Two Contrasting Reforms

Both the Netherlands and Sweden have historically practiced restrictive policies on dual citizenship and moved toward liberalization in recent decades. Sweden fully liberalized its dual citizenship policy in July 2001, while the Netherlands did so already in the early 1990s, halfheartedly, and with a subsequent reversal in June 1997. In this paper, we study the effect of the canonical liberalization in Sweden (in the sense of being an early example of contemporary dual citizenship acceptance in Europe) and the atypical reintroduction of the renunciation requirement in the Netherlands (including an exemption for foreigners with a Dutch citizen partner, further details in Supplementary Material SM3).

These two countries are well placed to compare the effect of these changes in the renunciation requirement on immigrant naturalization rates for three reasons. First, the citizenship policies in both countries around the respective reforms are among the most inclusive in Western Europe with a residence requirement of five years, no formalized language or civic integration tests, and no economic conditions (see Supplementary Material SM1, cf. Schmid Reference Schmid2021). This shared inclusiveness provides a comparable institutional context where changing dual citizenship requirements are likely to have the largest marginal effect on migrant naturalization decision-making. Note that we restrict our observation period to exclude potential confounding by subsequent restrictive changes which were introduced in the Netherlands in 2003, see “Data”). Second, both countries are advanced economies where citizenship comes with access to highly developed welfare regimes, free movement within the European Union, and facilitated global mobility (Henley & Partners 2023). This provides a setting where demand for citizenship should be comparable. Third, the availability of detailed micro-level administrative data allow us to compare the before-after differences around the reform years in the two countries across complete migrant populations.

Identification Strategy

We apply a DiD strategy to identify the effect of dual citizenship policy reform on naturalization rates among immigrants, by comparing rates among migrants from the treatment group, in the years before and after the respective reforms in Sweden and the Netherlands, compared to the before/after naturalization rates among migrants from the control group. The main intuition of our approach is that whether an immigrant is affected by dual citizenship reform in the destination country depends crucially on the type of citizenship legislation in the origin country. Given the dyadic nature of dual citizenship, if the origin country does not allow migrants to keep their citizenship when naturalizing abroad, the destination country reform is irrelevant to migrants from such countries. Since migrants’ origin country is exogenous to their individual motivation to naturalize, they cannot self-select into either the treatment or the control group; hence treatment allocation is plausibly exogenous, allowing a causal interpretation of the statistical findings.

Based on the citizenship legislation of the origin country, we allocate migrants to one of three groups:

• • Control group: migrants whose country of origin does not allow them to keep their country-of-origin citizenship if they naturalize in the destination country.

• • Treatment group: migrants whose country of origin allows them to keep their country-of-origin citizenship if they naturalize in another country but does allow them to renounce their country-of-origin citizenship if this is required by the destination country.

• • Placebo group: migrants whose country of origin does not allow them to renounce their country-of-origin citizenship.

Figure 1 visualizes our treatment identification strategy.

Figure 1.

Identification Strategy: Treatment Group Allocation

Hence, only a subset of migrants are affected by destination country dual citizenship reform: those who (a) do not lose their citizenship due to restrictive country-of-origin legislation, and (b) can renounce their country-of-origin citizenship if this is required by the destination country (treatment group). Only these migrants were (Sweden, until the liberal reform of 2001) or are currently required (the Netherlands, since the restrictive reform of 1997), after having acquired the destination country citizenship, to demonstrate that they renounced their origin country citizenship. If migrants lose their origin country citizenship automatically by effect of the origin country law (control group) or if they cannot renounce it even if they wanted (placebo group), the destination country authorities did/do not require them to renounce it upon naturalization (see Supplementary Material SM3 for details on the implementation of the renunciation requirement in both countries).

Note that in 1960 around 74% of origin countries practiced a restrictive approach to dual citizenship whereas only 31% of countries did so in 2022; a smaller, but still significant share of around 7% of countries does not allow its citizens to renounce their citizenship (Vink et al. Reference Vink, Van der Baaren, Baubock, Dzankic, Honohan and Manby2023). Those migrants that do not lose and can renounce the origin citizenship, but who are not required to do so due to destination country exemptions (as in the Netherlands, for migrants with a citizen partner) serve as an alternative placebo group, as illustrated in Figure 1.

Data

We draw on administrative register data from Statistics Sweden and Statistics Netherlands. These data include all legally registered individuals in these countries over time and are accessed through remote access procedures (see “Data availability statement”). Hence the analyses of both samples are run separately.

The research population consists of all foreign-born residents registered between 1998 and 2006 in Sweden and between 1994 and 2002 in the Netherlands, that is, eight years surrounding the dual citizenship reform in both countries. We do so in light of two data constraints in the Dutch sample: (a) we have administrative data available from 1994 onward; and (b) in 2003 the Netherlands introduced mandatory civic integration, which risks confounding the effect of dual citizenship restriction from that year. We only include foreign-born individuals both of whose parents were born abroad and migrants who are 18 years or older at the moment of migration. To exclude major post-WWII return migration flows with, especially in the Netherlands, very high citizenship acquisition rates we include only those who resided less than 40 years in the host country. We observe migrants once they are eligible for naturalization (see Supplementary Material SM3). As in both countries the renunciation requirement during the restrictive observation periods does not apply to migrants who have refugee status or are entitled to subsidiary protection, in line with international law, we exclude those migrants with a high probability to be exempted on asylum grounds (see Supplementary Material SM2). In Sweden, the sample consists of 484,008 migrants (3,191,258 observations). In the Netherlands, due to the partner exemption to the reintroduced renunciation requirement, we restrict the main analyses to migrants without a citizen partner (N = 136,521 individuals, 912,031 observations); we employ a sample based on migrants with a citizen partner in a placebo analysis (N = 148,511 individuals, 931,413 observations).

We subsequently link individual-level register data to origin country-level information on dual citizenship acceptance (Vink, De Groot, and Luk Reference Vink, De Groot and Luk2015). This dataset charts the rules of 200 countries over time with regard to the loss or renunciation of citizenship after a citizen voluntarily acquires the citizenship of another state (see details and allocation of migrants across treatment groups in Supplementary Material SM4).

Method

Since dual citizenship reforms do not affect all migrant groups at the same time, as also origin country policies change over time, our empirical strategy needs to deal with such staggered treatment. This problem has been overlooked so far in the literature on immigrant naturalization. In particular, while the destination country reform takes place in a single year, the timing of origin country policy change varies between migrants from different countries. As treatment depends on the policy conditions in both the origin and host country, some migrants receive delayed treatment.

To incorporate staggered treatment in a robust manner we draw on the DiD model developed by Callaway and Sant’Anna (Reference Callaway and Sant’Anna2021), which generalizes the canonical two-way fixed effects model by estimating the group-time average treatment effect reflecting the average treatment effect for group $ g $ at time $ t $ . The “groups” in our case represent units who, due to the citizenship policies of their origin countries, were first treated in a given year (see Supplementary Material SM5 for the trend in naturalization rates among these staggered treatment groups). In Sweden, this varies from 2001 (when the renunciation requirement was abolished) to the subsequent years 2002–2005 (there is no migrant group that becomes treated in 2006). In the Netherlands, there is only one group because there are no origin countries which abolished their “cannot renounce” provisions between 1998 and 2002 (which would cause migrants who were not affected by the restriction of 1997 to be affected later).

Under this specification, the conventional DiD design can be expressed as:

(1) $$ \mathrm{ATT}\left(g,t\right)=E\left[{Y}_t(1) - {Y}_t(0)\ |\ {G}_g=1\right], $$

where the average treatment effect $ \mathrm{ATT} $ for each group-time combination ( $ g $ , $ t $ ), is the difference in the naturalization rate as a result of the respective policy reform between migrants from treatment groups $ g $ and never treated migrants at every time $ t $ . The $ \mathrm{AT}{\mathrm{T}}_{\left(g,t\right)} $ allows for the estimation of heterogeneous treatment effects, that is, treatment effects that increase or decrease over time, or treatment effects that differ based on adoption time. The effect on the overall ATT is weighted as a function of the time that each unit spends in the pre-period and post-period (details in Supplementary Material SM6). We interpret the ATT as the average difference in the naturalization rate among migrants from the treatment group as a result of the removal or reintroduction of the renunciation requirement, all else constant. We report both the average treatment effect, as well as the dynamic event study to test whether the pre-reform trends meet the parallel trend assumption that the difference in naturalization rates between control and treatment group is constant in the absence of treatment.