Evaluating turnover intention as a proxy for behavior in the federal public service: Evidence from two surveys of senior civil servants

• Published date: 01 November 2024
• Author: Scott Limbocker,Mark D. Richardson
• Date: 01 November 2024
• DOI: http://doi.org/10.1111/puar.13748

EARLY CAREER INTEL

Evaluating turnover intention as a proxy for behavior in the

federal public service: Evidence from two surveys of senior

civil servants

Scott Limbocker

1

| Mark D. Richardson

2

1

United States Military Academy, West Point,

New York, USA

2

Department of Government, Georgetown

University, Washington, DC, USA

Correspondence

Mark D. Richardson, Department of

Government, Georgetown University, 37th and

O Streets, N.W., Washington, DC, USA.

Email: mark.richardson@georgetown.edu

Abstract

Studies of turnover in the federal public service often study intention rather

than behavior and studies that evaluate the reliability of intention as a proxy for

that behavior are rare. This article uses a novel data set that measures exit inten-

tion and behavior for the same senior civil servants to evaluate two aspects of

the reliability of turnover intention as a proxy for behavior: (1) prediction error

and (2) similarity of marginal effects. Overall, exit intention is a problematic

proxy for behavior because it consistently overestimates the probability of exit.

While coefficients from models of intent tend to have the correct sign, the mag-

nitudes of marginal effects can be misleading, tending to overestimate corre-

sponding marginal effects in models of exit behavior. Moreover, coefficients do

sometimes change signs between models. The article discusses the implications

of these findings and suggests the next steps to better use exit intention to

understand exit behavior.

Evidence for practice

•Exit intention is a problematic proxy for predicting exit behavior. Rates of exit

intention overestimate rates of exit behavior; however, most people who intend

to stay remain in public service two years later.

•While coefficients from models of intent tend to have the correct sign, the mag-

nitudes of marginal effects tend to overestimate corresponding marginal effects

in models of exit. Additionally, coefficients sometimes change signs between

models of exit and intention, which increases prediction error and yields incon-

sistent inferences about marginal effects.

•Results suggest changes in shared conditions (e.g., economic conditions) that

affect civil servants’turnover decisions between the measurement of intention

and observation of behavior can increase prediction error and result in substan-

tive changes in marginal effects. If such conditions at the time of exit are suffi-

ciently similar to prevailing conditions when intention was measured, exit

intention can be cautiously used to determine the directional effect of determi-

nants of exit when testing theories of exit behavior. More work is needed to pro-

vide additional evidence to evaluate how changes in shared conditions affect

the relationship between exit intention and behavior.

•Models of exit intention can be conceived of as models of exit behavior that suf-

fer from misclassification error, defined as intent that does not manifest in

behavior. This conceptualization lights a path forward for using exit intention as

a proxy for behavior when no measure of behavior is available.

Received: 5 August 2022 Revised: 3 August 2023 Accepted: 9 September 2023

DOI: 10.1111/puar.13748

Public Admin Rev. 2024;84:1067–1078. wileyonlinelibrary.com/journal/puar © 2023 American Society for Public Administration. 1067