Reputation-Driven Adoption and Avoidance of Algorithmic Decision Aids in Credence Goods Markets

TL;DR

This paper studies how reputation and obfuscated expert ability affect the adoption of diagnostic technologies in credence-goods markets with diagnostic uncertainty. It develops a model contrasting Skill investments with Algorithmic decision aids and tests predictions via two online experiments that manipulate future interaction and signaling opportunities. The results show that high-ability experts strategically forego algorithm adoption to signal their type, generating pooling or separation outcomes depending on reputation dynamics, whereas Skill investments do not differentiate types. The findings highlight how signaling, beliefs, and repeated interactions shape technology diffusion in professional services, with implications for policy and practice in sectors like health care and repair services.

Abstract

In credence goods markets such as health care or repair services, consumers rely on experts with superior information to adequately diagnose and treat them. Experts, however, are constrained in their diagnostic abilities, which hurts market efficiency and consumer welfare. Technological breakthroughs that substitute or complement expert judgments have the potential to alleviate consumer mistreatment. This article studies how competitive experts adopt novel diagnostic technologies when skills are heterogeneously distributed and obfuscated to consumers. We differentiate between novel technologies that increase expert abilities, and algorithmic decision aids that complement expert judgments, but do not affect an expert's personal diagnostic precision. When consumers build up beliefs about an expert's type through repeated interactions, we show that high-ability experts may strategically forego the decision aid in order to escape a pooling equilibrium by differentiating themselves from low-ability experts. Without future visits, signaling concerns cause all experts to randomize their investment choice, leading to under-utilization from low-ability experts and over-utilization from high-ability experts. Results from two online experiments support our hypotheses. High-ability experts are significantly less likely than low-ability experts to invests into an algorithmic decision aid if reputation building is possible. Otherwise, there is no difference, and experts who believe that consumers play a signaling game randomize their investment choice.

Figures (24)

• Figure A1: Timing of events in each round for Phase 2. In Phase 1, experts skip the investment decision.
• Figure A2: Left: Price setting under fully transparent and fully obfuscated expert abilities. The shaded areas represents the high-ability expert's loss. Due to consumer uncertainty, the high-ability expert cannot choose the profit-maximizing price vector $\mathbf{P}^e$ as often as they would want to. Variables for this figure: $q = 0.5$, $\gamma = 1/3$. Right: Variations for different consumer beliefs $\tilde{\gamma}$.
• Figure A3: Consumer Belief Updating with $Pr(L) = 0.4$ and $Pr(L) = 0.2$.
• Figure A4: Panels (a) and (b) show the separating equilibrium space in which only the high-ability expert invests if (i) $x_{ha} = 0$, $x_{la,i} = 3$, $x_{la,j} = 0$ or (ii) $x_{ha} = 1$, $x_{la,i} = 2$, $x_{la,j} = 0$. Panels (c) and (d) illustrate mixed equilibrium space where only one low-ability expert invests for $x_{ha} = 1$, $x_{la,i} = 2$, $x_{la,j} = 0$. In Panel (d), LA$_0$ invests. For all cases, experts do not observe the distribution of "other" consumers who do not approach them.
• Figure A5: Strategic expert separation in which only the high-ability expert foregoes the decision aid. Experts invest above their respective hyperbola, and do not invest below it. Panel (b) shows the separating space when $x_{ha} = 2$, Panel (e)= for $x_{ha} = 0$, $x_{la,i} = 2$ and $x_{la,j} = 1$.