The inconsistency of immigration policy: the limits of “Top-down” approaches

ABSTRACT

To what extent can we infer government objectives from policies on paper? We show that this assumption in migration scholarship is problematic because most states adopt immigration policies that are inconsistent, combining or alternating between contradictory objectives. Further, we develop a measure to track how immigration policy inconsistency varies over time. We use these methods to demonstrate that some of the main theories of policy inconsistency, which focus on variables located at the national scale, find limited empirical support. Based on these findings, we make the case for further research into the local scale of politics, focusing on the agency of street-level bureaucrats and migrants. We then discuss the potential for crossing quantitative and qualitative divides in order to further explore the impact of local factors on national immigration policies.

KEYWORDS:

• Migration policy
• Inconsistency
• State capacity
• Mixed methods
• Micro-politics
• Scale

Introduction

How do we know how freely people can move abroad? Migration scholars have taken written national immigration policies – which include policies on entry, exit, integration and return – as key indicators of how much and in what way states aim to restrict migrant rights and freedoms, such as the right to social protection and the right to reside. In this article, we argue that in fact, most written immigration policies are inconsistent, combining or alternating between measures to make policy more restrictive and measures that make policy less restrictive. We develop this claim through a critical engagement with quantitative studies of migration policy. Quantitative research on national immigration policies has significantly enhanced our ability to explore and test theories about the politics of migration (Beine et al. 2016; De Haas, Natter, and Vezzoli 2018; Helbling et al. 2017). This research has coded immigration policies on the extent to which different national governments aim to restrict immigration. They make sense of their data through a “top-down” model of migration politics. This model consists of two main assumptions: (a) immigration policies reveal the objectives of states or national elites; and (b) the behaviour of ordinary government officials are primarily efforts to implement these policies. In this paper, we show how these assumptions limit researchers’ ability to make sense of one of the main patterns in their data: the inconsistency of written immigration policies.

“Inconsistent immigration policies” combine or alternate between (a) measures to make immigration policies more restrictive; and (b) measures to make immigration policies less restrictive. To clarify, a law that introduces new penalties on unauthorized migrants but also features a provision that offers unauthorized migrants an amnesty is “inconsistent” because it combines restrictive and non-restrictive measures. Alternatively, a pair of laws which begins with a law that introduces new penalties on unauthorized migrants but is soon followed by another law offering unauthorized migrants an amnesty, may be deemed “inconsistent” because this pair of laws alternates between restrictive and non-restrictive measures.

To make sense of this data, quantitative studies have tended to (a) ignore evidence of inconsistency in written immigration policies; (b) disaggregate the data in ways that minimize its incidence; or (c) explain its incidence as an indirect outcome of policy objectives. Our study demonstrates the limitations of these moves, and provides an alternate approach. We begin by revealing that there is simply too much inconsistency in written immigration policies to ignore. More specifically, using data produced by the Determinants of International Migration (DEMIG) team, we demonstrate that a significant majority of immigration policy changes are inconsistent: occurring in the same country and same year as a policy with a conflicting objective (De Haas, Natter, and Vezzoli 2018). We then demonstrate that inconsistency can be found within policies designated for specific types of migrant (refugee, labor, family etc.) and policies designed for specific types of migration policy (border control, visas, integration etc.).

What explains this inconsistency? Some explanations of inconsistency focus on the national scale.¹ National scale explanations may focus on electoral-cycles where a new ruling party passes policies that conflict with the purposes of the old or “grand bargains” where “pro-migrant” and “anti-migration” lobby groups develop laws with some policies that each group prefers. Others focus on the local scale, suggesting that policies may instead be better understood as indicators of the agency of street-level officials and migrants who help to generate inconsistencies in written policy from the “bottom-up”. Our analyses offer only limited support for explanations of policy inconsistency located at the national scale. Unfortunately, existing data sets do not provide us with the information we need to fully explore or test theories located at the local scale. However, we argue that a third approach would integrate these two scales: the local scale is not only a site where national objectives are “implemented” – or not, but can also produce dynamics that ultimately shape policies and inconsistencies at the national scale. We conclude with the suggestion that we need to develop the methods and data required to explore the local scale as an independent political scale, and one with the potential to help explain the phenomenon of immigration policy inconsistency.

The paper develops this case in five steps. First, we critically analyze the assumptions of several prominent quantitative studies of immigration policies. Second, we survey a broader literature on migration politics, to identify alternate explanations of policy inconsistency. Third, we introduce the DEMIG data set and our techniques for measuring variation in policy inconsistency. Fourth, we present our findings and analysis. Our analyses reveal the limitations of conventional assumptions, while suggesting how to focus empirical inquiry into these theories. These analyses also suggest that there remains a great deal of unexplained policy inconsistency, which we take as grounds for research that pays more attention to the local scale. Fifth, we indicate what it might take to develop more compelling accounts of immigration policy inconsistency and re-situate the national and the local in the study of migration politics.

The top-down assumptions in quantitative research on immigration policy

There is an emerging stream of quantitative research into immigration policy. The main empirical aim of this research has been to understand variation in the degree to which national governments restrict migration from abroad. The main theoretical aims of this research have been to learn what factors determine whether a given government is more or less restrictive towards migrants, how restrictive different governments are at different moments in time, and how attempts to make immigration policies more or less restrictive impact migrant decision-making and migration flows.² The datasets compiled by these studies contain information indicating whether a given country’s written immigration policies are more or less restrictive (by asking, e.g. whether undocumented migrants pay fines, whether asylum seekers have the right to work, how much migrants pay to apply for specific types of permits etc.). The datasets also track changes in policy restrictiveness over time. As such, these databases also provide us with the material to discern whether and when national immigration policy is “inconsistent”: combining or alternating between restrictive and non-restrictive measures.

In this review, we specifically focus our attention on the quantitative studies by the International Migration Policy and Law Analysis (IMPALA), Immigration Policies in Comparison (IMPIC) and DEMIG teams, as these studies seek to generate and analyze data on all areas of immigration policy. These three studies all share a “top-down” image of the nature and effects of immigration policy. This image consists of two assumptions. The first assumption is that written policies and laws define the objectives of actors located at the national scale. This assumption is evident in the way these different teams define “migration policy” and analyze evidence of its variation. All three teams see policies as ways of discerning government objectives, and more specifically their objectives to restrict, or not restrict, migration (Czaika and De Haas 2013, 489; Gest et al. 2014, 262; Helbling et al. 2017, 82). The second assumption is that the behaviour of government officials and their agents can be defined as efforts to “implement”, or “not implement” the objectives that have already been defined in written policy. This assumption is evident in these teams’ efforts to differentiate written policies from other dimensions of migration governance. More specifically, all three teams regard “policies” as distinct from “implementation” and then define “implementation” as the work that puts policy objectives into action (Beine et al. 2016, 834; De Haas, Natter, and Vezzoli 2018, 329; Helbling et al. 2017, 83). None of these studies regard the behaviour of ordinary officials as unimportant or predetermined, but all of them limit their understanding of it to the “implementation or non-implementation of government objectives”, thereby implicitly denying it any agentic or dynamic qualities. Furthermore, these studies do not offer means to explore relationships between measures of policy restrictiveness and measures of “implementation”, because they do not collect data on implementation.

To clarify, we believe these assumptions are consistent with a “top-down” understanding of immigration politics, because they treat actors located at the “higher” national scale as the main creators and inventors of policy and actors located at the “lower” local scale as merely receivers and interpreters of national policy. This top-down image of migration politics constrains these teams’ ability to draw meaning from specific types of evidence in their data. The first problem is that inconsistent policy may not reveal discernible objectives and therefore appears to these researchers as indecipherable “noise”. To compensate for this, these researchers tend to reduce evidence of inconsistency in their data and instead shine the light on evidence of consistent policy objectives. They have at least two ways of doing so. The IMPALA and IMPIC teams developed indices of restrictiveness which scored countries’ policies for migration in general, or for specific migrant tracks at given moments in time. These indices would change substantially when a given legislative act changed a country’s policy in a way that was consistently more restrictive or consistently non-restrictive. However, individual laws that included restrictive and non-restrictive changes would cancel one another out and therefore be reflected as minimal or no change in a country’s final restrictiveness score (Beine et al. 2016; Helbling et al. 2017). The DEMIG team applied a different form of analysis to similar effect. They measured the average restrictiveness of all policy changes at the global, regional and national scales. Deploying this method, de Haas et al. inferred that immigration policy-making has been less restrictive in the 1945–2014 period because the average restrictiveness of all policy changes at the global and at most regional scales was below zero (2018, 332–333). Again, this form of analysis also exaggerated the value of the most common type of policy by removing inconsistency in the data. More specifically, when restrictive and non-restrictive policy changes are passed simultaneously, their values cancel one another out. While the DEMIG, IMPALA and IMPIC team deployed different analytical techniques, the effect of these techniques – to “mute” inconsistency in the data – was the same.

The other technique these teams used to isolate consistent objectives in the data was disaggregation. This involved separating immigration policy into a number of categories, pertaining to specific migrant groups or “tracks” (e.g. labour migrants, refugees, family migrants etc.) or different policy “areas” (e.g. visas, controls, integration etc.). The analysis would then downplay evidence of inconsistency between policies for these different groups and policy areas and emphasize the consistency of objectives for policies pertaining to individual population groups or policy areas (Helbling et al. 2017, 94). For example, in their preliminary analyses, the IMPIC team analyzed indices of family migration policies alone, revealing apparent consistent trends in the restrictiveness of policies towards this group. The DEMIG team also separated out policies into types and studied the averages for these separate types. This analysis suggested that what appeared to be inconsistency in the restrictiveness of policy changes at the aggregate level, was in fact the result of consistent but contrasting policy changes for different categories of migrant, e.g. more restrictive policy changes for refugees passed at the same time as less restrictive policies passed for skilled migrants (De Haas, Natter, and Vezzoli 2018).

The key point to note here is that when these researchers analyze their data, they first de-emphasize inconsistencies and emphasize consistencies. They do not assess the amount of immigration policy-making that might fall into the category that we have defined as “inconsistent”. This gives us the first gap that this study will seek to address: to shine a light on inconsistency itself, asking how prevalent it is in immigration policy-making and in what contexts is it commonly found.

Once we have more fully described policy inconsistency, we can then seek to better explain the phenomenon. The IMPIC, IMPALA and DEMIG studies have thus far offered two types of explanation for policy inconsistency, both of which involve drawing inferences from the results of disaggregations. First, these teams have inferred that evidence of inconsistent policies may reveal that policymakers’ aim to be “selective” in immigration policy-making – making it easier for some groups to migrate but more difficult for others. For instance, the DEMIG team disaggregated their data by the migration “group” that the policies referred to. Since family migration policies have on average been more restrictive and labor migration policies have been on average less restrictive, they inferred that governments have been attempting to be “selective” in favour of families and against workers (De Haas, Natter, and Vezzoli 2018, 342–352).

The IMPALA team took this idea of selectivity further. They separated out migration policies into discrete tracks, pertaining to different categories of migrants. In addition to suggesting that different levels of restrictiveness for different tracks are suggestive of discrete policies for specific groups, they also sought to measure selectivity by counting changes in the overall number of tracks a given country had: “[t]he number of entry tracks can be used as a raw estimate of the evolution of admission policies in terms of complexity. The multiplicity of these entry tracks shows that countries fine-tune their policies in order to target some specific categories of migrants” (Beine et al. 2015, 540). Crucially, the idea of selectivity holds true to the first assumption of this research, avoiding the possibility that policy-making lacks consistent objectives, by instead inferring that policymakers’ objectives are more sophisticated than we had previously thought or have been able to “see”.

The second line of interpretation involves treating evidence of inconsistency between categories as the outcome of policy manipulation. For example, on these grounds, the DEMIG team argued that the lower numbers of more restrictive border and land control policies at the global scale can be regarded as evidence that policies of this sort represent mere “tough talk” by policymakers, because in other areas (visa policy, integration etc.) there are much higher numbers of less restrictive policy changes (De Haas, Natter, and Vezzoli 2018, 354). In this case, instances of policy changes which contradict the intentions revealed by more prevalent policies are discounted as the result of propaganda techniques. Policymakers pass these policies to mollify aggrieved constituencies, but never intend to implement them. Again, the idea of policy manipulation remains consistent with conventional assumptions about the ability of policies to reveal policy maker objectives, albeit this time inferring that policymakers are more creative in the way they reveal their objectives than we had previously thought. Furthermore, it is worth noting how this interpretation preserves the “top-down” image of migration politics. What happens at the local scale can continue to be defined as the “implementation” of prior “objectives” that have already been established at the national scale.

Before moving on, it is important to clarify the line of critique being developed here. We are not arguing that evidence of inconsistency cannot ultimately be explained as a result of these more sophisticated and disaggregated objectives, but that this is an empirical question which needs to be demonstrated rather than assumed. At the same time, we also note that both lines of explanation lead us to expect that disaggregation into categories should largely eliminate evidence of policy inconsistency, revealing the deeper objectives and removing misleading data. Therefore, in order to explore the merits of these interpretations, we plan to ask: once we have separated out immigration policy into tracks, types or categories, do immigration policies become perfectly consistent? Alternatively, does some inconsistency remain, prompting us to question our assumptions and explore alternate lines of inquiry?

Theoretical expectations of policy inconsistency

If we move beyond purely quantitative research, we find several theories that treat policy inconsistency as a common characteristic of immigration policy-making, and explicitly set out to explain its incidence. We can broadly split these theories into two types: (1) those that abandon the assumption that policies reveal consistent objectives but continue to focus on factors located at the national scale; and (2) those that abandon both conventional assumptions and draw attention to factors located at the local scale.

Theories of inconsistency at the national scale

The idea that governments are continually torn between competing objectives has been a mainstay of research into migration control policies for some time. For example, Christian Joppke’s classic work described how, somewhat counter-intuitively, “states accept unwanted migration” (1998). For Joppke, while one part of the government might want to remove unauthorized migrants from state territory, other parts of the government frustrate measures to do so by defending such migrants’ rights to stay. In a similar vein, James Hollifield’s work referred to the “paradox” of the liberal state – which had an interest in remaining open to migratory processes which simultaneously challenged its territorial, sovereign, and national foundations (2004). Advocacy coalition research has further operationalized these theories, presenting immigration policy-making as the outcome of ongoing battles between organized pro-migrant/liberal and anti-migrant/nationalist lobby groups (e.g. Sabatier 1988; Tichenor 2002; Klotz 2013). According to advocacy coalition research, we shouldn’t necessarily expect immigration policies to reveal consistent objectives, but rather expect that the playing out of contests amongst organized interests at the national scale would tend to push policymakers to adopt contrasting policies at different times.

Other research builds on advocacy coalition theory, to develop more clear expectations regarding the incidence – in total and across time – of inconsistency in national immigration policies. The paradigmatic model of this type of inconsistent immigration policy-making is the idea of the “grand bargain” between opposing advocacy coalitions. According to this idea we should expect immigration policy reform to be very infrequently inconsistent, because the adoption of a non-restrictive reform to appeal to the liberal political coalition, requires a trade off with a restrictive policy reform designed to appeal to the nationalist political coalition. The paradigmatic grand bargain is the US Immigration Reform and Control Act of 1986, which combined the introduction of sanctions against employers for hiring unauthorized migrants, with a series of regularization and temporary work measures (Skrentny 2011). Such legislative bargains may account for a portion of the inconsistent policy that we see, as coalition governments and across-the-aisle maneuvers pave the way for laws consisting of what Kitty Calavita called “a hodgepodge of ad hoc measures designed to please everyone” (1989, 35). However, as the US example of the grand bargain aptly illustrates, cross-aisle legislative compromises are particularly difficult to achieve and tend to leave most parties dissatisfied and unwilling to return to the bargaining table soon after. We therefore expect them to be historically rare. In order to explore the merits of grand bargain theory, we will ask: does inconsistent policy appear infrequently and in legislative acts or policy-making processes that take place at the same time?

An alternative, historical institutionalist line of analysis suggests that inconsistency might be a more regular, albeit still episodic feature of immigration policy-making. Historical Institutionalists suggest that “oscillations” occur as policy goes through or is affected by regular external economic or political cycles. For example, James Hollifield et al. have noted how US immigration policy was affected by the business cycle, tending towards restrictiveness in growth periods and oscillating back towards restriction in times of recession (2011). This “business cycle” theory may require more specification and substantiation in case study research before its merits can be explored in larger data sets. For example, is there an inconsistency “window” in the economic cycle where tendencies towards restrictiveness and non-restrictiveness are in balance? If so, when will these windows appear and how long should we expect them to last?

Giovanna Zincone has noted how the electoral cycle might generate similar “oscillations”, “according to which immediately before and immediately after the elections policy makers respond to the electorate’s demands and as time elapses have to cope with objective needs and powerful lobbies’ pressures” (2010, 4). These theories of cyclical oscillation provide an alternate set of expectations regarding the incidence of inconsistency, specifically, that we should expect to see greater amounts of inconsistency in the year of, or immediately following, an election that leads to a transfer of power to a new administration. In order to explore the merits of policy-cycle theory, we will ask: does inconsistency tend to appear during years where there is a transfer of power between characteristically restrictive and non-restrictive parties, or the year that follows such a transfer of power?

Theories of inconsistency that incorporate the local scale

We now move to approaches that pay greater attention to the significance of factors located at the local scale. In contrast to the expectations of national scale explanations that inconsistency will be episodic and/or rare, these theories expect policy inconsistency to be a more continuous or regular feature of policy-making processes. We begin with “policy-cycle” theory. This approach differs to the rest of the work we have discussed to this point in that it pays far more attention to the “inner workings” of the state and particularly the work done by ordinary officials. In Lasswell’s classic model of the decision-making cycle, policy-making goes through a variety of independent and sequential phases – of which developing prescriptions is only one – that can shift policies in unexpected ways (1956). Giovanna Zincone and Luigi di Gregorio apply this model to explain inconsistency in Italian immigration policy formulation processes. They note the significance of “feedback effects” from policy implementation processes as the “failure of the previous regulations, both in terms of poor control of illegal entry and in terms of the failure to integrate legal immigrants” (2002, 9) became an “input” in subsequent policy-making rounds. They described how these dynamics played out in one case, wherein a law designed to notify unauthorized migrants of their expulsion failed because of an absence of enforcement mechanisms. This failure then resulted in greater demand for regularization measures, which then induced more unauthorized migration, creating further demand for heightened controls. According to a policy-cycle approach, we should expect inconsistency to be a regular feature of policy-making, as feedback effects from implementation efforts continuously generate inconsistencies.

Other literature on the micro-politics of migration suggest we may need to go beyond conceptualizing the behaviour of officials as “implementation” and instead begin to conceive of this behaviour as defined by objectives that are not outlined in policy at all. This research has emphasized the non-linear, multi-directional relationships between officials, non-state actors and migrants that structure migration governance practices. Different studies working in this vein have emphasized how capitalist imperatives (De Genova 2010; Heyman 2004), the interests of public-private rent-seeking coalitions (Doty and Wheatley 2013; Gammeltoft-Hansen and Sorensen 2013; Golash-Boza 2009a; 2009b), and global identity frameworks (Punter et al. 2019; Bakewell 2008; Hyndman and Mountz 2008) are grounded in the routinized practices of ordinary or “street-level” officials.

The truly dynamic nature of the micro-politics of migration becomes more evident when we take migrant agency more fully into account. This has been a central focus of the literature on the “Autonomy of Migration”. This work assumes that migration itself is an autonomous socio-political phenomenon and seeks to understand how the agentic behaviour of individual migrants, and the collective impacts of their autonomous decisions, shape political institutions (e.g. Franck and Vigneswaran 2021; Papadopoulos, Stephenson, and Tsianos 2008; Bojadžijev and Karakayalı 2010). When taken collectively, this literature suggests that micro-political interactions amongst officials and migrants constitute (a) a relatively autonomous socio-political realm; and (b) a potential site where government objectives may be constituted and reconstituted. Again, joining with the “policy-cycle” literature, these local scale approaches would lead us to expect immigration inconsistency to be a regular feature of immigration policy-making.

We have recently demonstrated how one might study the impact of local scale factors in a non-representative case study of inconsistent policy in Thailand (Vigneswaran 2020). This study focused on a nine-day period in 2017 when the Thai government passed a pair of highly contrasting changes to its laws. The first law was the very restrictive Royal Decree on Foreign Worker Management B.E. 2560 (2017), which included massive fines for employers illegally hiring undocumented workers. The second was the very non-restrictive NCPO Order 33/2560 (2017) which annulled penalties for illegal work and prepared the way for an immigration amnesty. Drawing on in-depth qualitative work with policymakers and with ordinary officials, employers, brokers and migrants, our research explained the passage of the second law, by delving into dynamics occurring at the local scale. This approach revealed how the combined effect of the decisions (a) by local employers to fire workers; (b) by local level police officers to conduct raids to enforce the new law; and (c) by tens of thousands of Myanmar and Cambodian workers to flee the country, generated pressure on policymakers to pass a second law that contrasted markedly with the first. In short, once a series of highly localized political reactions to the initial policy began to bring large sections of the Thai economy to a halt, the Thai government was compelled to introduce less restrictive policy measures. This case study demonstrated that local politics may not be a tabula rasa where policy initiatives succeed or fail, as “top-down” research might suggest. The local scale may also not be merely a site where policy initiatives are tried and tested – as a policy-cycle approach might suggest. Instead, the local scale may constitute a complex and dynamic theatre that may generate forces which drive policy-making processes in different directions, engendering more frequent inconsistency in policy outcomes. Unfortunately, the qualitative data contained in the databases we have reviewed is limited to the national scale and therefore does not provide the sort of data we need to fully explore connections between local agency and dynamics, and policy outcomes at the national scale.

According to this review, much quantitative research suggests that policy inconsistency reflects policymakers’ sophisticated objectives, or cunning strategies for concealing them. In contrast, other strands of research accept that written immigration policies will feature inconsistencies and then seek to explain why. Macro-political theories suggest that inconsistency ought to be episodic and/or rare: (a) either a small selection of grand bargains between competing advocacy coalitions; or (b) a larger number of policy turnarounds generated by electoral cycles. Theories focusing on what happens at the local scale suggest instead that inconsistency ought to be a more regular feature of written immigration policies. What we take from this review is that there may be merits in exploring the incidence of inconsistency in specific policy categories and areas, and measuring its variation across time, in order to explore the competing merits of these varying accounts.

Conceptualizing and analyzing inconsistency in the DEMIG dataset

We now move to our efforts to reveal, explore and explain evidence of policy inconsistency empirically. In order to do so, we need to first clearly differentiate the concept of policy “inconsistency” from related concepts like “ineffectiveness” or “inaccuracy”. Much empirical research into the contradictions in immigration politics has focused upon the “ineffectiveness” of policies: the failure of policies to shape what migrants do. This research points to the fact that policies on paper have limited impacts upon practices on the ground, because officials fail to implement written policies, and have considerable discretionary power (e.g. Ellermann 2008; Vigneswaran 2008; Vigneswaran et al. 2010, etc.) or because migrants, smugglers and agents consistently develop new ways of circumventing regulations (e.g. Schapendonk 2018; Tshabalala 2017). Other work focuses on immigration policy “inaccuracy”: the tendency of policies to misrepresent the intentions of powerful political actors. In this vein of research, policy is seen as a form of rhetoric that serves as an incomplete guide to what national policymakers want and attempt to achieve (Andreas 2012).

“Inconsistency”, by way of contrast, does not refer to the questions of whether policies match with the intentions that are formed prior to them or the outcomes that follow – but whether policies match one another. “Inconsistent policy” combines restrictive and non-restrictive policies, or alternates between restrictive and non-restrictive policies. Temporal proximity is central to our definition of “policy inconsistency”. Inconsistency refers to the simultaneous or near-contemporaneous passage of restrictive and non-restrictive policies, and not contradictory policies that are separated by many years, and that may be better regarded as outgrowths of entirely separate and/or sui generis policy-making processes and political dynamics. In this respect, our work may be seen as interested in inconsistency between policy changes, rather than inconsistency in policy per se.

We looked for evidence of policy inconsistency in the DEMIG policy database, which tracks changes in immigration policy across 45 countries from 1945 to 2014. The key components of interest to us in this data set were its classification of policy changes into restrictive or non-restrictive changes and its ranking of the significance of these policy changes on a 4-point scale from “fine tuning” changes (1) to major changes (4). These two variables can then be combined to produce an 8-point restrictiveness scale, ranging from major restrictive change (+4) to a major non-restrictive change (−4) (De Haas, Natter, and Vezzoli 2018).

The DEMIG data reveals the number and magnitude of restrictive and non-restrictive policies in a given year, allowing us to observe moments or periods of stasis (where no policy changes occur), periods where all policies are more restrictive, periods where all policies are non-restrictive, and periods where policies are inconsistent. By adding the values of individual policies, we could measure the volumes of these different policy types, specifically comparing the amounts of consistent and inconsistent policy for a given country, year and in total.

The policy changes recorded in DEMIG are also classified into four policy areas – Border and land control, Legal entry and stay, Integration and Exit – and 15 migrant sub-groups (asylum seeker, migrant worker etc.), allowing us to differentiate between consistent and inconsistent policies within a given policy area and/or group and to compare the amounts that fall into these different categories. As with the other databases we have mentioned, the DEMIG study did not include information on the behaviour of migrants or street-level officials. Therefore, while the database could be used to directly explore a selection of national scale explanations of policy inconsistency, it offered no data that could be used to measure local scale factors. We return to this limitation in our final section.

We adopted an exploratory approach to the prevalence of policy inconsistency within the DEMIG Policy Database, capturing inconsistency by counting and visualizing the policies coded in the database. We began with a simple measure of the total amount of inconsistent policies, regardless of policy area or group. Recognizing the importance of temporal proximity, we set a 2-year temporal threshold on policy inconsistency. We coded as “inconsistent” any restrictive policy that was passed in the same year as, or in the two years after or two years before a non-restrictive policy plus any non-restrictive policy passed in the same year as, or in the two years after or two years before a restrictive policy. The remaining policies we coded as “consistent”.³ We used these analyses to discern whether inconsistency is infrequent in immigration policy-making – appearing only occasionally, in a small number of countries’ policies, or for short periods of time – or whether it is more frequent in immigration policy-making. We then conducted the same analyses within policy areas and within migrant sub-groups. We used these counts and comparisons to help us discern whether inconsistency is primarily a reflection of differences in the restrictiveness of policies for different areas or different groups, or whether these written policies also reveal differences in the restrictiveness of policies for the same areas and groups.

To explore the merits of theories focusing on the national scale and more specifically their expectations regarding temporal variation, we needed more precise measures that would allow us to track variation more closely in policy inconsistency at the national scale across time. We began by measuring to what degree policies were inconsistent for each country within a given year. The most basic type of inconsistency can be called “ambiguity” – government policies enacted simultaneously which contradict each other in terms of their openness or restrictiveness. Grand bargains, where a single set of legislative measures are passed more or less simultaneously, fall into this category of “ambiguous” policy. We developed a ratio measurement of ambiguity: the ratio of the sum of all restrictive policies in a year to the sum of all non-restrictive policies. In order to situate the measure on a scale of 0 through 1, where 1 is complete inconsistency and 0 is complete consistency, the denominator in the ratio is always the highest value. Here, we also account for the impact of policies by using the DEMIG policy impact variable, as it makes little sense for policies with a small impact to have the same consideration as policies with large-scale impact. We express this relationship in the following way: $f\left(x\right)=\{\begin{array}{c}{\displaystyle \frac{\sum r_t}{\sum o_t},\sum r_t<\sum o_t}\\ {\displaystyle \frac{\sum o_t}{\sum r_t},\sum o_t<\sum r_t}\end{array}$where t is the given year, r is restrictiveness, and o is openness. Given a specific year, the equation returns the ratio of restrictive policies to open policies, always dividing by the highest value and therefore situating the inconsistency scale between 0 and 1.

While ambiguity is the most basic form of inconsistency, with contrasting policies coinciding in the same year, as we have suggested earlier, we also believe that inconsistency reflects developments across a longer timeframe. For example, a large number of restrictive policies in one year, followed by an equally large number of non-restrictive policies in the next should be considered as a form of inconsistency. We label this second type of inconsistency: “alternation”, where a country flips between restrictive and non-restrictive policies in short time periods. The theory of electoral cycles, which expects the outgoing government’s policies to contrast markedly with the incoming government, fits into this category of “alternation”. For the purpose of measurement, we conceptualize this short period to be a two-year window – two years into the future and two years into the past, a total of five years. To account for both alternation and ambiguity, we calculate the ratio of all restrictive policies to all non-restrictive policies in a given 5-year period. We express this relationship in the following way: $f\left(x\right)=\{\begin{array}{c}{\displaystyle \frac{{\sum }_{t=t-2}^{t+2}{r}_t}{{\sum }_{t=-2}^2{o}_t},\sum _{t=t-2}^{t+2}{r}_t<\sum _{t=t-2}^{t+2}{o}_t}\\ {\displaystyle \frac{{\sum }_{t=-2}^2{o}_t}{{\sum }_{t=-2}^2{r}_t},\sum _{t=-2}^2{o}_t<\sum _{t=-2}^2{r}_t}\end{array}$Where t is year, r is restrictiveness, and o is openness. Given a specific year, the equation returns the ratio of the sum of restrictive policies to the sum of open policies for a two-year margin (if t = 1990, then it would be the ratio from 1988 to 1992), always dividing by the highest number and therefore situating the inconsistency scale between 0 and 1.

Finally, we develop a measure that recognizes the arguments formulated by De Haas, Natter, and Vezzoli (2018): that policy inconsistency could potentially be attributed to differences between policy areas: a single-year could contain restrictive policies regarding integration and non-restrictive policies in the area of border control. To make our measurement more exhaustive, we construct a ratio which only factors in inconsistency within specific policy areas, essentially controlling for this differing policy area approach. For this purpose, we calculate the ratio of inconsistency within each policy area, and then make an average of this inconsistency for each year. We express this relationship in the following way: $f\left(x\right)={\displaystyle \frac{{\displaystyle \frac{\sum o_{\mathrm{ya}}}{\sum r_{\mathrm{ya}}}+{\displaystyle \frac{\sum o_{\mathrm{yb}}}{\sum r_{\mathrm{yb}}}+{\displaystyle \frac{\sum o_{\mathrm{yc}}}{\sum r_{\mathrm{yc}}}+{\displaystyle \frac{\sum o_{\mathrm{yd}}}{\sum r_{\mathrm{yd}}}}}}}}{4}}$Where y is year, r is restrictiveness, o is openness, and a through d are different policy areas. Given a specific year, the equation returns the mean ratio of restrictive policies to open policies within specific policy areas, always diving by the highest number and therefore situating the inconsistency scale between 0 and 1.

Exploring incidence and variation of inconsistency

We now turn to discuss the results of these analyses. Our simple measures suggest that policy inconsistency is not an irregular or isolated phenomenon. Adopting a minimal definition of inconsistency as “ambiguity” and excluding instances of “alternation” from the analysis, 3,066 out of 4,837 policies, 63.39 per cent of all immigration policies were inconsistent with policies passed in the same country, within the same year. These results are visualized in Figure 1 which plots policies for all 45 countries in the DEMIG dataset by their openness (above X-axis) or restrictiveness (below X-axis), and classifies them by whether they are inconsistent (an open policy introduced in the same year as a restrictive policy (yellow)), or a restrictive policy in the same year as an open policy (red), or consistent (a restrictive or open policy that is not introduced in the same year as a policy of the opposite type (blue)).

Figure 1. World-wide cumulative changes in immigration control policy (1945–2014).

Crucially, policy inconsistency was not isolated to a small number of countries or a small number of periods but was observed across most countries and many periods. Policy inconsistency is still observed when policies are disaggregated into respective migrant target groups: with this calculation, 893 of 4,837 migration policies in the DEMIG database, 18.46 per cent, are inconsistent. While inconsistency within migrant target groups makes up a little under a fifth of all policies in our data, this estimation is a conservative one, as DEMIG codes migrant target groups into 15 distinct categories, with several divisions for distinct types of family members, and migrant workers. If one were to group these similar categories, we would expect the proportion of inconsistency to rise.

Disaggregating policies into discrete policy areas also reduces, but does not eliminate, evidence of policy inconsistency. Again, looking at the entire sample of countries in the DEMIG dataset, we can see that a total of 1,410 policies, out of 4,837, were inconsistent within their policy area, a total of 29.15 per cent. The proportion of “within policy area” inconsistency varied from country to country. For Australia, 34.29 per cent of policies are inconsistent, for Canada, 42.89 per cent of policies were inconsistent, for the United States, 41.11 per cent of policies were inconsistent, and for the United Kingdom 23.13per cent of policies passed were inconsistent.

More focused exploration of the national scale data allows us to better isolate “when” this inconsistency occurs, and in which policy areas. Figure 2 visualizes this inconsistency within four policy areas (border and land control, legal entry and stay, integration and exit) for the case of Australia. This figure plots policy changes over time. The y-axis measures the impact and restrictiveness of policies passed that year. Large negative values indicate high-impact policies resulting in more restrictive measures. Large positive values indicate high-impact policies resulting in less restrictive measures. Policies are further disaggregated by their policy area through colour coding, allowing for a more detailed examination. This offers us a different means of exploring inconsistency. When bars of the same colour appear on both sides of the X-axis in a given year or in consecutive years, this represents an instance of inconsistent policy within a single policy area. For example, between 1988 and 2001 we see several years of inconsistency in “Legal Entry and Stay” policies (beige). By way of comparison, when bars of the same colour only appear on one side of the X-axis in a single-year or consecutive years, policies for that area may be deemed “consistent”. For example, we see only consistent policies in “Border and Land Control” policies. While we see orange bars on both sides of the X-axis, they do not appear on different sides in the same or in consecutive years.

Figure 2. Policy changes by policy area for Australia (1945–2014).

While these analyses suggest that inconsistencies cannot all be explained as a result of different policy strategies for different policy areas, it does not address the possibility that governments may be targeting policies at specific migrant groups. Perhaps, for example, the evidence of inconsistency in Australia’s “Legal Entry and Stay” policies have to do with the fact that they are being “selective”: passing restrictive laws for one group, while more open laws for another? To address policies aimed at different groups, we used the Target Group variable in the DEMIG database. Here, policies were collapsed into one of four groups conducted through broad topical grouping. The first group pertains to migration for employment – this included all types of migrant workers, high – as well as low-skilled. The second was policies towards migration as a form of aid, which included refugees, asylum seekers, and other vulnerable groups, as well as their families. Third are policies towards irregular immigrants, as this group receive a lot of policy and political attention. Lastly, we grouped together remaining large policy target groups: students and family members. While somewhat distinct groups, they are united in that their migration status is not tied to employment, are not a vulnerable group, and are not irregular. Target groups pertaining to “all migrants”, “investors and businesspeople”, “diaspora”, and “specific categories” were removed due to their relatively low occurrence in the database. Figure 3 aims at capturing inconsistencies within country, policy area, and target group by presenting policies regarding “Legal entry and stay” for Australia by Target group. As can be observed, there is significantly less evidence of inconsistency at this level of disaggregation. Nevertheless, evidence of inconsistency remains, for example in the years 2000 and 2001 for family and students and in the years 2012 and 2013 for workers.

Figure 3. Legal entry and stay policies for Australia by target group.

The finding that disaggregating the data into discrete areas reduces overall amounts of inconsistency suggests that conventional research may be able to explain some amount of policy inconsistency as the outcome of “selectivity” or “manipulation”. At the same time, the fact that disaggregation does not eliminate evidence of inconsistency suggests that we may need to move beyond their assumptions to fully understand inconsistency.

In order to explore our alternate explanations, we move to the results of our efforts to measure variation in inconsistency across time at the national scale. These analyses offered limited support for the “grand bargain” hypothesis. We explored this hypothesis by visualizing our measure of policy ambiguity – when countries introduced restrictive and open policies in a single-year – for the years 1945–2010 across all 45 countries (Figure 4). In this figure, scores close to 1 indicate a high level of ambiguity (every policy of one type is “matched” by a policy of the opposite type). Scores close to zero indicate a low level of ambiguity (most policies are of one type, but there is some evidence of policy of the opposite type). Scores of zero suggest that policies are consistent (indicated by a black dash on the x-axis), or there were no policies passed in that year (indicated by no mark on the x-axis).

Figure 4. Ambiguity across DEMIG database countries.

In this visualization, a few countries, such as Argentina, Brazil and China, exhibit the temporal pattern expected by a “grand bargain” hypothesis, with ambiguity appearing in a few isolated spikes. However, in most other countries, years with ambiguity are either immediately followed by another year of ambiguous policy, or a gap of one year, followed by more ambiguous policy. If the power of advocacy coalitions explains this ambiguity, then – contrary to our expectations – parties are more regularly returning to the table to make compromises than the “grand bargain” theory would have us expect.

Second, we considered whether inconsistency could be explained as the outcome of electoral cycles. Here, instead of using the single-year measure of ambiguity, we use the extended, 5-year measure of “ambiguity plus alternation” to explore whether changes in administration generate shifts between more restrictive and open policies. More specifically, we plotted these results across a sample of two-party, winner-take-all political systems (Australia, the United Kingdom, and Canada). These were “easy” cases for an electoral cycle theory because elections more commonly resulted in a change in the governing party instead of an adjustment in a ruling coalition (e.g. the Netherlands) or a partial shift in control of key governing bodies (e.g. the United States). We then added dotted red lines to years that experienced a change in government as the result of an election. The electoral cycles explanation of inconsistency would suggest that alternation is most-likely to cluster around these red lines.

Figure 5 identifies moments when a change of government may have accounted for inconsistency. For example, periods of inconsistency may be observed in the following transfers of power: in Australia in 1996 and 2007; in the UK in 1997 and 2010; and in Canada in 1979 and 1993. At the same time, the visualizations do not demonstrate a close correlation between transfers of power and immigration policy inconsistency, with ample inconsistency occurring two or three years away from a transfer of power, and several transfers of power generating no inconsistent policy.

Figure 5. Migration policy inconsistency with a 2-year window. Changes in government marked in Red.

To summarize this section so far, our analyses offer some support for national scale explanations of policy inconsistency and offer ways to further explore these ideas, through more finely grained analyses. First, the significant reductions in overall levels of inconsistency that resulted from disaggregation by area and group offers some support for “selectivity” and “manipulation” hypotheses. Our visualizations disaggregating the data for individual countries into policy areas and population groups offer the means to detect when inconsistency between areas and groups occurs. These observations could be the subject of more intensive case studies that examine whether policy-maker selectivity or manipulation account for these contrasting policies. Second, although our visualizations of country level temporal patterns in inconsistency suggest that ambiguity is not an isolated or rare occurrence, as grand bargain theories would expect, the same visualizations might be used to isolate countries with several brief moments of ambiguity – such as Argentina, Brazil and China – as “most-likely” cases for such a theory. Alternatively, our analyses may encourage consideration of a modified version of an advocacy coalition theory which posits that advocacy coalitions are in fact more prone to compromise than we have previously thought. Third, while our visualizations of variation in inconsistency over time do not suggest a strong correlation between transfers of power and policy inconsistency, scholars could use these same visualizations to identify moments and periods when electoral cycles may have generated expected and unexpected outcomes of policy-cycle theory. Importantly, while each of these three lines of empirical inquiry would require more in-depth qualitative case study research, they could all benefit from the raw data collected and published by the DEMIG team, which includes narrative descriptions of all the policy changes they observed.

At the same time as providing the tools to refine and further explore national scale explanations of inconsistency, our research also suggests the need to move beyond national scale explanations of policy inconsistency. Here we introduce the finding that inconsistency is a feature of immigration policies across time and space. The sheer volume of inconsistent policy, whether measured in the aggregate or disaggregated by policy area and population group, belies efforts to dismiss inconsistency as mere “noise” or reduce it to policy-maker selectivity. Inconsistency does not adopt the episodic temporal patterns expected by advocacy coalition theories about grand bargains or historical institutionalist theories of electoral cycles. Instead, inconsistency conforms more closely with the expectations of local scale theories: inconsistency appears regularly in most countries, and evenly across time. Figure 6 illustrates the prevalence of policy inconsistency across the 45 countries in the DEMIG sample. It suggests the wide variety of examples of inconsistent policy-making that researchers might use to gauge the impact of local scale factors vis-à-vis national scale variants.

Figure 6. Migration policy inconsistency with a 2-year window across DEMIG database countries.

Unfortunately, our ability in this paper to further explore the merits of theories that focus on the local scale is limited by the fact that – as noted above – quantitative researchers did not collect data on what they define as “implementation”. In our concluding remarks, we therefore move on to consider what techniques and resources would be required to overcome this limitation.

Concluding remarks

In summary, this study has engaged critically with a conventional “top-down” image of migration politics. We specifically problematize previous efforts to ignore or “mute” inconsistency in immigration policy data. We have developed new techniques to reveal and measure inconsistency and demonstrated how we might explore and begin to test the merits of competing explanations of this phenomenon. Contrary to conventional assumptions, we have suggested that the sheer amount and temporal patterns of immigration policy inconsistency suggest the need to critically interrogate analytical methods (disaggregation) and data collection strategies (excluding “implementation” data) that implicitly enforce a “top-down” image of migration politics, and the need to further consider how factors at the local scale shape developments at the national scale.

Over the longer term, and depending on the findings of these studies, the challenge will be to discover ways to more effectively engage with national scale data on written immigration policies. In this venture we want to note two issues. The first relates to the way we frame our inquiries. More specifically, we believe that research into the topic of policy inconsistency ought to take an equal interest in the topic of policy consistency. Due largely to the problematic assumptions that we have identified, policy consistency has tended to be assumed in policy analysis and not taken seriously as an object of study. Our study has, to a large degree, continued in this vein, seeking to isolate inconsistency as a variable and explain why it appears in specific places and times, while largely ignoring “consistency”. Yet, if our study has shown anything, it is that policy consistency is an equally variable phenomenon and not simply the status quo. Part and parcel of understanding why policies are often inconsistent, will be efforts to analyze why they are sometimes consistent.

The second challenge relates to the need to combine quantitative and qualitative research in the study of inconsistency. Our example of the Thailand case study suggests that there are concrete ways of tracing how factors located at the local scale cause inconsistent policy at the national scale. The DEMIG data set, which includes brief qualitative accounts of all the policy changes listed in the data set, provides a unique opening for further mixed method research. These qualitative accounts can be used to develop more rich descriptions of instances of inconsistency in the data, particularly cases within “target group” and within “policy area”. Further qualitative inquiry, using policy documents, newspaper reports and key informant interviews, could help us understand whether the data is in fact revealing inconsistency, or merely unrelated sets of policies with no significant contradiction. This could lead to further quantitative work, by cleaning the data of false positives and refining the measures of policy inconsistency that we have provided here. Alternatively, this approach could provide the basis for further explanatory endeavor, using case study and process tracing methods to gauge the merits of competing accounts located at the national and local scales. Our study has provided the opening for such work, by suggesting that inconsistency in immigration policy is a puzzle that cannot be easily solved using existing “top down” assumptions, theory and data.

Acknowledgements

The authors would like to thank Evelyn Ersanilli, Anja Karlsson Franck, Audie Klotz, Agnes Kvistborg and Abbey Steele for their generous comments on previous drafts.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This paper was developed through funding received from the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (406.22.SW.046), Vetenskapsrådet (DNR: 2019-05443), and the Transnational Configurations of Conflict and Governance Programme Group at the University of Amsterdam.

Notes

1 “Scale” refers to the nested hierarchy of bounded spaces of differing size, such as the local, regional, national and global (Delaney and Leitner 1997, 93). Scales are relatively discrete spatial containers of political behaviour and phenomena, which tend to feature specific sets of actors, dynamics and norms (Brenner 2001, cf. Marston, Jones III, and Woodward 2005).

2 Helbling et al. define “restrictiveness” as “the extent to which a written policy mandates the limitation or liberalization of the rights and freedoms of immigrants” (2017, 88).

3 We also are not interested in contrasts between the restrictiveness of policies in different countries. Hence, we did not interpret differences between aggregate measures of policies at the regional or global scale as inconsistency.