When do municipalities smooth access for new social assistance claimants? Local determinants of municipal welfare agency ’ s implementation practices in Belgium

In many Western countries local welfare agencies retain large responsibilities in the practical organisation of means-tested support. While this may lead to substantial differences in the application trajectories of prospective claimants, what has garnered little empirical research is how local agencies use their discretion to implement practices that may encourage or discour-age take-up. In addition, we know little on the extent to which variation in local implementation is attributable to differences in local conditions. Focusing on Belgium as an interesting case, this article addresses this gap by mapping the variation in local implementation practices in specific aspects of social assistance that are relevant in terms of limiting non-take-up. We assess whether certain local political, socio-demographic or economic characteristics favour implementation practices, using structural equation modelling on custom-collected data on the practical organisation of social assistance application trajectories in  Belgian municipalities. The results show that local welfare agencies implement coherent practices regarding information provision, accessibility, trust and the locus of initiative in the claiming process. We find the political orientation of the municipality and the share of foreigners in the municipality to be related with, respectively, shifting the locus of initiative away from the welfare claimant and a focus on information provision.


Introduction
In many Western countries local welfare agencies have considerable freedom in the implementation of means-tested support (Sabatinelli, ), likely leading to variation in the treatment of and interaction with (prospective) claimants (Trydegård and Thorslund, ). Studies on local welfare systems have found that this variation is often associated with the local context in which the welfare office operates. Local variation has, for instance, been demonstrated with regard to elderly care (e.g. Jensen and Lolle, ; Trydegård and Thorslund, ), direct payments for health and social care (e.g. Fernández et al., ), different local services (e.g. Kriisk, ) and the way minimum income support is organized across European cities (e.g. Oberti, ). Despite the fact that in many Western countries social assistance implementation is to a large extent devolved to the local level, relatively few studies have focused on the relation between the local context and local social assistance implementation. Particularly noteworthy is the study of Minas (), who looks specifically at the relationship between the local context and the way welfare offices organize their interaction with new social assistance clients in Sweden, focusing on the existence of special intake units within the welfare office. Yet, how local welfare agencies use their discretion to implement a coherent set of practices in a wider range of aspects related to social assistance provision, has gained little attention.
How local welfare agencies structure access to social assistance is however no trivial matter. Scholars have highlighted the importance of the interaction with the local welfare agency (van Oorschot, ), and the need for a welcoming environment in order not to deter potential claimants from filing a claim, known as non-take-up (Dubois, ; Hasenfeld, ). Especially, in the case of social assistance, available estimates indicate that in many Western European countries non-take-up is a serious problem (Fuchs et al., ). Together with both more systemic factors (such as the guaranteed benefit level) and more individual-level factors, such as potential claimants' characteristics and the interaction with the social worker (De Wilde and Marchal, ), the local welfare agency has an important role to play in removing potential thresholds for taking up benefits by eligible persons in need.
In this article, we look into the relation between local characteristics and local implementation practices, focusing on those aspects of social assistance provision that have the potential to improve take-up of social assistance benefits. We focus on Belgium, as municipal welfare agencies have large autonomy in the practical implementation of social assistance, while benefit levels and eligibility conditions are nationally determined. For the purposes of this article we collected new data on the practical organisation of social assistance applications in  Belgian municipalities and used structural equation modelling to assess the relation between local characteristics and combinations of implementation practices at the level of the welfare agency.
Section  shortly summarizes the reasons behind local social assistance implementation. Section  discusses the local levers welfare agencies have at their disposal to facilitate benefit take-up, as well as the local conditions that may be responsible for a greater focus on these levers. Section  outlines the institutional organisation of social assistance in Belgium, while Section  describes the data and methods used. Section  contains the main findings. Section  concludes.

Multi-layered social assistance
Most welfare states have a means-tested minimum income scheme, commonly known as social assistance. Due to, among other things, their roots in local charity relief, municipalities generally retain important competences in the field of social assistance (Sabatinelli, ). Whereas eligibility conditions and benefit levels are often decided upon at the national or the regional level, in many countries the local level is in charge of the implementation of these regulations, including checking the eligibility of new claims and assessing the need of beneficiary households, often through a detailed means-test, social contacts and a home visit (Frazer and Marlier, ).
Apart from the historical roots, local social assistance implementation is based on the premise that municipalities are better placed to assess both need and willingness to work. Local implementation would allow for more individually and locally tailored support and practices (Powell and Boyne, ).
Local social assistance implementation is also believed to increase the legitimacy of anti-poverty policies (Kazepov, ), as local social workers may be more suited to identify the truly needy, fostering the support's perceived legitimacy. In addition, anti-poverty policies can also be more adapted to local political preferences. This has been demonstrated for both sanction rates and benefit generosity in different national contexts (Fording et al., ; Lien and Pettersen, ).
Finally, variation in local implementation may open up opportunities for policy innovation (Kazepov, ). Each local welfare agency will develop practices on a history of trial-and-error (Künzel, ; Trydegård and Thorslund, ). To the extent that local welfare agencies and national governments monitor the variation in local practices and outcomes, local implementation can act as a laboratory for testing best practices (Vandenbroucke et al., ).
Self-evidently, the economic and socio-political considerations that are taken into account at the local level may not necessarily lead to the best practices in terms of poverty reduction or, as is the focus of this contribution, on reducing risks of non-take-up. The local level may simply lack the means to design and execute effective poverty reducing strategies in its social assistance implementation, and the aforementioned "laboratory" of local social policy implementation only works if sufficient care is taken to both identify and disseminate best practices (Vandenbroucke et al., ). Moreover, some municipalities may have more financial room to manoeuvre or administrative capacity than others (Carpentier, ).

Local implementation levers facilitating take-up
Research into the causes of non-take-up usually focuses on the trade-off between the costs and benefits a potential claimant faces when claiming (social) assistance. Costs relate to the time and effort it takes to gather information on a benefit, to realize one's eligibility and to take all necessary steps to start and finalize a claim, whereas benefits refer to the expected benefit level and duration of benefit receipt (Fuchs et al., ). Scholars have extended this framework with insights from behavioural economics, stressing the role of cognitive biases and behavioural barriers (Baicker et al., ). Others highlighted the importance of prohibitive access thresholds, and of feelings of stigma (Baumberg, ) when discussing the reasons for non-take-up. In this regard, some see non-take-up as an additional targeting instrument that will make policies more efficient, as potential claimants will only self-select in support schemes when their need justifies the costs that one faces in applying for a benefit (Kleven and Kopczuk, ). Whereas this element is certainly at play, it does not unequivocally mean that we should not be concerned about non-take-up. This argument supposes that all potential claimants have the same competences to make this trade-off, whereas it has been demonstrated that this is not the case, especially not for those confronted with the highest need (Mani et al., ).
Also, an unequivocal focus on costs and benefits from an individual perspective neglects the importance of policy design and policy implementation (Van Mechelen and Janssens, ). On a very practical level, the trade-off that a potential claimant faces is heavily influenced by how client-friendly local welfare agencies are and by practical administrative arrangements that may lay stumbling blocks in clients' paths (van Oorschot, ). A rare empirical study found that in the US up to a quarter of those initiating a welfare claim do not finalize it because of hurdles resulting from practical implementation decisions of the local welfare agency (Bartlett et al., ). In this article, we consider the practices local welfare agencies can assume in minimizing thresholds to take-up.
We identify five aspects of social assistance provision mentioned in qualitative and quantitative literature where local administrations may have an impact on non-take-up: () physical accessibility; () information provision; () streamlining the administrative procedure () installing an appropriate balance between trust and control elements; and () placing the locus of initiative with the local welfare agency.
The first aspect refers to the accessibility of the local welfare agency. The target population of local welfare agencies is often confronted with multiple challenges. Potential claimants are likely not only income poor, but may also be confronted with transport poverty (Lucas, ) or time poverty (Mullainathan and Shafir, ). Local welfare agencies may take these deprivations in multiple areas into account in ensuring the physical accessibility of their agency. Think, for instance, of satellite offices at different locations in the municipality or the options that are available to less-abled persons to go to the welfare agency, but also of accessibility in terms of time. Empirical research demonstrated that limited opening hours (i.e. only during office hours) are an important reason for potential claimants to postpone their claim (Bartlett et al., ).
A second aspect that lies within the realm of local possibilities is to focus on information provision and communication (Finn and Goodschip, ). Whereas large-scale information campaigns may fall beyond the competences and budgetary possibilities of local welfare agencies, they can provide a consistent and clear documentation of the types of support that can be expected. An added value of a local communication strategy is that the format and language can be tailored to local circumstances. Unfamiliarity with the supply of services and in particular with the application procedure is one of the most prominent thresholds to claim (Okbani, ).
To assess the eligibility of claims is an important task of local welfare agencies in countries with local social assistance implementation, in line with relevant legislation and access conditions. Designing the administrative procedure in a way that minimizes a client's time and effort to apply can therefore be an important local lever to reduce non-take-up. It has been shown that setting up an online application process (Kopczuk and Pop-Eleches, ), providing personal enrolment assistance or simplifying or translating application forms (Bertrand et al., ) reduce the complexity of the application process.
Rather than being discouraged during a process, potential claimants can also be deterred by what they may consider a stigmatizing procedure or treatment. Each system that is partly based on difficult to observe eligibility criteria needs to implement some checks. Adding control elements may however increase stigmatisation among recipients (Altreiter and Leibetseder, ). Intrusive elements in an application procedure, such as obtaining documents from other administrations or (prior) employers or taking potential claimants fingerprints, likely deter potential claimants (Bartlett et al., ). A welfare agency seeking to limit non-take-up could therefore aim for a balance between trust and control, with as little (repetition of) control elements as possible.
The administrative process of collecting all necessary documents from the potential claimant and to establish eligibility usually takes time. A potential claimant may effectively drop out during this period. Behavioural economic research shows that smartly designed deadlines and reminders are effective in retaining claimants throughout an application procedure (Farrell et al., ). These shift the burden of taking initiative to some extent from the claimant to the welfare agency. Interpreted even more broadly, locus of initiative can also refer to the very first step of filing a claim. Within social assistance, it is usually the potential client that has to approach the welfare agency. Shifting this burden from the claimant to the welfare agency can lower the threshold for clients in making their request for help. A randomized control trial in Flanders shows that sending outreaching case managers to potential claimants' homes had an impact on the take-up of additional financial support by disadvantaged families (Van Lancker et al., ).

Local characteristics and coherent implementation practices
We expect variation in local practices in line with the reasons for local social assistance implementation discussed in Section . First, in a context where the organisation and resources devoted to social assistance are to an important extent (co-)determined by local political authorities, we can assume that local political preferences have an impact. In general, we could expect that parties that traditionally focus on combatting poverty and are in favour of a more generous redistribution will strive for maximal effectiveness of the policy levers at their policy level. At the national level, the presence of leftist parties in the government links to increases in social spending or more generous social benefits (Wang et al., ). Similarly Nordic municipalities with a more leftist composition of the local political assembly are found to be related with longer periods of support and larger amounts of money than other political constellations (Lien and Pettersen, ). We therefore expect that a stronger presence of left-wing parties at the local level is related to actions at the level of the welfare agency that facilitate the uptake of benefits (H).
Second, local policy implementation should allow to develop practices that are in line with the local socio-economic and demographic situation (Kazepov, ). This means that local welfare agencies should tailor their services to the vulnerable population at hand. An obvious expectation is that cities confronted with a large population of migrants will be more likely to have developed a set of practices taking account of language differences and information provision (H). The local situation may however have an impact in a less purposeful way. It can be expected that practices that ask more time and labour resources than others, such as outreach efforts, will be harder to implement in municipalities that are confronted with high numbers of vulnerable persons (H).
In line with the concern that local policy implementation may also give rise to variation due to practical concerns, rather than through alignment with local socio-demographic and political circumstances, we include both the size of the municipality (as a proxy for administrative capacity) and the financial situation of the municipality in the analysis. It is plausible that municipalities with higher administrative capacity will have more standardization, and perhaps on the whole, more consistent practices implemented in a particular domain (H). Financial concerns may lead to less investment in implementation practices that have the potential to reduce non-take-up (H). Table  summarizes our hypotheses.

Social assistance implementation in Belgium
We study the empirical presence of coherent local implementation practices with the potential to reduce non-take-up and their drivers through a survey of local welfare agencies situated in the Flemish region of Belgium. As in a number of other European countries, social assistance for individuals at working age in Belgium is a shared responsibility between the national state and the local municipalities (Frazer and Marlier, ). Municipalities have considerable freedom in the implementation of social assistance (De Wilde and Marchal, ). The federal level legislates the right to social assistance by detailing the eligibility conditions, the benefit level and a number of minimal procedural requirements. The financial burden is shared with the local municipalities' budget.
The procedural requirements determined at the federal level stipulate that the welfare agency should conduct a social assessment and deliver a summative report with a recommendation from a social worker on the claimant's eligibility. Based on this report, a board of local politicians will decide upon awarding the benefits. This social assessment should be conducted by a licensed social worker, who collects socio-demographic information on the claimant and her household members, on the economic means available and her work willingness (De Wilde and Marchal, ), in the course of at least one mandatory home visit. The social worker may additionally rely on conversations with the client, electronically available administrative data on social security benefits and employment, or additional documents.
These federal requirements leave quite some room for manoeuvre. For instance, the intake procedure and number of social workers involved in this procedure are not described at the federal level. Each agency can organize this procedure in a unique way. Municipalities have furthermore large autonomy in the actual organisation of their local welfare agencies, including their internal organisation, opening hours, specific social projects and their collaboration with external agencies or organisations. It should be noted that the federal legislation leaves ample room for discretion not only to the local level, but also to the individual case worker. Prior research has highlighted the importance of both the local and the case manager's level, and the interaction between both (De Wilde and Marchal, ). Here, we limit ourselves to the measures taken at the local level, as this allows us to quantitatively explore the real-world implementation practices in a large number of municipalities.

Data
We collected new data on the local social assistance claiming process and on the broader organisation of local welfare agencies in Belgium. We conducted an online survey sent to all Flemish municipal welfare agencies, asking after the usual practices in their local welfare agency and organisational features. We received complete surveys for  Flemish municipalities (i.e. a response rate of %) filled out by  qualified social workers responsible for carrying out social assessments on claimants' eligibility. A comparison of the characteristics of the municipalities in our non-probability sample with the total population of Flemish municipalities shows that the sample is relatively representative in terms of size, number of social assistance beneficiaries, share of foreigners and median income in the municipality (see the online Supplementary Materials).
Our survey consisted of two parts. A first part inquired after the local welfare agency's habits and practices regarding the reception and further treatment of first-time entrants to the social assistance system during the  day claiming process. This part of the survey was structured around relatively detailed stories of four hypothetical first claimants (so-called vignettes) (de Wilde, ) asking after the nature and type of interactions these hypothetical individuals would have with the staff of the local welfare agency, before and during the processing of their claim (see the online Supplementary Materials). A second part inquired after more general organizational features such as infrastructure and opening hours.

Analytical strategy
Our survey collected ample information on implementation practices regarding different aspects of social assistance provision. We reduced this information using Exploratory Factor Analysis (EFA), a statistical technique that allows to discover the unobservable explanatory variables (so-called latent factors) of a set of observed variables. The presence of these latent factors indicates that welfare agencies adopt coherent practices in specific aspects of social assistance provision. Subsequently, we applied Confirmatory Factor Analysis (CFA) to assess whether the final set of identified factors was indeed a reliable reflection of the observed variables. In particular, we examined whether the factor structure derived from EFA was a good fit for the data and whether each of the items loaded significantly on its hypothesized factor. The model fit was evaluated by looking at the Standardized Root Mean Squared Residual (SRMR) which is an absolute measure of fit and is appropriate for survey data. The SRMR can take a range of values between . and .. A value less than . is generally considered a good model fit (Browne and Cudeck, ).
To examine the relationship between the coherent sets of practices and local characteristics, we used Structural Equation Modelling (SEM) with the coherent sets as the dependent variables. We included the following municipal characteristics as independent variables. We used the political party of the mayor to measure local political preferences with two dummy variables: centre-oriented and right-oriented, with left-oriented parties as the reference group. The demographic variables used are the percentage of foreigners in the municipality and the share of children born into underprivileged families  . Finally, we include both the number of residents and the self-financing margin of each municipality as proxies for the administrative capacity and the economic situation. The "selffinancing margin" shows how much budget the municipality has left to do additional investments after both income (from daily operations) and expenses (on loans) are taken into account. With the exception of the political variables, the information on local conditions was obtained from the Flemish statistical agency and concerned the most recent year available. The political data were compiled based on different data sources and concerned the legislature -  .
The simultaneous estimation of the different structural paths (i.e. from the municipal variables to the coherent sets of practices), together with the estimation of the measurement models (i.e. the factor loadings of the indicator items), puts higher constraints on the items used. For some aspects it is hence not possible to include all municipal characteristics at the same time, while maintaining a sufficiently good measurement model (defined as significant factor loadings above . for all indicators of the latent factor). We therefore step-wise deleted the municipal variables with the lowest level of significance in relation to the dependent variable while comparing the model fit of the successive models. Once no significant improvement in model fit is gained by omitting structural paths and variables compared to the previous model, while obtaining an adequate measurement model, the step-wise process of omitting variables stops. Again, the SRMR was used to evaluate the model fit of the models.  For the purposes of factor analysis and SEM, interval or ratio data are required. Where this was appropriate, we created count variables of categorical information (e.g. the number of options available to claimants), or recoded the information based on insights from the literature on non-take-up. To estimate the structural equation models, we used a robust maximum likelihood estimator (MLMV) which corrects for the non-normality of our data.

Identifying coherent practices
The EFA showed several items to cluster logically together in latent factors. We interpret these latent factors as aspects of social assistance provision, in which a coherent set of practices that have the potential to reduce non-takeup is implemented by the welfare agency. We ultimately identified four coherent sets of practices on the social assistance provision aspects information provision, accessibility, locus of initiative and trust.
Three items clustered together that all relate to information provision, indicating that municipalities adopt coherent practices on this aspect: i) the number of informative documents handed out during the first meeting (info intake), ii) the number of different media through which information in foreign languages was made available to potential claimants (info languages), and iii) the number of elements in the application procedure for which standardized, informative documents are available in the welfare office (standardized documents). The use of standardized documents can indeed be a significant help for the client during the application process: checklists with required forms or leaflets outlining the specifics of the home visit are examples of standardized documents that lower information costs. High values for the resulting factor indicate a stronger focus on information provision by the welfare agency. Welfare agencies hence generally combine a focus on these three local practices (or, in contrast, are relatively negligent regarding all three).
Four items related to accessibility also clustered: i) the frequency of late opening hours in the welfare office (late opening), ii) the number of infrastructural features that facilitate access to less abled persons (infrastructure), iii) the presence of infrastructural features that might make a visit more or less stigmatizing, such as a glass divide between the receptionist and the claimant during the first contact (feel welcome) and iv) whether or not the welfare agency shares its infrastructure with other organisations (shared location). Where the first two items increase the physical accessibility of the office, the latter two are believed to increase the psychological accessibility for prospective clients (Sannen, ).
The EFA further showed three items related to the degree of trust vs. control during the application process to cluster together: i) the possibility of an additional, unannounced home visit in case of an unexpected event during a first visit (unexpected event), ii) the number of documents that are 'double-checked' by the administration (documents double-check), and iii) whether, in the extreme case that a home visit could not take place, the application for a social assistance benefit could still be processed (missed home visit).
A final cluster of three items related to the locus of initiative within the welfare office: i) the potential outreach efforts that the welfare agency might take after receiving alarming information about a household in their municipality (outreach), ii) the potential contact efforts that the welfare agency takes after a client does not open the door during a planned home visit (closed door), and iii) the different possibilities for making an appointment at the welfare office (appointment). The inclusion of the third item can potentially be explained by the fact that one of the possibilities to make an appointment was by referral of another person or organization. Table  in the online Supplementary Materials gives an overview of the items and how they were coded. The questionnaire also included items on the administrative procedure for claiming social assistance in order to measure practices that may be implemented by local welfare agencies if they would strive to reduce administrative process costs for their clients (e.g. the number of documents claimants need to provide themselves, or the number of appointments they typically need to have before their claim is processed). However, we were not able to find a sufficiently large set of practices implemented by local welfare agencies related to this aspect based on the results of the EFA (see the online Supplementary Materials for the results of the EFA).
To support the validity of our results, we examined the clusters of items identified in the EFA by conducting a confirmatory factor analysis (CFA). In particular, we examined whether the grouping of the items in four distinct factors was a good fit for the data and whether each of the items loaded significantly on its latent factor. The results in Table  show that all factor loadings are statistically significant at the  percent level and within the minimum acceptable rangeexcept for the item 'closed door', which is only significant at the  percent level. Nevertheless, we decided to retain this item for further analysis, as the significance of the indicators should not be used as the only criterion for selecting items in case of a non-probability sample and exploratory research. Instead, retaining items that maintain content coverage is important as well (Koran, ). In addition, a minimum of three indicators per factor is generally considered to avoid model identification problems (Kenny, ). When we look at the overall goodness of fit of the integrated model, the theoretical distinction in four separate aspects of social assistance provision seems to fit the observed data well (SRMR = .). Therefore, we can be confident that we are able to use the corresponding latent variables in structural equation models. Additionally, we assessed the correlation between the different aspects, in order to examine whether welfare offices that implement a coherent set of practices related to one aspect also implement coherent practices related to other aspects. The correlation estimates between any pair of social assistance provision aspects are shown in Table . In general, most correlations between the different aspects are low and insignificant. This may indicate that even though municipalities do appear to take coherent actions on individual aspects that are relevant from a non-take-up perspective, they do not consistently do so. Rather, these results indicate that the focus of municipalities may be on the aspects separately, e.g. municipalities may focus on increasing accessibility as a goal in itself, rather than from a broader concern with non-take-up.

Structural equation models
To assess the relation between the local conditions and the welfare office's focus on the different aspects of social assistance provision, we estimated and interpreted for each aspect separately a structural equation model in which we test for its association with municipal characteristics (see the online Supplementary Materials for the full structural model). For two aspects, i.e. locus of initiative and trust, we had to stepwise delete municipal variables and their structural paths from the model until we obtained a model with an adequate measurement model and good model fit. Table  reports the estimated path coefficients between the municipal variables and each of the identified aspects.  The full models, including the factor loadings of the indicators for the latent constructs, can be found in the online material. For each of the four aspects, the specified model fits the empirical data adequately well (SRMR < .) so that we can be confident in interpreting the results substantively. As we use a nonprobability sample, in the discussion of the results, we focus on the magnitude of the effect, rather than solely considering their statistical significance. We find a strong positive association between the percentage of foreigners in the municipality and the degree of information provision by the welfare agency. The standardized coefficient of . shows the degree of information provision in the welfare agency is to a large extent related with the share of foreigners in the population. In addition, we find a moderate negative association between the total number of residents in the municipality and the degree of information provision.
When estimating the full structural equation model for accessibility, the model shows a relatively small, yet statistically significant association between the self-financing margin of the municipality and the physical accessibility of the welfare agency. The effect points towards a positive association between municipalities with a more healthy financial situation and consistent practices implemented at the level of the welfare agency to improve its accessibility.
We furthermore find that the political constellation in the municipality is relatively strongly related with the latent factor locus of initiative. The strong negative effects of right-and centre-oriented municipalities indicate that in comparison with left-oriented municipalities, welfare agencies operating in rightand centre-oriented municipalities leave the initiative of claiming much more in the hands of their claimants and undertake relatively little action to shift the burden of taking initiative to the local welfare agency.
In the final model with trust as the dependent variable, the standardized coefficient of centre-oriented is relatively large in size and points towards a negative association with the extent of trust demonstrated by the welfare agency.

Discussion and conclusion
This article is a first exploration of the variation in implementation practices at the level of the welfare agency, focusing on local welfare agencies in the Flemish region of Belgium. We examined the relation of coherent sets of practices with local contextual factors. We focused on implementation practices in aspects of social assistance provision that are likely important to foster benefit take-up.
Using data from a purpose-collected survey, we found evidence for coherent practices in four domains: information provision, accessibility, locus of initiative and trust established during the application process. An exploratory factor analysis showed local practices on these four fronts to cluster logically together in latent factors. We did not find a similar set of coherent practices with regard to the administrative procedure. Despite many references in the literature regarding the importance of automating or simplifying the administrative procedure, our analysis did not reveal a sufficiently large set of practices related to the administrative procedure. Possibly, our quantitative approach failed to capture the full range of possibilities available to local welfare agencies. In addition, some of the "wins" identified in the literature, such as automated procedures or personalized assistance, are outside the legal scope of variation between Belgian municipalities.
Associations between the aspects in which we did find evidence for coherent actions were mostly statistically insignificant, suggesting that when local welfare offices in our sample implement practices in a specific domain, this occurs relatively independently of practices in other domains that are in the literature known to be important for boosting take-up. This may indicate that local policy actors focus on specific aspects of social assistance provision for its inherent value (e.g. of being accessible), rather than as an instrument in order to improve benefit take-up. Self-evidently, this does not negate the broader implications of local actions in the identified aspects in limiting non-take-up, rather it shows that non-take-up is presumably not the driving concern by local policy makers.
Additionally, we explored whether the political, economic and sociodemographic context in which local welfare agencies operate is related to the presence of a specific and coherent set of implementation practices at the level of the welfare agency, using structural equation models. Table  summarizes our hypotheses and findings.
We found that the political orientation of the municipality relates to the practices implemented at the local welfare office. We found a stronger focus on outreaching practices in municipalities with a left-oriented mayor relative to municipalities with a right or center-oriented mayor. In addition, we found that welfare agencies in left-oriented municipalities are more likely to focus less on control elements throughout the application process than agencies in centeroriented municipalities.
Also, we found a positive relation between the share of foreigners in a municipality and the degree of information provision. Municipalities with a large population of foreigners tend to devote more attention to different practices focused on standardized and multi-modal communication in different languages. Alternatively, we found no indications that a more vulnerable population leads to less attention to the reduction of non-take-up.
Finally, we did not find consistent confirmation of our hypotheses on the relation between more economic characteristics and coherent practices implemented in specific non-take-up reducing implementation domains. We only found a small positive association of a healthy financial situation with more accessible local welfare agencies. Given the local practices that load on our latent factor accessibility, it is indeed likely that certain infrastructural features will require some investment. Relationships with the other aspects of social assistance provision we looked into are small and non-significant. In this case study for Belgium, we highlight the influence of the political context in which the local welfare office operates on a take-up friendly implementation. This finding is in line with evidence for Northern welfare states that the political composition of the local government is among the most important determinants for the variation in implementation of local social security programs (Hanssen et al., ; Lien and Pettersen, ). It is telling that we find in our results mainly evidence for a relation between the local political inclinations and those aspects that reflect specific ideas on the role and organization of the welfare state, such as the extent of trust placed in clients, and how outreaching a local welfare agency has to be. Coherent actions on other aspects of social assistance provision, such as information provision and accessibility, appear to be mainly driven by practical concerns, such as the local socio-demographic context and local finances. Although this study focuses on Belgium, it can be expected that also in other countries local variation in implementation practices exists. Frazer and Marlier (, p. ) highlight the large role of the local level in delivering minimum income protection in sixteen other EU countries, including the Netherlands and Sweden.
Our study has a number of limitations. First, our approach only allows the analysis of associations between the contextual municipal variables and relevant sets of coherent practices, and it is therefore not possible to say anything definite about causality. We do not establish a precise direction, although implicitly, we expect the context factors to (co-)determine the efforts that local welfare offices invest in the different aspects of social assistance provision. For instance, regarding the association between the share of foreigners in the municipality and the degree of information provision in the welfare agency, one cannot exclude that the direction is ultimately reverse: residents with a foreign nationality may selfselect in municipalities with local welfare agencies that communicate better. Given the high costs of moving, the fact that social assistance benefit levels are equal over the Belgian territory, and other reasons migrants take into account when deciding to move (such as the presence of family and job opportunities (Åslund, )) we deem this less likely. But additional research is needed in order to pin down the exact direction of the relationship. Second, with our survey we tried to capture subtle differences in approach and treatment of clients. For some actions, this is more obvious than for others. Essentially, this leads our analysis to be based on those indicators that are more easily quantifiable, and perhaps neglecting other, equally important practices. In addition, the use of variables that are often based on qualitative information in combination with the relatively low number of municipalities have an impact on the strength of our models. Finally, we limit ourselves to local practices at the welfare agency's level, whereas it has been shown that interaction with case managers and the degrees of freedom a case manager is awarded by her local welfare agency is of primordial importance for the individual claimants' experience. This limitation in scope allowed us however to quantitatively zoom in on aspects solely or mainly in the realm of local welfare agencies (such as their opening hours), and how these are part of a coherent approach.
Overall, our findings do provide insight in the large local variation in social assistance provision. We focused on those aspects that may have an impact on the ways potential claimants experience their interaction with the local welfare agency. Whereas literature on the drivers of non-take-up does acknowledge the importance of the local welfare agency in discouraging potential claimants, our study adds to this research by quantitatively illustrating the variation in local practices that may impact on prospective clients' claiming behaviour. In addition, we explored how this variation relates to local characteristics.
An exciting agenda for further research can look into the relation between local social assistance provision practices and local non-take-up rates. In addition, we should look further into the degrees of freedom left to social workers by different local approaches regarding social assistance provision. Finally, repeating this investigation in countries with a different division of competences between the national and the local level should give us more insight in the extent to which the institutional set-up of a country's welfare system is tied to local variation in implementation.

Supplementary material
To view supplementary material for this article, please visit https://doi.org/. /S