Till taxes keep us apart? The impact of the marriage tax on the marriage rate

Individuals pay both income and wealth taxes in Switzerland. Since wealth is per default considered to be the property of both spouses for married couples and couples in a registered partnership⁸, it is taxed jointly. Individuals pay income tax on their income from employment, wealth (movable and non-movable) and replacement income (pensions, unemployment insurance, etc.) (ESTV, 2018). In what follows, I will focus on the income tax from employment only, as wealth data are not easily available and the wealth tax burden is relatively small (Brülhart et al., 2022).⁹

Any individual living and working in Switzerland needs to pay taxes on their income. Income taxes are levied on three levels: the federal, cantonal and municipal level.¹⁰ The tax is paid in the canton and municipality of main residence, i.e., a person living in the city of Zurich and working in Basel would pay income tax to the federal government, the canton of Zurich and the city of Zurich, even though the place of employment is Basel. Income that is generated through usage of property or businesses, however, is taxed in the municipality and canton where the property or business is located (so-called limited taxation). The revenue from individual income taxation is the biggest source of tax revenue across all levels of government (2015: \(38.8\%\)), of which around 20% go to the federal government, 47% to the cantons and 33% to the municipalities (ESTV, 2018). Tax schedules differ across cantons, while within cantons tax rates also differ across municipalities. Cantons present a tax schedule each year from which a simple income tax is calculated depending on an individual’s taxable income. The final tax owed to the canton and municipality is then determined by applying the current cantonal and municipal tax rates, which are usually quoted in percentage or a multiple of the simple income tax. The federal direct income tax is determined in a similar fashion and the same across all cantons.

Swiss nationals and individuals with a long-term residence permit (C-permit) pay taxes by submitting a tax declaration in the calendar year after the respective tax year. They submit information on their income and wealth in the concerned year, but also have the option of making deductions for expenditures. The most common deductions are for work-related expenditures, commuting (by train or by car), children, donations and insurance payments. Most deductions are either fixed or capped. The tax office then issues a tax bill based on the tax declaration. Foreign residents with residence permits other than a C-permit are taxed at the source. Their tax contributions are directly deducted from their monthly paycheck and transferred to the resident canton. The size of the tax depends on their income level, marital status, age and municipality of residence. It is calculated using assumptions about the wealth of these residents, based on similar individuals who submit a tax declaration (ESTV, 2019). While the fashion of tax collection is differently for these residents, the size of their tax burden is still the same as for those who submit a tax declaration.

Switzerland follows the concept of joint (family) taxation. Married couples are taxed jointly for both their income and their wealth. In addition, any non-employment income generated by children under the age of 18 is also included in the joint taxation.¹¹

The federal government and the cantons aim to take the marriage tax into account by either applying different tax rates for married couples or splitting the income (ESTV, 2019).

Table 1 gives an overview of the procedure in each canton, where “Double” stands for separate tax rates for married couples, and “Splitting” stands for a splitting of the overall income, where the divisor represents the amount by which the overall income is divided to determine the relevant tax rate. Three cantons (Uri, Obwalden and Valais) apply the single tariff to the joint income of married couples, but allow for large deductions or a tax rebate on the final tax bill. For the income tax owed to the federal government, there are different tariffs for singles, married couples and families (in the form of an additional deduction from the tax). All cantons except for Vaud allow for child deductions. Depending on the canton, these deductions can be progressive in the number of children or not. Most cantons and the federal government have introduced an indexation of the tax tariffs, rebates and/or deductions based on the development of the inflation during the respective tax period (ESTV, 2018).

Nevertheless, a marriage tax can arise due to different reasons: In case of splitting factors lower than 2, it is evident that the taxable income is higher under joint than individual taxation. Similarly, this is the case for cantons applying a double tariff, where the two tariffs do not co-move. However, even in cases where the splitting factor is equal to two, it is possible that a marriage tax exists due to the nonlinearity of the Swiss tax system, as it is progressive in income. The progressivity of the tax code is the main driver of the marriage tax both in splitting- and double tariff systems. Table 2 presents examples of the marriage tax for different types of couples and income levels¹² for the three largest cities: Zurich, Lausanne and Geneva. A dual-earner¹³ couple with two children and high income in Geneva (Panel D), which employs a splitting factor of 2, pays considerably more when married compared to when unmarried, but the same couple with low income receives a marriage subsidy. This is due to the progressivity of the tax code. Conversely, single-earner couples without children tend to profit financially from marriage from an income tax perspective (Panel A).

Figures 1 and 2 show the variation in the marriage tax across Swiss municipalities for different types of hypothetical couples. The figures therefore are representative of the tax structure of the cantons and give a first impression of which types of couples tend to experience a marriage penalty or subsidy depending on their municipality of residence. In each figure I consider high- and low-income dual- and single-earner households, respectively. Figure 1 shows couples with two children, while Figure 2 assumes that households are childless. Single-earner households tend to experience a marriage subsidy, while dual-earner households are financially penalized by marriage. However, as can be seen in Panel (c) of both Figures 1 and 2, this is not necessarily true in all cantons. Those with a higher income tend to experience a higher marriage penalty than those with low income, especially when children are present and in dual-earner households (Panels (a) and (c) of Figure 1). Single-earner households with children, on the other hand, tend to always experience a small subsidy (Panels (b) and (d)).

Interestingly, the tax system leads to a much lower marriage tax for couples without children (Figure 2). In fact, the couples who enjoy the highest subsidies are childless couples where only one individual works (Panels (b) and (d)). This is likely due to the way single parents are treated in almost all cantons. A single person with a child gets very often treated the same as a married couple, i.e., a splitting factor is applied to their income or the married tariff is applied. Hence, the marriage tax as a percentage of net joint income is naturally smaller for these couples, as the difference between the tax burden when they are married compared to unmarried is smaller. Similarly, for dual-earner couples without children in the same household, the marriage penalty is relatively smaller than for dual-earner couples with children.

I obtain four different datasets from three different sources: The Swiss Federal Statistical Office (FSO), the Swiss Federal Tax Administration (ESTV) and the Central Compensations Office (CCO).

Population and Households Statistics (STATPOP). I obtain both the individual-level and household-level datasets from the Population and Households Statistics (STATPOP) of the FSO. Individual-level data are available since 2010 and are a stock dataset, containing demographic variables such as age, gender, marital status, residence status and parents on each individual living in Switzerland.¹⁴ It also contains a household identifier (ID), which allows me to obtain further information on who this person lives with by combining it with the household-level dataset (available since 2012). It is therefore possible to locate unmarried couples who are living together, but due to their marital status are not filing taxes jointly. Additionally, the household-level dataset contains information on the household age group and gender composition, as well as the number of children (Bundesamt für Statistik, STATPOP, 2021).

Income. I receive information on the gross income of each employed individual in Switzerland by combining the STATPOP dataset with the individual social security accounts (AHV) from the CCO. It includes the exact income based on which the social security contributions are calculated over the course of a year. Since contributions to the AHV are not topcoded in Switzerland, I can see the actual income and there is no distortion at the top. However, this only includes income through legal earnings from employment, i.e., income from capital is not included (Zentrale Ausgleichsstelle, 2021).

Tax data. Income tax data are on municipal and cantonal level and can be downloaded from the website of the ESTV. This dataset contains the tax schedule for each canton-year pair, as well as within each canton the tax rate of each municipality. It also includes the different tax tariffs depending on the income level, which can vary with marital status in cantons that apply different tax rates (the “Double” cantons in Table 1), as well as deductions which can be made from net joint income. I obtain the same information for the federal income tax (ESTV 2021a).

Municipal data. Municipal data are from the FSO and include both demographic (population density, sex ratio) and socio-economic (social assistance rate) variables on municipal level (FSO , 2021).

I merge the STATPOP and income datasets on an individual-year level such that each individual appears only once per year. I then add the remaining two datasets based on the municipal and cantonal identifiers, which are registered in the STATPOP data for each observation.

In order to estimate the impact of joint income taxation on the marriage rate, I need to identify two groups of individuals in the overall dataset: married and unmarried couples. Identification of married couples is simple enough, as the marriage status of each individual as well as the ID of their spouse is registered in the STATPOP dataset. I therefore include anyone in my analysis who is married during the sample period and who is living with their spouse. I identify unmarried couples living together by employing an algorithm which is described in detail in Appendix A.¹⁵ To treat both types of couples the same, I drop married couples if their age difference is more than 12 years.¹⁶ In addition, I limit my analysis to couples where both are in working age (18–65 for men and 18–64 for women) and drop those where at least one individual is self-employed. This is necessary, as the AHV income data are not reliable when it comes to the self-employed. Finally, I drop couples whose’ gross joint income is zero, as they do not pay any income tax.¹⁷ As the entire analysis is done at couple level, I reduce the dataset to one observation per couple-year pair and assign each couple an ID which depends on both individuals’ personal IDs.

Divorced and widowed individuals are not part of the married dataset, but they may show up in the unmarried dataset if they are living together with a new partner.

The income as reported from the CCO is the gross income of individuals, i.e., the income based on which the social security contributions are calculated. For taxes, however, net income is relevant. I therefore deduct from the gross income the social security contributions: old-age insurance, unemployment insurance and accident insurance. Switzerland’s pension system relies on a three-pillar approach such that in addition to the old age insurance, individuals also save through a pension fund. I subtract the mandatory contributions to the pension fund system, which are dependent on age and income level to reach the net income. As mentioned in Section 2.1, it is possible to further deduct certain expenditures from the net income for tax purposes. I apply the most common deductions, which are dependent on work status and/or observable without further knowledge of the couple¹⁸ to the net joint income before applying the tax calculator: deductions for children, secondary earner, work-related expenditures, as well as personal deductions based on marital status. The latter two are calculated as a percentage of income with minimum and maximum values in most cantons and on the federal level. As deductions for children above the age of 18 are dependent on the child’s work status (i.e., is she still in school/university or already earning her own income) in most cantons, I limit the child deductions to children below the age of 18 living in the same household. While this might overestimate the taxable income in some cases, this should not influence the results too much as the magnitude of these deductions is limited and generally not very large for adult children.

To analyze the effect of the marriage tax, I calculate for each couple the tax burden under joint taxation and under individual taxation. I do this by applying the tax code and rate of the respective year in the couple’s canton and municipality of residence, as well as the respective tax code and rate of the federal income tax to the taxable income, i.e., net income after deductions for expenditures.¹⁹ The couple then faces a marriage tax, \(mtax\), which is the difference between the tax burden under joint and individual taxation as a percentage of net joint income:where \({\text {tax}_{\text {individual}}=\text {tax}^{\text {female}} + \text {tax}^{\text {male}}}\) is the sum of the tax burden based on each individual’s separate taxable income.²⁰\({\text {tax}_{\text {joint}}}\) is the tax burden based on the joint taxable income, and \({Y_{\text {joint}}}\) is the couple’s net joint income. A negative value of \({\text {mtax}}\) indicates a marriage subsidy, and a positive value a marriage penalty.

In what follows, I briefly describe the current state of marriage in Switzerland in general and the marriage tax in particular. This analysis is purely descriptive as the results arise both due to the tax system and due to the composition of couples in the sample, as all results are unweighted.

Figure 3 shows that while the marriage rate has decreased by roughly 4.5pp over the analyzed time period (Panel (a)), the average marriage tax has risen up until 2018 (Panel (b)). The drop in the average marriage tax in 2019 is likely due to a simultaneous drop in the share of couples where both individuals work and therefore a reduction in the average net joint income²¹, which is in line with a relative drop of the average marriage tax for single-earner households in 2019 (Panel (f)). Panel (c) shows that while couples from the first and second income quintile²² mostly experience a marriage subsidy, those from the middle to upper half of the income distribution face a marriage penalty.

Households with children on average face a marriage penalty, while those without children face a marriage subsidy (Panel (e)) and so do single-earner households (Panel (f)). The findings for hypothetical couples in Section 2.2 that single-earner and childless households tend to profit from marriage are therefore confirmed when looking at the data. Finally, the average marriage tax for unmarried and married couples has converged over the analyzed time period, as it has risen more for the married than the unmarried couples (Panel (d)).

Panel (a) in Figure 4 shows that while they develop similarly over time, the average marriage tax is roughly 0.5pp higher in cantons with an income splitting system than those which apply a double tariff. The marriage tax has remained the same in most cantons, with the exception of a large reduction in the canton of Jura (JU) due to a change in the applicable child deductions (Panel (b)).

Panels (c–d) show the average marriage tax and marriage rate by within household income distribution, where households are clustered in five distribution bins²³ with the most equal distribution on the right and the most unequal distribution on the left of the graph. While the average marriage tax rises the more equal the within-household income distribution, the marriage rate declines.

As previously mentioned, the Swiss tax system depends heavily on the specific cantonal tax schedules in general, and the applied tax rates in the municipalities, in particular. As a result, there is considerable variation in the marriage tax couples face across municipalities. Figure 5 presents the average marriage tax paid by married couples per municipality in Switzerland in the form of a heatmap. It is evident that there is quite some variation across municipalities in the magnitude and direction of the average marriage tax. While it is small in most municipalities, with the majority of municipalities showing an average marriage tax of \(+/- 0.5\%\), it can be as low as \(-5.78\%\) and as high as \(2.98\%\). Note that this is a purely descriptive result, meaning that the differences may arise due to compositional effects of certain types of couples choosing to move into specific municipalities in addition to the differences in the tax schedules and rates.

Recall from Section 3.2 the definition of the marriage tax:where a negative value of \(\text {mtax}\) indicates a marriage subsidy, and a positive value a marriage penalty. The straightforward approach to analyzing the impact of the marriage tax on the marriage rate is to estimate the following linear probability model:where \(y_{\text {ict}}\) is a binary outcome variable for couple²⁴\(i\) living in canton \(c\) at time \(t\), indicating whether a couple is married or not. The coefficient of interest is \(\beta\), which shows the effect of the marriage tax on the marriage rate. I include a vector of couple-level controls \({X_{\text {ict}}}\) which could potentially affect the marriage decision: age and age squared (per person), age difference, net income and net income squared, nationality, years in municipality, and number of children in the household. \(Z_{\text {mt}}\) is a vector of municipal-level controls that can influence the local marriage and labor market: population density, sex ratio²⁵ and social assistance rate. \(\alpha _i\), \(\zeta _{c}\) and \(\eta _{t}\) are couple, canton and time fixed effects, respectively. \(\varepsilon _{\text {ict}}\) denotes the error term.

As Fisher (2013) points out, it is possible that both the marriage tax a couple faces and the decision to get married are jointly influenced by unobserved characteristics. For example, preferences for marriage and education may be correlated, but so may be education and within-household income distribution, which directly affects the size of the marriage tax. However, education may change over time and is therefore not necessarily captured by fixed effects. Another bias may arise due to measurement error: while income should be correctly reported in the administrative dataset, I cannot rule out measurement error in the size of the marriage tax, as it is estimated based on household characteristics. The calculation for the taxable income, for example, is only using deductions that can be observed from the data (see Section 3.2 for a discussion on the limitations of the tax burden estimation). Equation (2) therefore suffers from omitted variable- and measurement error bias. In order to eliminate this bias, I use a simulated instrumental variable approach to estimate the causal effect. The underlying idea is to use the institutional characteristics which are a strong predictor of the size of the endogenous regressor (here tax schedule in canton of residence) to create an exogenous variable which does not suffer from this omitted variable bias.²⁶ Specifically, I estimate the following first stage:where the instrument is the average marriage tax per canton as predicted by the tax schedule, \(\text {avg}(\text {mtax}_{\text {ct}})\). I simulate the marriage tax a random sample of 1 000 couples²⁷ faces in each canton-year pair by applying the respective tax code for each hypothetical couple in each canton and year.²⁸ I use the cantonal tax rate and the population-weighted average municipal tax rate per canton and year to estimate the tax burden under joint- and individual taxation per hypothetical couple. I then average the marriage tax as a percentage of net joint income over each canton-year pair²⁹ and merge this as an instrument to each observation in the initial dataset on a yearly basis. See Appendix A for a detailed discussion of the instrument construction and validity, including a comparison to Borusyak & Hull (2021) and the grouping IV.

To estimate the impact of the marriage tax on the marriage rate at the time of marriage, I only include couples who get married during the sample period, i.e., between 2012 and 2019.³⁰ This circumvents a potential downward bias due to the inclusion of couples who were married decades before my sample period, when other (now unobserved) factors may have influenced the decision to get married and the marriage tax may have been very different. The same is true for couples who never get married, as it is not clear whether they remain together after my sample period. The best sample would likely be the baseline sample in addition to those who as of 2019 are not yet married but will get married in the future. As it is not possible to observe this group, I also present results when adding those who remain unmarried in 2019 to those who get married after 2012 as a robustness check in Section 4.3.³¹ The true effect is likely somewhere in between the baseline effect and the effect when including the not yet married.

Table 3 presents descriptive statistics for the baseline sample both by marriage status and in total. The marriage rate within the sample is \(81.5\%\) and the average marriage tax is 1446 CHF, or \(0.64\%\) of net joint income. This value is slightly larger for couples which are actually married than those who are still unmarried. Both men and women in unmarried couples are more likely to work and be Swiss nationals than in married couples, and they also tend to have a larger income, driven by a larger female income. In addition unmarried couples tend to have fewer children and be younger than their married counterparts. \(61.1\%\) of couples do not have any children when they get married. The municipal-level control variables are similar across the two types of couples.

The marriage tax in Switzerland on average presents itself as a marriage penalty, while it tends to be a marriage subsidy in other countries such as the USA and Germany (Fink, 2020; Friedberg & Isaac, 2022). However, when considering the absolute value of the marriage tax, estimates for Switzerland are well within the range of other countries: The average marriage subsidy in Germany according to Fink (2020) is 893 EUR, or \(2.67\%\) of net joint income, which is around four times larger than the average marriage penalty in Switzerland. Additionally, the marriage tax for both unmarried and married couples is roughly two times larger in Switzerland than in the USA (Friedberg & Isaac, 2022): US married and unmarried couples enjoy an average marriage subsidy of 442 USD (\(0.35\%\) of net joint income) and 264 USD (\(0.25\%\)), while their Swiss equivalents experience an average marriage penalty of 1493 CHF (\(0.67\%\)) and 1234 CHF (\(0.51\%\)), respectively.

I conduct a heterogeneity analysis based on a couple’s overall income level, within-household income distribution and type of couple (age, child status). Section 2.2 shows that there are differences in the tax regimes across cantons, which could be mirrored in the impact of the marriage tax on the marriage rate. I therefore estimate equations (2) and (3) by taking into account these different types of joint taxation through interaction terms. Finally, couples may experience an increase in the marriage tax in different ways depending on whether it presents in the form of a marriage penalty or a marriage subsidy. An increase in the marriage tax for couples facing a marriage subsidy is likely to be experienced as a reduction in a subsidy, i.e., having less than previously, rather than an increase in a penalty. The theory of loss aversion shows that individuals react differently to loosing a privilege than they do to gaining something they did not have, which may play a role in the marriage tax. For example, Benzarti et al. (2020) show that the pass-through of value added tax (VAT) changes is asymmetric. The behavioral response to an increase in the VAT is much larger than to a decrease of the same size. I therefore also present results when considering the type of marriage tax, i.e., whether couples (would) pay a penalty or receive a subsidy.

Table 4 presents the results from estimating equations (2) and (3) using simulated instruments. The sample only contains those couples who are getting married during the sample period, i.e., between 2012 and 2019. I therefore analyze the effect of the marriage tax on the marriage rate at the time of marriage.

I present results with alternative samples in Section 4.3 and the first stage and reduced form results in Appendix B. The marriage tax is measured in percent of net joint income. A 1pp higher marriage tax decreases the probability that a couple is married in a given year by \(13.7\%\) (column (5)). As all couples in the sample are getting married at some point during the sample period, one can also think of this effect as a delay in marriage. The control variables influence the probability to be married as expected: age, children, income and the sex ratio in a municipality positively influence the probability to get married, while couples where at least one has Swiss nationality are less likely to get married. Columns (4) and (5) show that it is imperative to consider cantonal- and time variation by including canton and time fixed effects, respectively. Column (6) presents the reduced form regression of the marriage rate on the simulated marriage tax. Together with the first stage results (cf. Appendix B), it confirms the importance of using an instrumental variable approach and the strength of the instrument.

Section 2.2 shows that the marriage tax can vary considerably both across type of couple and geographically. This implies that couples may react differently to the marriage tax when it comes to the decision to get married depending on their characteristics. I therefore present the differential impact by couple- and geographical characteristics in Figure 6. It plots the point estimate and the \(95\%\) confidence interval of the interaction coefficient between the marriage tax and different indicators: income quintile, child status, age group, within household income distributions, type of joint taxation and type of marriage tax. All controls and fixed effects from the baseline estimation in column (5) of Table 4, as well as the base effects of the indicators are included. The coefficients in Figure 6 therefore show the impact of the marriage tax by subgroup. The average marriage rates per subgroup are displayed in Table 7 in Appendix A.

When considering couples from different income quintiles, we can see that those in the second to fifth income quintile react very similarly to a higher marriage tax. Their probability to be married is \(7.3-10.0\%\) lower when the marriage tax is 1pp higher, while those in the first income quintile are \(28.3\%\) less likely to be married. Couples from the lowest income quintile are likely to have a tighter budget constraint both due to lower income and wealth levels. They are also likely to consume a larger share of their income compared to couples from higher-income quintiles, and therefore save less. Put together, this indicates a higher sensitivity to both taxes in general, and to a higher marriage tax in particular for this group, as a higher marriage tax indicates a stronger reduction in consumption than for couples from higher-income quintiles.

As shown in sections 2.2 and 3.3, the tax system is constructed in a way such that couples without children in the same household tend to experience a marriage subsidy. As evident in Figure 6, this group also reacts much stronger to a higher marriage tax, being \(22.3\%\) less likely married when the marriage tax increases by 1pp. In fact, the probability that a couple with children is married is only reduced by very little when the marriage tax increases, indicating that other factors than the marriage tax are much more important in this group when deciding to be married or not. One possibility here may be that once children are present, the insurance character of marriage outweighs the potential negative effects due to a higher tax burden. For example, if a couple has children it may have a different intra-household labor supply and task distribution than without children as one individual stays home more than the other. Marriage can then act as an insurance mechanism against adverse income shocks in the case of the death of the main earner, both in terms of public and private survivor’s insurance.³²

While there is little overall variation across different age groups, those in the youngest (18–30 years old) and oldest (51–64 years old) age groups react stronger to a higher marriage tax, although the difference is not statistically significant. These couples are likely to be similar in their characteristics as the couples without children in the same household. Interestingly, the analysis by within household income distribution also shows that those who should in theory profit the most—couples with the most unequal distributions—are those who react the strongest to a change in the marriage tax. It points to the possibility that a reduction in the size of the marriage subsidy, i.e., a reduction in the amount a couple can save with the marriage, is a stronger incentive than the punishment, i.e., a higher marriage penalty. This is confirmed when looking at the last group of estimates, which divides the impact by whether couples face a marriage penalty or subsidy: couples who would profit from getting married react much more strongly to an increase in the marriage tax (decrease in the subsidy they receive). The impact of the marriage tax is therefore asymmetric in the type of marriage tax, as couples are shown to be loss-averse.

I expect individuals to be more aware of the marriage tax in cantons that apply income splitting than those that apply a double tariff, i.e., that the marriage tax is more salient in the former cantons. It should be easier for individuals to estimate the effect of getting married on their tax burden if they simply have to add up their income, split it by the divisor and then apply the same tax rate as if they were single, than if they have to apply a completely separate tariff. However, Figure 6 shows that the type of joint taxation does not matter much. In fact, couples living in cantons with an income splitting approach are not significantly less likely to get married when their marriage tax increases. The results are driven by those living in double tariff cantons, who are \(5.9\%\) less likely to be married when the marriage tax increases. This implies that the salience of the marriage tax does not vary much across type of joint taxation, i.e., that couples living in double tariff or splitting cantons are roughly similarly aware of the effect of marriage on the tax burden.

I present robustness checks in Table 5 where I vary the sample used for the estimation. Column (1) present the baseline results from Table 4. I additionally include all unmarried couples in column (2). This sample therefore also includes couples who as of 2019 still remain unmarried. Column (3) presents the results when using the full sample, i.e., everyone in the sample who is married irrespective of marriage date, as well as the unmarried.

As expected, the results show the same sign with smaller magnitude the more couples are included. The impact of the marriage tax on the marriage tax decreases by roughly 11.5pp when moving from the baseline sample in column (1) to the full sample in column (3). This is likely due to the fact that including couples who got married a long time ago biases the results downward as other factors may have played a role when these couples made the decision to get married. Additionally, their income may have changed considerably since then, and with it the marriage tax they face. These results, however, confirm the baseline results and underline the importance of measuring the marriage tax at the time when the marriage decision is made. Column (4) only includes couples who submit a tax declaration, i.e., Swiss nationals and foreigners with a C-permit. The magnitude of the effect is around \(40\%\) lower than the baseline effect, but the sign remains the same.

In Section 4.1 I discussed the question of which sample is the “true” sample. The baseline sample excludes relevant individuals from the analysis (those who are planning on getting married but have not done so as of 2019), while the Baseline with Never Married sample includes irrelevant individuals (those who will grow apart after 2019 and never get married). The coefficient of column (1) in Table 5 likely includes an upward bias, while the coefficient in column (2) is downward biased. The “true” effect of a 1pp higher marriage tax in the marriage rate is therefore somewhere in between \(-13.7\%\) and \(-8.2\%\).

A couple can avoid a higher tax burden after marriage either through a change in income (for example through a change in labor supply) or by moving to an area with a lower marriage tax. In fact, Panel (a) in Figure 7 shows that the share of couples in which both individuals work is stable leading up to marriage, but declines gradually thereafter, indicating a potential avoidance of the marriage tax through a reduction in labor supply rather than a delay of marriage.³³ Similarly, even though the share of couples moving to a new municipality is low on average, there is a slight increase in the years leading up to the marriage. Most couples move in the year after the marriage (Panel (b)). To address these two concerns, I show the estimation results when limiting the sample to couples who did not change either their work status (columns (1—3)) or their residence (columns (4–5)) between 2012 and 2019 in Table 6. The impact when only looking at couples who did not change the extensive margin of their labor supply is similar to the baseline result (column (1)). Moreover, the reduced marriage rate due to the marriage tax seems to be driven by those couples where always both individuals are working over the entire sample period (column (2)), while those where only one individual is working are not significantly less likely to marry (column (3)). This result is to be expected, considering that the marriage tax is stronger for double earner couples than for single-earner couples.

Column (4) of Table 6 shows the results conditional on either individual of a couple living in the municipality already before 2012. Column (5) additional conditions on both individuals living in said municipality before 2012. These results therefore exclude couples who move as a result of the marriage tax. While slightly smaller in magnitude, the results are in line with the baseline results.³⁴

One concern may be that only sampling couples from the 2012 dataset is not a suitable choice for an instrument, as the couples may change over time. I therefore estimate the baseline IV regressions using different versions of the instruments, where for each version I draw the random sample from the data of a different year. Figure 8 plots the point estimate and the \(95\%\) confidence interval for each version. The estimates are very similar to each other, with all results being significantly different from zero.