Mandating vaccination with unknown indirect effects

• DOI: http://doi.org/10.1111/jpet.12234
• Date: 01 June 2017
• Author: Charles F. Manski
• Published date: 01 June 2017

Received: 14 March 2016 Accepted: 10 August 2016

DOI: 10.1111/jpet.12234

ARTICLE

Mandating vaccination with unknown indirect

effects

Charles F.Manski

NorthwesternUniversity

CharlesF. Manski , Department of Eco-

nomicsand Institute for Policy Research,

NorthwesternUniversity, Evanston, IL

(cfmanski@northwestern.edu).

Social interactions make communicable disease a core concern of

public health policy.A prevalent problem is scarcity of empirical evi-

dence informative about how interventions affect illness. Random-

ized trials, which have been important to evaluation of treatments

for noninfectious diseases, are less informative about treatment of

communicable diseases because they do not fully reveal the indirect

preventive(herd immunity) effect of vaccination on persons who are

not vaccinated or are unsuccessfully vaccinated. This paper studies

the decision problem faced by a health planner who observes the ill-

ness rate that occurs when persons make decentralized vaccination

choices and who contemplates whether to mandate vaccination. The

planner’s objective is to minimize the social cost of illness and vac-

cination. Uncertainty about the magnitude of the indirect effect of

vaccination implies uncertainty about the illness rate that a mandate

would yield. I first study a simple representative-agent setting and

derive conditions under which the planner can determine whether

mandating vaccination is optimal. When optimal policy is indeter-

minate, I juxtapose several criteria for decision making—expected

utility,minimax, and minimax-regret—and compare the policies they

generate. I then extend the analysis to a more general setting in

whichmembers of the population may have heterogenous attributes.

I have benefitted from the opportunity to present this work in sem-

inars at the Booth School of Business, University of Chicago, the

Department of Economics, University of California at Santa Barbara,

and the Schaeffer Center for Health Policy and Economics, Univer-

sity of Southern California. I havealso benefitted from the comments

of an anonymous reviewer and associate editor.

1INTRODUCTION

Social interactions in treatment response make communicable disease a core concern of public health policy.Spread

of infection creates a negative externality. Preventive administration of vaccines, therapeutic administration of

Journal of Public Economic Theory.2017;19:603–619. wileyonlinelibrary.com/journal/jpet c

2017 Wiley Periodicals,Inc. 603

604 MANSKI

antimicrobial drugs, and separation of infected persons from the general population mayreduce disease transmission.

In a decentralized health care system, infected and at-risk persons may not adequately recognize the social implica-

tions of their actions. Hence, there may be a rationalefor government to seek to influence treatment of communicable

disease. In practice, policies range from quarantines of infected persons to mandatory vaccination to subsidization of

vaccines and drugs.

A prevalent problem in policy choice is scarcity of empirical evidence that are informative about how interventions

affect illness. Randomized clinical trials (RCTs),which have been important to evaluation of treatments for noninfec-

tious diseases, are less informative about treatment of communicable diseases. The classical argument for inference

from RCTs assumes that the outcome experienced by each person or other treatment unit may vary only with his

own treatment, not with those of other members of the population. This assumption—variously called noninterference

(Cox, 1958), the stable unit treatment value assumption (Rubin, 1978), or individualistic treatment response (Manski,

2013)—does not hold when treating communicable diseases. RCTshave limited power to identify the effects of treat-

ments for such diseases.

A leading case is the decision problem of a health planner who must choose a vaccination policy for a population.

Vaccination of a particular person may benefit that person directly by generating animmune response that reduces

his susceptibility to the disease. It may also reduce the infectiousness of this person and thereby inhibit transmission

of the disease to others who are unvaccinated or unsuccessfully vaccinated. Thus, vaccination may have both a direct

preventive effect on the person vaccinated and an indirect preventive effect on other persons. The indirect effect is

often called herd or community immunity.See Fine (1993) and Fine, Eames, and Heymann (2011) for discussions of the

long history and use of this concept in epidemiology.1

An RCT that randomly vaccinates a specified fraction of the population may enable evaluation of the direct effect

of vaccination on illness, but it does not reveal the indirect effect. The trial only revealsthe illness outcomes that occur

with the vaccination rate used in the trial. The outcomes that the population would experiencewith other vaccination

rates remain counterfactual. Yetpolicy choice requires comparison of alternative vaccination rates.2

Attempting to cope with the dearth of empirical evidence, researchers studying vaccination policy have creatively

used epidemiological models of disease transmission to forecast the outcomes that would occur with counterfactual

policies. See, for example, Brito, Sheshinski, and Intriligator (1991), Becker and Starczak (1997),Ball and Lyne (2002),

Scuffham and West (2002), Hill and Longini (2003), Patel, Longini, and Halloran (2005), Boulier, Datta, and Goldfarb

(2007), Althouse, Bergstrom, and Bergstrom (2010), and Keeling and Shattock (2012). However,authors typically pro-

vide little information that would enable one to assess the accuracy of their assumptions about individual behavior,

social interactions,and disease transmission. Hence, I think it prudent to view their forecasts more as informative com-

putational experiments than as accuratepredictions of policy impacts.

With this background, I consider the decision problem of a health planner who observes the illness outcomes that

occur when persons make decentralized vaccination choices and who contemplates whether to mandate vaccination.

In the U.S. federal system, the health planner who decides whether to mandate vaccination typically is a state public

health agency.States may mandate that specific populations be vaccinated against specific diseases.3The usual alter-

native to a mandate is decentralization, with vaccination decisions being made by families, schools, care facilities, and

employers. While it is natural for economists to think of subsidies as an alternativeto mandates, c onsiderationof vac-

cine subsidies as a practical policy has been rare.

1While the usual epidemiological presumption is that vaccines havebeneficial indirect effects, they can in principle have negative effects. Vaccinesbasedon

attenuated live pathogens mayinfect persons other than the recipient. Vaccination may also encourage selection of pathogens towards variants resistant to

thevaccine. These negatives forces are thought to be weak in practice. See Mishra, Oviedo-Orta, Prachi, Rappuoli, and Bagnoli (2012).

2Vaccinetrials can reveal indirect effects in special cases where the population partitions into many isolated groups (also known as clusters) of persons. Then

the members of each group may infect one another but not the members of other groups. In such cases, one can define treatment units to begroups rather

than persons, randomly assign varying vaccination rates to different groups, and use the trial to learn about illness outcomes under alternative vaccination

rates.Hudges Hudgens and Halloran (2008) develop methodology for analysis of RCTs performed in such settings. Loeb et al. (2010) report a trial performed

onisolated Hutterite communities in Canada. However, populations rarely partition in modern societies. RCTshave no identifying power in the polar case of a

fullyconnected society, where social interactions are global rather than local (Manski, 2013).

3Thewebpage www.immunize.org/laws/ describes the mandates in each state.