—
Mismatching in Long-Term Relationships∗
Robert Ullmann‡
March 1, 2019
Abstract
We use administrative micro-level data of all 1,799,066 divorce
cases in Germany over the 2006 to 2015 period to empirically show
that marriages with weddings at year end last considerably shorter
than does the average (median) marriage. Specifically the average
(median) difference (conditional on divorce) between marriages with
December weddings and marriages with weddings in all remaining
months is 500.4 (654) days with overall length of marriage being
5223.6 (4498) days. We attribute this empirical observation to the
German marriage tax benefit which is granted for the entire
calendar year if the couple is married for at least one day
therein. In response to this incentive, couples use forward
shifting of the ’legal’ wedding date as a tax planning strategy to
collect the marriage tax benefit for the year(s) prior to their
’real’ (i.e. in the absence of any tax benefit) date of the
wedding. We theoretically model this decision and argue that, in
essence, couples forego sufficient courtship time to collect the
marriage tax benefit, and hence, mismatching in long-term
relationships results.
—First Discussion Draft—
∗The authors thank Stefan Weil and Michelle Wittler (RDC of the
German State Rhineland- Palatinate, Bad Ems) for their invaluable
support in facilitating analysis of the confidential administrative
data. Data source: RDC of the Federal Statistical Office and
Statistical Office of the German States, full population divorce
data (EVAS 12631) for the period 2006-2015. We further thank Mark
Trede for helpful comments. †Email:
[email protected]. ‡Email:
[email protected].
1
Contents
1 Introduction 1
2 Literature Review and Institutional Details 2 2.1 Literature
Review . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 2.2
Marriage Tax Benefit . . . . . . . . . . . . . . . . . . . . . . .
. . . 2 2.3 Wedding Month Distribution . . . . . . . . . . . . . .
. . . . . . . 4
3 Theoretical Model and Hypotheses 6 3.1 The Setup . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . 6 3.2 Couple’s
Optimization Problem . . . . . . . . . . . . . . . . . . . . 7 3.3
Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . 8
4 Empirical Strategy 8
5 Data 9
6 Results 13 6.1 Probability of Divorce Conditional on Wedding
Month . . . . . . . 13 6.2 Length of Marriage Conditional on
Wedding Month . . . . . . . . 14
6.2.1 Graphical Analysis . . . . . . . . . . . . . . . . . . . . .
. . 14 6.2.2 Regression Analysis . . . . . . . . . . . . . . . . .
. . . . . 17
7 Robustness and Caveats 18 7.1 General . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . 18 7.2 Family Structure . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 18 7.3 Time
Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
8 Conclusion 21
A Tables 24
B Figures 28
1 Introduction
Since Becker’s (1973; 1974) pathbreaking articles on the theory of
marriage, economists have been interested in studying the driving
forces of the formation and dissolution of couples. At the latest
since Alm and Whittington’s series of papers starting around 1995
one key factor that has received more attention is the differential
tax treatment of single and married individuals. The lack of
marriage neutrality in the income tax scheme changes the combined
tax liabilities of two individuals after they marry. Depending on
the specifics of the tax scheme in a country the married couple may
either face a “marriage tax benefit” implying a lower combined tax
liability, or a “marriage tax penalty” implying a higher tax
liability after marriage. The former as well as latter change the
gains from mar- riage and have an influence on the formation and
dissolution of couples (see Alm, Dickert-Conlin and Whittington
1999 for an overview).
In Germany, married couples profit from a marriage tax benefit. The
marriage tax benefit enables married individuals to transfer tax
thresholds and tax exemp- tion limits between one another, and
hence, unused excess amounts are less likely. Moreover, married
couples can apply a beneficial tax rate schedule. Notably, the full
annual marriage tax benefit is granted for each calender year in
which the couple was married for at least one day. Given these
benefits, it is empirically well established that the German
marriage tax benefit impacts on the choice of wedding dates.
Specifically, it creates incentives to use forward shifting of the
wedding date as a tax planning strategy. However, albeit peculiar,
wedding date shifting is widely viewed as a rather minor distortion
to welfare. Contrary to this, we provide empirical evidence that
wedding date shifting is in fact strongly indicative of mismatching
in long-term relationships (measured as lower duration of
marriage).
This paper makes several contributions to the literature. First, we
provide new evidence on the unintended consequences of a
distortionary marriage tax benefit in Germany. So far, the
literature has concentrated on analyzing the effects of
distortionary income taxes (be it a marriage tax benefit or
penalty) on the decision to marry and to divorce in separation. To
our knowledge we are the first who show empirically that a marriage
tax benefit is associated with a mismatching of couples as measured
by lower marriage duration. Second, we estimate the magnitude of
these effects in a large full population data set. Finally, our
results are interesting for policy-makers who typically aim at
encouraging marriage and at preventing divorce. We demonstrate that
offering a marriage tax benefit is a delicate endeavor that in the
end might lead to welfare losses.
The paper is organized as follows: Section 2 provides a short
literature review and describes the relevant institutional details
of the German income tax scheme. Section 3 establishes a simple
theoretical framework that shows how a marriage tax benefit can
lead to hasty marriages. Section 5 presents the data and Section 6
shows the main results. Finally, Section 7 provides a discussion of
our results before Section 7 concludes the paper.
1
2.1 Literature Review
Alm and Whittington (1995; 1999) are among the first who studied
how taxes affect the probability of marriage. They use data for the
US where, in contrast to Germany, marriage comes with a tax
penalty. Even though the effect is small, they find statistically
significant evidence that the larger the tax penalty on marriage,
the less likely an individual is to marry. Obviously, a tax system
that distorts the decision to get married likely also distorts the
decision to divorce. Alm and Whittington (1997a) and Dickert-Conlin
(1999) present evidence for this and show that in the US couples
with more to gain from separating are more likely to divorce.
Sjoquist and Walker (1995), Gelardi (1996) and Alm and Whittington
(1997b) specifically analyze for different countries the influence
of the income tax scheme on the timing of marriage. All papers find
a significant positive relationship between the marriage penalty in
a year and the probability of delaying marriage until the following
year.
Also in other areas of family life the tax scheme has been found to
affect the timing of actions. Dickert-Conlin, Chandra (1999) and
LaLumia, Sallee and Turner (2015) reveal that the (not prorated)
child-related tax benefits in the US increase the probability of a
late-December birth. LaLumia et al. (2015) also document a large
reporting response of self-employment income since a newborn
changes the location of a household’s Earned Income Tax Credit
(EITC) sched- ule. Similarly, Gans and Leigh (2009) show that in
Australia the (unexpected) introduction of a USD 3,000 “Baby Bonus”
for Babies born on or after July 1st 2004 caused over 1,000 births
to be moved to the respective date. Most of the effect was due to
changes in cesarean procedures or birth inductions. Not only the
timing of birth has been found to be influenced by financial
incentives but also the timing of death. Kopczuk and Slemrod (2003)
find that changes in the US estate-tax system caused people to
prepone or postpone their death in order to reduce their heirs tax
liability.
2.2 Marriage Tax Benefit
The design of the German marriage tax benefit creates incentives to
use forward shifting of the wedding date as a tax planning
strategy. Specifically, the marriage tax benefit is granted for
each calender year in which the couple was married for at least one
day. The marriage tax benefit allows spouses to transfer tax
thresholds and tax exemption limits between one another, and hence,
unused excess amounts are less likely. Moreover, liquidity benefits
can arise temporarily during the calendar year if parameters are
optimized between the spouses that impact on wage withholding tax.
Finally, married couples can apply a beneficial tax rate
schedule.
Figure 1 shows the income tax rate schedule for Germany for the
year 2015. The solid line depicts the marginal income tax rate
while the dashed line depicts
2
the average income tax rate. The marginal income tax rate for
individuals is zero up to the tax exemption limit which was EUR
8,472 and then increases up to 45% for incomes higher than EUR
250,731. Due to the tax exemption limit and the increasing marginal
tax rate, the German income tax scheme is clearly
progressive.
0
10
20
30
40
50
nt )
0 200 400 600 800 1000 Taxable Income of Married Couple (Joint
Taxation, kEUR)
0 100 200 300 400 500 Taxable Income of Individual (Individual
Taxation, kEUR)
Marginal Tax Rate Average Tax Rate
Figure 1: Tax Rate Schedule for Individual Taxation vs. Joint
Taxation
Married couples can choose between separate and joint assessment.
Joint as- sessment represents the default option and is also the
option that most couples in fact choose. Only under very rare
circumstances can separate assessment be more beneficial for the
married couple. This includes certain cases of losses with one
spouse or of tax free income that nonetheless impacts on the tax
rate of taxable income (exemption with progression). Empirically,
the number of married couples with separate assessment is
negligible.
If couples choose the separate assessment, each spouse’s yearly
gross income (Yj with j = f,m) is taxed according to the income tax
schedule T (Yj) as illustrated in Figure 1. Abstracting from
possible deductions, their common tax burden with separate taxation
simply amounts to T (Yf ) + T (Ym) which equals the sum of taxes
they have to pay when they are not married. In other words, with
separate assessment the marriage does not change their common tax
burden. If, instead, the couple opts for joint assessment, then the
couple’s common income tax burden is given by 2 · T (0.5(Yf + Ym)).
So the tax rate schedule applies to half of the sum of the couples’
gross income; and the tax burden for this amount is then paid
3
twice. Since the income tax scheme is progressive, joint assessment
in almost all cases comes with a lower overall tax burden. The
marriage tax benefit is nil only when the marginal tax rates do not
differ between spouses in separate assessment. Otherwise, the
couple profits in that income is automatically shifted to a lower
income tax rate.
In cases with different income of spouses, the marriage tax benefit
ceteris paribus increases in the difference of spouses’ income.
However, the marginal increase in overall marriage tax benefits
also decreases with higher overall income of the couple. The
maximum tax benefit, for instance in 2015, asymptotically
approaches EUR 15,784. This maximum tax benefit is obtained when
one spouse earns zero and the other spouse earns EUR 501,462 (or
more), i.e. when the couple first enters the bracket of the highest
marginal tax rate. However, EUR 501,462 is a high income in the
German income distribution. Altering the income of the earning
spouse to EUR 43,344, which is the German average wage income in
2015 (Statista 2018), the marriage tax benefit is EUR 4,265. We
argue that neither of the two amounts would be enough to enter into
a marriage in which the misfit between the spouses is known, but
rather that the amount nodges couples to marry prematurely.
Utilization of the marriage tax benefit in year t has no impact on
the tax benefit in year t + 1, and hence, couples do not face an
intertemporal decision problem. For couples at the beginning of
their marriage, the institutional framework is such that the exact
wedding date in year t is irrelevant for a couple’s tax burden in
that year. Specifically, the full marriage tax benefit for year t
is granted if the couple was married for at least one day therein.
I.e. the same marriage tax benefit can be collected when the couple
marries on January 1st in year t, and when it waits and marries on
December 31st in year t, or any time in between - and hence
shifting the wedding date within a year is irrelevant from an
income tax perspective. However, postponing the wedding date from t
to t+ 1 can, of course, come with tax losses. Specifically, the
couple forgoes the option for joint assessment in year t and can
not collect the marriage tax benefit for that year. Acting
according to the well- known saying “so test therefore, who join
forever” by Friedrich Schiller may thus cause tax losses.
Ultimately, when December 31st nears, couples may fear to lose the
marriage tax benefit for the current year and decide to marry
before the year end threshold cliff. This may distort the couple’s
optimal courtship time decision. We refer to this behavior as
’wedding date (forward) shifting’.
Sections 3.1 and 3.2 below show with a simple theoretical model how
a hasty marriage in order to collect the marriage tax benefit can
affect a couple’s matching quality and thus their divorce
probability.
2.3 Wedding Month Distribution
It is well investigated in the literature that couples react to the
marriage tax ben- efit by shifting their ’legal’ wedding dates
forward (relative to their unobservable counterfactual ’real’
wedding date in the absence of tax benefits). This leads to
4
tax avoidance as it allows these couples to collect the full
marriage tax benefit for one additional year. Arguably, wedding
date shifting occurs mostly into Decem- ber, as then the relative
amount of time shifted is lowest, c.p., and hence the costs of
shifting (i.e. foregone courtship time) are minimal while
collecting the full mar- riage tax benefit. Henceforth, we denote
marriages with weddings in December as “December marriages”; and
correspondingly for all other wedding months.
To show that wedding date shifting occurs, we obtain data from the
National Office of Statistics which holds monthly numbers of
weddings from the full popu- lation of ’legal’ weddings in Germany
over the 1990 to 2015 period, i.e. covering 10,608,220 weddings. We
refer to this data set in the following as ’wedding data’. We
convert the monthly number of weddings into fractions per year and
report the available data in Table 1.
[Insert Table 1 about here]
We then treat this data as cross-sectional, by summing the number
of weddings per month over all years, and create a histogram in
Figure 2. We additionally include the median and its non-parametric
binomial interpolated 95% CI for the monthly values of fraction per
month over all years.
0
5
10
15
Figure 2: Wedding Month Distribution Over All Years (Wedding
Data)
5
We report that differences between median and mean are
non-significant for all months. Considered jointly, Table 1 and
Figure 2 show that the distribution of wedding dates across all
years is as expected from ’common sense’, i.e. that the majority of
weddings occur in summer and early fall. However, we also observe a
notable peak in number of weddings in December (where the
counterfactual should arguably be at max 5%). This peak is strongly
indicative of wedding date shifting, particularly of wedding date
forward shifting. The pattern submerges virtually identical in all
individual years in our wedding data and is well investigated in
the literature (for an overview see the dissertation of Scholz
(2015)). Wedding date shifting is unanimously attributed by
researchers to the German marriage tax benefit. However, the extend
of wedding date shifting is not the topic of this paper. Instead,
we ask if wedding date shifting has an impact on long-term
relationships.
3 Theoretical Model and Hypotheses
3.1 The Setup
Assume the following situation. A couple meets in period t. They do
not yet know about their matching quality θ ∈ {θ`, θh} which can be
either low θ = θ` < 0 or high θ = θh > 0. In period t+ 1 the
couple receives a signal σ ∈ {pos,neg} about their matching
quality. When the signal is positive then the matching quality is
θh and when the signal is negative the matching quality is θ`. The
signal is thus perfect. The positive signal has probability p. The
expected matching quality in period t is thus:
Et[θ] = pθh + (1− p)θ`. (1)
Denote the net income of the female partner yf and the net income
of the male partner ym. Precisely, we define yj = Yj − T (Yj) for j
= f,m. In other words, net income is the income of each spouse if
they were individually taxed. Marriage (with joint taxation) comes
with a tax benefit B given by:
B = T (Yf ) + T (Ym)− 2 · T (Yf + Ym) ≥ 0,
that is the difference in taxes they have to pay when they were
individually taxed and the taxes they have to pay with joint
assessment.
Each partner derives utility from consumption of a numeraire
commodity x. Partners make all decisions cooperatively, that is
they maximizes their common utility uf + um which is simply a
linear function of their consumption and the matching quality, that
is uj = xj + θ. A couple’s intertemporal preferences in period t
are given by:
Ut = uf,t + um,t + δ[uf,t+1 + um,t+1]
where δ ∈ [0, 1] reflects the degree of a couple’s
impatience.
6
The timing is as follows: in period t the couple decides if they
want to marry, or if they prefer to postpone marriage to the next
period. In period t + 1 the couple receives the signal σ about
their matching quality. If the couple married in period t, they
choose if they want to remain married, or if they rather opt for a
divorce. And, if the couple did not marry in period t + 1, they
decide if they want to get married, or if they prefer to split up.
In case the couple opts for a divorce they have to pay the divorce
costs c. These costs include the costs of a lawyer, possible costs
for alimony, or psychological and stigma costs.
3.2 Couple’s Optimization Problem
The game needs to be solved backwards, so that we first have to
analyze a couple’s optimization problem in period t+ 1.
Period t + 1. Assume the couple married in period t. Then, their
common utility in period t+ 1 if they stay married after they have
received the signal is:
UM t+1 = yf + ym +B + 2θ. (2)
If, after observing the signal, they decide to divorce, they loose
the marriage benefit and have to pay the cost of divorce, but the
matching quality become irrelevant. Hence, their common utility
amounts to:
UD t+1 = yf + ym − c.
A couple will divorce if doing so gives them a larger joint utility
than remaining married.1 Thus, a divorce occurs if:
UD t+1 > UM
t+1 ⇔ −θ > 0.5[B + c].
The above inequality can only be true when the matching quality is
low, i.e., θ = θ` < 0. Lets simply assume that this is the case,
that is −θ` > 0.5[B + c].
Assume the couple did not marry in the period t, but postponed
their decision to period t+1. Then, after they received the signal
they decide whether to marry, or to split up. If the couple marries
their utility in period t+ 1 is again given by Expression (2). In
case the couple splits up, their common utility simply amounts
to:
US t+1 = yf + ym.
That is, when the couple splits up the marriage tax benefit and the
(negative) matching quality are lost, but unlike in the case of a
divorce they do not have to pay the additional costs. A couple
splits up in period t+ 1 if
US t+1 > UM
t+1 ⇔ −θ > 0.5B.
The above inequality is true when the realized matching quality is
low.
1Peters (1986) provides evidence that couples divorce if it is
efficient to do so.
7
Period t. Lets turn to the couple’s period t decision problem.
First, assume the couple “tests before joining forever”, then their
expected joint utility in period t when they only cohabitate in
period t is given by:
UC t = yf + ym + δ
[ pUM
] .
With probability p the signal is positive and the couple marries in
period t + 1, while with probability 1− p the signal turns out
negative and the couple splits up in period t+ 1.
Now, assume the couple “rushes into marriage” in period t, then
their expected utility amounts to:
UM t = yf + ym +B + δ
[ pUM
] .
UM t ≥ UC
3.3 Hypotheses
An early marriage is more likely if
(i) the degree of a couple’s impatience δ is lower,
(ii) the costs of divorce c are lower or
(iii) the probability of a bad matching quality (1− p) is
lower.
Additionally, it is obvious that couples who marry early have a
higher risk of divorce since those who wait and find out that they
indeed are a bad match will not marry in the first place. We
ultimately formulate the following two hypotheses to test in our
data (alternative form):
• H1: December marriages are significantly more likely to be
divorced than are marriages with wedding dates in all other
months.
• H2: December marriages are significantly shorter than are
marriages with wedding dates in all other months.
4 Empirical Strategy
We focus on descriptive analyses in our empirical strategy. Most of
these descrip- tive analyses are trivial graphical consolidations
of the data. We use this graphical approach because i) arguably the
research question is straightforward and ii) the data can be used
in obvious ways to show the effects in our hypotheses. Graphical
analyses also enable us to probe for potential drivers of the
effects.
8
However, we also include regression analysis to allow for more
elaborate con- trols and to provide an economic estimate of the
effect. Our baseline specification covers different versions of the
following model.
LMarriagei = α+ β1DecWedi + β2NovWedi + β3JanWedi
+
n∑ k=1
µkControlki + ε. (4)
As the endogenous variable, we use LMarriagei, which measures the
number of days between the date of the legal marriage and the date
of the final court decision to divorce for couple i. In alternative
specifications, we use LLMarriagei as the endogenous variable,
which is the natural logarithm of LMarriage.
As our main exogenous variable of interest, we include a
dichotomous vari- able DecWedi, which takes value one if divorce
case i is a December marriage, i.e. has a wedding in December in
any wedding year t, and zero otherwise. We correspondingly include
NovWedi, for November marriages, and JanWedi, for January
marriages.
Controls are MAgeWedi and FAgeWedi which measure age of both male
spouse and female spouse, respectively, at the time of wedding. For
mere pur- poses of descriptive analysis, we similarly compute age
at divorce, MAgeDiv and FAgeDiv. We also use as a control Children
which holds the number of joint children that are below the age of
18 at the date of divorce. As dichotomous con- trols, we include
MForeigni and FForeigni indicating if either the male spouse or the
female spouse, respectively, are foreign nationals at the time of
divorce. In alternative specifications, we include fixed effects
for each possible nationality of the spouses, i.e. MNationi and
FNationi. We also include fixed effects for the spouse that
initiated the divorce proceedings (Litiganti, also including
informa- tion about whether or not the other spouse agreed to the
divorce proceedings) and for the district in which the couple was
living at time of divorce (Districti). We argue that District is a
geographic measure to proxy for differences in religion, historical
and cultural background and also for financial stability of the
region, and hence, ultimately between couples. Even though District
is measured at the time of divorce, it is also indicative of the
area of living at the time of the wedding as mobility within
Germany is usually modest. We include fixed effects indicating year
of wedding t for observation i (Y earWedi).
5 Data
We use administrative family court data on all divorce cases over
the period of 2006 to 20152. Henceforth, we refer to this data set
as ’divorce data’. In Ger- many, a marriage can only be divorced by
family court decision, and thus, our
2Data made available by the Federal Statistical Office and
Statistical Office of the German States (data project
identification code: EVAS 12631).
9
data set includes all possible cases of divorce in Germany. Note,
however, that we do not have corresponding data on all weddings
during the period that is covered by the weddings in our divorce
data, as the wedding data only includes informa- tion on number of
weddings per month and year. In consequence, all our more detailed
analyses are conditional on a marriage being divorced. A marriage
is also terminated by death of one spouse but these cases are not
included in our data set.
We have available the exact date of legal wedding, the exact date
of legal divorce and the exact date of birth of the two spouses. We
use these variables to compute LMarriage, LLMarriage, DecWed,
NovWed, JanWed, MAgeWed, FAgeWed, MAgeDiv and FAgeDiv as discussed
above. In addition, we take all remaining variables presented in
Section 4 from the divorce data set. We are unable to see if a
spouse was divorced prior to the divorce that we observe in our
data set. Hence, it is possible that we unknowingly observe
multiple divorces of individual spouses in our data.
We point out that the data does not have actual panel structure, as
we nat- urally observe only one divorce per court case.
Nonetheless, we do have divorce data available for multiple years
with the time variable being the year of the court decision. This
divorce data naturally includes also a wide range of years of
wedding dates Y earWedi.
We are unable to merge financial data to individual court cases
because of legal limitations in data usage.
We show in Table 2 descriptive statistics of December marriages
(Panel A), jointly for all marriages between January and November
(Panel B) and jointly for all marriages (Panel C).
[Insert Table 2 about here]
We observe that roughly 9.4% of weddings in our data set of overall
1,799,066 divorce cases occur in December. We also find that -
without including any covari- ates - the average (median)
difference between December marriages and marriages with weddings
in all remaining months is 500.4 (654) days with overall LMarriage
on average (median) being 5223.6 (4498) days. Already this naive
descriptive in- vestigation indicates that differences in LMarriage
might be substantial between wedding months.
We also note that mean (median) age of the spouses, i.e. MAgeWed
and FAgeWed, is considerably larger in December marriages relative
to marriages with weddings in all remaining months. Over all months
the average male spouse is 1051.6 days older than the average
female spouse at time of marriage. This dif- ference between
MAgeWed and FAgeWed is also larger in December marriages relative
to marriages with weddings in all remaining months. Corresponding
dif- ferences are naturally also found with MAgeDiv and FAgeDiv.
The distribution
10
of all the age related variables appears symmetric around the mean
(median), al- beit slightly less so, i.e. with mean greater than
median, for age at wedding than for age at divorce. Given that our
divorce data is conditional on divorce, this implies that spouses
in the right tail of the distribution of MAgeDiv or FAgeDiv are
more likely to have a lower LMarriage. Finally, the variables
Children, MForeign and FForeign show no notable differences
conditional on wedding month.
We then create histograms of LMarriage conditional on wedding
months in Figure 3. Each bin covers 1095 days, i.e. approximately
three years.
11
Length of Marriage (Days)
Month 2 (N = 82944)
Length of Marriage (Days)
Month 3 (N = 111306)
Length of Marriage (Days)
Month 4 (N = 122594)
Length of Marriage (Days)
Month 5 (N = 204433)
Length of Marriage (Days)
Month 6 (N = 187764)
Length of Marriage (Days)
Month 7 (N = 195844)
Length of Marriage (Days)
Month 8 (N = 224153)
Length of Marriage (Days)
Month 9 (N = 185686)
Length of Marriage (Days)
Month 10 (N = 143653)
Length of Marriage (Days)
Month 11 (N = 106616)
Length of Marriage (Days)
Month 12 (N = 168257)
Length of Marriage (Days)
Median Mean Kernel Density
Month 1 (N = 65816)
Length of Marriage (Days)
Median Mean Kernel Density
Figure 3: Distribution of LMarriage Conditional on Wedding
Month
We observe that the distributions are right skewed and of similar
shape for all wedding months. We also note differences in sample
size per wedding month which correspond to our discussion on the
wedding data in section 2.3.
12
6.1 Probability of Divorce Conditional on Wedding Month
To investigate our first hypothesis H1, we use the wedding data as
described in Section 2.3 above with the fractions as presented in
Figure 2. We then corre- spondingly compute fraction of weddings
per wedding month for each Y earWed observable in our divorce data.
We then plot the median of fraction of weddings in a given wedding
month over all years in the wedding data against its corre-
sponding value in the divorce data. Figure 4 shows the results. We
treat data as cross sectional and do not weigh for number of
weddings or divorces in any given year.
A dot above (below) the diagonal indicates that weddings in the
respective months are observed more (less) often in the divorce
data set than in the wedding data, and hence, that divorces are
more (less) probable for this wedding month.
Jan
Feb
a)
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Fraction of Weddings per Month (Percent, Wedding Data)
Median 95% CI
Figure 4: Median Scatter Plot of Fraction of Weddings Conditional
on Wedding Month (Wedding Data vs. Divorce Data)
For the months January, February, November, March and April we
observe a significantly higher fraction of marriages in the divorce
data than in the wed- ding data. For the months December,
September, June and May we observe a significantly lower fraction
of marriages in the divorce data than in the wedding
13
data. Hence, we report that marriages with the former (latter)
wedding months are significantly more (less) likely to be divorced.
Notably, it seems to be system- atic that months with a lower
(higher) fraction of annual weddings belong to the former (latter)
group. We have no explanation for this peculiarity.
Specifically with regard to H1, we report that December weddings
occur sig- nificantly less often in the divorce data than in the
wedding data. Hence, the data does not speak to the directional
hypotheses H1. The data even provides stark indication that the
Null of H1 is true.
6.2 Length of Marriage Conditional on Wedding Month
6.2.1 Graphical Analysis
We begin our investigation of H2 using graphical analyses. First,
we overlap all kernel distributions for all months from Figure 3
into one graph in Figure 5. The distribution of December marriages
is shown in black, all other months’ distribu- tions are shown in
grey.
0
.00005
.0001
.00015
.0002
Length of Marriage (Days)
December All Other Months
Figure 5: Kernel Density of LMarriage Conditional on Wedding Month
(Over- lapping)
We observe that the distribution of LMarriage for all wedding
months January to November is rather similar while December
marriages have a higher maximum
14
with a relatively lower right tail. The maximum of the distribution
function of LMarriage for the December marriages occurs at about
the same LMarriage as with all other wedding months. The next
highest maximums are observed in the distribution functions of
November marriages, February marriages and January marriages (in
that order); the lowest maximum is observed in the function of May
marriages.
Investigating the distribution further, we separate our entire data
set into 3- day intervals conditional on wedding date within year.
We plot the Median (and its 95% CI) of each 3-day interval in
Figure 6, vertically standardized to the annual median (dashed
horizontal line), i.e. the median over all observations regardless
of wedding date.
-1000
-500
0
500
1000
n
3-Day Intervals
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110
115 120
Median 95% CI Significant
Figure 6: Median of LMarriage Over 3-Day Wedding Date
Intervals
We observe that LMarriage fluctuates with large dispersion around
the annual median. However, specifically for December marriages, we
observe that observa- tions are significantly lower than the annual
median. Considering particularly observations near the year end
threshold, we observe that of the 20 last obser- vations before
year end, 19 are significantly negative and none are significantly
positive. This is indicative of wedding date forward shifting not
only into De- cember but even into November of the previous
year(s). Moreover, medians for adjacent 3-day intervals at year end
are relatively close to one another, indicating systematic effects.
To the contrary, when extending our analysis mirror-inverted
15
for the first 20 observations after year beginning, we observe that
9 are significantly negative and 2 are significantly positive.
Moreover, confidence intervals are larger and median point
estimates are closer to the annual median for observations at year
beginning compared to observations at year end. Replicating Figure
6 with the 25% quartile and the 75% quartile instead of the median
generates similar patterns (not graphed).
We then use week-like intervals and focus on December and January
as wedding months. Figure 7 shows interquartile ranges of LMarriage
for week-like intervals around the year-end threshold. Note that
the non-standardized y-axis ranges from 2000 to 8000 days, hence we
report stark dispersion in quartiles of LMarriage.
2000
4000
6000
8000
95% CI 95% CI
Figure 7: Interquartile Range of LMarriage Over Week-Like Wedding
Date In- tervals
First we note that variation in median of LMarriage between the
wedding date intervals decreases considerably relative to Figure 6,
indicating noise in Fig- ure 6 owing to small sample sizes per
3-day interval. In Figure 7, we observe that quartiles are
relatively similar within months. Between months, we observe sig-
nificantly higher medians and 75% quartiles for all week-like
intervals in January compared to December. This effect is by far
largest in economic magnitude with the 75% quartiles. To the
contrary, the 25% quartiles of the week-like intervals are about
equal in December marriages and January marriages with particularly
small confidence intervals. Specifically short-term and medium-term
marriages
16
at the 25% quartile and the median seem to be less affected by the
marriage tax benefit while mismatching occurs specifically with
long-term marriages in the 75% quartile of LMarriage.
6.2.2 Regression Analysis
We report results of our baseline regression in Table 3. These
include different specifications of control variables and fixed
effects. We recall that we are partic- ularly interested in the
dichotomous variable DecWed which indicates marriages with weddings
that occurred in December.
[Insert Table 3 about here]
Considering first the most naive specification in (1), we find that
December marriages are on average 512.11 days shorter than in the
omitted category wedding months. November marriages are 213.09 days
shorter while January marriages are 56.55 days longer than in the
marriages in omitted wedding months. However, as we include
additional control variables we observe that effects for December
mar- riages and November marriages decrease in absolute economic
magnitude while the effect for January marriages increases in
economic magnitude. In terms of economic magnitude, in the most
extensive specification (10) we find that Decem- ber (November)
marriages are 139.14 (133.78) days shorter than marriages in the
omitted category wedding months. January marriages are 115.08 days
longer than marriages in the omitted category wedding months. The
opposite mathematical sign between year end and year beginning with
similar economic magnitudes in- dicates wedding date shifting out
of year beginning into year end. This leaves on average better
matched couples in the January marriage category, and hence leads
to relatively greater LMarriage there.
Considering the control variables in the specifications with fixed
effects, we find that Children have a positive association with
LMarriage. It is not surprising that this effect only emerges when
including especially Y earWed fixed effects because Children is
also in some ways a proxy for the available time to have children
within the marriage. Marriages with at least one non-German spouse
are significantly shorter than is the average marriage. This effect
is relatively larger for a non-German male spouse, i.e. MForeign,
relative to a non-German female spouse, i.e. FForeign. The age of
the spouses at time of wedding, MAgeWed and FAgeWed, explains a
large part of the difference in LMarriage between year end and year
beginning, as is evidenced by comparing column (1) and (3).
Coefficients are negative in the majority of specifications
indicating that older spouses are more likely to have a lower
LMarriage, conditional on divorce.
F-statistics do not compute in some specifications because of
singleton dum- mies in the variable Y earWed, i.e. because of
individual years of wedding only occurring exactly once in our
data.
17
We repeat the baseline regression using LLMarriage as the
endogenous vari- able, including the same specifications of control
variables and fixed effects, below in Table 4. We recall that we
are particularly interested in the coefficient of DecWed which
indicates December marriages.
[Insert Table 4 about here]
We find similar effects in mathematical sign and magnitude as in
Table 3, indi- cating that our functional form assumptions are
valid in both regression analyses. Specifically, we find in column
(10) that December marriages are 3.6% shorter than the weddings in
the omitted category wedding months.
Overall, our results indicate that December marriages and November
marriages are significantly shorter than marriages in the omitted
category wedding months, and hence that LMarriage is negatively
associated with wedding day forward shifting. We thus confirm H1.
Moreover we do find that January marriages are significantly longer
than marriages in the omitted category wedding months. This is
somewhat contrary to the graphical findings in Figure 6. However,
we note that Figure 6 does not include any control variables and
hence regression results provide more detailed understanding. Our
regression results are consistent with wedding date forward
shifting out of January as a wedding month, i.e. out of the earlier
months in the year t+ 1 into the preceding months, i.e. into the
later months in year t.
7 Robustness and Caveats
7.2 Family Structure
We investigate how marriages with different months of wedding
differ in family structure. We note that our regression results
explicitly control for number of children below the age of 18 at
date of divorce (Children). However, we still provide more
descriptive discussion here. As a naive analysis, we show in Figure
8 the variable Children, conditional on wedding month.
18
50.2 50.4 50.0 51.3 50.8 50.0 52.9 51.3 51.1 50.7 50.0 49.8
26.5 26.4 26.4 26.0 26.4 26.8 25.0 26.0 26.0 26.0 26.4 26.9
18.7 18.6 19.0 17.7 17.6 18.2 16.6 17.5 17.7 18.3 19.1 18.8
3.7 3.6 3.7 3.8 3.9 3.9 4.1 4.0 4.0 3.9 3.6 3.6 1.0 0.9 0.9 1.1 1.3
1.1 1.3 1.3 1.2 1.1 0.9 0.9
0
20
40
60
80
n Feb Mar Apr
Wedding Month
0 1 2 3 >=4 Number of Children < 18 yrs at Date of
Divorce
Figure 8: Categories of Children Conditional on Wedding Month
We find that 49.8% to 52.9%, 25.0% to 26.9%, 16.6% to 19.1%, 3.6%
to 4.1% and 0.9% to 1.3% of divorces result from couples with zero,
one, two, three, or four and more children, respectively. This data
diverges from available data on children per woman in Germany,
which shows that approximately 20%, 26%, 38%, 12% and 4% of women
have zero, one, two, three, or four and more children, respec-
tively (National Office of Statistics 2016, data on childbearing by
women born in 1967-1971). These data sets are naturally not
directly comparable. However, re- calling that the average (median)
wedding lasts for 5223.6 (4498) days, i.e. would theoretically
allow for four and more children to be born, and given that
children tend to be born in the earlier years of the marriage a
naive comparison of these data sets provides some indication that
marriages resulting in divorce have rela- tively less children than
do non-divorcing marriages of equal length. This finding is also
mirrored in our analysis on LMarriage in Table 3 by the positive
coefficient on Children with specifications (8) to (10), i.e. when
including Y earWed fixed effects.
Comparing the distribution conditional on month of wedding, we
observe no particularly stark difference over the 12 wedding months
in overall family struc- ture. However, we do note that January
marriages present the maximum fraction of cases with zero children
(52.9%) and the minimum fraction of cases with one child (25.0%)
and two children (16.6%). Somewhat reversed, December
marriages
19
are near the lower bound of fraction of cases with zero children
and near the upper bound of fraction of cases with one child.
However, we also recall that descriptive analyses in Table 2 show
that spouses of December marriages are considerably older than are
spouses in all other wedding months. It seems plausible that older
couples are more likely to have children faster after their wedding
than do younger couples or that older couples relatively more often
already have children at time of wedding.
7.3 Time Variation
In our baseline specification, we ignore time variation in our main
effect. However, respective trends may exist. Correspondingly, we
plot median length of both December marriages and January marriages
as well as difference between the two in Figure 9, conditional on
year of divorce.
288 201 440 410 549.5 643 757
525 712 736
95% CI Jan 95% CI Dec Diff Jan-Dec
Figure 9: Difference in LMarriage between December Marriages and
January Marriages over Time
First, we observe an increase in LMarriage over time for both
December mar- riages and January marriages. The time period of 2006
to 2015 is characterized by an overall strong economic development
in Germany. A positive association between LMarriage and economic
development is well-documented. Second, and more importantly, we
observe that the difference in median LMarriage between
20
January marriages and December marriages is significant in all
divorce years avail- able in our data. Moreover, and possibly more
important, we observe a stark trend towards larger differences in
median over time. For instance, in 2015, the differ- ence in median
LMarriage between December marriages and January marriages is 736
days, relative to 288 days in 2006.
We note that the years of wedding are controlled for with fixed
effects Y earWed in our baseline specification. As a
robustness-test, we include fixed effects for the year of divorce
instead of Y earWed in column (8) and (10) and find results simi-
lar to our baseline specification results without Y earWed in
column (4) and (9). Standard errors are robust to
heteroskedasticity in all specifications. However, nei- ther of
these specifications assess time trends in differences in LMarriage
between December marriages and January marriages. In consequence,
the coefficients in our baseline specification might underestimate
(overestimate) the economic mag- nitude of the effect for the more
recent (earlier) years.
8 Conclusion
We show that December marriages last considerably shorter than does
the aver- age (median) marriage. Specifically the average (median)
difference (conditional on divorce) between December marriages and
marriages with weddings in all re- maining months is 500.4 (654)
days with overall length of marriage being 5223.6 (4498) days. When
controlling for potential confounders, we find that December
marriages are 139.14 days or 3.6% shorter relative to marriages
with weddings in February to October. Similar results as for
December marriages can be found for November marriages. To the
contrary, January marriages are significantly longer than are
marriages with weddings in February to October.
We attribute this empirical observation to the German marriage tax
benefit which is granted for the entire calendar year if the couple
is married for at least one day therein. We argue that this
marriage tax benefit creates incentives for couples to marry
prematurely, i.e. to forego sufficient courtship time in exchange
for the marriage tax benefit of one additional year. Our results
indicate that wedding date shifting occurs particularly into
November and December, i.e. that couples feel pressured to marry
when they see the year end threshold cliff approaching.
One policy advice that derives from our analysis could be that
marriage tax benefits should not have a year end threshold cliff.
Having such a stark incentive induces couples to ’jump ahead’ to
avoid foregoing an entire year of marriage tax benefit. If the
marriage tax benefit was instead granted, for instance, pro rata
for each month in the year of marriage, such an institutional setup
would provide a much smoother incentive scheme. Ultimately, an
improved institutional setting would lead to less tax-induced
wedding date shifting, and thus, to less mismatching in long-term
relationships.
21
References
[1] Alm, J., Dickert-Conlin, S., and L.A. Whittington, “Policy
Watch,”Journal of Economic Perspectives, 1999, 13 (3),
193-204.
[2] Alm, J., and L.A. Whittington, “Does the income tax affect
marital decisions?,”National Tax Journal, 1995, 48 (4),
565-572.
[3] Alm, J., and L.A. Whittington, “Til death or taxes do us part:
the effect of income taxation on divorce,”The Journal of Human
Resources, 1997a, 32 (2), 388-412.
[4] Alm, J., and L.A. Whittington, “Income taxes and the timing of
marital decisions,”Journal of Public Economics, 1997b, 64 (2),
219-240.
[5] Alm, J., and L.A. Whittington, “For love or money? The impact
of income taxes on marriage,”Economica, 1999, 66 (263),
297-316.
[6] Becker, G.S., “A theory of marriage: part I,”Journal of
Political Economy, 1973, 81 (4), 813-846.
[7] Becker, S., “A theory of marriage: part II,”Journal of
Political Economy, 1974, 82 (2), 11-26.
[8] Chade, H., and G. Ventura, “Taxes and marriage: a two-sided
search analysis,”International Economic Review, 2002, 43 (3),
955-985.
[9] Dickert-Conlin, S., “Taxes and transfers: their effects on the
decision to end a marriage,”Journal of Public Economics, 1999, 73,
217-240.
[10] Dickert-Conlin, S., and A. Chandra, “Taxes and the timing of
births,”Journal of Political Economy, 1999, 107, 161-177.
[11] Gans, J.S., and A. Leigh, “Born on the first of July: an
(un)natural experiment in birth timing,”Journal of Public
Economics, 2009, 93, 246-263.
[12] Gelardi, A.M.G., “The influence of tax law changes on the
timing of mar- riages: a two-country analysis,”National Tax
Journal, 1996, 49 (1), 17-30.
[13] Kopczuk, W., and J. Slemrod, “Dying to save taxes: evidence
from estate-tax returns on the death elasticity,”The Review of
Economics and Statistics, 2003, 85 (2), 256-265.
[14] LaLumia, S., Sallee, J.M., and N. Turner, “New evidence on
taxes and the timing of birth,”American Economic Journal: Economic
Policy, 2015, 7 (2), 258-293.
22
[15] National Office of Statistics, “Frauen haben im- mer weniger
Kinder”, Data on Childbearing by Women Born in 1967-1971, URL:
https://www.demografie-
portal.de/SharedDocs/Informieren/DE/ZahlenFakten/Kinderzahl.html,
2016.
[16] Peters, H.E., “Marriage and divorce: informational constraints
and private contracting,”American Economic Review, 1986, 76 (3),
437-454.
[17] Scholz, C., “Empirische Untersuchung des Einflusses der
Ehegat- tenbesteuerung auf die Verteilung der Anzahl der
standesamtlichen Eheschlie- ungen in Deutschland und den Laendern
der EU 15”, Dissertation at the University of Illmenau (Germany),
2015.
[18] Sjoquist, D.L., and M.B. Walker, “The marriage tax and the
rate and timing of marriage,”National Tax Journal, 1995, 48 (4),
547-558.
[19] Statista, “Average Gross Income of Full Time Wage Earners
1991-
2017,”https://de.statista.com/statistik/daten/studie/237674/umfrage/durch-
schnittlicher-bruttomonatsverdienst-eines-arbeitnehmers-in-deutschland/,
2018.
23
Fraction of Weddings (Percent)
Year Total Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1990 516388 3.02 4.00 7.02 7.22 13.38 13.06 9.42 12.43 10.14 7.37
5.71 7.24 1991 454291 3.12 3.91 6.46 6.46 14.83 9.98 9.65 12.56
10.68 8.59 5.58 8.19 1992 453428 3.22 4.22 5.83 7.78 12.10 11.14
11.30 11.19 10.82 8.56 5.45 8.38 1993 442605 2.99 3.77 5.84 7.29
12.82 10.97 11.23 11.31 11.22 8.26 5.33 8.99 1994 440244 3.04 3.75
6.47 6.56 13.48 11.10 10.92 11.29 12.14 6.87 5.27 9.11 1995 430534
3.07 3.66 6.32 6.45 13.73 12.70 9.99 11.21 11.65 7.32 4.91 8.99
1996 427297 2.75 3.82 5.92 6.26 13.70 11.45 9.90 12.80 10.34 8.35
5.55 9.15 1997 422776 3.30 3.97 5.28 6.88 13.25 10.65 11.20 12.77
9.89 8.45 5.14 9.19 1998 417420 3.12 3.79 5.32 7.26 11.61 9.95
12.10 12.70 10.31 8.55 5.50 9.80 1999 430674 2.90 3.63 5.40 6.53
11.23 9.80 11.09 10.86 15.59 7.78 5.38 9.80 2000 418550 2.75 5.13
5.52 6.20 11.63 11.99 10.58 11.77 11.34 7.47 5.23 10.41 2001 389591
2.98 3.83 5.71 6.37 11.13 11.64 9.71 12.79 9.69 7.79 6.61 11.77
2002 391963 2.61 6.10 5.32 5.93 12.72 10.23 9.85 13.11 9.33 8.25
5.76 10.79 2003 382911 2.94 3.87 6.05 6.21 11.99 10.66 10.19 13.06
9.78 8.55 5.45 11.26 2004 395992 2.65 3.64 4.70 8.15 10.89 10.12
11.38 11.12 10.02 8.45 5.91 12.96 2005 388451 2.92 3.55 4.77 5.84
14.69 9.61 11.30 11.53 11.58 7.05 5.35 11.80 2006 373681 2.43 3.26
5.08 5.73 10.48 14.39 10.42 12.07 11.69 7.66 5.22 11.56 2007 368922
2.26 3.39 5.19 5.48 10.00 11.31 15.66 12.40 10.46 7.75 5.69 10.42
2008 377055 2.14 3.83 4.47 5.79 11.61 9.83 9.93 18.45 9.58 8.21
5.18 10.96 2009 378439 2.21 3.37 4.46 6.46 11.55 10.78 11.70 13.03
12.53 8.31 5.12 10.47 2010 382047 2.37 3.15 4.59 6.37 11.73 10.73
13.12 12.48 10.54 10.29 4.98 9.65 2011 377816 2.30 3.21 4.08 6.54
10.42 12.40 13.18 12.06 11.42 7.32 7.34 9.73 2012 387423 2.14 3.33
4.73 6.37 11.29 12.56 10.93 13.28 10.51 7.70 5.23 11.93 2013 373655
2.11 2.88 5.43 5.94 12.31 11.13 11.48 14.11 11.01 8.38 5.15 10.07
2014 385952 2.30 3.54 4.08 7.53 11.47 11.51 11.50 14.46 10.46 8.49
5.12 9.54 2015 400115 2.36 3.17 4.50 6.37 13.36 10.68 12.07 12.99
10.74 9.00 5.22 9.55
’Year’ shows the year of the wedding. ’Total’ shows the total
number of weddings in the respective year. Months ’Jan’ to ’Dec’
show the fraction of weddings per month in the respective year (in
percent).
Table 1: Wedding Month Distribution per Wedding Year (Wedding
Data)
24
N Mean StDev p5 p10 p25 p50 p75 p90 p95
Panel A: December
LMarriage 168257 4770.0 3241.7 1010 1400 2280 3914 6601 9342 11183
LLMarriage 168257 8.227 0.735 6.918 7.244 7.732 8.272 8.795 9.142
9.322 MAgeWed 168257 11803.2 3179.9 7757 8264 9443 11214 13522
16152 17921 FAgeWed 168257 10675.4 2931.8 6998 7425 8456 10119
12295 14711 16332 MAgeDiv 168257 16573.1 3709.2 10782 11835 13957
16416 18873 21423 23149 FAgeDiv 168257 15445.4 3594.9 9829 10758
12800 15403 17770 20024 21573 Children 168257 0.797 0.964 0 0 0 0 1
2 2 MForeign 168257 0.098 0.297 0 0 0 0 0 0 1 FForeign 168257 0.099
0.299 0 0 0 0 0 0 1
Panel B: January - November
LMarriage 1630809 5270.4 3411.1 1042 1475 2485 4568 7470 10017
11674 LLMarriage 1630809 8.328 0.748 6.949 7.296 7.818 8.427 8.919
9.212 9.365 MAgeWed 1630809 11094.8 2888.4 7659 8084 9039 10472
12456 14971 16800 FAgeWed 1630809 10051.1 2667.9 6933 7303 8137
9449 11265 13683 15419 MAgeDiv 1630809 16365.3 3598.8 10707 11717
13820 16270 18618 20973 22593 FAgeDiv 1630809 15321.5 3516.2 9793
10703 12714 15322 17621 19749 21225 Children 1630809 0.788 0.959 0
0 0 0 1 2 2 MForeign 1630809 0.104 0.305 0 0 0 0 0 1 1 FForeign
1630809 0.094 0.292 0 0 0 0 0 0 1
Panel C: All Months
LMarriage 1799066 5223.6 3398.8 1040 1467 2462 4498 7397 9964 11634
LLMarriage 1799066 8.319 0.747 6.947 7.291 7.809 8.411 8.909 9.207
9.362 MAgeWed 1799066 11161.1 2924.2 7668 8098 9069 10534 12562
15100 16928 FAgeWed 1799066 10109.5 2699.8 6938 7307 8161 9499
11365 13804 15525 MAgeDiv 1799066 16384.7 3609.8 10715 11727 13833
16284 18641 21014 22648 FAgeDiv 1799066 15333.1 3523.8 9796 10708
12722 15330 17634 19775 21259 Children 1799066 0.789 0.959 0 0 0 0
1 2 2 MForeign 1799066 0.103 0.304 0 0 0 0 0 1 1 FForeign 1799066
0.095 0.293 0 0 0 0 0 0 1
Table 2: Descriptives (Divorce Data)
25
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
DecWed -512.11 *** -503.13 *** -166.44 *** -137.16 *** -118.87 ***
-140.21 *** -140.47 *** -142.63 *** -123.60 *** -139.14 ***
(-61.04) (-60.27) (-21.27) (-17.72) (-15.35) (-18.20) (-18.16)
(-57.62) (-16.06) (-56.81)
NovWed -213.09 *** -161.77 *** -52.83 *** -43.98 *** -26.38 ***
-25.57 *** -42.94 *** -141.95 *** -7.11 -133.78 *** (-19.60)
(-15.12) (-5.31) (-4.49) (-2.70) (-2.62) (-4.39) (-46.54) (-0.73)
(-44.46)
JanWed 56.55 *** 205.91 *** 179.09 *** 166.32 *** 183.82 *** 202.44
*** 171.71 *** 103.13 *** 223.97 *** 115.08 *** (3.93) (14.64)
(13.80) (13.01) (14.41) (15.96) (13.46) (26.67) (17.71)
(30.17)
Children -227.65 *** -515.76 *** -514.97 *** -527.54 *** -510.13
*** 54.71 *** -520.44 *** 54.40 *** (-97.35) (-217.02) (-214.87)
(-221.49) (-214.46) (63.91) (-216.56) (63.87)
MAgeWed -0.256 *** -0.259 *** -0.259 *** -0.259 *** -0.257 ***
-0.004 *** -0.258 *** -0.004 *** (-260.65) (-262.34) (-262.63)
(-261.01) (-261.23) (-11.77) (-259.96) (-11.11)
FAgeWed -0.246 *** -0.282 *** -0.282 *** -0.279 *** -0.286 ***
0.001 *** -0.283 *** 0.001 ** (-237.65) (-262.52) (-261.95)
(-259.03) (-265.72) (2.65) (-261.72) (2.34)
MForeign -1060.89 *** -1247.37 *** -1249.15 *** -1243.78 ***
-1194.82 *** -130.08 *** (-142.36) (-169.94) (-165.01) (-161.86)
(-156.84) (-52.35)
FForeign -929.55 *** -717.72 *** -743.91 *** -741.13 *** -766.52
*** -34.51 *** (-121.41) (-96.36) (-96.65) (-94.67) (-99.56)
(-13.28)
Intercept 5282.08 *** 5650.16 *** 10773.00 *** 11579.00 ***
11498.90 *** 11194.10 *** 11381.80 *** 23746.50 *** 10759.90 ***
22260.70 *** (1876.21) (1360.32) (1034.96) (963.14) (228.70)
(32.88) (654.51) (56.67) (31.46) (51.44)
Fixed Effects Included
District NO NO NO NO YES NO NO NO YES YES MNation NO NO NO NO NO
YES NO NO YES YES FNation NO NO NO NO NO YES NO NO YES YES Litigant
NO NO NO NO NO NO YES NO YES YES YearWed NO NO NO NO NO NO NO YES
NO YES
SE robust robust robust robust robust robust robust robust robust
robust
N 1799066 1799066 1799066 1799066 1799066 1799066 1799066 1799066
1799066 1799066 Adj. R2 0.0021 0.0261 0.1674 0.1876 0.1908 0.1954
0.1900 0.9189 0.2007 0.9214 F-Stat 1335.47 8147.51 53423.70
48814.80 1097.75 7842.04 33219.20 1.3e+07 1031.79 2.1e+06
Endogenous variable is LMarriage in all models. T statistics in
parentheses (two-sided). * p<0.1, ** p<0.05, ***
p<0.01.
Table 3: Regression Results LMarriage
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
DecMar -0.105 *** -0.104 *** -0.036 *** -0.036 *** -0.032 ***
-0.037 *** -0.037 *** -0.038 *** -0.033 *** -0.036 *** (-55.31)
(-55.15) (-20.11) (-20.04) (-17.74) (-20.50) (-20.42) (-47.38)
(-18.43) (-45.87)
NovMar -0.052 *** -0.041 *** -0.019 *** -0.019 *** -0.015 ***
-0.016 *** -0.019 *** -0.037 *** -0.011 *** -0.035 *** (-21.71)
(-17.56) (-8.76) (-8.74) (-6.97) (-7.12) (-8.58) (-39.27) (-5.22)
(-37.14)
JanMar -0.005 * 0.028 *** 0.021 *** 0.021 *** 0.025 *** 0.028 ***
0.022 *** 0.024 *** 0.033 *** 0.027 *** (-1.66) (9.42) (7.48)
(7.46) (8.83) (10.22) (7.95) (20.23) (11.97) (23.09)
Children 0.050 *** -0.002 *** -0.001 *** -0.004 *** -0.001 0.048
*** -0.002 *** 0.048 *** (100.81) (-4.55) (-2.92) (-9.10) (-1.62)
(193.51) (-4.27) (195.38)
MAgeWed -0.0001 *** -0.0001 *** -0.0001 *** -0.0001 *** -0.0001 ***
0.0000 -0.0001 *** 0.0000 (-214.57) (-214.52) (-214.83) (-214.73)
(-213.33) (0.78) (-213.61) (-0.04)
FAgeWed 0.0000 *** 0.0000 *** 0.0000 *** 0.0000 *** 0.0000 ***
0.0000 *** 0.0000 *** 0.0000 *** (-179.43) (-176.22) (-175.98)
(-173.29) (-179.22) (27.12) (-176.04) (26.92)
MForeign -0.231 *** -0.271 *** -0.271 *** -0.271 *** -0.258 ***
-0.040 *** (-127.60) (-147.90) (-147.81) (-145.94) (-140.58)
(-45.71)
FForeign -0.178 *** -0.137 *** -0.137 *** -0.137 *** -0.142 ***
-0.008 *** (-95.77) (-73.55) (-73.54) (-72.67) (-76.20)
(-8.23)
Intercept 8.332 *** 8.331 *** 9.426 *** 9.430 *** 9.394 *** 9.413
*** 9.373 *** 10.044 *** 9.283 *** 9.719 *** (13481.16 ) (9450.26 )
(4182.25 ) (3704.65 ) (808.29 ) (111.68 ) (2457.43 ) (596.09 )
(109.87 ) (250.06 )
Fixed Effects Included
District NO NO NO NO YES NO NO NO YES YES MNation NO NO NO NO NO
YES NO NO YES YES FNation NO NO NO NO NO YES NO NO YES YES Litigant
NO NO NO NO NO NO YES NO YES YES YearWed NO NO NO NO NO NO NO YES
NO YES
SE robust robust robust robust robust robust robust robust robust
robust
N 1799066 1799066 1799066 1799066 1799066 1799066 1799066 1799066
1799066 1799066 Adj. R2 0.0018 0.0239 0.1390 0.1390 0.1425 0.1463
0.1415 0.8446 0.1520 0.8478 F-Stat 1119.72 8247.03 41060.10
36627.00 830.01 5964.46 25077.00 . 794.11 .
Endogenous variable is LLMarriage in all models. T statistics in
parentheses (two-sided). * p<0.1, ** p<0.05, ***
p<0.01.
Table 4: Regression Results LLMarriage
B Figures
Figures are included in the main body of the text. Below we show
alternative graphical analyses for a more in-depth documentation of
the data and results.
28
0
.00005
.0001
.00015
Length of Marriage (Days)
Figure 10: Replication of Figure 5 with Color-Coding
Figure 10 replicates Figure 5 with color-coding of kernel density
distributions per wedding month.
29
-400
-200
0
200
400
n
3-Day Intervals
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110
115 120
25%-Quartile 95% CI Significant
Figure 11: Replication of Figure 6 for the 25% Quartile
Figure 11 replicates Figure 6 for the 25% quartile. We observe that
general patterns of distribution around the annual 25% quartile
(dashed horizontal line), i.e. particularly December marriages (and
to some extent November marriages) have a significantly lower 25%
quartile than marriages in all other wedding months. However, we
note that the y-axis range is smaller than in Figure 6, indicating
that difference between months of wedding are relatively less
pronounced in the 25% quartile.
30
-1500
-1000
-500
0
500
n
3-Day Intervals
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110
115 120
75%-Quartile 95% CI Significant
Figure 12: Replication of Figure 6 for the 75% Quartile
Figure 12 replicates Figure 6 for the 75% quartile. We observe that
general patterns of distribution around the annual 75% quartile
(dashed horizontal line), i.e. particularly December marriages (and
to some extent November marriages) have a significantly lower 75%
quartile than marriages in all other wedding months. However, we
note that the y-axis range is larger than in Figure 6, indicating
that difference between months of wedding are relatively more
pronounced in the 75% quartile.
31