Anonymous or personal? Repeated personalized advice in the lab

How to cite

Gramb, Marius; Schottmüller, Christoph. 2021. "Anonymous or personal? Repeated personalized advice in the lab." AEA RCT Registry. AEARCTR-0008682. https://doi.org/10.1257/rct.8682-1.2000000000000002

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Marius Gramb* , Christoph Schottmüller

* Primary investigator

Registry ID
AEARCTR-0008682
DOI
https://doi.org/10.1257/rct.8682-1.2000000000000002
First registered
Last updated
Status
In development
Trial dates
2021-12-14 – 2022-02-28
Countries
Private
JEL codes
D82, D83, D86
Registry record
http://www.socialscienceregistry.org/trials/8682

Abstract

In many situations, consumers seek advice from more knowledgeable advisers/experts. However, it is very often the case that the adviser gets a certain bonus depending on the product the consumer ends up buying. In these situations, it is a priori unclear why the adviser should give good advice rather than collecting the highest possible bonus every time. We analyze a model of repeated advice in which the possibility of future bonus payments can motivate the adviser to give good advice even if this does not maximize his present payoff as good advice increases the chance that the consumer seeks advice from the adviser later on. We are particularly interested in the effect of learning in the sense that repeated interactions with a consumer allow the adviser to learn the consumer's preferences and therefore give better advice in the future. In a particular class of equilibria, both adviser and consumer benefit from this learning possibility.
With this laboratory experiment, we aim at testing our theoretical predictions with actual decisions by subjects who are assigned the role of the adviser/expert or consumer, respectively. In particular, we are interested in (i) the distribution of payoffs between adviser and consumer (ii) the length of advice relationships and (iii) how (i) and (ii) depend on the adviser's learning.

Intervention

The treatment is given by the opportunity for the signal quality to increase based on the amount of good advice produced.

Experimental design

In the experiment, there will always be two options out of which the adviser can recommend one to the consumer. The adviser gets a bonus for exactly one out of two options and has noisy private information about which option yields a higher payoff for the consumer. Subsequently, the adviser chooses an option for the consumer. After observing his resulting payoff (0 points in the bad case, 1 point in the good case), the consumer can decide whether or not to dissolve the relationship. This procedure is repeated until the consumer dissolves the relationship or it is ended exogenously (after a randomly predetermined number of rounds). Whenever the consumer dissolves the relationship, he gets an outside option of 5 points added to his collected points from the current adviser-consumer relationship.
When all relationships are ended (either exogenously or by the consumers), all consumers are randomly assigned a new adviser and they play another round of the same game. All in all, each consumer will be matched to seven advisers in total (which do not have to, but are likely to be distinct). In the sessions conducted after January 13, 2022, each player plays ten rounds of the above game, so that each consumer is matched to ten advisers in total.

The difference between the control group and the treatment group is in the information the adviser gets about the good option for the consumer (meaning the option that pays 1 point to the consumer). In both groups, the adviser will always get a signal indicating the good option for the consumer. In the control group, this signal is always correct with a probability of 82% and incorrect with a probability of 18%. In the treatment group, this signal quality also starts off being 82%, but it increases by 2% whenever the adviser generates good advice (in the sense that the consumer received one point from the recommended option). The maximal signal quality is, however, 90%. Hence, after improving four times, the signal quality stays at 90% until the end of the specific adviser-consumer relationship.

After these games we elicit gender, risk attitude and trust attitude. The elicitation of risk attitudes is incentivized: Subjects make six binary choices where each choice is between a risky lottery and a safe payment. Trust attitude is measured by the extent to which subjects agree to four statements about trust (statements are as in Naef and Schupp, 2009, SOEP paper 167).

All individuals are paid a show-up fee of 4 €. Depending on how they do in the experiment, they can earn more money: one out of the seven (or ten in the sessions after January 13, 2022) subgames is randomly picked and paid out. Furthermore, one of the six choices in the risk elicitation part is randomly picked and paid out. The overall payments per person are expected to vary between 4 and 30 €.

The experiment is conducted at the CLER (Cologne Laboratory for Economic Research) using oTree (Chen et al., 2016).

Design details

Randomization method
Randomization done by a computer
Randomization unit
Experimental lab sessions. Within each session, participants are assigned one of two possible roles. These roles are randomly assigned by the software used.
Clusters
8 sessions
Observations
272 subjects (8 sessions with 34 subjects per session). Due to uncertainties in recruitment linked to the pandemic, some sessions might have less than 34 subjects (and consequently a lower total number of subjects as well).
Arms
3 control sessions, 5 treatment sessions
Primary outcomes
- the experts' advice quality - the payoff distribution between experts and consumers - the lengths of the advice relationships
Secondary outcomes
- Risk attitude - Trust attitude

Cited by

No citation data available for this registration yet.

Keywords