Through a large-scale telephone interview, we elicited risk attitude metrics for each doctor using direct stated preferences to take risks in four different contexts. We measured the patient-regarding preferences of GPs by asking similar questions involving their own health and the health of their patients.
Using this specific population of GPs, we checked and verified results that were already found in the literature. However, we also obtain results that demonstrate the particularities of the physisian population. We comment in more details these results with regard to gender, age and health domains.
Gender difference in risk attitudes have been the object of numerous empirical findings in behavioral economics (for literature reviews, see [22, 23]) and psychology . In most of these studies, women are found to be more risk averse than men . This result holds when using Domain-Specific Risk-Taking Scale, DOSPERT  or self-assessed Likert measures of risk attitudes in cross-sectional surveys . For GPs, we observe significantly more risk aversion among women in the general and financialFootnote 9 contexts and extend this result to the patient’s health domain. However, we do not observe such a gender effect in the physician’s health domain, unlike in  where women are more risk averse than men concerning their own health. This confirms that the personal health domain seems to be a context in which the GPs substantially depart from the general population with respect to risk attitudes. Such differences with the general population are also observed for specific populations when risk attitudes are measured in their domains of expertise or practice .
There are several studies in behavioral economics [17, 27] revealing a gradually lower willingness to take risks across the life span in cohorts, suggesting that individuals become more risk-averse as they grow older. Recent cross-cultural psychological meta-study  confirms this empirical result that appears to be robust to the risk attitude elicitation method [26, 29].
In our GPs’ panel we oberve a significant age effect in the general and patient’s health domains. Interestingly, in these two domains, older GPs are less risk averse than are younger ones. No significant effect of age is found in the financial and GP’s own health domains. Regarding the general domain, this result contradicts the common finding in the literature and highlights the particularities of the GP population. For example, the hypothesis proposed by  or , which explains this age effect on risk attitudes by a decrease in cognitive abilities due to aging, may not apply to our surveyed population of GPs since only 5% of the interrogated GPs are over 70. It should also be noted that the mean age in our sample is 50.1 years, std 9.6, min 29, max 76 which is a rather limited age range. In addition, all the interrogated GPs are active, even the older ones (who also have the higher volume of activity and revenue).
Concerning the effect found in the patient’s health domain, two interpretations are possible: on the one hand, the older physicians may be willing to take more risks for their patients because they are more experienced and potentially aware of the upsides of, occasionally, pursuing risky options. On the other hand, a generational effect could explain this result. Older GPs may be less concerned by possible lawsuits in the event of medical errors or may have an extremely self-centered approach to the doctor-patient relationship due to their antiquated medical education, which induces a greater willingness to take risks involving their patients.
It is however important to note that whatever age category is considered, GPs take significantly more risk regarding their health than regarding the health of their patients; this point is addressed in greater detail in the next section.
The within discrepancy among domains of health
A troubling result of the study is that GPs’ willingness to take risks involving their own health appears to be much higher than for the general population and appears much higher than the risk they declare to be willing to take for their patients, creating a discrepancy between self and others. In addition, unlike in , we do not find any gender nor age effect in this domain. The particularity of the GP population is, thus, really striking.
The domain in which GPs are the most reluctant to take risks is their patient’s healthFootnote 10. The discrepancy between the GPs’ risk attitudes regarding their own healthFootnote 11 and that of their patients is extremely large and might have significant consequences for medical behaviors. Indeed, we can legitimately imagine that a GP would not have the same prescriptive behavior for himself as for his patient when facing the same medical symptoms, although patients often expect their physician to make the same decision for them as he would for himself. This result showing that GPs are more risk averse when they are addressing the health of their patients could be interpreted in different ways:
An economic rational A first explanation of this result could be that physicians take less risk when the health of their patient is involved because this type of risk taking might lead to lawsuits and potential monetary losses. Given the French medical system, in which such lawsuits are extremely rare, this interpretation can be excluded. However, we conjecture that the gap we found is likely to be more pronounced in countries where the legal system is harsh towards medical caregivers.
A social desirability bias During interviews, doctors intentionally may have reported a discrepancy between patient’s health and own health risk attitudes because they believed they “politically” had to. In our view, this interpretation can be eliminated because the order of questions regarding “own health” and “patient’s health” was randomized. Because we find no order effectFootnote 12, the measured gap cannot be a consequence of “desirable” response behaviorFootnote 13 intended to create a contrast (when GPs respond to the first question, they do not know that they will have to respond to the second one)
A paternalistic bias In their daily practice -and not only in interviews- doctors do not act for their patients as they would act for themselves. In our framework, this paternalistic attitude is not what  call “asymmetric” or “libertarian paternalism”, which would be desirable for the patient. In fact, our finding may imply a real decline in opportunities and medical options proposed to patients. Indeed, GPs explicitly recognize that they support an higher degree of risk for themselves that they would suggest their patient to take, potentially reducing the scope of medical options they will advise them to consider. That is why we would rather qualify this discrepancy between own’s and patient’s risk attitudes of “paternalistic biais” rather than libertarian paternalism.
A self-correction of the pre-existing gap The fourth interpretation assumes that GPs substantially differ from the general population with respect to risk attitudes in the health domain (GPs are more willing to take risks). Thus, an (highly optimistic) interpretation could be that GPs, reporting safer attitudes for patients than for themselves, tend to rectify the gap that they know there exists between them and the general population. This gap could have several origins: GPs differ from patients in their medical knowledge and informationFootnote 14, as well as in their access to care. Thus a physician may take greater risk for his health, as he is certain to receive appropriate assistance in the event of an adverse outcomeFootnote 15. Aware of this gap, the GP would then be attempting to act as a “perfect agent” of the patient  and establish the proper attitude using a sophisticated adjustment. In this situation, the GP would dodge any preferences’ diagnosis and the gap highlighted would have adverse consequences for the healthcare system. Indeed, following the paradigm of health professionals based on patient-centered communication and shared decision-making, the physician should not make the decision based on what seems as the proper decision for himself but based on what seems right for the patient.
Naturally, this last interpretation would require the investigation of risk attitudes in a representative sample of the French population to confirm the existence of the GP/patient risk attitude gap in the health domainFootnote 16.
Another limitation of our study is the type of metric we used to measure risk attitudes, i.e., a self-assessed measure of willingness to take risks on a 0-10 scale. These survey measures are, by construction, not incentivized, which could be an important concern regarding the measurement of risk attitudes in the monetary domain. However,  showed that, in a survey-experiment using a sub-sample of their population, their general risk question was significantly correlated with the measure derived from incentivized binary lottery choices, suggesting that these two methods indeed measured the same psychological trait. We therefore assume that this relationship holds in our survey and allows for a meaningful interpretation of our results in terms of risk attitudes.
Concerning the health domains, incentivization of the questions is impossible both in survey measures and in “quantitative” measures of risk attitudes involving lottery choices with consequences framed in terms of health. In a companion paper,  elicited GPs risk attitudes using hypothetical binary lottery choice questions with three different attributes (money, own health and patient’s health). In the two health domains, GPs had several binary choices to make between two therapies: one safe therapy that provides a certain amount of additional years of living in good health and a risky therapy that provides an higher number of additional years of living in good health with probability p and nothing with probability 1−p. Using a between-subject analysis,  find a similar systematic discrepancy between the GPs’ willingness to take risks for their patients’ health and for their own. Although their measures are also based on hypothetical choices, they rely on experimental technics that are commonly used in behavioral and experimental economics for measuring risk attitudes in various domains . It is thus reassuring to find the same qualitative result with both types of measures since this study is the first to collect attitudes in the patients’ health domain using a Likert scale measure of willingness to take riskFootnote 17.
Finally, [36,37] use the risk attitudes measures presented in this paper as explanatory variables of actual medical behaviors. Michel-Lepage et al.  find that risk-averse GPs use more Rapid Antigen Diagnostic Tests (RADTs) in tonsillitis in children, and  find that risk-averse GPs were more often vaccinated against seasonal and pandemic influenza, more often recommended the pandemic influenza vaccination and were also more in favor of vaccination in general. Concerning the measure of willingness to take risks for the health of their patients (that has been introduced in this study for the first time),  find that it is associated with two medical practices (no prescription of antibiotics and update of a booklet) which suggests a good external validity of this question. The statistical significance of these risk attitude scales in the explanation of professional medical behaviors emphasizes the utility of the study and measures of this personality trait –risk attitudes– for the medical professions and for the population they address. Furthermore, the potential influences of GPs’ individual characteristics on clinical decision making , also confirmed by our study (through age, gender and activity effects on GPs’ risk attitudes), may lead to individual physician practices variation part of which are (probably) not desirable . In highlighting the specificities of the GPs’ population (especially an unconventional age effect ), our study may help to understand and prevent potential medical practice variations among the French GPs’ population and to provide solutions for targeting “at risk” GPs with personalized practice recommandations.
Our main point concerns the self vs patients gap found in this study which extends the existing findings in self versus others medical decision making [10,11] and in other domains [12,14]. In fact, this gap could be detrimental when the gap is not justified by a difference of preferences between GPs and the general population. In this case, the GP tend to underestimate their patients’ willingness to take risks and medical conflicts that may arise between doctors and patients , as the core of the patients’ preferences i.e. true risk attitudes, would be neglected. The GPs may recommend them medical options that do not match their risk attitudes, involving a deep public health issue . To account for the relevance of this issue, we must check the reality of the gap and evaluate clearly its precise magnitude; only a systematic study of the difference between actual patients’ risk preferences and GPs’ risk preferences for their patients would be able to provide this precise measurement (see for example , which used mirrored questions).