DISC Research Seminar Series

TBD

TBD

Problem definition: This study explores the impact of initial and in-process waiting times in asynchronous telemedicine on subsequent online and offline visits, and the moderating effects of the consultation fee. Methodology/results: We focus on three measures of waiting time in asynchronous telemedicine: initial waiting time (the duration from a patient’s initial request to admission into an online consultation session), average in-process waiting time (the average time between a patient’s question and the doctor’s response during the session), and the variability of in-process waiting times, which together capture both the initial and ongoing responsiveness to patients during asynchronous care delivery. We use 42,111 patients’ online and offline consultation records from a primarily text-based, asynchronous telemedicine platform affiliated with a top-ranked hospital system (February 2021-April 2024). Our results show that patients with a longer (above median) average in-process waiting time (≥0.75 hours) have 14.53%, 16.47%, and 13.41% lower odds for subsequent all visits, online visits, and outpatient visits in the next 30 days, respectively. Patients with a higher (above median) variability of in-process waiting times (≥0.40 hours) have 9.70% and 12.72% lower odds of subsequent all visits and offline outpatient visits in the next 30 days, respectively. Surprisingly, the initial waiting time shows no significant effect. Results remain consistent when considering whether the subsequent visits are with the same doctor. Finally, the consultation fee negatively moderates the relationship between in-process waiting time and subsequent visits. Managerial implications: Average in-process waiting time in asynchronous telemedicine has the most significant impact on both subsequent online and offline patient visits among the three waiting time metrics. The findings highlight reducing in-process waiting in asynchronous telemedicine as a viable means of enhancing patient engagement and ensuring continuity of care across channels.

Firms increasingly use Artificial Intelligence (AI) enabled forecasting engines ("prediction machines") to augment their managers' own forecasting capabilities and thus improve sales-and-operations planning outcomes. Deployment of a prediction machine may cause an unintended reduction in a manager's own forecasting effort which in turn diminishes the value of machine adoption. We model a firm facing uncertain demand that delegates a procurement quantity decision to a human manager who can exert effort to generate a demand prediction. The firm deploys a machine that provides the manager with a demand-prediction signal. We establish the conditions under which managerial effort reduction occurs and thus reduces the machine's potential value. Adopting a Bayesian persuasion approach, we show that partially disclosing the machine's prediction, either downplaying high predictions or exaggerating low predictions, can be optimal, depending on the product's cost-to-revenue ratio. A strategy of minimal obfuscation (to achieve effort) is optimal if the machine is more accurate than the human however, maximal obfuscation (while maintaining effort) can be optimal if the human is more accurate. Our results imply that the firm may be better off tuning a machine to be less informative than its maximum capability.

Risk assessment algorithms are increasingly being implemented to inform judicial decisions, yet their welfare and fairness consequences depend not only on algorithm design but also on how judges exercise discretion when faced with algorithmic recommendations. Using a decade-long administrative dataset of 32,879 felony drug cases in Virginia from 2013 to 2024, we examine how judges discretionarily respond to algorithmic advice from a sentencing risk assessment instrument that issues a Recommendation for Alternative Punishment (RAP). Using a regression discontinuity design around the RAP cutoff, we show that receiving a RAP increases the likelihood of alternative punishment by 16.7% and reduces average logged sentence length by 31%. More importantly, these effects are distributionally uneven. Judges’ responses attenuate the gender disparity (driven by sizable sentence reductions for men) while widening the racial disparity (larger reductions for White than Black offenders) in the recommended cases. To interpret these patterns, we propose a dual-process mechanism, where the algorithmic recommendation can either trigger deliberation when it conflicts with judges’ prior beliefs or reduce deliberation when it aligns with those beliefs. To further test the mechanism, we leverage the Black Lives Matter movement as an exogenous increase in external scrutiny, showing that heightened attention to racial bias induces greater deliberation, which, in turn, dampens the differential leniency toward White offenders observed earlier. Lastly, we link upstream sentencing to downstream recidivism to assess when discretion is difficult to rationalize on risk grounds, highlighting fairness concerns and potential efficiency losses from misallocated incarceration.

Referral programs are widely used by firms as a tool for new customer acquisition. In practice, most referral programs reward existing customers for the acquisition of their referrals (i.e., referred customers). Although such acquisition-based referral rewards incentivize existing customers to refer new customers, they can be ineffective in generating high-value referrals. To increase the value created by referral programs, we propose that firms complement their acquisition-based referral reward with an additional reward to existing customers contingent on the short-term performance of their referrals. We tested our proposal using a randomized field experiment conducted at a Chinese firm offering financial deposit services. During a 30-day experimental campaign, existing customers in both the control and treatment conditions were offered the same acquisition-based referral reward, while only those in the treatment condition could additionally receive a performance-based referral reward for each referral whose total investment made during the campaign in selected financial deposits met a predefined threshold. Assessing the value of referred customers acquired during the campaign based on their investment behavior over a 480-day period, we find that the introduction of the performance-based referral reward increases the total value of referrals by more than 110%, and this effect is driven primarily by the acquisition of higher-value referred customers rather than more referred customers. We propose two mechanisms for the acquisition of higher-value referred customers, including the performance-based referral reward (1) motivating existing customers to screen their friends and refer good matches to the firm and (2) providing referred customers with an additional incentive to invest. Our data provide suggestive evidence for both mechanisms.

Recommender systems and advertising are independently considered among the most successful digital persuasion techniques frequently employed by platforms and marketers, owing to their significant influence on consumer behavior. Yet, our scientific understanding of the interplay of these business tactics remains rather limited and underexplored. This study investigates the relationship between advertising and the effectiveness of recommendations addressing an important gap in the literature. Specifically, we examine whether advertising expenditure relates to the effectiveness of system-generated recommendations towards increasing consumer demand. Interestingly, our results reveal that even though both recommendations and advertising independently have a positive impact on demand, combining these different digital persuasion techniques might lead to advertising reducing the effectiveness of recommendations. However, advertising on different channels, earned media, and marketing promotions that increase the transaction utility of consumers do not cannibalize the positive effect of recommender systems on product demand. This discovered interplay of the advertising and recommender systems demand effects likely stems from consumers’ elevated levels of persuasion knowledge when exposed to multiple same-channel consumer persuasion techniques. The findings are supported by various identification strategies and robustness checks. We further validate the findings and underlying mechanisms in a randomized lab experiment with an online recommendation platform. Our study enhances the current understanding of the demand effects of recommendations by incorporating marketing communications, and extends the persuasion knowledge theory to recommender systems, while having important implications for platforms and marketers and suggesting new directions for future research.

The rise of large language models (LLMs) raises important questions about their impact on political discourse. This study investigates how the public release of ChatGPT influenced ideological and emotional dynamics in online political discussions. Using a comprehensive dataset from Reddit’s largest political discussion forum, we trace user-level ideological shifts by measuring the political slant of millions of comments. We find that ChatGPT's release significantly intensified political polarization: liberal-leaning users became more liberal, and conservative-leaning users became more conservative. This effect was primarily driven by the emergence of previously inactive users who began posting more ideologically extreme content with ChatGPT’s assistance, which in turn prompts greater engagement from active users. Interestingly, while ChatGPT-generated comments aligned with a user's own ideology reinforced polarization, cross-partisan content generated by ChatGPT had a moderating effect — though insufficient in scale to counteract the overall trend. Despite the rise in ideological polarization, we observe a decline in affective polarization, characterized by a reduction in hostility and toxicity in online political discourse. These findings challenge the conventional view that political extremity often coexists with incivility and suggest that while LLMs intensify ideological divides, they hold the potential to foster more civil political engagement.

Retailers are shifting from brick-and-mortar to omnichannel systems. Ship-from-store (SFS) is a prominent yet risky integration because its value depends on product attributes and local shopping frictions: the same program can expand demand for some SKUs while cannibalizing store sales for others. Using transaction and inventory-order data from a large pharmacy chain, we document SKU-level patterns and analytically model how store-only customers and omni-customers choose between visiting the store and ordering online, incorporating shortened online waiting time under SFS and different cross-selling profits by channel. The model’s equilibrium links product characteristics and local geography to demand pooling, in-store fill rates, and profit, yielding testable predictions that separate market expansion from cannibalization and inform SKU selection and inventory targets. Testing these predictions across pre- and post-SFS periods, we find: (i) stores with higher store visit costs realize larger post-SFS profit gains (ii) SKUs with lower baseline in-store availability see larger improvements and (iii) among cannibalization-prone SKUs, order profitability still rises when post-SFS inventory targets are sufficiently high. Together, the model and evidence explain when SFS raises profit and why effects vary across products and locations.