to a household survey, for example, is highly feasible.
One use for these data is estimating welfare effects associated with the quality gap between the private and public hospital systems (e.g. disparate waiting times). By combining subjective expectations with a simple theoretical model for the choice to privately insure, Kettlewell estimated how much money a person would need to be indifferent between treatment in the public versus private system. While Kettlewell warns that the data in his small survey are not rich enough to provide a credible estimate, the exercise demonstrates the value of including these data in large household (ideally longitudinal) surveys. Subjective expectations data circumvent the challenge of estimating hospitalisation risk and can aid the estimation of important, policy-relevant metrics.