Journal of Economic Behavior & Organization

Vol. 48 (2002) 57–69

Marital investments, time consistency and emotions

Martin Dufwenberg∗
Department of Economics, Stockholm University, SE-10691 Stockholm, Sweden
Received 26 October 1999; accepted 10 November 2000

Abstract
A benchmark model of a married couple’s educational investment yields an inefficient outcome
due to the possibility of opportunistic divorce. Motivated by findings in social psychology, I use
psychological game theory to incorporate belief-dependent guilt feelings. Multiple equilibria be-
come possible. Some marriages have inefficient under-investment. Some have efficient outcomes
and preclude divorce. If guilt is sufficiently important, a life-long efficient marriage is implied be-
cause a spouse may signal a trust so strong as to force the partner to hold beliefs that make divorce
unattractive. © 2002 Elsevier Science B.V. All rights reserved.

JEL classification: C72; J12

Keywords: Marriage; Marital investment; Divorce; Time consistency; Emotions; Guilt; Trust; Psychological
game theory; Forward induction

1. Introduction

Married couples making human capital investments often concentrate on developing one
spouse’s skills more than the other’s. At first glance this may seem profitable. 1 However,
a closer game theoretic scrutiny suggests that such arrangements are hard to sustain: sup-
pose a wife supports her husband towards a very lucrative specialist education, instead of
making a moderately remunerative investment in her own schooling. Once the husband
receives his valuable degree he has an incentive to divorce his wife, reaping the benefits
from his enhanced earnings capacity all by himself. Of course, a clever wife anticipates

∗ Tel.: +46-8-163547; fax: +46-8-159482; URL: http://www.ne.su.se/∼md/.

E-mail address: md@ne.su.se (M. Dufwenberg).

1 There may be increasing returns to scale in the production of household welfare (Weiss, 1997, Section 2.1); dif-

ferential investments may bring about comparative advantages in household production (Becker, 1991, Chapter 2);
credit from external sources may be more expensive than what the spouses can provide for each other (Borenstein
and Courant, 1989, Section 1 and Weiss, 1997, Section 2.2).

0167-2681/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 7 - 2 6 8 1 ( 0 1 ) 0 0 2 2 5 - 6
58 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

such opportunistic behavior and, therefore, invests in her own schooling. Specialization is
not “subgame perfect’, and an inefficient outcome may result.
This dismal story reflects an implicit assumption that the spouses are motivated solely
to maximize personal income. Such an assumption is often accepted in economics, but in
the current context two sets of results suggest that it needs to be appended. First, some
recent work in experimental economics concern games in which the structure resembles the
situation with the investing spouses, and subjects often nevertheless manage to bring about
efficient outcomes (see, e.g. Berg et al., 1995; Fehr et al., 1997; Dufwenberg and Gneezy,
2000). It seems clear that people do not selfishly maximize their personal income, but rather
have some complementary objectives. The second set of results is suggestive of just what
motivational force may be relevant in the context of marriage, or at least one such force.
Recent findings in social psychology based on autobiographical accounts by Baumeister
et al. (1993) document that a person who rejects a relationship partner often suffers from
guilt. Baumeister et al. (1994, 1995, p. 174) report that: “the prototypical cause of guilt is
inflicting harm or distress on a relationship partner”.
The purpose of this paper is to investigate how guilt may affect the problem of efficient
marital investment in a model which explicitly admits opportunistic divorce. My aim is
to isolate and highlight interesting connections that follow solely from this focus, and I
shall, therefore, abstract from many other aspects that may be relevant for understanding
marriage, divorce, and marital investment as a general matter (cf. Section 4). My model
explicitly describes both, how decisions are taken across time and how guilt affects payoff
perceptions and behavior. The first issue can be handled using game theory. However,
accommodating the second one is more tricky. Geanakoplos et al. (1989) argue that standard
games are unsuitable tools for incorporating emotions in strategic situations. The problem
is that standard games mandate that players’ payoffs depend only on what strategy profile
is played, while many emotions depend directly on the beliefs that the players harbor. In
response, Geanakoplos et al. develop the ‘theory of psychological games’. I shall now argue
that the guilt felt by a divorcer is likely to be a belief-dependent emotion, and propose that
psychological game theory provides adequate tools for treating the problem at hand.
Suppose a husband divorces his wife, despite each spouse having vowed not to pursue
opportunistic divorce. 2 In line with the findings by Baumeister et al. cited the previous
paragraph, it is reasonable to expect the husband to feel guilty to the extent that he believes
he “inflicts harm” on his wife. That harm should depend on the wife’s beliefs, since the
stronger she believes that the marriage will last the worse off she will be following a divorce
relative to her expected payoff before divorce. 3 Effectively, this leads to the assumption
that ‘the stronger the husband believes that his wife believes that he will stay in the marriage,

2 I will throughout phrase the presentation in terms of an educated husband with a supporting wife. This asym-

metric treatment of the sexes is consistent with Weitzman’s (1986, p. 67) observation that: “[h]usbands and wives
typically invest in careers—most particularly in the husband’s education and career—and the products of such
investments are often a family’s major assets”, with Borenstein and Courant’s (1989; Footnote 3) observation that
a medical student with a supporting spouse typically is a husband with a wife, with evidence concerning divorce
cases decided in US courts (Polsby and Zelder, 1994, Footnote 4), and with Cohen’s (1987) general finding that
nuptial gains tend to accrue to men early on in a marriage and to women towards the end.
3 Baumeister et al. (1995, p. 173) write: “Feeling guilty [is] associated with . . . recognizing how a relationship

partner’s standards and expectations differ from one’s own” (my emphasis).
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 59

the more guilty he feels by divorcing her’. Note that a belief about a belief here directly
motivates the husband in his decision making. Psychological game theory is called for!
As a starting point for my analysis, I design a ‘marital investment game’ in which the
given payoffs reflect the spouses’ personal incomes only. The unique subgame perfect
equilibrium is inefficient and entails low marital investment. I then incorporate guilt along
the lines described earlier, creating a ‘psychological marital investment game’. A multitude
of equilibria are possible if the husband is moderately sensitive to guilt. Who obtains depends
crucially on what expectations the spouses harbor. One equilibrium is still inefficient, with
low marital investment and a husband ready to divorce his wife if given the chance. However,
there is also an efficient equilibrium with high marital investment and justified full belief
in a life-long marriage.
If the husband’s guilt sensitivity is very high, I argue that only the equilibrium with a
life-long marriage is viable. This is because by agreeing to support her husband the wife
“signals” a certain degree of belief that the husband will not divorce her. The husband
realizes this and is “forced” to hold a corresponding belief about his wife’s belief about
his choice. With a high enough guilt sensitivity he would feel exceedingly guilty pursuing
divorce. The wife understands this logic of “psychological forward induction”, and hence
she is quite happy to support her husband . . . . In the end, a happy outcome is brought about
which reflects on Leith and Baumeister’s (1998, p. 1) assertion that: “guilt serves many
adaptive, beneficial, and prosocial functions”, and that “guilt helps strengthen and maintain
close relationships” (p. 2).
These results may have some bearing on other partnerships than marriage. A guilt effect
similar to that modeled here can potentially explain the aforementioned experimental re-
sults. 4 Moreover, such an emotional “safeguard” (to allude to a term used by Williamson
(1989, p. 167) against opportunistic behavior may also be relevant in many other real world
contexts. Situations in which the strategic possibilities and monetary incentives resemble
the marital investment game may include business ventures, employment relationships,
when an inventor presents a new idea to a potential producer, or in athletic or musical spon-
soring when a young athlete is financially supported in his early career with the implicit
understanding that the sponsor will get reimbursed if the prodigy becomes a successful
professional. However, emotional concerns are more likely to be important in some part-
nerships than in others. Perhaps in marriages, where monetary transfer between the partners
is not very prevalent, these are particularly salient.
As regards to related literature, Borenstein and Courant (1989) analyze how various
kinds of divorce legislation affect marital investments, efficiency, and equity. As their focus
is on the legal issues, they make less elaborate behavioral assumptions. By assumption, the
supporting spouse forms quite naive expectations, acting as if the probability of divorce
is 0. In contrast, in the model below assumptions regarding the spouses rationality have not
been relaxed. They make marital investments or divorce whenever such actions maximize
expected utility. Each decision is required to be time consistent, meaning that it is optimal at
the time it is taken for some belief with a reasonable foundation. Outside family economics,
Huang and Wu (1994) use psychological game theory to study corruption in principal-agent

4 Dufwenberg and Gneezy measure beliefs (about actions and beliefs) in an experimental game which resembles

the marital investment game, and report evidence which is consistent with this claim.
60 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

Fig. 1. The marital investment game Γ 1 .

relationships. They incorporate “remorse” similarly as I model guilt, but they do not examine
the phenomena of belief signaling and psychological forward induction.
Section 2 contains the benchmark model of a marital investment game with selfish spouses
who are unaffected by guilt. Such an emotion is incorporated in Section 3, which is the main
part of the paper. Section 4 comments on the limitations of the model. Section 5 concludes.

2. The marital investment game

Consider the extensive game Γ 1 of Fig. 1 which models the following situation: first,
nature determines the quality of a married couple’s match (according to some exogenously
given probability distribution). With a bad match, the spouses agree to separate and the end
node where payoffs have not been specified is reached. With a good match, the spouses
must decide on investments in education, and the wife is temporarily in control. The default
outcome is that each spouse makes a moderate investment in her or himself, but the wife
may also let her husband pursue a very profitable education. If she says no to supporting her
husband, each spouse’s lifetime income will be one monetary payoff unit. If she says yes
to supporting her husband, she foregoes the opportunity to invest in herself. Her personal
lifetime earnings will then be zero, but the husband’s lifetime earnings are quadrupled. To
say yes is a good choice for the wife if the husband subsequently decides to stay with her
and split his income. However, the husband also has the option to divorce his wife and reap
all earnings by himself, in which case the wife would be best off saying no. Of course, when
they married he vowed not to act so opportunistically. However, it is here understood that a
“no-fault” divorce legislation applies under which the husband can walk out of the marriage
without his wife’s consent, and that in the benchmark case there is no sanction involved. 5

5 Over the past 25 years, no-fault has become the most common sort of divorce legislation in the western world.

This trend is reflected, e.g. in Price and McKenry’s (1998, Chapter 6) account of the history of US divorce
legislation. See Nakonezny et al. (1995) and Friedberg (1998) for empirical studies of how the introduction of
no-fault divorce law has affected divorce rates in the US. For a penetrating discussion of the virtues and drawbacks
of many different kinds of divorce legislation, see Cohen.
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 61

While the possibility of divorce figures prominently in Γ 1 , the purpose of this paper is
not to explain why and when divorces come about as a general matter. The aim is rather to
analyze how marital investments are made when the associated gains accrue asymmetrically
to the spouses across time, and when the possibility of opportunistic divorce is taken into
account. The first stage of Γ 1 , where nature moves, is included just to underscore what is
and what is not the paper’s objective. Many divorces come about because of the resolution
of some kind of uncertainty. Nature’s choice in Γ 1 corresponds to an exogenously realized
“quality of match”—a shock often modeled in theoretical approaches to marriage and
divorce (see, e.g. Weiss and Willis, 1993). Such a shock plays a role in Γ 1 , yes, and this
may be realistic. However, from now on I will leave this shock aside and focus on what
happens in the subgame reached ‘after nature has selected a good match’.
The subgame of Γ 1 , where the wife starts to move has a very simple structure, yet it is
rich enough to capture the basic tension between the efficient marital investment and time
consistent decision making discussed in the introduction. There are potential efficiency
gains from the educational investment that takes place if the wife says yes. Both spouses
prefer the strategy profile (yes, stay) to any profile where the wife says no. In this respect,
Γ 1 is related to other models of marriage (see, e.g. Weiss, 1997; Section 2.2). However, a
key feature of Γ 1 is that, unlike many other models in the marriage and divorce literature,
unilateral divorce appears as an explicit choice opportunity. When this is taken into account,
Γ 1 has an obvious time consistent solution. If the husband is called upon to play, to divorce
is a dominant choice. The wife figures this out, and hence says no to supporting her husband.
This argument is captured by the solution concept of ‘subgame perfect equilibrium’. In Γ 1 ,
there is a unique such strategy profile: (no, divorce). This outcome is inefficient. Note that
the solution is robust in the sense that (no, divorce) remains the unique solution even if a
small change is made to any payoff parameter in Γ 1 .
It may seem that the spouses could profit from signing some clever prenuptial agreements.
I will, however, in this paper not discuss such contracts. This can be justified in at least
two ways: first, contracts may be avoided because they are too costly, financially or even
emotionally. As noted by (Cohen, 1987, p. 291), explicit discussion of marriage contracts
may be considered “indelicate during courtship”. Second, as argued by Ulph (1988), even
if one wishes to admit (Nash) bargaining it seems reasonable that the relevant threat points
are determined by a non-cooperative solution, a clear understanding of which then is crucial
to the bargaining process.

3. Adding guilt

In standard games, a player’s payoff depends only on what strategy profile is played, and
not on the player’s belief. Geanakoplos et al. argue, however, that when emotions matter
in strategic situations it is often the case that beliefs directly influence payoffs. In order to
deal with such issues, they develop the theory of psychological games.
It is easy to imagine ways in which a divorce may evoke emotions which depend
on beliefs: when a husband suddenly divorces his wife it is possible that the stronger the
wife believed that her husband would stay, the more disappointed she is. The stronger the
husband believes that his wife believes that he will stay, the more gratifying he may find
62 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

it to do so. The husband may be averse to letting a trusting wife down, and the stronger
he believes that she believes that he will stay the more guilty he feels by forcing divorce.
In the Introduction it was argued that the last of these cases may be relevant to the situation
modeled in Γ 1 . In Section 3.1, the marital investment game Γ 1 will, therefore, be modified
into a psychological marital investment game Γ 2 which incorporates an effect of divorcer
guilt. The first stage where nature moves is no longer explicitly given. Γ 2 is solved in
Sections 3.2 and 3.3.

3.1. A psychological marital investment game

In the psychological marital investment game Γ 2 to be constructed, the spouses have the
same strategy sets as in Γ 1 and they move in the same order. In Γ 1 , the unique solution
involved pure strategies but in what follows mixed strategies may be relevant. Moreover,
beliefs in the form of certain expectations are important. Some new notation is needed in
order to represent mixed strategies and beliefs, to formalize the psychological assumption
that will be used, and to calculate equilibrium behavior. The spouses’ actions will be denoted
as follows:
σ ∈ [0, 1]: the probability with which the wife says yes;
τ ∈ [0, 1]: the probability with which the husband stays.

Some data concerning the spouses’ beliefs, will be denoted as follows:

τ  ∈ [0, 1]: the wife’s expectation of τ (her trust);
τ  ∈ [0, 1]: the husband’s expectation of τ  .

These expectations are part of the beliefs the spouses hold when making their respective
choices. They play a crucial role when the psychological marital investment game below is
solved. Note that τ  is interpreted as the wife’s trust. There is a large literature (spanning
many fields) which attempts to define and analyze the notion of trust. The usage of the term
here (recall the husband’s vow from the presentation of Γ 1 ) is consistent with that of Rotter
(1980) who defines (interpersonal) trust as an “expectancy held by an individual that the
word, promise, oral or written statement of another individual or group can be relied on”
(p. 1).
The second-order expectation τ  , interpreted as the husband’s belief in his wife’s trust,
is used to model an emotion. Specifically, the following assumption will be made.

Assumption 1 (psychological). When the husband makes his choice, the stronger he
expects that his wife trusts him to stay the more disutility of guilt he experiences by choos-
ing divorce. That is, if the (yes, divorce) profile is implemented, the husband’s utility is
decreasing in τ  .

Assumption 1 reflects the idea that the husband is averse to letting his trusting wife down.
I argued in the Section 1 that this idea is consistent with findings in social psychology. One
can certainly conceive of some other motivational concerns that may have a bearing on
the problem under study, and some of these are discussed in Section 4. However, since my
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 63

Fig. 2. The psychological marital investment game Γ 2 .

aim in this paper is to isolate and highlight interesting connections concerning how guilt
affects marital investment in a model which admits opportunistic divorce I shall abstract
from other aspects that may be relevant in many marriage contexts. The only way in which
the transition from Γ 1 to Γ 2 involves a change in the spouses motivation is via Assumption
1. The following assumption only governs more specifically how this is implemented.

Assumption 2 (technical). The “guilt effect” of Assumption 1 enters additively into the
husband’s utility function. The husband’s “guilt sensitivity”, is constant at γ ≥ 0.

Now an extensive psychological marital investment game Γ 2 (see Fig. 2) can be con-
structed which captures the Assumptions 1 and 2. It is convenient to indicate explicitly
in connection to the game tree that σ ∈ [0, 1] are probabilities chosen by the respective
spouses.
Given any γ > 0, Γ 2 is not a standard game in which a unique payoff vector is associated
with each strategy profile. The husband’s payoff following the profile (yes, divorce) rather
depends on τ  , his expectation of his wife’s trust, which is in line with the Assumptions 1
and 2. To illustrate, assume the marginal guilt sensitivity equals 3, i.e. γ = 3. Say the profile
(yes, divorce) occurs. Depending on τ  , his payoff may vary from 1 to 4.
Note that if γ = 0, the payoffs in Γ 2 collapse to those in Γ 1 .

3.2. Solving Γ 2 : preliminary observations

With a subjective belief affecting payoffs in Γ 2 , at first glance, this psychological game
may seem difficult to analyze. However, careful inspection suggests that for some parame-
terizations sharp predictions appear quite reasonable. First, consider the cases where γ < 2.
Since τ  ∈ [0, 1], it must hold that 4 − γ τ  > 2, and hence the husband will rationally
choose to divorce irrespective of his beliefs. The wife should figure this out, and hence say
no to supporting her husband, just like in the subgame perfect equilibrium of Γ 1 .
What happens for larger values of γ ? Leave intermediate cases aside for the moment
and consider the case where γ = 5. Suppose that the wife says yes. She then maximizes
her expected payoff only if she expects to get at least a payoff of 1. This means that her
expectation of τ is at least 1/2, or equivalently that her trust τ  is at least 1/2. Hence, if the
64 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

husband believes that the wife is rational in this sense, he must believe that τ  ≥ 1/2 if he
is called upon to play. But this means that τ  ≥ 1/2, and since this belief affects his payoff
he should stay (since 4 − 5 × 1/2 < 2). So, if the wife believes that her husband believes
that she is rational, then she must believe that τ = 1, in which case she will of course say
yes. A life-long efficient marriage with full trust is guaranteed!
What goes on in this example is an instance of what might be dubbed “psychological
forward induction”. With γ = 5, the husband is so sensitive to feeling guilty that when
his wife says yes she signals a trust so strong as to force the husband to hold a belief that
makes staying a dominant choice. Rabin (1993) raises the issue that effects of this nature
may be obtained in psychological games, although he does not deal explicitly with games
with a dynamic structure. He asks: “can players ‘force’ emotions, i.e. can a first mover
do something that will compel a second mover to regard him positively?” The example
discussed here illustrates that the answer to this question may be affirmative.
The reader may verify that an analogous psychological forward induction argument can
be applied whenever γ > 4, but not for lower values of γ . Consider, e.g. the case where
γ = 3. Again, the trust signaled by the wife equals 1/2. However, the husband is now not
forced to hold a belief that makes stay a dominant choice (since 4 − 3 × 1/2 > 2), so it
seems that a divorce is not out of the question. On the other hand, nothing seems to exclude
the possibility that τ  takes a value such that 4 − 3 × τ  < 2, so it seems that also his choice
to stay can be justified.
By analogous reasoning the reader may verify that whenever γ ∈ [2, 4]] some belief
τ  ∈ [1/2, 1], impregnable to a psychological forward induction argument, can be found
such that any particular choice is optimal for the husband. Hence, it is not obvious what the
wife should do. In order to get more definite conclusions in the cases where γ ∈ [2, 4], it is
necessary to introduce some techniques which are inspired by Geanakoplos et al.’s theory.
This will be done in the Section 3.3 where a general solution is proposed for Γ 2 .

3.3. Solving Γ 2 : marital equilibria

In the previous section, it was suggested that no marriage obtains if γ < 2, that a life-long
marriage obtains if γ > 4, and that it is unclear what happens for intermediate values of
γ . So far, no presumption of equilibrium has been made. Here, a solution will be presented
which formally captures the heuristic arguments brought forth in the previous section as
well as adds some structure to the cases where γ ∈ [2, 4].
In standard games payoffs depend on actions chosen only, and it is therefore not neces-
sary to explicitly write down the requirement that beliefs are correct in equilibrium, even
if this is usually implicitly understood to be the case. The spirit of subgame perfection
entails that players optimize at all decision nodes given their correct beliefs about one
another’s actions. In Γ 2 , however, the husbands’ payoff depends directly on τ  , and it is
necessary to impose explicitly that the beliefs τ  and τ  are correct in equilibrium. To this
end, note that given τ  , Γ 2 has real numbers characterizing payoffs at each end-node. In
this sense, it reduces to a “standard game”, to be denoted Γ 2 (τ  ). A time consistent equi-
librium solution of Γ 2 must fulfill three requirements: first, the spouses must optimize their
beliefs and choices at all given decision nodes. This means that they must play a subgame
perfect equilibrium in Γ 2 (τ  ). Second, the beliefs must be consistent what is actually
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 65

happening, so that τ  = τ  = τ . 6 Third, the equilibrium must be impregnable to a

psychological forward induction argument as sketched in the previous section.
The following definition, in which (σ , τ ) denotes the strategy profile in which the wife
says yes with probability σ and the husband divorces with probability τ , imposes these
requirements formally.

Definition 1. The profile (σ , τ ) is a marital equilibrium in Γ 2 if:

1. (σ , τ ) is a subgame perfect equilibrium in the standard game Γ 2 (τ  );
2. τ  = τ  = τ and
3. 4 − γ × 1/2 < 2 ⇒ τ = 1.

Condition 1 is the “subgame perfection” requirement. Condition 2 guarantees that the

husband is motivated by a belief which is consistent with what is actually happening.
Condition 3 requires robustness against a psychological forward induction argument as
sketched in Section 3.2. If the husband is called upon to move he must believe that τ  ≥
1/2. Therefore, τ  ≥ 1/2, and so condition 3 captures the idea that he must choose to stay
whenever γ > 4.
The definition will now be used to solve Γ 2 for different values of γ . It is convenient to
group the marital equilibria into three qualitatively different cases.
1. The suspicious spouses (no, divorce): In this equilibrium the spouses choose the same
strategies as in the subgame perfect equilibrium of the game Γ 1 . The equilibrium exists
whenever γ ∈ [0, 4], and it is the unique equilibrium whenever γ ∈ [0, 2]). The wife
does not trust her husband at all (τ  = 0) and she does not support him. The equilibrium
entails that τ  = τ  = τ = 0.
2. The trusting twosome (yes, stay): This equilibrium exists whenever γ ≥ 2 and it is the
unique equilibrium whenever γ > 4. It entails that τ  = τ  = τ = 1. There is full trust
(τ  = 1) and the spouses live happily ever after. The payoffs are (2, 2), which Pareto
dominates the suspicious spouses equilibrium in which payoffs are only (1, 1).
3. The mixed matrimony: This equilibrium exists whenever γ ∈ (2, 4) and entails that
τ = τ  = τ  = 2/γ . The wife supports her husband (says yes) in all the cases where
γ ∈ (2, 4), since then τ = 2/γ > 1/2.
These equilibria match the outcomes hinted at in Section 3.2: if γ < 2, only the suspicious
spouses equilibrium is possible. Multiple types of equilibria are possible when γ ∈ [2, 4],
and which one is relevant depends on the spouses beliefs. If γ > 4, only the trusting
twosome equilibrium is viable. 7

6 Geanakoplos et al. impose explicitly the strong restriction that equilibrium profiles be common knowledge. It

simplifies the presentation, and affects no conclusion in the current context, to be explicit only about those parts
of the players’ beliefs which have a direct bearing on some players’ payoff perception. Only τ  appears at an end
node in G2 , and therefore τ  = τ  = τ is the only explicit restriction on beliefs that is made.
7 Note that this conclusion is robust in the sense that small changes to any payoff parameter in G will (making
2
obvious changes to condition 3 of the definition) qualitatively leave the pattern of possible equilibria unaffected:
for γ small enough, only the suspicious spouses equilibrium is possible, for γ large enough only the trusting
twosome equilibrium is viable, and in between there are multiple equilibria.
66 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

This solution invites two further comments. First, the emotional effects modeled in
Section 3 could not have been adequately captured using standard (non-psychological)
game theory. In any standard extensive game with the same tree as Γ 1 or Γ 2 , subgame
perfect equilibrium choices can be determined by backward induction. If in such a game
the husband strictly prefers to divorce, there cannot also be a subgame perfect equilib-
rium where he stays. Contrast this with the case of Γ 2 with γ ∈ (2, 4)). In the suspicious
spouses equilibrium, the husband strictly prefers to divorce, and yet there is also the trusting
twosome equilibrium where he strictly prefers to stay. Neither of these equilibria can be
identified using backward induction, since the optimal choice for the husband at his node
depends on τ  . In Geanakoplos et al.’s words (p. 63), “in psychological games . . . a node
. . . does not capture adequately the state of a game: the node identifies a history of play,
but not the players’ beliefs”.
Second, it is noteworthy that a unique equilibrium outcome is implied when γ > 4. It is
the psychological forward induction requirement imposed via condition 3 of the definition
of a marital equilibrium that is responsible for this. The reader familiar with the work of
Geanakoplos et al. may have wondered why their notion of a “subgame perfect psychological
equilibrium” is not applied to Γ 2 . The answer is that their theory is not adequate for capturing
the logic of psychological forward induction. They restrict attention to psychological games
where only initial (pre-play) beliefs are allowed to directly affect the payoffs (although they
mention on p. 78 that it may be desirable to consider alternatives). This restriction makes a
forward induction argument inconceivable—if a player is forced to revise his beliefs as play
proceeds this will have no bearing on his relative valuation of different strategy profiles.
By contrast, the psychological forward induction argument is built around the idea that the
husband’s payoff depends on his belief ‘at the time he moves’ (as specified in Assumption
1), and the solution concept of marital equilibrium takes this into account.

4. Extensions

The purpose of this paper is not to propose a fully realistic account of marriage, divorce
and marital investment. I have focused on how the presence of a psychological guilt effect
influences marital investment behavior in a context which admits unilateral divorce and
where the investment gains accrue asymmetrically to the spouses across time. For the sake
of clarity, I have abstracted from all other issues. This section offers some brief comments
on aspects of reality that I have disregarded, but which it may be interesting to consider in
future work.
There are emotional or other motivational concerns besides guilt that may influence
behavior. Sociologists typically emphasize various non-monetary considerations seldom
touched upon by economists. See, e.g. Price and McKenry (1998, Chapter 2) or Collins and
Coltrane (1991, Chapters 8, 9 and 12) for interesting discussions of companionship, esteem
for spouse, erotic ties, love, etc. Elster (1998) discusses how a vast array of emotions may
influence economic behavior. I refer the reader to his Section 2 for an extensive listing and
classification, and here mention only a specific emotion which may have particular relevance
in the current context. Baumeister et al. (1995) report evidence that: “people who induce
guilt sometimes seem to feel guilty themselves over doing so” (p. 188), a phenomenon they
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 67

term meta-guilt. It may be worthwhile to investigate how behavior in the marital investment
game is affected if such an emotion influences the wife’s motivation.
One assumption which is made rather often in the economics of the family literature is
that decision makers are altruistic; their well-being depends positively on other persons’
well-being. Altruism typically does not depend on beliefs, and can, therefore, be handled
using conventional tools of economics or game theory (see, e.g. Becker, 1991, Chapter 8;
or Stark, 1995). However, it is conceivable that altruism can interplay in interesting ways
with a belief-dependent emotion like the guilt feelings modeled in this paper, and that it
could be interesting to incorporate these things simultaneously in one model.
My model fails to capture why in some marriages divorce occurs after marital investments
have taken place. Consider the following story told by Takas (1986, p. 48):
[W]hen I got divorced my husband and I quickly agreed on financial arrangements . . . .
We split everything down the middle. Everything, i.e. except what could be the single
most lucrative asset of our marriage—his newly earned post-professional degree. I’d put
him through school, yet he would keep an earning power that had doubled while my own
stood still.
It is true that the model of Section 3 permits a limited possibility of post-investment
divorce, as part of a mixed matrimony equilibrium. However, such an explanation may
appear contrived, as it portrays the husband as indifferent between divorcing or staying in
marriage. I suggest that this points to the following problem: it is implicitly assumed in
Section 3 that all parameters of Γ 2 are common knowledge between the two spouses. It
may be more realistic to analyze a game of incomplete information. For example, one may
assume that while the husband knows the value of γ , the wife has incomplete information
about this parameter. With proper adjustments, one can, e.g. develop an equilibrium where
the wife’s a priori expectation of a life-long marriage is good enough that she agrees to
support her husband, yet she may at times be unlucky and catch a husband with little guilt
sensitivity who will force divorce (details of an explicit example are available on request).
There are some findings in social psychology which lend support to such an approach.
According to Tangney (1995, p. 1138), “there are stable individual differences in the degree
to which people are prone to shame and guilt”. If a person’s trait is not perfectly observable
to others, this justifies assuming incomplete information.
Finally, I note that my model abstracts from many other strategic considerations that may
be of relevance in a context of marriage. It may thus be interesting to modify the spouses’
strategy spaces in a marital investment game in order to analyze more complicated family
situations. For example, child production could be incorporated. The presence of children
may affect the payoffs that the spouses will get if they if they stay married or divorce each
other. This, in turn, may have an impact on the decision to enter wedlock, on fertility, and
on welfare.

5. Concluding remarks

This paper may be viewed as a contribution to a research program recently promoted by

Elster (1998). He asks (p. 48), “[H]ow can emotions help us explain behavior for which
68 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69

good explanations seem to be lacking?” I have used psychological game theory to ana-
lyze how guilt may help sustain efficient marital investment in marriage where the gains
from this investment activity accrue asymmetrically to the spouses across time, and there
is the possibility of opportunistic unilateral divorce. The introduction of guilt affects the
nature of equilibrium play dramatically, as it may allow the spouses to benefit from efficient
investment activities which would not materialize were each spouse selfishly maximiz-
ing monetary income. Behavior-wise, I have argued that both experimental evidence and
casual observation of real life marriages suggests that people indeed manage to reach effi-
cient outcomes in similar situations. Motivation-wise, the specific guilt effect modeled was
motivated with reference to findings in social psychology.
As can be sensed from Elster’s article, not much research has been done which introduces
emotions in economic analysis. While pinpointing which motivational amalgam really ap-
plies in some situation may be one of the ultimate goals of this line of research (if that goal is
at all attainable), when the first steps are taken into this arena it may be wise to go slowly and
introduce only a limited number of carefully motivated emotional elements. Accordingly,
in this paper only money and the guilt effect influence payoffs. I hope there are two lessons
to learn. First, there are insights directly related to the specific belief-dependent guilt effect
being modeled. The way, this arguably important emotion influences the outcome sheds
light on interaction in real life marriages and possibly also other partnerships. Second,
I also hope to have promoted the idea that Geanakoplos et al.’s theory of psychological
games is a useful tool for work on incorporating emotions in economic analysis. Many
emotions depend on beliefs. I have given some examples, Geanakoplos et al.’s text con-
tains several others, as does Ruffle’s (1999) psychological games-based study of gift-giving
with emotions. Elster (1998, p. 49) writes that: “[e]motions . . . are triggered by beliefs”.
Psychological game theory, unlike standard game theory, goes to the heart of this matter
by using utility functions that are defined on domains that specify beliefs in addition to
strategic choices.

Acknowledgements

I am most grateful to Yoram Weiss who introduced me to this topic and has given me
helpful advice. I thank Jan Bouckaert, Gary Charness, Eric van Damme, Eddie Dekel, Tone
Dieckmann, David Frankel, Uri Gneezy, Bo Larsson, Michael Lundholm, Niels Noorder-
haven, Roald Ramer, Phil Reny, Bradley Ruffle, Frank Verboven, Jörgen Weibull, Shi-Young
Lee, Robin Pope and participants in many seminars for valuable suggestions and discussion.
The paper is a revised version of a chapter in my thesis.

References

Baumeister, R.F., Wotman, S.R., Stillwell, A.M., 1993. Unrequited love: on heartbreak, anger, guilt, scriptlessness,
and humiliation. Journal of Personality and Social Psychology 64, 377–394.
Baumeister, R.F., Stillwell, A.M., Heatherton, T.F., 1994. Guilt: an interpersonal approach. Psychological Bulletin
115, 243–267.
 M. Dufwenberg / J. of Economic Behavior & Org. 48 (2002) 57–69 69

Baumeister, R.F., Stillwell, A.M., Heatherton, T.F., 1995. Personal narratives about guilt: role in action control
and interpersonal relationships. Basic and Applied Social Psychology 17, 173–198.
Becker, G.S., 1991. A Treatise on the Family. Harvard University Press, Cambridge, MA.
Berg, J., Dickhaut, J., McCabe, K., 1995. Trust, reciprocity, and social history. Games and Economic Behavior
10, 122–142.
Borenstein, S., Courant, P.N., 1989. How to carve a medical degree: human capital assets in divorce settlements.
American Economic Review 79, 992–1009.
Cohen, L., 1987. Marriage, divorce, and quasi rents; or, I gave him the best years of my life. Journal of Legal
Studies 16, 267–303.
Collins, R., Coltrane, S., 1991. Sociology of Marriage and the Family: Gender, Love, and Property. 3rd edition.
Nelson-Hall, Chicago.
Dufwenberg, M., Gneezy, U., 2000. Measuring beliefs in an experimental lost wallet game. Games and Economic
Behavior 30, 163–182.
Elster, J., 1998. Emotions and economic theory. journal of economic literature 36, 47–74.
Fehr, E., Gächter, S., Kirchsteiger, G., 1997. Reciprocity as a contract enforcement device: experimental evidence.
Econometrica 65, 833–860.
Friedberg, L., 1998. Did unilateral divorce raise divorce rates? Evidence from panel data. American Economic
Review 88, 608–627.
Geanakoplos, J., Pearce, D., Stacchetti, E., 1989. Psychological games and sequential rationality. Games and
Economic Behavior 1, 60–79.
Huang, P.H., Wu, H.-M., 1994. More order without more law: a theory of social norms and organizational cultures.
Journal of Law, Economics and Organization 12, 390–406.
Leith, K.P., Baumeister, R.F., 1998. Empathy, shame, guilt, and narratives of interpersonal conflicts: guilt-prone
people are better at perspective taking. Journal of Personality 66, 1–37.
Nakonezny, P.A., Shull, R.D., Rodgers, J.L., 1995. The effect of no-fault divorce law on the divorce rates across
the 50 states and its relation to income, education, and religiosity. Journal of Marriage and the Family 57,
477–488.
Polsby, D.D., Zelder, M., 1994. Risk-adjusted valuation of professional degrees in divorce. Journal of Legal Studies
33, 273–285.
Price, S.J., McKenry, P.C., 1998. Divorce. Sage, Beverly Hills, CA.
Rabin, M., 1993. Incorporating fairness into game theory and economics. American Economic Review 83, 1281–
1302.
Rotter, J.B., 1980. Interpersonal trust, trustworthiness, and gullibility. American Psychologist 35, 1–7.
Ruffle, B.J., 1999. Gift giving with emotions. Journal of Economic Behavior and Organization 39, 399–420.
Stark, O., 1995. Altruism and Beyond. MIT Press, Cambridge, MA.
Takas, M., 1986. Divorce: who gets the blame in ‘no-fault’. Ms. Magazine, Feb., 48–52, 82–83.
Tangney, J.P., 1995. Recent advances in the empirical study of shame and guilt. American Behavioral Scientist
38, 1132–1145.
Ulph, D., 1988. A general model of household consumption behavior. Discussion Paper 1988/205, University of
Bristol.
Weiss, Y., 1997. The formation and dissolution of families: why marry? who marries whom? and what happens
upon divorce? In: Rosenzweig, M.R., Stark, O. (Eds.), The Handbook of Population and Family Economics,
Vol. 1A. North-Holland, Amsterdam (Chapter 3).
Weiss, Y., Willis, R.J., 1993. Transfer among divorced couples: evidence and interpretation. Journal of Labor
Economics 11, 629–679.
Weitzman, L.J., 1986. A new look at career assets. Ms. Magazine, Feb., 67–68.
Williamson, O.E., 1989. Transaction cost economics. In: Schmalensee, R., Willig, R.D. (Eds.), The Handbook of
Industrial Organization, Vol. 1. North-Holland, Amsterdam (Chapter 3).