Cheating the Rating Systems in Healthcare Credence Goods Markets

We experimentally investigate the impact of cheating a public rating system in healthcare credence goods markets. Therefore, we plan to employ a laboratory experiment framed in a healthcare context, where experts are called physicians and consumers are called patients, using a student sample from the University of Innsbruck.

To start with, we plan to run three experimental conditions. In the baseline condition, there is no feedback mechanism in place. Next, we introduce a public rating mechanism into the market, where patients can rate the interactions with physicians on a five-star-rating-scale. Given that the feedback mechanism enhances market outcomes, we plan to run a follow-up treatment where we investigate the robustness of the feedback mechanism to cheating, i.e. we allow physicians to buy up to three additional five-star ratings per period to improve their public rating.

Our design allows us to investigate the robustness of public rating mechanisms to fraud by introducing the possibility to cheat.

The experiment will be conducted at the Econ Lab at the University of Innsbruck. We build our experimental design on the credence goods framework of Dulleck and Kerschbamer (2006) and slightly adapt it to better resemble healthcare markets. Throughout the experiment, we implement a health care framing in which we refer to consumers of the credence good as patients and sellers as physicians, respectively.

Experiment

The basic set-upput and parameterization:

In our basic set-up, patients and physicians are grouped in a market of 8 subjects (4 patients & 4 physicians). Patients suffer from a major health problem with probability h = 0.5 and a minor one with probability (1-h). The probability h=0.5 is common knowledge. Patients decide whether to consult a physician knowing that they suffer from some health problem in every period. They do not get information about the severity of their health problem. Physicians diagnose their patients’ health problem with certainty and at zero costs. They provide one of two treatments, a major or a minor treatment. The cost for the physician to provide the major treatment (cH) is 4 ECU (Experimental Currency Unit). The cost for the minor treatment (cL) is 2 ECU. Treatment prices, paid by an insurance company (not represented by participants in the lab), are either 8 ECU (pH) or 4 ECU (pL). The major treatment cures both, the major and the minor health problem, while the minor treatment only cures the minor one. Patients obtain 6 ECU (v) if cured, and zero if treated insufficiently. The payoff for patients consulting a physician is the difference between the obtained value v and a disutility, which depends on the type of treatment. If patients have to go through major treatment, they bear a disutility of 4 ECU (dH), while the disutility for a minor treatment is 1 ECU (dH). Hence, the payoff for patients is 2 ECU if undergoing major treatment and 5 ECU for undergoing minor treatment. For physicians, the payoff is the spread between the price charged (pH or pL) and the cost for the chosen treatment (cH or cL). In case a patient decides against consulting a physician, the patient receives an outside option of (-0.5) ECU (oPat). Physicians receive oPhy = 0 if they do not interact with any patient in a given round. Compared to the framework of Dulleck and Kerschbamer (2006), our basic model differs in three dimensions. First, the outside option of patients is negative (oPat = -0.5) illustrating the disutility of an uncured health problem. Second, pH and pL are exogenously fixed, which is common in highly regulated health care markets. Third, patients are insured, i.e. they do not have to pay the price for the treatment. However, unlike in other credence goods markets (e.g. car repair), the patients themselves have to undergo the treatments, and thus, compared to a minor treatment, a major treatment results in a higher disutility, irrespective of the type of health problem. Throughout our experiment, we implement verifiability, that is, physicians can only charge the price for the treatment they perform (i.e. overcharging is ruled out by design) and participants are not identifiable such that reputation building in the treatments without feedback mechanism is excluded.

The structure of the stage-game is as follows:

1) For each patient, nature draws the type of health problem. With probability h patients have a major health problem, and with probability (1-h) patients have a minor health problem.
2) Patients decide whether to consult a physician. If patients decide not to visit a physician, the period ends. Otherwise, they choose one physician from a list of four.
3) Physicians are informed about the health problem and provide a treatment (q_H or q_L).
4) Patients and physicians observe their payoff in the respective period. Note that patients cannot infer whether their physician treated them appropriately, they only learn which treatment was chosen and whether it was sufficient to cure the health problem.
5) In the conditions with a public rating system: After learning the payoff for the respective period, patients decide whether to rate the interaction with their treating physician. If they decide to rate the interaction, they choose the rating on a scale between 0 and 5 stars which is shown to the treating physician afterward.
6) In the conditions where physicians may buy ratings: After learning the payoff for the respective period, physicians decide whether to buy up to three additional 5-star-ratings. For each additional rating, the physician has to pay 1 ECU .

[Treatment Variation]
As explained above, we plan to run (at least) four treatments:

[Experimental Condition 1] — No Feedback-Mechanism
[Experimental Condition 2] — Public Feedback-Mechanism
[Experimental Condition 3] — Public Feedback-Mechanism + Buy Rating 3

2021-11-16

2022-01-31

Overtreatment-rates

Overtreatment is characterized by the fact that the patient needs the mild treatment (𝑞l) but receives the major treatment (𝑞H). The overtreatment rate is the number of actual overtreatment decisions divided by the number of interactions with patients with a mild health problem.

Market Efficiency as the sum of patient, physician and insurance payoffs.

We plan to use a student sample from the University of Innsbruck and run each experimental condition with 48 subjects (as suggested by our power analysis). Therefore, we plan to run two sessions with 24 subjects each in every experimental condition. All sessions are run computerized using z-Tree and students are recruited using hroot. Participants do not know which experiment they are going to participate in when they register. They only receive information about the expected duration of the experiment (2h).
Our experiment is structured as follows for all our conditions:

Stage 1: The experimenter explains the experiment and participants read the instructions.
Stage 2: Participants answer several control questions to ensure they understood the game.
Stage 3: The computer randomly assigns roles and markets to participants.
Stage 4: Participants play the game for 16 periods.
Stage 5: Participants participate in additional games: an individual risk preference task, a dictator game, a lying task, and a trust game.
Stage 6: Participants fill out a questionnaire.

Randomization is carried out in the experiment by a computer.

at the session level

Yes

6 clusters á 8 individuals per experimental condition.

48 (6 x 8) individuals per experimental condition.

192 (3 x 48) individuals (students at the University of Innsbruck).

Based on previous findings, we performed a power calculation, indicating that we need six clusters á 8 subjects per experimental condition when aiming for a power of 80%.