’Will You Marry Mein December?’ Tax-Induced Wedding Date

—
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 marriage 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 exemption 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) schedule. 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 assessment 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 benefit 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 marriage 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 population 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 descriptive 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 controls 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 variable 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 controls, 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 naturally 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).
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 investigation 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 difference between MAgeWed and FAgeWed is also larger in December marriages relative to marriages with weddings in all remaining months. Corresponding differences are naturally also found with MAgeDiv and FAgeDiv. The distribution
10
of all the age related variables appears symmetric around the mean (median), albeit 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)
Month 3 (N = 111306)
Month 4 (N = 122594)
Month 5 (N = 204433)
Month 6 (N = 187764)
Month 7 (N = 195844)
Month 8 (N = 224153)
Month 9 (N = 185686)
Month 10 (N = 143653)
Month 11 (N = 106616)
Month 12 (N = 168257)
Median Mean Kernel Density
Month 1 (N = 65816)
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 correspondingly 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 corresponding 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 wedding 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 systematic 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 significantly 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’ distributions are shown in grey.
0
.00005
.0001
.00015
.0002
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 observations are significantly lower than the annual median. Considering particularly observations near the year end threshold, we observe that of the 20 last observations 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 significantly 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 particularly interested in the dichotomous variable DecWed which indicates marriages with weddings that occurred in December.
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 marriages 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 indicates 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 variable, 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.
We find similar effects in mathematical sign and magnitude as in Table 3, indicating 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, respectively (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 relatively 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 structure. 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 marriages 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 available 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 difference 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 similar to our baseline specification results without Y earWed in column (4) and (9). Standard errors are robust to heteroskedasticity in all specifications. However, neither 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 magnitude of the effect for the more recent (earlier) years.
8 Conclusion
We show that December marriages last considerably shorter than does the average (median) marriage. Specifically the average (median) difference (conditional on divorce) between December marriages and marriages with weddings in all remaining 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 marriages: 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
Panel C: All Months
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
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

Documents

’Will You Marry Mein December?’ Tax-Induced Wedding Date