Public Outrage and Criminal Justice: Lessons From The Jessica Lal Case
Public Outrage and Criminal Justice: Lessons From The Jessica Lal Case
Public Outrage and Criminal Justice: Lessons From The Jessica Lal Case
BRENDAN O’FLAHERTY
Department of Economics, Columbia University
Email: bo2@columbia.edu
RAJIV SETHI
Department of Economics, Barnard College, Columbia University
Email: rs328@columbia.edu
Witness tampering and public outrage have combined to affect judicial out-
comes in a series of high-profile criminal cases in India. We study how these
phenomena operate together in a country with extremes of wealth and poverty,
but with functioning judicial and political systems. Bribes and threats are intri-
cately linked in the strategic interaction between offenders and witnesses. Not
only do bribes provide a direct incentive that can suppress testimony, they also
signal a greater likelihood of retaliation and hence serve as implicit threats to
witnesses. The possibility of public outrage turns out to be an effective con-
straint on witness tampering. In many situations, greater media effectiveness
can improve the administration of justice, even when more obvious improve-
ments in judicial effectiveness cannot.
1. Introduction
During the early hours of April 30, 1999, a thirty-four year old model named
Jessica Lal1 was shot and killed at a private party in a South Delhi restau-
rant, allegedly for refusing to serve liquor to a guest after the bar had
closed. The man in question was identified by three eye witnesses as Manu
Sharma, the son of a senior Congress Party politician and former Union
minister. After several days in hiding, Sharma surrendered to authorities in
Chandigarh. In an interview with police that was subsequently broadcast
on national television, Sharma confessed to the murder. This confession was
later retracted, and a plea of non-guilty entered at trial. During the trial the
three critical eye witnesses recanted earlier statements made to the police,
and twenty-nine witnesses of lesser importance did the same. One of the
eye witnesses, Shyan Munshi, changed his testimony so completely that his
revised statement was used as exculpatory evidence by the defence.
145
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
a For a collection of contemporaneous news reports from the earliest stages of the
case to the present day, see http://www.rediff.com/news/jessica.html. The Delhi
High Court decision is available online as Criminal Appeal No. 193 of 2006 at
http://delhihighcourt.nic.in/.2
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
judicial systems, unlike firms, are not subject to the pressures exerted by
competitive markets, voice takes on even more importance in determining
their level of functionality.
This paper explores the manner in which voice interacts with more fa-
miliar incentives in determining the incidence of witness tampering. We
begin with the observation that the offer of a bribe can affect witness in-
centives in two quite distinct ways. There is a direct monetary inducement
to remain silent (or give false testimony), and there is a more subtle effect
on witness beliefs regarding the likelihood of retaliation. If the offer of a
bribe is a signal that the offender is more likely to retaliate, bribes can be
far more effective in reducing testimony than the monetary inducements
alone would predict. We show how such effects can arise in equilibrium,
and how the possibility of public outrage in response to a miscarriage of
justice affects beliefs and behavior.
Our model is characterized by unobserved heterogeneity among both
witnesses and offenders. Witnesses differ in the satisfaction they experience
from testifying, and offenders differ in the costs they would incur to harm
witnesses who testify, and the cost of giving bribes. Some offenders have
powerful revenge motives and gain from attacking witnesses who have tes-
tified against them, even after the fact, and even when those witnesses have
accepted no bribes. This motive is amplified if a witness testifies even after
accepting a bribe.
We identify the existence of equilibria in which offenders with strong
revenge motives are disproportionately more likely to offer bribes. Bribes
therefore serve as a (noisy) signal of the viciousness of the offender: on
average, offenders who offer bribes are more likely to attack witnesses than
those who do not. In addition, the acceptance of a bribe changes offender
payoffs in such a manner as to make retaliation more likely. Both of these
effects result in an increased subjective evaluation on the part of witnesses
of the likelihood that testimony will be met with retaliation. Bribes are not
simply bribes; they are also veiled threats.
Greater judicial effectiveness and greater media effectiveness have very
different equilibrium effects in this model. If the likelihood and severity of
punishment faced by offenders (conditional on truthful witness testimony)
is increased, more offenders offer bribes, and these are perceived by wit-
nesses to be less threatening on average. The rate of testimony conditional
on a bribe offer therefore increases, although overall levels of testimony may
decline since attempted bribery is more common. In contrast, an increased
likelihood of public outrage (for instance due to greater media effective-
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
ness) results in fewer offers of bribes but a greater perception on the part
of witnesses that testimony will be met with retaliation. Bribes become less
frequent but more threatening. The likelihood of testimony conditional on
a bribe offer may decline, but the overall rate of testimony could never-
theless rise since in incidence of attempted bribery declines. Interestingly,
greater media effectiveness increases the rate of testimony whenever judi-
cial effectiveness fails to do so. Finally, more witness tampering occurs when
wealth differences between offenders and witnesses are great, and favor of-
fenders. Lowering inequality can therefore reduce the incidence of witness
tampering.
This paper is a theoretical contribution to a growing body of litera-
ture that finds real effects flowing from the organization of communications
media. For instance Besley and Burgess (2002)4 show how newspaper cir-
culation affects the distribution of food aid and calamity relief in India,
Gentzkow, Glaeser, and Goldin (2006)5 relate the reduction in corruption
in US politics between 1870 and 1920 to changes in the newspaper busi-
ness, and Djankov et al. (2003)6 relate state media ownership to various
measures of poor government performance. Our result that greater press
freedom can be effective in securing convictions even when greater judicial
effectiveness cannot do so may perhaps be viewed as a theoretical corrobora-
tion of the empirical finding of McMillan and Zoido (2004)7 that Fujimori’s
government in Peru was willing to pay vastly more in bribes to influence
a television station than to influence a judge. Note, however, that notional
freedom of the press need not imply substantive freedom, and effective con-
trol of the media by governments can arise endogenously without formal
censorship; see Besley and Prat (2006)4 for theoretical insights into this
process.
Before proceeding to the formal model and analysis, we survey four ad-
ditional cases involving prominent defendants, witness reversals, significant
media attention, and public outrage.
2. Other Cases
2.1. The Priyadarshini Mattoo Case
Priyadarshini Mattoo was a twenty-five year old law student when she was
raped, brutally beaten and strangled to death at her New Delhi residence on
January 23, 1996. Although there was no eye witness to the murder, physical
and circumstantial evidence pointed immediately to Santosh Kumar Singh,
the son of a senior police officer then posted in Pondicherry. Singh had been
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
stalking Mattoo for over a year at the time of the murder, with multiple
instances of harassment having been reported to the police. He was seen
outside her residence by a neighbor immediately prior to the attack, and
blood stained pieces of his motorcycle helmet vizor were found beside the
body. DNA tests confirmed the presence of his semen on her clothes and
her blood on his helmet.
What seemed like an open and shut case, however, ended in an acquit-
tal in 1999. Defence claims that the physical evidence had been tampered
with while in police custody were given enough credence by trial judge to
allow for reasonable doubt. Suspicions were raised by the judge regarding
deliberate police misconduct, including false depositions, traced to the in-
fluence of the father of the accused: by the time of the trial Singh’s father
was among the most senior police officers in Delhi.
The acquittal triggered massive public outrage, and the case was ap-
pealed to the Delhi High Court in 2000. Little action was taken for several
years, until the 2006 acquittal of Manu Sharma for the Jessica Lal murder
led to renewed scrutiny and a sense of urgency in the part of the Court.
The verdict of the lower court was reversed in October 2006, with the High
Court finding that the “circumstantial evidence in the case is absolutely in-
consistent... with the innocence of the respondent.” Justices RS Sodhi and
PK Bhasin observed that the acquittal by the trial judge had “shocked the
judicial conscience” of their Court. Singh was sentenced a few days later to
death by hanging.
b “BMW witness takes U-turn; says it was truck”, Indian Express, October 2, 1999.8
c “KeyWitness Identifies Nanda in BMW Case”, The Hindu, May 18, 2007;“BMW Case:
Another U-Turn by Kulkarni”, The Times of India, July 20, 2007.9
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
2.5. Overview
Two common themes run through these cases: dramatic changes in witness
testimony in response to threats or bribes, and significant media and pub-
lic involvement. In the Lal, Mattoo, and Best Bakery cases, public action
played a major role in the reversal of earlier verdicts. In the Best Bakery
case, such action was mediated through an NGO, Citizens for Justice and
Peace. Media outlets such as Tehelka magazine and NDTV aired confessions
which clearly galvanized public opinion.d We explore these common themes
more formally below, using a model of witness intimidation and public out-
rage which highlights strategic aspects of the interaction between offenders
and witnesses.
d Even some of the main protagonists made an appearance in multiple cases: Delhi High
Court Justices RS Sodhi and PK Bhasin handled the cases of both Lal and Mattoo, and
Vikas Yadav was among the accused in the cases of Lal and Katara.
e The model developed here builds on our earlier work on witness intimidation
(O’Flaherty and Sethi, 2007),11 which was motivated by interactions between offend-
ers and witnesses in the United States. In that work we considered explicit threats but
not bribes, and did not take into account the possibility that public outrage could result
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
3.1. Preliminaries
Consider a crime involving one offender and one witness. The testimony of
the witness is crucial to the case against the offender. If the witness does
not testify, the offender will be acquitted. If the witness testifies truthfully,
the offender suffers expected punishment ρ > 0. The magnitude ρ reflects
both the probability that the offender will be convicted conditional on the
witnesses testimony, and the severity of the punishment that will ensue
upon conviction. Thus ρ could increase either because of changes in the way
prosecutors and judges operate that increase the probability of conviction
contingent on witness testimony, or because of harsher sentences. We will
call changes in ρ changes in judicial effectiveness.
In order to deter the witness from testifying (or more generally to induce
false testimony) the offender may offer the witness a bribe, which the wit-
ness may either reject or accept. Acceptance of the bribe has value α > 0
to the witness. Following this decision, the witness may either testify or
decide against doing so. If the witness testifies, the criminal may retaliate
by harming the witness. In this case the witness suffers damages δ, and the
offender incurs a cost which is assumed to be private information and, for
reasons discussed below, contingent on whether or not a bribe was accepted
prior to the testimony.
An offender who does not attempt to bribe the witness, or whose bribe
is rejected, pays a cost γ to harm a witness who testifies. This cost is drawn
from a continuous distribution F (γ) with support [γmin , ∞), where γmin <
0. Hence there are some offenders who take pleasure in harming witnesses
who have testified against them, even if the witnesses have rejected a bribe
or were never offered one. We refer to γ as the offender type.
The cost of harming a witness who has accepted a bribe is dependent on
the offender type, and denoted c(γ), assumed to be continuous and strictly
increasing with c(γ) ≤ γ, with strict inequality holding for all γ > γmin . In
other words, the cost of harming a witness who testifies after accepting a
bribe is lower than the cost of harming a witness who has not accepted a
bribe. This is motivated by the idea that testimony following the acceptance
of a bribe is a form of betrayal, or a violation of an implicit contract, which
further inflames the revenge motive for harming witnesses who testify.
in the reversal of an acquittal. Akerlof and Yellen (1994)12 have also explored the man-
ner in which a fear of reprisals affects community cooperation with law enforcement and
the equilibrium level of crime, but without considering the strategic effects of bribery or
voice.
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
Since c(0) < 0, there is some subset of offenders who are prepared
to harm witnesses who testify after having accepted a bribe, even if such
offenders would not harm witnesses who either reject the bribe or were not
offered one. Let γ̄ > 0 denote the offender type for which c(γ̄) = 0. Then
the proportion of offenders who are prepared to harm witnesses who testify
after accepting a bribe is simply F (γ̄); we refer to these offenders as violent.
Since γ is private information, a witness cannot be certain that an offer of
a bribe comes from a violent offender.
The cost of bribing a witness is also dependent on the offender type,
and given by a non-negative, continuous, and strictly increasing function
b(γ).Hence bribing a witness is cheapest for those offenders for whom harm-
ing the witness is also cheap.
Witnesses are also heterogeneous, and vary in the payoff they obtain
from testifying against offenders. A witness who testifies receives a (pri-
vately observed) payoff β > 0, drawn from a continuous distribution G(β)
also having support [0, ∞). We refer to β as the witness type. This payoff
may be interpreted as the satisfaction obtained from an act of retribution,
or simply from performing one’s civic duty. Which of these motives drives
witnesses to testify depends on the context; retribution may have been par-
ticularly important in the Best Bakery case. Since offenders cannot observe
β, they cannot know whether or not the offer of a bribe will induce a witness
to refrain from testifying.
We introduce voice into this simple model of witness tampering by al-
lowing for the possibility that public outrage can result in the reversal of an
acquittal that is perceived to be unjust. Specifically, if the witness accepts
a bribe and turns hostile, there is a probability η that the acquittal will
be reversed upon appeal. In this case the offender receives the same pun-
ishment, ρ, that would have followed an immediate conviction. In addition,
the witness receives punishment σ. This makes it more costly for witnesses
to accept bribes, and makes bribes less effective in securing acquittals. We
assume that α > ησ, otherwise bribes would never be accepted even by
witnesses who refuse to testify.f
The parameters α, δ, ρ, η and σ, the functions b(γ) and c(γ), and the
distributions F and G are all assumed to be commonly known.
None of our results are sensitive to the manner in which ties are broken,
f Oneway to interpret the condition α > ησ is that the size of the bribe is chosen by
offenders to be large enough to outweigh the expected penalty from accepting it. A fully
endogenous treatment of α beyond the scope of this paper.
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
since the set of witnesses and offenders who are indifferent in equilibrium
is of zero measure. We adopt the convention that ties are broken in favor of
legality: witnesses indifferent between acceptance and rejection of a bribe
will reject, witnesses indifferent between testifying and not doing so will
testify, offenders indifferent between offering a bribe and not doing so will
not do so, and offenders indifferent between harming a witness and not
doing so will not do so.
We focus below on six key probabilities, all endogenously determined
in equilibrium. For a witness who has not been offered a bribe, let qn de-
note the likelihood of testifying, and pn the likelihood of being harmed
conditional on testifying. For a witness who has been offered a bribe, let
qr denote the likelihood of rejecting it and testifying, and pr the likelihood
of being harmed conditional on having done so. Finally, let qa denote the
likelihood of accepting and testifying, and pa the likelihood of being harmed
conditional on this.
A witness will accept the bribe and refuse to testify if and only if β <
min{pa δ − ησ, pr δ + α − ησ}. Hence the likelihood of testimony conditional
on a bribe being offered is
qa + qr = 1 − G(min{pa δ − ησ, pr δ + α − ησ}). (3)
What if β ≥ min{pa δ − ησ, pr δ + α − ησ}? In this case the witness will
reject the bribe and testify if
α ≤ δ (pa − pr ) , (4)
and accept the bribe and testify if this inequality does not hold. Since (4) is
independent of β, it will never be the case that both AT and RT are selected
with positive probability. Hence either qa = 0 or qr = 0 in equilibrium.
4. Equilibrium
Since the costs of bribery and retaliation both rise with γ, it may be conjec-
tured that there exists an equilibrium with a partitional structure, in which
offenders with high costs choose not to bribe and those with low costs bribe.
The bribe serves also as a threat since it signals a higher likelihood of retal-
iation. Witnesses are therefore less likely to testify when offered the bribe.
We identify conditions under which such an equilibrium does indeed exist.g
We focus on equilibria with the property that there exists some γ̃ > γ̄
such that all offenders with γ < γ̃ offer a bribe in the equilibrium and
rium in which no intimidation occurs and witnesses who are offered a bribe (off the
equilibrium path) believe that the offer is from a nonviolent offender. Such equilibria are
discussed in the appendix, together with the restrictions on out-of-equilibrium beliefs
that are sufficient to rule out their occurrence.
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
pn = 0,
F (0)
pr = , (7)
F (γ̃)
F (γ̄)
pa = , (8)
F (γ̃)
so pa > pr > pn . Those who are offered a bribe face a greater likelihood
of violence conditional on testimony, and this likelihood is highest if they
accept the bribe before testifying. We show below that the following con-
ditions are sufficient for the existence of such an equilibrium:
These two conditions require that the costs of bribing witness are not too
large relative to the other specifications of the model, in the case of offenders
with γ = 0 and γ = γ̄. (Note that since γmin < 0, the second condition is not
implied by the first.) Condition (9) ensures that offenders are not so strongly
retaliatory that they want witnesses to testify against them simply to get
the satisfaction of harming them, while (10) ensures that the threshold γ̃
exceeds γ̄, so all potentially violent offenders offer bribes.
In addition, we assume the following:
We show below that this condition ensures that (4) holds in equilibrium, so
witnesses who testify after being offered a bribe do so after rejection rather
than acceptance. As a consequence,
0 = qa < qr < qn = 1
5. Comparative Statics
5.1. Greater Judicial Effectiveness
Note that the marginal offender type γ̃ must be indifferent between bribing
the witness and not doing so: πb = πn for this type. Since qn = 0, πn = −ρ.
And since γ̃ > γ̄ > 0 and qa = 0, (6) simplifies to
b(γ̃) = (1 − η) ρ. (12)
qr = 1 − G(pr δ + α − ησ),
An increase in F (γ̃) , arising for instance from harsher penalties for con-
victed offenders, raises t if and only if
g F (γ̃) 1
> = , (14)
G δF (0) pr δ
where g is the density function derived from G, and g/G is evaluated at the
initial equilibrium. Thus harsher penalties are effective when the elasticity
g/G is high and the danger of turning down a bribe pr δ is great.
Even though both may lead to greater convictions and overall testimony,
the channels through which they operate are very different. Voice is not
simply another form of deterrence, it leads to qualitatively different out-
comes.
6. Conclusion
In many ways the witness tampering cases that have motivated our analysis
reveal the strength of the Indian judicial system. Civilians were bribed and
threatened, not prosecutors or judges. Poor people were bribed because it
was feared that their testimony would be credible. And the media exposed
it. A lot has to be going right for witness tampering to be publicly acknowl-
edged problem. We have shown, moreover, that effective legal institutions
do not eliminate witness tampering and may even exacerbate it.
A major open question that remains is the following: what determines
the willingness of the media to expose witness tampering and the willingness
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
Acknowledgements
We thank participants at the ISI Conference on Comparative Devlop-
ment, especially Bhaskar Dutta, Parikshit Ghosh, Arunava Sen, and E.
Somanathan, for suggesting that our earlier work on witness intimidation,
suitably modified and extended, could be relevant to the Indian context.
We also thank Kai Friese for helpful comments and suggestions, and Om
Prakash Shokeen for assistance with sources. This material is based upon
work supported by the Behavioral Sciences Program at the Santa Fe Insti-
tute.
Appendix
Proof of Proposition 1.
Consider a candidate equilibrium such that, for some γ̃ > γ̄ all offenders
with γ < γ̃ offer a bribe and those with γ ≥ γ̃ do not. At any such equilib-
rium, pn = 0 and qn = 1. Assume for the moment that (4) holds, so qa = 0.
(We show below that this is consistent with equilibrium). From (7) and (8),
and the fact that F (γ̃) < 1, we have
pr > F (0),
and
pa > pr .
Hence from (3) and the assumption that (4) holds and qa = 0,
qr = 1 − G(pr δ + α − ησ) < 1 − G(F (0)δ) (15)
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
offender type offers a bribe, and a witness who received a bribe (off the
equilibrium path) believes that it comes from an offender with γ > γ̄. In
this case the witness will simply accept the bribe and testify, believing that
pa = 0. Knowing this, no offender will offer a bribe. While this can be an
equilibrium, the beliefs that sustain it are entirely implausible: since b(γ)
is increasing, it would be far more reasonable to suppose that the offender
making the offer had a very low value of γ. We make the conservative as-
sumption that a witness receiving an offer of a bribe off the equilibrium
path believes that the offender type is drawn from the distribution F (γ),
which corresponds exactly to the prior belief of the witness. In other words,
the witness believes that the offer was made in error, and all types are
equally likely to make such an error.
This is sufficient to rule out the equilibrium in which no bribes are
offered. To see why, note that in any such equilibrium pn = F (0) and hence
qn = 1 − G(δF (0)) from (1). Hence the offender payoff, conditional on no
offer of a bribe, is given by
πn = − (1 − G(δF (0))) (ρ + min{0, γ}) .
If a bribe is offered, witness beliefs are pa = F (γ̄) and pr = F (0). Hence
from (3),
qa + qr = 1 − G(min{δF (γ̄), δF (0) + α}).
Now consider an offender with γ = γmin . If δF (0) + α < δF (γ̄), then qa = 0
and this offender’s payoffs satisfy
πb − πn = (G(δF (0) + α) − G(δF (0))) (ρ + γmin ) > 0,
so this offender type prefers to offer a bribe. Similarly, if δF (γ̄) ≤ δF (0)+α,
then qr = 0 and this offender’s payoffs satisfy
πb − πn = (G(δF (γ̄)) − G(δF (0))) (ρ + γmin ) > 0.
Again this offender prefers to offer a bribe, contradicting the hypothesis
that no bribes are offered in equilibrium.
References
1. The Jessica Lal murder case rediff.com(December, 2007).
2. State v. Sidhartha Vashisht Delhi High Court. Criminal Appeal No. 193 of
2006(December, 2006).
3. A. Hirschman, Exit, Voice, and Loyalty: Responses to Decline in Firms, Or-
ganizations, and States (Cambridge: Harvard University Press, 1970).
4. T. Besley and A. Prat, American Economic Review 96, 720 (2006).
March 16, 2009 12:22 WSPC - Proceedings Trim Size: 9in x 6in JessicaLal
5. E. G. Gentzkow M. and C. Goldin, The Rise of the Fourth Estate: How News-
papers Became Informative and Why it Mattered, in Corruption and Reform:
Lessons from America’s Economic History, eds. E. Glaeser and C. Goldin
(Chicago: University of Chicago, 2006), pp. 187–230.
6. T. N. Djankov S., McLeish C. and A. Shleifer, Journal of Law and Economics
46, 341 (2003).
7. J. McMillan and P. Zoido, Journal of Economic Perspectives 18, 69 (2004).
8. BMW case: Another u-turn by Kulkarni The Times of India(July, 2007).
9. Key witness identifies Nanda in BMW case The Hindu(May, 2007).
10. Best Bakery case: Dateline(December 2007).
11. B. O’Flaherty and R. Sethi, Witness Intimidation, Tech. Rep. 0708-07, De-
partment of Economics, Columbia University (2007).
12. G. Akerlof and J. Yellen, Gang behavior, law enforcement, and community
values, in Values and Public Policy, eds. T. M. Henry Aaron and T. Taylor
(Washington, D.C.: Brookings, 1994), pp. 173–209.