Transcript Aucun titre de diapositive - University of New South Wales

“You can’t be happier than your wife”
Divorce and the distribution of life satisfaction across spouses
Cahit Guven (Deakin University) and Claudia Senik (Paris
School of Economics)
September 4, 2009
What this paper does
• Ask:
 Does the distribution of life satisfaction across spouses matters per
se?
 Does it predict divorce?
 (beyond the level of individual satisfaction of each spouse)
• Try to answer this question using the GSOEP panel data 1983-2007
• 359958 observations, 45225 individuals, 13456 couples
2
Motivation: 1. Economic consequences of divorce
• Impact of actual and expected divorce on factors of GDP growth:
 Fertility and number of children
 Capital accumulation in marital specific assets (Becker, 1974)
 Human capital of children (education, care, expenditure)
 Houses
 Specific human capital of spouses
 Labor market force participation of women
• Implications for public policy concerning family and women’s labor
force participation
• Generalize the evidence on aversion to inequality?
 to “other contracts of indefinite duration where the parties involved
have the option of termination, perhaps with a penalty” (Becker et
al. 1977)
Motivation: 2. aversion to inequality in households?
• Literature on income distribution and subjective well-being
 Negative association between income inequality and SWB
• Literature on income comparisons and well-being
 income comparisons and other types of comparisons inside the
household (Clark, 2005) associated with lower levels of happiness
• Literature on marriage and divorce
 Essentially self-centered decision of getting/remaining married
• But no literature on whether the distribution of subjective wellbeing inside the household matters.
4
Motivation: 3. Reliability of subjective data
• Show impact of subjective variables on actual choices, decisions
and actions
 Inequality in Subjective Well-Being Divorce
The economics of marriage and divorce
• Marriage is viewed as a means to maximize individual welfare and
collective output (Becker, 1974, 1991)
• Joint production, joint consumption (e.g. children)
 Increasing returns, division of labor, risk pooling, coordination
• Rational individuals:
 look at her level of well-being inside marriage versus outside and
decides whether to become/remain married or not (Becker)
• Other compatible assumptions:
 Altruism, intra-household externalities of welfare (Powthdawee, 2004)
6
Unitary models of household
• Basic unitary model:
 One decision-maker
 Consider only aggregate utility for all members
• More sophisticated models (Becker 1974, 1991)
 Head of household is altruistic: takes into account individual
preferences of household members
 Gains of marriage shared among members of family depending on
marriage market (sex ratio)
 Upfront payments in traditional societies: dowries or bride-price
 Division of labor in modern families
Unitary models of household (continued)
• Income pooling
 behavior of spouses (labor supply, expenditures) only depend on
aggregate exogenous income
 Does not depend on the distribution of income across members
 But unitary model of household rejected by empirical tests
 Phipps and Burton (1992)
Collective models of the household
• Cooperative models (following Chiappori, 1992):
 1) Sharing rule depends on individual preferences and individual
bargaining power (distribution factors)
 Bargaining power depends on outside wage, divorce legislation, child
custody rules, remarriage market, etc.
 2) Each individual maximizes his utility under the budget constraint
defined in first stage
 Pareto efficiency of all decisions
• Non cooperative models of Nash bargaining
 not necessarily Pareto-efficient
The economics of marriage and divorce
• But are all equilibria in terms of distribution of welfare across
spouses stable?
• Beyond purely self-regarding motives, are there also concerns
for the distribution of well-being?
Concerns for the distribution of well-being across
spouses?
• We try to answer this question, controlling for the classical
correlates of the value of marriage/ value of outside options (Weiss
and Willis, 1997)
 Income, education, age, of each spouse, children, etc.
• We take life satisfaction as given, as the result of bargaining and all
intra-household decisions and allocations (chores, etc.)
• We find a positive statistical association between the difference in
life satisfaction across spouses and the probability that they will
divorce in later years.
11
Possible mechanisms
• Aversion to inequality in terms of happiness inside couples
• The gap in satisfaction is a sign of the degrading quality of the
marriage technology
 altruism, sharing, spillovers of SWB, pooling
 Impossibility to transfer well-being between spouses
 Makes compensation of the less happy spouse impossible
• Positive assortative mating in terms of life satisfaction more stable
 Matching on the set-point of happiness (Lucas and Schimmack, 2006),
Fujita and Diener (2005), Lucas et al. (2003)
12
Other alternative explanations
• Reverse causality: the perspective of divorce makes one spouse
more unhappy and creates the happiness gap that we observe
• Infidelity: One of the spouses is contemplating (or experiencing)
forming another couple, and this creates the gap between him and
his spouse
•  We try to rule out these mechanisms using long distance lagged
variables, pre-marital life satisfaction levels and other strategies.
Some related papers on marriage and divorce
using subjective happiness data
• GSOEP:
 Lucas et al. (2003), Stutzer and Frey (2006), Zimmermann and Easterlin
(2006): Marriage makes people happy (beyond happier people getting
married)
 Lucas and Schimmack (2006): Similarity of happiness of spouses
• BHPS:
 Gardner and Oswald (2002): Marriage increases life expectancy
 Gardner and Oswald (2005): Divorcing couples become happier
 Powdhtavee (2009): Happiness spillover effect between spouses
Data
• GSOEP panel data 1983-2007
 Individual and partner identification variable for 45226 people and
252753 observations
 Number of couples: 13456
 Number of divorces : 4074
• GSOEP includes a separate spell dataset for marital status.
• Constructed dataset: sample of women with all socio-demographic
variables pertaining to themselves and their husband. Before, during
and after marriage.
• Symmetrically: sample of men with all variables pertaining to
themselves and their wife.
15
Attrition
• % 10 of couples in the sample for the whole period (23 years)
• Average duration of a couple in the sample is 13.4 years
• By men: 13.3 years, by women: 13.5 years
• Characteristics of those who are more likely to leave the sample:
men, non-German, young, unmarried, seperated
(Kroh and Spieß, 2008)
• We weight the observations by the inverse of the probability to
remain in the sample.
Estimates
• We run a dprobit estimate of the probability to divorce
• Divorce t+1 = f (total happinesst, absolute value of happiness difference
between spousest; age t, age differencet, household incomet, number
of childrent)
(1)
• Controls =classical determinants of marriage and divorce (Weiss and
Willis, 1997)
• Cluster standard errors at individual level
Comparability of self-declared happiness of
spouses?
• Individual fixed effects or couple fixed effects controls for the
anchoring effect
 Interpretation: probability of divorce depending on the evolution
of the gap in SWB
• Impact of subjective representation of happiness rather than
objective happiness
Description of the data and main variables
How happy are you? (scale: 0-10)
Not weighted
20
Absolute difference in happiness across spouses,
1984-2007
21
Couples who marry and do not divorce throughout
the sample (1984-2007)
Total happiness, happiness gap around the year of
divorce
Married and partnering together
Individual happiness and happiness gap around
the year of divorce
Married and partnering together
Total happiness and happiness gap around the
year of divorce
Legally Married Only
Total residual happiness, residual happiness gap
around the year of divorce
Married and partnering together
Residuals of equation (1)
% of divorces depending on happiness differences
Married and partnering together
Residuals of equation (1)
OLS estimates of the % of people who divorce
T-statistics are reported in absolute values.
The second column is estimated only at the first year of marriages.
Number of observations=number of years.
Results
Probability to divorce and absolute value of
happiness difference
One row per control
show only wife results during the whole presentation
put interesting coefficient in bold
Standard errors clustered at individual level
Happiness difference as a categorical variable
Standard errors clustered at individual level
Hapiness difference and marriage duration:
Only for those who married in the sample
• Do for those who marry in the sample
Standard errors clustered at individual level
Avoid the risk of reverse causation or infidelity
Happiness gap in the first year of marriage predicts
divorce
Write Dprobit
Lagged values of absolute happiness differences
Write Dprobit
Controls: total happiness, age, age difference, number of children, ln
household income. Each coefficient corresponds to a separate
regression.
Robustness: additional controls
Sample of wives
Write Dprobit
Controls as usual, cluster(individual)
Robustness continued. Sample of wives
Write Dprobit
Split into several tables
Controls: as usual. Cluster(individual).
Robustness continued. Sample of wives
Write Dprobit
Split into several tables
Controls: as usual. Cluster(individual). Column 5: omitted category: one
spouse born in Germany and the other is not.
Robustess continued. Sample of wives
Write Dprobit
Self-reported health : 5 is very good health; 1 bad health. Individual
fixed effects is estimated using conditional logit.
Write Dprobit
Omitted: 1) different nationalities, 2) German origin and living inWest-Germany, 4)
Each manages own money separately. Specification 3 is estimated by weighting with
the inverse of the individual longitudinal staying probabilities which
is provided in the GSOEP. In specification 5, importance of family:1 very
unimportant; 4 is very important and, is treated as a continuous variable.
Robustness
Same results obtained on the sample of husbands
Unexpected income shocks, such as disability and unemployment
increase the absolute value of happiness difference but,
can not predict divorce.
Interpretation
• Aversion for inequality of happiness
• Positive assortative mating in terms of happiness
• Indeed there are signs of assortative mating in the data
Assortative mating by happiness level
in the first year of marriage
1 if happiness<5
2 if happiness=5, 6, 7
3 if happiness>7
Assortative mating by residual happiness
in the first year of marriage
1 if residual happiness>0
0 if residual happiness<0
Happiness gap between divorced people remains
higher than between married people
(although it decreases)
Attention: need to take absdslife not dslife
Conclusions
• Some evidence suggestive that more equal distributions of
subjective well-being are more favorable to marriage continuation.
• Reflects bargaining inside household or assortative matching.
• One additional motive of marriage/divorce beyond purely self-
regarding motives.
• Predictive power of SWB variables.
Descriptive statistics of the main variables
Descriptive statistics of the main variables
Transition matrix of partnership
The estimates excludes people who lose partners due to death. 2.02
is the probability of separation from partner conditional on having
the same partner in the previous period. Ratios are in percentages.
Correlation matrix of happiness variables
Happiness difference =Happiness of husband – happiness of wife
Transition matrix of marital status
We do not differentiate between separations and divorces in the paper. Hence
separation/divorce probability for marital relationships is 0.93+0.31=1.24.
Ratios are in percentages.
Correlation matrix of lagged absolute value of
happiness differences