Trading places: An experimental comparison of reallocation mechanisms for priority queuing

• Author: Sander Onderstal,Anouar El Haji
• Published date: 01 November 2019
• Date: 01 November 2019
• DOI: http://doi.org/10.1111/jems.12314

Received: 12 May 2017

|

Revised: 25 January 2019

|

Accepted: 25 February 2019

DOI: 10.1111/jems.12314

ORIGINAL ARTICLE

Trading places: An experimental comparison of

reallocation mechanisms for priority queuing

Anouar El Haji

|

Sander Onderstal

International Strategy & Marketing,

Faculty of Economics and Business,

University of Amsterdam, Amsterdam,

The Netherlands

Correspondence

Anouar El Haji, Faculty of Economics and

Business, University of Amsterdam,

Plantage Muidergracht 12, Amsterdam

1018 TV, The Netherlands.

Email: a.elhaji@uva.nl

Funding information

University of Amsterdam Research

Priority Area Behavioral Economics

Abstract

In a laboratory experiment, we compare two auction mechanisms that are

designed to improve a queue's efficiency by allowing customers to trade places.

In the server‐initiated auction, the server, when idle, sells the right to be served

next to the highest bidding customer in the queue and distributes the proceeds

among the remaining customers. In the customer‐initiated auction, new arrivals

can sequentially trade places with queued customers. We use two novel

experimental protocols to examine the behavioral properties of both auction

mechanisms. We find that both auction mechanisms improve a queue's

efficiency on average and that both perform equally well in terms of efficiency

gain. We also find evidence of the sunk‐cost effect but not of the endowment

effect. Participants indicated that they found the server‐initiated auction a fairer

mechanism than the customer‐initiated auction. When voting between the two

auctions, the participants tended to favor the server‐initiated auction.

KEYWORDS

endowment effect, queuing, sunk‐cost effect

1

|

INTRODUCTION

It is well known that queues where customers are served on a first‐come first‐served basis are an inefficient way to

ration scarce service time. The reason is that the queuing order does not guarantee that customers with high waiting

costs are served before those with low waiting costs.

1

Policies that allow customers with high waiting costs to get ahead

for a price might improve the experience for customers who can afford this service but harm the other customers in the

queue (Zhou & Soman, 2008). In this paper, we explore the possibility of allowing customers to, literally, trade places to

increase a queue's efficiency. This would allow customers to get ahead in a queue while compensating other customers

for their longer waiting times. Kleinrock (1967) shows that a queue's efficiency may be restored if customers' positions

depend on how much they “bribe”the server.

2

Using a laboratory experiment, we study the efficiency‐enhancing properties of two auction mechanisms that

facilitate customers' trading places: the server‐initiated auction and the customer‐initiated auction. In the server‐

initiated auction, the server, when idle, invites each queuing customer to submit a bid. The server will then serve the

customer who has submitted the highest bid. This customer shares her bid equally among each of the remaining

customers in the queue. In the customer‐initiated auction, a new arrival can sequentially trade places with customers

currently in the queue by offering money to the queuing customers, from the back to the front.

© 2019 The Authors. Journal of Economics & Management Strategy Published by Wiley Periodicals, Inc.

J Econ Manage Strat. 2019;28:670–686.670

|

wileyonlinelibrary.com/journal/jems

-----------------------------------------------------------------------------------------------

This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any

medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.

The two auction mechanisms share the following attractive properties. First of all, an efficient ordering is feasible for

both mechanisms if customers act nonstrategically.

34

Moreover, both mechanisms can be straightforwardly used in a

dynamic setting where customers arrive while the server is busy. In addition, both mechanisms are budget‐balanced

from the viewpoint of the customers, in contrast to Kleinrock's (1967) “bribing mechanism.”As both auction

mechanisms have the potential to decrease total waiting costs, they increase the “pie”compared to a setting where

customers cannot trade places. Because all gains from trade remain in the customers' hands, entry into the queue is not

discouraged, in contrast to a mechanism where customers pay the server to obtain priority.

5

Comparing the two mechanisms allows us to study a potential trade‐off between efficiency and fairness. Using an

independent private waiting costs model, we show that the server‐initiated auction has an efficient (Bayesian–Nash)

equilibrium, in contrast to the customer‐initiated auction. On the other hand, users may perceive the customer‐initiated

auction as a fairer mechanism than the server‐initiated auction because only the former grants them ownership rights

over their initial position. So, we compare a mechanism that is theoretically efficient but that does not protect

customers' positions to a mechanism that protects customers' positions but that is inefficient in theory. Of course,

alternative auction mechanisms may exist that have an efficient equilibrium or that protect customers' position rights.

In this study, we focus on first‐price mechanisms in the sense that customers, when winning, pay their own bid. Other

auction mechanisms can be studied in future research.

We compare the behavioral properties of the two mechanisms in a laboratory experiment. To do so, we use two novel

experimental protocols. Our first protocol implements induced waiting costs. The efficiency gain resulting from the

auctions can be readily measured because the induced waiting costs are known to the experimenter. The second protocol

involves actual waiting. We used this protocol to determine the order by which participants could leave the laboratory.

Participants vote for either of the two auction mechanisms and a majority rule determines which auction is actually

implemented. In addition, participants were asked to rate the auctions in terms of fairness on a five‐point Likert scale.

Besides studying the efficiency and fairness of the auction mechanisms, we also check whether psychological biases

like endowment and sunk‐cost effects have an impact on bidding behavior. The endowment effect occurs when the

sheer possession of an object increases a person's value for it.

6

Anecdotal evidence suggests that people in a queue feel

entitled to their position,

7

which in turn could result in the endowment effect in the sense that customers submit

relatively high bids when their position is auctioned. Someone falls prey to the sunk‐cost bias if her decision depends on

unrecoverable costs that are irrelevant for the decision at stake.

8

Time spent waiting in a queue is such a sunk cost.

Our main results are the following. First of all, the two auction mechanisms considered do not differ in a statistical

meaningful way with respect to the average efficiency gain. This is surprising in light of our theoretical findings that the

server‐initiated auction has an efficient equilibrium while the customer‐initiated auction does not. In a deeper

examination of our data, we do observe differences between the auctions in terms of efficiency gains: Efficiency gains

are significantly greater [lower] in the server‐initiated auction than in the customer‐initiated auction if the initial

queuing order is relatively inefficient (efficient). Neither auction comes close to always reaching an efficient outcome.

For the server‐initiated auction, this result is rooted in noisy individual bidding behavior that is partly explained by a

sunk‐cost effect but not by a noticeable endowment effect. In the customer‐initiated auction, the queuing order remains

relatively inefficient because customers bid more aggressively for their current position than arriving bidders do. In

addition, we find evidence of a sunk‐cost effect in the customer‐initiated auction. Finally, when given the choice

between the two auction mechanisms, participants tended to favor the server‐initiated auction. This may be partly

explained by participants evaluating the server‐initiated auction as fairer than the customer‐initiated auction.

Our paper speaks to several literatures. First of all, we add to the behavioral operations literature.

9

Several

papers within this literature examine queuing processes in the lab. Rapoport, Stein, Parco, and Seale (2004), Seale,

Parco, Stein, and Rapoport (2005), and Stein, Rapoport, Seale, Zhang, and Zwick (2007) study participants'

decisionsastowhentoenteraqueue,ifatall,totestwhether participants' arrival times are consistent with Nash

equilibrium predictions. Kremer and Debo (2012) examine queue herding in a setting where participants can enter

a queue to obtain a good of an uncertain quality. Second, we contribute to the behavioral economics literature by

examining the endowment effect and the sunk‐cost effect in a setting involving queues. Third, our paper

contributes to the experimental literature on multiunit auctions.

10

In our setting, the positions in the queue are the

objects up for auction. Our setting differentiates itself from most of this literature in that in our setting, (a) bidders

are preassigned objects in an initial allocation, (b) the objects' values are ordered in that the earlier the position the

higher its value for all bidders, and (c) the auction mechanisms studied are budget‐balanced. The server‐initiated

auction closely resembles a right‐to‐choose auction, that is, an auction in which rights to choose are auctioned

EL HAJI AND ONDERSTAL

|

671