Screening Talent for Task Assignment: Absolute or Percentile Thresholds?

• Date: 01 September 2020
• Author: HAIJIN LIN,RAMJI BALAKRISHNAN,KONDURU SIVARAMAKRISHNAN
• DOI: http://doi.org/10.1111/1475-679X.12327
• Published date: 01 September 2020

DOI: 10.1111/1475-679X.12327

Journal of Accounting Research

Vol. 58 No. 4 September 2020

Printed in U.S.A.

Screening Talent for Task

Assignment: Absolute or Percentile

Thresholds?

RAMJI BALAKRISHNAN ,∗HAIJIN LIN ,†

AND KONDURU SIVARAMAKRISHNAN ‡

Received 19 October 2018; accepted 9 June 2020

ABSTRACT

Matching talents to tasks is an important part of job design. Organizations

routinely use performance thresholds to group agents by talent. We see

thresholds defined both in terms of an individual’s own performance (ab-

solute value) and in terms of peer performance (percentile). Intuition sug-

gests a preference for percentile thresholds because the resulting rank-order

statistic is sufficient to assess relative talent. Yet, in the context of a task as-

signment problem in which the objective is to match talent with task type

(using two agents and two task types), we show that absolute thresholds can

dominate percentile thresholds under either of two conditions. First, flex-

ibility in task assignment tilts the balance toward absolute thresholds. Sec-

ond, performance manipulation can adversely affect the inherent advan-

tage of percentile thresholds because they motivate agents to invest relatively

more in personally costly influence activities to cast their performance in a

∗Tippie College of Business, University of Iowa; †C. T.Bauer College of Business, University

of Houston; ‡Jones Graduate School of Business, Rice University

Accepted by Haresh Sapra. We thank Rabah Amir, George Drymiotes, Nisan Langberg,

Mark Penno, Suresh Radhakrishnan, Jack Stetcher, Devika Subramanian, Nicholas Yannelis,

and workshop participants at the University of Iowa and at the 2015 MAS conference for

helpful comments and discussions. We are particularly grateful for insightful comments

from an anonymous referee. An online appendix to this paper can be downloaded at

http://research.chicagobooth.edu/arc/journal-of-accounting-research/online-supplements.

831

© University of Chicago on behalf of the Accounting Research Center, 2020

832 r. balakrishnan, h. lin, and k. sivaramakrishnan

favorable light. We examine how these results hold up when there are count-

ably large number of agents and discuss empirical implications.

JEL codes: D82, M40, M41, M51, M53, M55

Keywords: incentives; measurement system; relative performance evalua-

tion

1. Introduction

Successful organizations excel in identifying talented individuals in their

workforce and putting their skills to best possible use (Allgood and Farrell

[2003], Jovanovic [1979a, 1979b]). Some judge talent by focusing solely on

an individual’s ability to meet or exceed prespecified absolute standards

while others evaluate an individual’s performance relative to a peer group.

Libby and Lindsay [2010] report that over 80% of surveyed firms use abso-

lute targets (e.g., in budget-based evaluations). Grote [2005] reports that

many Fortune 500 firms use relative performance to group employees into

three to five categories, each with differing career prospects. Likewise, in

an academic context, we observe some instructors setting grade thresholds

in terms of a student’s own score (all scores over 90% earn an A) while oth-

ers prefer percentile cutoffs that reflect relative performance (the top 10%

of students earn an A).

Why do some organizations prefer absolute value-based cutoffs while oth-

ers rely on percentile cutoffs in screening talent? What are the relative mer-

its and weaknesses of these choices? How does the importance of matching

talent with task attributes affect this choice? Relatedly, how does the choice

of screening mechanism affect employee incentives? Companies such as

GE and Microsoft abandoned the use of tournament-type relative ranking

mechanisms because these mechanisms allegedly fueled some employees

to engage in “influence activities” and even sabotage (Harbring and Irlen-

busch [2011]). Burgess [2005] provides several examples of sabotage when

firms use relative evaluation for screening and promotion. This deleteri-

ous aspect of relative performance evaluation has received little attention

although there is considerable research on the appeal of relative perfor-

mance evaluation as a way of filtering out common uncontrollable factors

in group settings (Holmstrom [1982]).1Moreover,we are not aware of work

that directly addresses the relative efficacies of absolute and relative perfor-

mance thresholds in screening talent. In this paper, we analytically examine

these issues in a setting in which screening talent is necessary for task assign-

ment.

An absolute performance threshold system (APTS) and a percentile-

based threshold system (PBTS) differ fundamentally in how they help an

1Holmstrom [1982] is among the first to provide a rationale for relative performance eval-

uation. Antle and Smith [1986], Janakiraman et al. [1992], and Gibbons and Murphy [1990]

provide empirical evidence of relative performance evaluation.

absolute versus percentile thresholds 833

employer (the principal) assign employees (agents) to talent pools when

outcome measures are noisy indicators of talent. With APTS, the princi-

pal can influence the probability that any given agent will be assigned to

a particular talent pool by adjusting the performance threshold. Although

increasing the threshold score to earn an “A” reduces the probability that

a student will receive this grade, it increases the confidence that anyone

who crosses the threshold is indeed talented. However, the principal can-

not fix the number of agents crossing a set threshold because outcome is

stochastic. On the other hand, under PBTS, the principal can fix the pro-

portion of agents in each talent pool by adjusting the percentile thresholds

but is inhibited in her ability to assess the absolute talent level of any given

agent. We show these differences subtly determine the principal’s ability to

achieve talent-task fit and, consequently, the relative efficacies of the two

approaches in specifying performance thresholds.

Prima facie, it would seem that the principal should prefer PBTS over

APTS. After all, PBTS assigns a unique rank to each agent, thus sorting them

by relative talent, whereas APTS lumps the agents into discrete groups. Yet,

we uncover two aspects of the task assignment problem that can lead to

APTS dominating PBTS. The first relates to the principal’s flexibility with

respect to task assignment. If filling all tasks is paramount (full staffing),

the principal has little flexibility— she has no choice but to tolerate some

mismatch of talent and task to staff all tasks. With flexible staffing, the prin-

cipal is not constrained to fill all tasks. Although such flexibility is valuable

under APTS, it has no benefit under PBTS because it constrains the task

assignment strategy by fixing the proportion of agents in each talent pool.

The second aspect stems from the recognition that employees also care

about the tasks to which they are assigned and therefore have an incentive

to engage in performance manipulation (Fudenberg and Tirole [1995]).

This incentive matters in our context because of the competition induced

by relative assessment. We show that greater levels of manipulation by agents

under PBTS can negate its inherent advantage over APTS even under

full staffing.

We consider a model with two agents and two types of tasks. Task type

1 (2) is more suited for the agent with high (low) talent. A mismatch of

talent with task type is therefore costly to the principal. Only the agent

knows his talent type—high or low. The principal uses an aptitude test to

update beliefs about talent type. This test produces a performance outcome

that serves as a noisy signal about the agent’s talent. As there are only two

task types, the principal must use thresholds in assigning tasks. Ex ante, she

chooses whether to set performance thresholds using absolute scores or

percentiles (i.e., chooses between APTS and PBTS).

Consider task assignment strategies under the two systems. PBTS pro-

duces a rank ordering of the two agents. Thus, the principal’s task-

assignment strategy is immediate—assign the agent ranked first (second)