Assessment and Prediction in Homelessness Services and Elsewhere

• DOI: http://doi.org/10.1111/1467-8462.12221
• Date: 01 June 2017
• Author: Brendan O'Flaherty
• Published date: 01 June 2017

Policy Forum: Australian Homelessness—Research and Policy

Insights

Assessment and Prediction in Homelessness Services and

Elsewhere

Brendan O’Flaherty*

Abstract

This article cautions against the na€

ıve use of

predictive techniques by homelessness service

providers. Prediction is hard and may be

impossible. Prediction does not automatically

lead to socially optimal assignments. On the

other hand, mechanism design, which does not

rely on top-down predictions, can be used to

good effect. Moreover, fairness counts as well

as optimality—and may be easier to achieve. I

use examples from f‌ire service, central bank-

ing, police patrol, tax auditing, art, f‌inance,

soup kitchens and cadaveric organ allocation.

1. Introduction

Assessment is popular these days. Cheap

information processing and new algorithms

make it easier than ever to predict how things

will act. So, we plan our life around up-to-the-

minute weather forecasts, choose our mate

through online dating algorithms, make our

investment decisions around what the polls or

the betting lines say about the outcome of the

Brexit referendum and let our global position-

ing systems tell us whether to go left or right at

the next intersection. This is all f‌ine and good:

the less energy I have to spend making

decisions about my life, the better.

But, the way we live our life has spilled over

into what governments and non-prof‌it organisa-

tions do, almost automatically. Thus, in the United

States, for instance, whether your tax returns are

audited depends on what your occupation is, how

long you spend in prison depends not so much on

the crime you were convicted of but on whether

you are predicted to get caught for another crime

after you leave, whether you are searched and

arrested for marijuana possession in New York

City depends on what your sex is and, in many

jurisdictions, whether you get pulled over for

modest speeding depends on what your race

appears to be. Also, commentators seem vora-

cious for more of this: they are distraught that the

authorities cannot predict who the next mass

murderer will be. This hunger for prediction is part

of the reason why assessment gets so much

attention in homelessness services.

My primarymessage is to caution against this

trend in homelessness services, especially the

na

€

ıve versions of it. Prediction is sometimes a

* Department of Economics, Columbia University, New

York 10027 United States; email <bo@columbia.edu>.I

thank Guy Johnson for suggesting this question and for

many helpful comments. I have benef‌ited from comments

at the Forum on Homelessness Policy and Practice in the

United States on 20 July 2016 and I am grateful to the

following sponsors: Centre for Applied Social Research at

RMIT University, The University of Melbourne, the

Victorian Council of Social Service and the Council to

Homeless Persons. The hospitality of the Melbourne

Institute of Applied Economic and Social Research, The

University of Melbourne, is gratefully acknowledged. I

have benef‌ited from discussions with David Ribar,

Rosanna Scutella, Yi-Ping Tseng and participants in the

National Alliance to End Homelessness Research Council.

The errors are my own.

The Australian Economic Review, vol. 50, no. 2, pp. 229–35

°

C2017 The University of Melbourne, Melbourne Institute of Applied Economic and Social Research

Published by John Wiley & Sons Australia, Ltd