Housing quality and homelessness among people who beg

Abstract

The poverty literature often overlooks the relationship between housing insecurity and income-generating activities. The often informal sources of income among the homeless strongly connect to other aspects of their lives. Nonetheless, these groups of severely deprived people are often missing from large-scale studies and official statistics. Using an original and comprehensive survey, we document the complex interaction between living conditions and begging, a most visible form of extreme poverty and a central concern for urban planning and social assistance. We interviewed 358 people who were begging between 2021 and 2022 in Brussels. We estimate that approximately 80% of them are Romanian Roma. Our findings suggest that around half of the people who beg sleep rough, while some occupy relatively insecure and low-quality dwellings. Further differences relate to gender and ethnicity, with women and those with children more often seen in emergency accommodations and non-conventional dwellings and Romanian Roma occupying, on average, more problematic housing.

Keywords:

• Homelessness
• begging
• panhandling
• ethnicity
• Roma
• Romanian

Introduction

This contribution aims to document the living conditions of a group of severely deprived people. In particular, it seeks to connect their insecure and problematic housing situation to several other aspects of their lives, focusing on begging as a means of subsistence. This approach is called for, and it does so for several reasons concurrently. We argue that there are at least three reasons why we need to document connections between the income and housing situations of the most deprived: surveys on poverty too often lack groups like the roofless or people who beg; the relationship between the housing situation and income-generating activities is often neglected in studies into the living situation of the poor; some stakeholders have recently cast doubt about the housing problems of people who beg (Hermer, 2019).

First, this contribution covers a group often not represented in large-scale datasets aiming to capture poverty in society. There are several legitimate, or at least understandable, reasons why this gap between large-scale data and forms of more extreme deprivation is happening. Take, for instance, EU-SILC, the European general population dataset that is most often used to measure poverty, but often misses the poorest in society. First, the sampling basis is the household (Bradshaw & Movshuk, 2019), but some, such as the roofless, do not live in stable and registered households. Some people are also unregistered, for instance, people in insecure housing situations, undocumented migrants, and groups that travel around. Most sampling frames are unable to identify or locate these individuals or groups. This systematic lack of representation of the extremely deprived makes instruments such as the EU-SILC less than ideal for documenting the lives of the most excluded. Some scholars have started collecting survey data comparable to the EU-SILC with these groups (e.g. Nicaise et al., 2019).

Secondly, the scholarship on homelessness, on the one hand, and documenting income-generating activities, on the other, are too often in separate research universes. This particularly holds for advanced economies. Homelessness is quite often treated as if it is void of people who desperately need an income, while lack of income is a route into homelessness of one kind or another. But once people are roofless or suffer from a lack of secure and quality housing, the connected mobility typically spills over into flexible and short-term income-generating activities (e.g. Venkatesh, 2006, pp. 168–169). This connection between housing insecurity and means of subsistence is too often disregarded. Nevertheless, the typical sources of income among the homeless may be quite diverse, with a concentration on the informal economy and a strong connection between both aspects of their lives (Ferguson et al., 2018; Flåto & Johannessen, 2010). To better understand people’s (lack of) housing and their income-earning choices, the connections may bring important insights.

Finally, begging has been at the centre of some debates depicting absolute poverty, although it was usually shrouded in a strange mix of city marketing and social assistance discourses related to the required begging regulation and policy in cities and towns. Some scholars have argued that these regulations, on average, are uninformed because they are based on very thin, if not non-existent, evidence about the group whose behaviour is regulated (Leeson & Hardy, 2022). For instance, in English and Welsh towns, a wave of ‘diverted giving’ campaigns started in the mid-1990s, built on the idea that people who begged were usually not deserving. They argued that the public should give money to social assistance organisations that help “real” homeless people (Hermer, 2019). One important element was the recurring assertion in the campaigns that those who begged and those who were roofless were not the same populations. These assertions were built on rather sandy ground. This public debate about the housing status of people who beg provides a third rationale for empirical studies of the connection between begging and housing.

These arguments together warrant the central goal of this contribution, which is to study people who beg from the perspective of their housing situation. We will document the interaction between begging and housing in the begging population in Brussels, Belgium, and answer the following research questions:

1. What is the housing situation of people who beg?

2. To what extent do the characteristics of people who beg and their income-earning strategies (such as gender, age, or health) relate to housing situation and quality?

We do so by collecting original data through an exceptional questionnaire survey from the population of people begging in Brussels.

The rest of the paper is organised as follows. The next section proposes a literature overview and presents the case of Brussels as the setting for our study. Section 3 introduces our primary data and methodology. Section 4 encompasses the results and their interpretation. Finally, our paper concludes by discussing and summarising our findings.

Literature review

In both the popular imagination and academic literature, the line between begging and homelessness is often blurred (Mansour, 2017). Empirical research suggests that most people who beg are homeless and have lived longer on the streets than others (Leeson & Hardy, 2022). The reverse does not hold: most people without a home do not engage in panhandling activities. A large-scale representative 1996 survey of the US population estimates that 10% of the homeless engage in begging (Lei, 2013). Estimates seem to vary, though. Ferguson et al. (2016) studied 600 homeless people in 3 US cities and found that one in two beg. Conversely, in Toronto, Bose and Hwang (2002) found that within a sample of 54 panhandlers, 65% did not have a home. Studies show that out of all types of homelessness, begging is most strongly associated with sleeping rough (Dean & Melrose, 1999; Fitzpatrick & Kennedy, 2001). In a mixed sample of 66 rough sleepers, people who beg and street magazine vendors in Edinburgh and Glasgow, 77% have engaged in both rough sleeping and begging; for most, rough sleeping happened first (Kennedy & Fitzpatrick, 2001). In Oslo, out of 374 homeless Romanians who reported earning an income from begging and other street work, 69% spent the night before the interview outdoors (Tyldum & Friberg, 2023).

This strong connection and frequent overlap between homelessness and begging may pose difficulty in delineating the two empirically. Begging can take different forms, so inconsistencies can arise between who we think is a person that begs and who actually is; this often leads to conflating homelessness with begging and vice versa (McIntosh & Erskine, 1999).

One broad way to define begging is asking for alms in the public space (Dean & Melrose, 1999; Fitzpatrick & Kennedy, 2001). This usually takes the form of a one-sided, non-reciprocated exchange between almsgivers and those who beg. There is often variation in terms of the begging technique, ranging from vocally addressing passers-by to using a sign or a cup, but also in terms of begging profiles (Bose & Hwang, 2002; Ferguson et al., 2016). Activities like busking or peddling, implying some form of market exchange, are sometimes considered begging (Hawkins, 2012; Östman, 2023), but some scholarship delineates them from this category (Dean & Melrose, 1999). While the lack of apparent reciprocity makes the typical begging encounter an ambivalent experience (McIntosh & Erskine, 1999), there may still be a more general type of reciprocity at play: a sense of satisfaction ("warm glow") after giving alms to a person in need (Andreoni, 2006; Muñoz & Potter, 2014).

There have been debates on whether to label begging as work. Some associate begging with idleness and avoiding (formal) labour (O’Neill, 2017). Work, however, in its fundamental meaning, is a purposive activity aimed at providing. Along this logic, other scholars define begging as work, albeit an informal economic activity (Adriaenssens & Hendrickx, 2011; Dean & Melrose, 1999). Individuals who beg themselves often see begging as work, not just because of its income-generating logic, but also because they associate it with work routines, such as a place of work and a working schedule, a feeling of self-worth and a stepping stone to better prospects (Kassah, 2008; Lenhard, 2021; Swanson, 2010). For these reasons, we refer throughout this study to begging as work.

Despite being a means of subsistence, arguably a way of earning an income like many others, begging is also an activity that carries a stigma. Debates on the legitimacy of those who beg have been around for centuries (Hitchcock, 2005), swinging back and forth between a social and a criminal policy problem (Hermer, 2019). Policy responses have ranged from diverted giving campaigns in the UK to outright bans and criminalisation (Kassah, 2008), whereas in some countries, begging has become recognised as a human right (Commère et al., 2023). Notwithstanding a widespread belief that people who beg are conmen, most associate the phenomenon of begging with the extreme poor (Adriaenssens & Hendrickx, 2011; McIntosh & Erskine, 1999).

Homelessness, on the other hand, is broadly defined as a lack of access to a stable residence. Aside from sleeping rough in the street or other public spaces, people who squat, rely on night shelters and other types of emergency accommodation or couch surf with relatives and friends can also be considered homeless (Pleace & Hermans, 2020). Often insecure or inadequate, these living situations are not mutually exclusive, given the ever-changing dynamic. Personal characteristics such as gender can play a role, where women tend to have more diverse survival strategies (Bretherton & Mayock, 2016). Disabled individuals, one of the vulnerable groups within the homeless population (Baker Collins et al., 2018), are often associated with chronic homelessness, with more frequent stays in homeless shelters (Stover-Wright, 2022). Homemaking and homelessness can be related to ethnic groups, too. For example, many Roma live in close-knit communities sharing lodging with kin (Smith & Greenfields, 2013), with frequent overcrowding (Szabó, 2018). This pattern is also observed among other groups such as First Nations people (Dockery et al., 2021). Besides the poor-quality housing, Roma families may suffer from exploitative landlords and high levels of stress and discomfort (Cioarta, 2023).

It should come as no surprise then that reaching an agreement on the definition of homelessness has been virtually unattainable internationally. Nonetheless, Europe’s most influential conceptualisation of homelessness is the European Typology of Homelessness and Housing Exclusion (ETHOS). ETHOS proposes a standardised way of understanding, categorising, and measuring the concept of being homeless. The development of this conceptual model allows for an operational definition adaptable to location-specific circumstances (Edgar, 2012). ETHOS defines three dimensions: the physical domain (adequate private living space), the social domain (privacy and the space needed to live as a household in the way of their choice) and the legal domain (a legal title to the occupation) (Edgar et al., 2007). These three domains jointly define what a home is. Nonetheless, one of the shortcomings of this typology is its lack of exhaustiveness. Criticism is often pointed towards not accounting for differences in similar living situations across countries, such as weaker or stronger tenancy rights in boarding houses or camping grounds (Amore et al., 2011). Whether someone has insecure housing might depend more on the local and personal context and less on the typology of their living situation. This ultimately makes the point as to why the definition of homelessness has been a long-debated topic and a fundamental problem with critical implications for its measurement. As Somerville (2013) emphasises, homelessness needs to be understood as multidimensional and storied and research needs to look at the whole life of a homeless person, rather than just at selected episodes of rooflessness.

Access to stable housing may be influenced by the individual’s labour market situation and general labour market conditions (Gowan, 2010; Johnson et al., 2019). Rowe and Wolch (1990) argued that becoming homeless will accommodate the individuals’ earning strategies – sometimes towards the street economy. Research supports this idea of a push towards informal activities such as begging, petty crime, sex work and the drug trade (Adriaenssens & Hendrickx, 2011; Ferguson et al., 2018). The relatively low level of generated income and prestige combined with easy access to the activity may explain why predominantly homeless people self-select into begging (Smith, 2005). Leeson and Hardy (2022) argue that begging also brings some advantages – people who beg can control their schedules and have no superiors to answer to – so that for some low-skilled workers, begging may be "the least-bad employment option".

Nonetheless, there is some debate on how begging affects the housing situation of those in precarious conditions. Lenhard (2021) theorises begging as a "labour of hope" as it also integrates daily routines and develops skills. Others label begging as a dead-end earning strategy which impedes escaping homelessness. Reinhard (2021) reviews the literature on begging income and concludes that most studies report it to be only a fraction of median income in formal labour markets and below the average rental cost of a studio apartment (see also Adriaenssens & Hendrickx, 2011). Bose and Hwang (2002) find that respondents who beg spend all their income on consumption. Those who beg may consider begging to be a dead-end themselves (Gloria & Samuel, 2012). Moreover, adaptation to a street lifestyle and economy may make it more difficult to escape homelessness (Tyler & Johnson, 2006).

Migration is vital in understanding the dynamics between begging and homelessness. One pattern is Central and Eastern European economic migrants seeking low-skill income-earning opportunities to send remittances home (León-Ledesma & Piracha, 2004). Migrant networks in receiving countries play an important role, as they reduce the costs of obtaining information on employment opportunities, the social system and so on (Munshi, 2003; Pedersen et al., 2008). However, these migrants often face unstable employment situations, such as informal labour, low-quality housing and poor working conditions (Hermans et al., 2020). This pattern is similar to the fate of those engaging in informal street-level economic activities, particularly begging (Tyldum & Friberg, 2023). Among the begging populations across the urban centres in Western and Northern Europe, Central and Eastern European Roma are consistently present (Hansson & Mitchell, 2018; Matras & Leggio Daniele, 2018; Tyldum & Friberg, 2023). The lack of formal opportunities and the heavy discrimination in their home countries make begging an attractive income-earning activity. It is often practised akin to seasonal work by the Roma (Friberg, 2018), notwithstanding the poor living conditions these migrants have once abroad (Cioarta, 2023; Tyldum & Friberg, 2023).

Who is Roma and who is not is a thorny question. As an ethnic-linguistic minority, one should first and foremost distinguish being Roma from being a travelling or itinerant population. Many of these groups are defined by outsiders as ‘Gypsies’ (Matras, 2013). Further, it is far from a straightforward or ‘natural’ categorisation (Ivanov, 2012). For instance, some Central European groups refer to themselves with other group concepts, sometimes denying their Roma roots (Marushiakova & Popov, 2016). Depending on the measurement approach and the criterion used, the outcomes of any survey-based measure of Roma populations can diverge widely. Most surveys gauge whether respondents are Roma through self-reports, while others let fieldworkers or geographical aspects (living in a hamlet that is deemed ‘Roma’) decide. Janka et al (2018), in a recent health survey in Hungary, compare both approaches and conclude that the number of Roma increases by almost 50% in the latter case.

The case of Brussels

The Brussels-Capital Region (from here on referred to as Brussels or the region) is the largest urban area in Belgium, standing out in its ethnically and linguistically more diverse social structure than other urban areas. It has high levels of both urban flight and external immigration (Observatorium voor Gezondheid en Welzijn in Brussel, 2018). Brussels is also an example of the "urban paradox": a centre of capital growth, income acquisition, and innovation, but also harsh living conditions, substandard housing, and poverty (Eurostat, 2016). As one of Europe’s main migration hubs, Brussels has attracted low-income immigrant workers, often Eastern Europeans, since the enlargement of the European Union (Verhoest et al., 2022). As Brussels is an arrival city, a third of the population exchange between Belgium and other countries occurs through Brussels.

One result is the visibility of extreme poverty amid abundance, amongst others, in the forms of begging and rough sleeping. This regularly leads to public outcries and considerable debate about the exclusion, neglect and abuse of vulnerable groups such as minors (Boutte & Erpicum, 2017). Children in these situations, often of a migrant background, are exposed to the dangers of life on the streets and sleeping rough (Van Houcke, 2005). The Brussels public transport network (STIB) banned begging in 2007. The City of Brussels passed a regulation prohibiting begging for children in 2022, which attracted sharp criticism from poverty experts and organisations working with Roma communities. The most prominent points were the inefficiency caused by shifting the problem to other municipalities within the region and the harshness in punishing poor migrant families that abide by different cultural norms when it comes to child-rearing ("La Ville de Bruxelles fait une erreur", 2022). Belgium plans to allocate more budget to intercultural mediators between the local Roma communities, authorities, social services, and support in navigating the job searching process (European Commission, 2023).

The only available systematic research on begging is a survey in Brussels conducted 17 years ago (Adriaenssens & Hendrickx, 2011; Clé & Adriaenssens, 2007). One of the notable conclusions of this survey was that the living conditions and background are fundamentally different between two groups of people who beg: one is a predominantly male, roofless local population and the other one is comprised of Roma individuals, who were primarily female, sometimes accompanied by children and who migrated recently from Romania and remained quite mobile between the countries. These results have led us to believe that there is a connection between these profiles and their typical housing situation, with the Roma having access to low-quality, insecure and overcrowded housing. The relationship with passers-by was ambivalent: respondents reported encountering both friendly and negative reactions. The relationship with the authorities was equally complicated: respondents felt they could not rely on police to protect them besides avoiding them due to previous negative experiences.

Data and methods

In this research, we are using original data collected via a standardised, face-to-face survey directed at people begging in Brussels in November 2021-January 2022 and April-May 2022.

Survey design

The survey (available in the online appendix) aimed to collect information on our respondents’ socio-economic characteristics, begging behaviour and quality of life indicators, such as health, living conditions and well-being. The questionnaire was based on two main sources: the 2005-6 survey of people begging in the region and the European Survey on Income and Living Conditions (EU-SILC). The former originates from a research project led in 2005 and was mostly used to replicate questions on begging activities (Adriaenssens & Hendrickx, 2011; Clé & Adriaenssens, 2007). From the latter, we sourced items gauging respondents’ housing situation, living standards and multidimensional deprivation. Other sources for our questionnaire include international household surveys such as DHS and MICS and the European-level Fundamental Rights survey.

To capture respondents’ living situation, we distinguished in our survey those who had a "fixed" place of residence from those who did not. Even though possibly unstable, for example, in the case of illegal sublets, we aimed to create a clear distinction between respondents who had the same place of accommodation to go to at the end of each day, albeit potentially insecure or inadequate, and respondents who had to figure out a sleeping arrangement on a regular basis, often resorting to public places or night shelters. Secondly, respondents in the former category were inquired about the type of accommodation they occupied, while those in the latter category were asked about their most frequent sleeping place in the previous two weeks.

The source questionnaire was developed in English, translated, and back-translated to the most prevalent languages among the begging population in Brussels: French and Dutch, the region’s official languages, and Romanian. Our results from 2005-6 and initial exploratory observations before our fieldwork started indicated the Romanian Roma to be the largest group in this population. In addition to the questionnaire, we developed an observation sheet (to be filled in by the fieldworker before and after each interview), which systematically collected qualitative information about the respondent and their surroundings.

As indicated, it is hard to measure validly and reliably who among the respondents can be categorised as Roma. Self-reports have the distinct advantage of capturing people’s identities as part of an ethnic group. The major downside of self-identification is that they may underestimate their actual status due to social desirability bias, as supported by a recent study comparing expert local estimations and self-reported census data (Csata et al., 2021). For our research, we chose not to ask respondents about their ethnic identity directly. Several arguments support that people who beg in Brussels of Romanian nationality are predominantly, if not solely, of Roma descent. Virtually all respondents (98%) who spontaneously reported speaking Romanes were respondents of Romanian nationality. Further, Brussels experts on Roma matters and social workers dealing with street populations in Brussels confirmed that the Romanians begging are predominantly of Roma descent. Finally, anecdotally, some respondents indicated they were Roma, while none of the Romanians interviewed argued that they were not Roma. It remains important to remind the reader that Roma identity is multifaceted, often a mix of ethnical, cultural and national dimensions (Gkofa, 2017).

Sampling strategy

The exceptional nature of the data collection imposed several challenges to its course. As the literature section discusses, multiple definitions can be attached to the concept of begging. Inevitably, we faced difficulty deciding on a uniform way of identifying possible respondents. We focus on the population asking for a non-reciprocated gift and thus exclude busking, peddling or other street activities, implying a market exchange (Adriaenssens & Hendrickx, 2011). For ethical reasons, we only considered respondents of 18 years and older.

Developing a sampling strategy is of key importance in any survey, especially in research into hidden or hard-to-reach populations. A random sample of the population of people begging in Brussels is complex due to the absence of a pre-existing sampling frame. Moreover, begging is quite a volatile activity. Because people who beg are foremost identifiable by their activities in the public space, a so-called "venue-based sampling" is in place (also named location, time-space, centre, and intercept sampling, see Kalton, 2014). As the sampling frame, a list was constructed of locations where begging occurred a few weeks before each data collection wave.

The list was built with the help of two working students and volunteers who were asked to input in a smartphone app each time they observed begging, documenting the exact date, hour, address, number of people and gender. The hired students were required to follow walking and subway routes across the region, alternating the starting hour and day of the week. Over 500 observations of people begging were registered, typically in proximity to commercial areas, places of worship, public transport stops, touristic attractions and high-traffic areas. The data were securely stored and solely accessed by the research team.

The central challenge is to infer conclusions about the population of people who beg from locations where begging occurs (the sampling frame) in the period where these locations are monitored. As begging is often transitory and mobile, the probability of a person who begs to be interviewed depends very much on the timing of the fieldwork. To maximise the chance of being selected throughout the year, we conducted two census-like survey waves in different seasons (November 2021-January 2022 and April-May 2022), where all people encountered begging were asked to participate. During each data collection wave, fieldworkers were sent to each begging location in the sampling frame list. We maximised the chance of being found by covering a begging location at least twice and at different moments during the day and the week. This strategy approximates the representativeness of the population of people who beg throughout the year.

Survey distribution

Because of the vulnerability of the group and the overall expectation of increased levels of distrust towards outsiders, the approach of respondents and the potential incentives for cooperation were carefully designed. Given the possibility of contacting illiterate and semiliterate respondents, we conducted face-to-face interviews prompted by oral informed consent from each participant.

Monetary incentives can effectively incentivise take-up and completion in hard-to-reach respondents (Singer & Kulka, 2002). In line with the 2005–6 survey, respondents were paid €10, €5 for agreeing to participate (before the survey started), and another €5 if the survey was completed, as prepaid incentives have the most substantial effect (Toepoel, 2012). The rationale was also to compensate for the working time lost by participating. In the case of group begging, each adult person was interviewed and compensated. In light of the vulnerability of the population and the sensitivity of some of the data, respondent anonymity was ensured at all costs.

The recruitment of interviewers was of key importance due to the linguistic and risk profile imposed by the work’s nature and the perceived social distance toward individuals who beg. In total, ten fieldworkers collected data. The training was aimed at following protocol to reduce interviewer effects but also at increasing motivation through information about the population, the social relevance of the study, and ethical dimensions regarding informed consent and non-disclosure (Berry & Gunn, 2014). The university’s ethical review board approved the project before data collection started.

Data

In total, 358 respondents were interviewed, and 186 individuals were approached but did not or could not participate. The refusal rate is low, but the average age is significantly higher for non-participants (Table 1). Only 2% of interviews were interrupted due to the respondent losing interest or a complete lack of comprehension between the interviewer and the respondent. Fieldworkers recorded whether a respondent was (potentially) under the influence of substances at the time of the interview. They continued the interview if the respondent could understand and answer the questions. 9% of respondents were reported as either having an alcohol smell and/or bottle at the time of the interview. Only 2 of the interviews had to be interrupted due to comprehension problems. Most nonresponses were related to the inability to communicate due to unforeseen language barriers, as we encountered small minorities of Bulgarian, Polish, Turkish and Arabic speakers or cognitive impairments. Despite occasional linguistic barriers, we are confident that our sample is representative of the begging population in Brussels. The comprehensive sampling frame and extensive variation in interview location and moments lead us to believe that we approached almost every person begging in the region who fit our definition during the two data collection waves.

Table 1. Response rates by subgroups.

	In the sample	Unable to respond	Refusal	Other reason	Total
Gender
Male	199	65	17	35	316
	63.0%	20.6%	5.4%	11.1%
Female	158	39	15	14	226
	69.9%	17.3%	6.6%	6.2%
Average age	41	45	50	39	534

Note: Chi-square value (χ²) is 5.54 for gender (p=.136). Anova F = 6.19 for average age (p<.001).

The interview duration was, on average, 50 min, with some respondents taking more than 2 h to complete the survey. The respondents’ degree of talkativeness, literacy level and language comprehension were the most influential factors. Certain questionnaire items required further clarification for some respondents, and there were occasional item nonresponses among respondents.

Table 2 shows descriptive statistics for the main variables. The distribution among male and female respondents is almost equal. Among the age groups, we notice that those older than 50 make up only 27% of our sample size. It is worth noting that, frequently, some respondents had trouble recalling their exact age or birth year due to an overall lack of access to formal education throughout their lives. In such cases, we relied on approximations provided by the respondent and the fieldworker’s estimation. 77% of respondents were of Romanian origin, while the others were primarily Belgian or Central European. 61% were begging alone at the time of the interview. The others were accompanied by a partner, a larger group, a child or a pet. It is also common that the respondent was part of a close-knit group of people begging in close proximity. They often checked in on each other during working hours, sharing food or changing locations together. Among those who beg with children, 97% are of Romanian origin (illustrated in Table A3 in the online appendix). Almost 13% of respondents reported receiving some form of benefit, either in Belgium or in their countries of origin. 34% of our sample reported suffering from long-term health problems, 11% were observed with a physical disability, while 15% struggled with alcohol or drug addiction. Disability was often observed as a physical or mobility impairment, with respondents usually relying on canes and, in the extreme case of upper or lower limb loss, on wheelchairs.

Table 2. Personal characteristics of respondents.

	Frequency	Per cent
Male	199	55.7%
Female	158	44.3%
Gender: Total	357	100.0%
Age: 18–30	117	32.7%
Age: 31–50	145	40.5%
Age: 51+	96	26.8%
Age: Total	358	100.0%
Health problems: no	224	65.9%
Health problems: yes	116	34.1%
Health: Total	340	100.0%
Disability: no	319	89.1%
Disability: yes	39	10.9%
Disability: Total	358	100.0%
Addiction: no	304	85.4%
Addiction: yes	52	14.6%
Addiction: Total	356	100.0%
Other than Romania	81	22.6%
Romania	277	77.4%
Country: Total	358	100.0%
Begs alone	218	60.9%
Begs in a group (no children)	70	19.6%
Begs with children (possibly in a group)	70	19.6%
Group: Total	358	100.0%
8 h or less	193	60.7%
More than 8 h	125	39.3%
Hours: Total	318	100.0%
Benefit: no	299	87.2%
Benefit: yes	44	12.8%
Benefit: Total	343	100.0%

More than half of the respondents reported that they begged fewer than eight hours on a typical working day (Figure 1). The lowest value in our sample was 45 min. Three extreme values (fewer than 1% of the observations), with more than 16 reported begging hours, were recoded as 16. We decided to categorise this variable in order to have a clearer overview of the effect of working long hours (>8 per day).

Figure 1. Distribution of the number of hours begged.

Note: The mean number of hours begged is 7.8 (n = 318).

We categorised our respondents using the ETHOS light framework, which has six operational categories: rough sleeping (also referred to as "rooflessness", sleeping in external or public spaces), emergency accommodation (overnight shelters), accommodation for the homeless (homeless hostels, temporary accommodation, women’s shelters or refugee accommodation), institutions (healthcare or penal), non-conventional dwellings (mobile homes/caravans, non-conventional buildings such as squats or garages, temporary structures) and temporary conventional housing with family and friends.

We ran into the difficulty of appropriately classifying in ETHOS the respondents that had what we referred to in our questionnaire as a "fixed" place of residence but did not rely on forms of municipal or social housing. They often occupied housing such as apartments and single-family houses for longer. Concurrently, our fieldwork notes suggest a considerable chance of predominant illegal sublets among the population, especially for foreigners, who often rely on personal networks to find accommodation. We strongly suspect a significant proportion of these respondents do not have a valid residence permit or the means to obtain one, as they need a (formal) source of income for that. Begging is an informal activity often not supplemented with formal employment. This could be a typical impediment for whoever seeks a legal rental agreement.

Moreover, due to the sensitivity of such questions and the related risk of eliciting socially desirable answers, we decided not to include specific questions on the legality of their stay or their rental arrangements. Based on this line of thought, we considered the legal domain of ETHOS to be virtually absent, rendering these respondents as non-conventional dwellers and their dwellings as insecure. We find it reasonable to assume that the overwhelming majority of survey participants falling under this category experience insecure housing conditions, typically as illegal subtenants.

Statistical analysis

For most of our analyses in the Results section, we use an indicator variable for rough sleepers and, later on, illustrate the housing conditions of respondents with insecure or inadequate housing. We generally focus on bivariate relationships, significance tests and multivariate regression analyses. We present linear models and, as a robustness check, non-linear (logistic) specifications. When the dependent variable is binary, it is implied that the linear specification is a linear probability model (LPM), allowing for a straightforward interpretation of the estimated effect sizes. In that case, we always use robust standard errors. In most regression analyses, we add control variables on gender, age, health problems, disability, addiction, group begging and data collection wave. To control for bias in sampling probability, we also account for the self-reported average number of hours spent begging on a typical day. Even though we used a standardised questionnaire and our quantitative analyses rely on the resulting dataset, our fieldwork notes have the potential to shed more light on and nuance our results. Therefore, we strive to provide some qualitative information derived from our fieldwork notes in our next section whenever relevant.

Results

Table 3 illustrates the number of respondents in each ETHOS operational category. Within our sample, 188 (53%) sleep rough, most often in the streets or in other public spaces such as parks, shopping centres, train stations and metro stops. Twenty-nine respondents (8%) fall under the second category of emergency accommodation, and four respondents (1%) live in accommodation for the homeless, more specifically in hostels, shelters (maison d’accueil) or medical centres (maison medicale). During the interview, none of our respondents reported living in healthcare or penal institutions (the 4th ETHOS light operational category). The second largest category in our dataset consists of 116 people (33%) living in non-conventional dwellings due to a lack of housing or insecure or inadequate living conditions. We refer here to squats (10), garages, personal cars and other places unintended for habitation (6), mobile homes (1) and, most notably, apartments and single-family houses (99). The latter, making up the majority of this category, is not considered conventional housing due to the respondents’ assumed lack of a legal title to the occupation and their consequent exclusion from the legal domain, rendering it insecure housing. Lastly, only 10 respondents (3%) reported temporarily living in conventional housing with family and friends and 7 (2%) do not fall under any of the ETHOS light operational categories, officially qualifying as stable dwellers. Accounting for less than 2% of our sample, 4 of them occupied municipal or social housing, 2 obtained legal rental agreements with the help of local organisations and 1 respondent was living in a nursing home. 4 respondents were Belgian, the remaining 3 having other European citizenships.

Table 3. Distribution of respondents per ETHOS-light category.

	Frequency	Per cent
1. People living rough	188	53.1%
2. People in emergency accommodation	29	8.2%
3. People in accommodation for the homeless	4	1.1%
5. People living in non-conventional dwellings due to lack of housing	116	32.8%
6. Homeless people living temporarily in conventional housing with family and friends (due to lack of housing)	10	2.8%
7. Stable housing	7	2.0%
Total	354	100.00%

Table 4 entails the proportions of those sleeping rough, in emergency accommodations and in non-conventional dwellings grouped by the categories previously presented in Table 2. The larger proportion of female respondents occupying non-conventional dwellings could be related to the subjective and objective risks of experiencing rooflessness as a woman. As such, women might avoid situations that would expose them to rooflessness.

Table 4. Relationship between personal characteristics and sleep location.

	Rough sleeping	Emergency accommodation	Non-conventional dwelling	Elsewhere	Total	N
Gender
Male	64.3%	6.1%	21.9%	7.7%	100.0%	196
Female	38.9%	10.8%	46.5%	3.8%	100.0%	157
Age
Age: 18–30	55.6%	12.0%	29.1%	3.4%	100.0%	117
Age: 31–50	55.6%	9.0%	30.6%	4.9%	100.0%	144
Age: 51+	46.2%	2.2%	40.9%	10.8%	100.0%	93
Health problems
Health problems: no	55.8%	9.8%	29.5%	4.9%	100.0%	224
Health problems: yes	50.0%	5.2%	37.9%	6.9%	100.0%	116
Disability
Disability: no	56.2%	8.9%	30.5%	4.4%	100.0%	315
Disability: yes	28.2%	2.6%	51.3%	17.9%	100.0%	39
Addiction
Addiction: no	52.6%	8.6%	33.8%	5.0%	100.0%	302
Addiction: yes	56.0%	6.0%	26.0%	12.0%	100.0%	50
Country
Other than Romania	53.2%	6.5%	27.3%	13.0%	100.0%	77
Romania	53.1%	8.7%	34.3%	4.0%	100.0%	277
Group
Begs alone	52.6%	4.7%	34.9%	7.9%	100.0%	215
Begs in a group (no children)	65.2%	8.7%	23.2%	2.9%	100.0%	69
Begs with children (possibly in group)	42.9%	18.6%	35.7%	2.9%	100.0%	70
Benefit
Benefit: no	55.2%	9.0%	31.8%	4.0%	100.0%	299
Benefit: yes	34.1%	4.5%	40.9%	20.5%	100.0%	44

Note: Chi-square values (χ²) are 31.1 for gender (p<.001), 13.3 for age (p=.038), 4.7 for health problems (p=.194), 22.1 for disability (p<.001), 4.8 for addiction (p=.187), 9.5 for country (p=.023), 21.8 for group (p=.001), 22.1 for benefit (p<.001).

A similar pattern is observed among disabled and, to a lesser extent, ill individuals, who also have a lower incidence of sleeping rough. During our data collection process, we often noticed that those affected by an illness keep the medicine they take next to or inside their cups. Among the older respondents, health is often cited as a reason to beg, as they cannot afford their prescribed medication. This was also recurrent among Romanians receiving sickness benefits for themselves or a family member from their local authorities. Virtually all reported health problems were physical ailments, and in the more extreme cases, some respondents offered to show their scars or surgical wounds to justify themselves.

In addition, although only a small minority of those who beg use emergency accommodations (shelters), the share of women using them is almost double the share among men. This could also be related to different experiences in overnight shelters. On average, male respondents reported more incidents of overcrowding, unsanitary conditions and unpleasant atmosphere caused by conflicts with other residents, whereas women, especially when accompanied by children, report highly positive experiences, often pleased by the space, clean facilities and toiletries.

Respondents who received a benefit often reported the amount to be too low to support one individual, let alone a family, leading them to resort to informal activities such as begging as a means of living. Nonetheless, more than half in this category live in non-conventional dwellings, temporary accommodation or more stable residences such as municipal and social housing.

Furthermore, a strong relationship between the number of hours worked and the respondents’ housing situation is apparent: rough sleepers beg 8.8 h on average compared to 6.6 h for others (results not in the table; t = 5.6; p<.001). This relationship between sleeping rough and working hours may be due to the blurriness of the barrier between working and off-hours. Rough sleepers were observed to place their cups in a visible spot next to their sleeping arrangement, resulting thus in longer but less strainful working hours from early morning until late evening. Some of these respondents remain mobile and roam during the day on very short distances with their belongings packed in suitcases. Our field notes suggest that respondents from ETHOS categories 1, 2, 3 typically had a sleeping location close to their place of work or at least within the region. Those with a "fixed" address (ETHOS categories 5-6 and those in formal housing) reported postal codes within the region. Virtually no respondents face long or difficult commutes to their place of work.

The environment friendliness is also an element of importance. Our results show that 76% of Roma respondents were chased away while begging, compared to 57% of the non-Roma respondents (χ²=10.35; p = 0.001), most often by the police and passers-by. Only 12% of the Roma respondents and 34% of the non-Roma (χ²=20.93; p < 0.001) reported physical aggression. Regarding police satisfaction, 62% of the Roma respondents were very or extremely satisfied, compared to 82% of the non-Roma (χ²=9.42; p = 0.002).

The proportion of Romanian individuals who beg and sleep rough (53%; n = 277) is almost the same as the proportion among those of other origins, something which is strikingly different from the situation in 2005 and 2006, when only 8% of Romanian respondents were rough sleepers (n = 144). This difference in proportions is statistically significant (${\chi }^2$=84.4; p<.001).

Table 5 illustrates the results of a linear regression analysis that aims to reveal which personal characteristics or begging strategies (e.g. gender, age, health, working hours, etc.) are related to the likelihood of sleeping rough. The linear regression results confirm the findings in the literature, showing that both women and those with a physical disability are less likely to be rough sleepers compared to their counterparts by 20 percentage points (ppt). Moreover, respondents accompanied by children are 24 ppt less likely to be rough sleepers. This result could be partly explained by the precedence that families take in accessing shelters or the struggle of parents to provide a safer living environment for their families.

Table 5. Regressing rough sleeping on personal characteristics.

	(1) Linear probability model	(2) Logit model
Gender: female	−0.200**	−0.940**
	(0.063)	(0.291)
Age	−0.001	−0.005
	(0.002)	(0.010)
Health problems	−0.007	−0.055
	(0.069)	(0.319)
Disability	−0.205	−1.043
	(0.109)	(0.559)
Addiction	−0.006	−0.034
	(0.094)	(0.453)
Country: Romania	0.038	0.233
	(0.083)	(0.412)
Group (ref = In group without children)
Begs alone	−0.077	−0.394
	(0.077)	(0.370)
Begs with children (possibly in group)	−0.236**	−1.152**
	(0.090)	(0.434)
Benefit (ref = no)
Benefit: yes	−0.119	−0.573
	(0.090)	(0.420)
Begging hours (ref: <8h)	0.242***	1.115***
	(0.057)	(0.264)
Wave (ref = 1st wave)	−0.060	−0.309
	(0.055)	(0.270)
Intercept	0.450*	−0.218
	(0.177)	(0.804)
N	304	297
(Pseudo) R-square	0.178	0.125

Note: Linear probability model (model 1) and logit regression (model 2) with robust standard errors (in parentheses). *p<.05; **p<.01; ***p<.001. The dependent variable is a dummy indicator for rough sleeping. A missing category for receiving benefits was included (coefficient not shown). The mean VIF is 1.34.

Next, we test, with the help of a regression analysis, the previously illustrated correlation between the number of hours worked and housing situation. Table 6 indicates a strong and positive relation between sleeping rough and the number of hours worked.

Table 6. The relationship between rough sleeping and begging hours.

	Hours
Rough sleeping	1.635***
	(0.390)
Controls	Yes
N	304
R-squared	0.152

Note: Linear regression. *p<.05; **p<.01; ***p<.001. The controls include gender, age, health problems, disability, addiction, country, group, benefits and wave. Full results in online appendix, Table A1.

Table 7 presents an overview of available facilities and dwelling problems for respondents with access to housing. As previously illustrated, we refer here to ETHOS categories which include various types of accommodation for the homeless and non-conventional dwellings (the vast majority being apartments and single-family houses), as well as respondents with officially stable housing who do not fall under the ETHOS framework. Approximately 68% have access to all facilities listed, while 32% report that at least one is missing – most often the double-glazed windows. Respondents usually reported complete satisfaction and often expressed gratitude for their living conditions. However, those occupying squats had a higher incidence of lacking facilities. Half (51%) of the respondents in these housing categories experience at least one problem in their dwelling. This typically is a shortage of space or noise. The average number of reported issues among Romanian respondents is 0.95, while for the others, it is 1.43 (p = 0.128). In terms of self-reported shortage of space, there is once again a discrepancy, but not statistically significant (33% of Romanians, 48% of the other respondents; results not shown).

Table 7. Housing quality indicators. Note: The composite indicators are the percentage of respondents where at least one facility is absent or where at least one problem is present.

	Observations	Per cent
Facilities absent
Running water	114	3.5%
Hot water	114	11.4%
Bath/shower	114	1.8%
Indoor toilet	113	1.8%
Double glazing	106	26.4%
Facilities: composite indicator	106	32.1%
Problems present
Shortage of space	112	35.7%
Leaking roof	113	8.8%
Damp walls	113	15.9%
Rot in window frames	112	8.0%
Darkness	112	11.6%
Noise	112	26.8%
Problems: composite indicator	111	50.5%

In Figure 2, we illustrate that, on average, more than two individuals share a room. Most respondents pay rent monthly, with the mean amounting to €385 (Figure 3). The mean rent among Romanians (€353) is substantially lower than that among others (€516) (t = 2.3; p=.023). Some respondents admit to having to pay a fixed price per person, and many do recall the benevolence of their landlords, who, for instance, allow them to live rent-free or pay their rent at a later time if they face financial difficulty or offer them odd jobs around the rest of the building to decrease their rental cost, for example by cleaning the communal areas. On the other hand, those occupying squats reported a fee of approximately €5 to €10 per night, as well as overcrowding and relatively unsafe living conditions, with reports of conflicts and petty theft.

Figure 2. Distribution of the number of people per room.

Note: The mean number of people per room is 2.6 and ranges from 0.5 to 11 (n = 106).

Figure 3. Distribution of the monthly rent.

Note: The mean rent is 385 euros and ranges from 0 to 1450 (n = 106).

Table 8 shows the results of several linear regression models that aim to determine the relationship between personal characteristics, income-earning strategies, and housing quality. Our results imply that those of Romanian origin are 45 ppt more likely to have access to all listed facilities and typically share their room with one more person than those of non-Romanian origin, something considered to be usual among the Roma community, which tends to share accommodation with kin (Bodner & Leininger, 1992).

Table 8. Relationship between housing quality and personal characteristics.

	(1) Facilities	(2) Problems	(3) People per room	(4) Rent
Country: Romania	0.453**	−0.139	0.962*	−23.363
	(0.156)	(0.162)	(0.370)	(89.484)
Controls	Yes	Yes	Yes	Yes
N	96	101	98	97
R-squared	0.149	0.104	0.161	0.178

Note: Linear probability models (models 1–2) and linear regressions (models 3–4) with robust standard errors (in parentheses). *p<.05; **p<.01; ***p<.001. The dependent variables are (1) a facilities indicator for having access to running water, hot water, bath/shower, indoor toilet and double glazing; (2) a problems indicator for having either a shortage of space, a leaking roof, damp walls, rot in window frames, darkness or noise; (3) the number of people per room in their dwelling; (4) the monthly rent (in euros). The dependent variables are binary in models 1 and 2 (linear probability models). The controls include gender, age, health problems, disability, addiction, group, hours of begging and wave (results in online appendix, Table A2).

Half of the rough sleeping respondents have slept in a shelter at least once, and 56% of those that did were satisfied or extremely satisfied with their experience (5-point Likert scale; results not shown). The results of our linear regression model presented in Table 9 suggest that those of Romanian origin are 46 ppt less likely to have slept in a shelter, yet those who have slept in a shelter are 40 ppt more likely to be satisfied with them. Begging with a child seems to increase the chance of having previously slept in a shelter by 38 ppt. Respondents reported that overnight shelters are often fully occupied, thus decreasing their chance of finding accommodation there. The lower access of Romanian individuals who beg to shelters could be due to the language and information barriers that newly arrived foreigners typically experience. On the one hand, those accompanied by children, who are almost uniquely of Romanian origin, might have a higher likelihood of sleeping in shelters, which provide better living conditions for this subgroup, and thus might report higher satisfaction levels. On the other hand, the higher rate of satisfaction could be in line with a perceived higher rate of adaptation among this particular subgroup.

Table 9. Relationship between shelter usage/satisfaction and personal characteristics.

	(1) Uses shelter	(2) Satisfied with shelter
Country: Romania	−0.458***	0.404**
	(0.089)	(0.129)
Group (ref = In group without children)
Begs with children (possibly in group)	0.375**	−0.045
	(0.115)	(0.157)
Controls	Yes	Yes
N	201	98
R-squared	0.166	0.266

Note: Linear probability models with robust standard errors (in parentheses). *p<.05; **p<.01; ***p<.001. The controls include gender, age, health problems, disability, addiction, begging alone, benefits, hours of begging and wave (results in online appendix, Table A4).

Discussion

Through an exceptional quantitative survey, we manage to capture the characteristics and living situation of the begging population in Brussels, a largely misunderstood and under-documented group of poor people. This unique dataset helps to shed light on the mechanisms that drive the dynamic between housing and begging as an income-earning activity.

Virtually all people who beg are facing homelessness as per the ETHOS classification. Most are roofless, while approximately 30% of the population lives in insecure housing. Although the self-assessed quality of these dwellings is on average satisfactory, recurring issues are reported by respondents. Almost 50% mentioned at least one problem with their dwelling, typically a shortage of space or noise. Over 30% lacked access to at least one facility, mostly double-glazed windows. Different (sub)groups can be distinguished within the begging population, and the disparities in living strategies become evident through our analyses and support the literature (Bretherton & Mayock, 2016). Our findings suggest that those that are more vulnerable, such as women, people with disabilities, older age groups and those begging with children, are more often in emergency accommodations (e.g. shelters) and non-conventional dwellings than in the street. This may be related to avoiding the increased safety risks and detrimental effects of rough sleeping. Further qualitative research could help better elucidate these findings.

The most evident connection between homelessness and begging is the positive correlation that exists between the average number of hours begged per day and the incidence of sleeping rough. It is open for debate whether long working hours are a means of passing time in the street or a necessity. For those who beg where they also sleep, there is virtually no reason to stop as long as they are awake. Rough sleepers may also have the incentive to work longer to earn more to escape homelessness. Nevertheless, previous literature has mixed findings on whether begging can really help here (Gloria & Samuel, 2012; Lenhard, 2021). Bidirectional causality cannot be excluded either. Migrants who, akin to seasonal workers, have travelled abroad to beg may be interested in maximising their earnings before returning home by working long hours and avoiding rental costs.

This is especially true for the Roma, a prominent ethnic minority migrating from Central and Eastern Europe to Western and Northern European countries to earn an income. Accounting for more than three-quarters of the begging population in Brussels, this subgroup has a different relationship with housing than other subgroups. We would argue that several elements call for attention. For example, there are strong indications that Roma respondents, on average, adopt different reference categories when assessing a situation. If one compares the responses to questions that require some sort of assessment (e.g. is there a problem of overcrowding?) and the more unambiguous (or ‘objective’) states of affairs (e.g. the number of inhabitants per room in the housing unit), we see a profile occurring. In objective measures, Roma have a more problematic status. Concurrently, Roma respondents with housing do not indicate more often that their housing is overcrowded.

Moreover, the fact that Roma rent at a lower price might suggest that the housing quality is worse, while this does not show in their reporting of self-assessed problems with the habitation (our results show that they are 45 ppt more likely to claim to have access to all facilities listed in our survey). This pattern hints that Roma have adapted to their living conditions and adopted a reference category of normalcy that is much more substandard than other groups. One point of further reflection could be that objective definitions and frameworks on homelessness and insecure housing may be less meaningful for certain ethnic and cultural groups.

Furthermore, it is striking that roofless Roma in the begging population have increased from virtually none to almost 50% over the last 17 years. Why this happened is open to debate. One interpretation can be that in the past two decades, travelling from Romania to Brussels has become much less costly regarding organisation, border control, and legal matters. In the early 2000s, Romanians could not travel freely, and formal and informal cheap transport was in scarcer supply. This may have made the access of very poor people travelling back and forth from Belgium to Romania more achievable. We would cautiously suggest that an even poorer stratum of destitute people now found the way to Brussels. This group is poor enough not to be able to pay for housing.

We also draw attention to the issue of child begging. While there are well-documented differences among the Roma community when it comes to child rearing and education (Van Houcke, 2005), the existing connection between begging and sleeping rough implies that the children of those who beg are exposed to rooflessness. Within our sample, one in five respondents was accompanied by a child while begging, and one in ten had a child with them while also sleeping rough. The problem is not only restricted to children having to sleep rough but also to them facing the dangers and risks of living in the streets and increasing the barriers to school. This reasoning ultimately gave rise to the 2022 regulation that bans begging with children in the City of Brussels.

The connection between begging and homelessness has always had a degree of ambiguity. Our study contributes to the literature by quantitatively analysing this relationship using a representative sample of people who beg in Brussels. We arrive at the conclusion that the number of hours spent begging and migration background play an important role in determining the degree of homelessness and housing conditions. We do acknowledge that our quantitative approach comes with certain limitations. Further research relying on mixed methods could enrich and provide depth to some of our findings (for example, by adding visits to respondents’ places of residence to corroborate the self-reported data on living situations).

Disclosure statement

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

Additional information

Notes on contributors

Teodora Soare

Teodora Soare holds degrees in Business Administration and Advanced Economics and is currently pursuing a PhD at the Faculty of Economics and Business of KU Leuven, documenting the lives of people who beg in Brussels, Belgium. She thereby not only analyzes the life chances and experiences of those who derive an income from begging, but also studies the deservingness criteria that society at large adopts to judge this group.

Stef Adriaenssens

Stef Adriaenssens is a sociologist teaching courses in Economic Sociology and Policy. His core research interests involve underground, informal, and poorly protected economic activities and groups. Those involved sometimes are mainstream populations, such as those evading taxes or employing black market workers. However, his core interest is in excluded and hidden activities and groups that are often hardly recognized as work(ers), such as those begging or selling sex.

Koen Hermans

Koen Hermans is an Associate Professor of Social Work and Social Policy at the Centre for Sociological Research of KU Leuven. He is also a project leader at LUCAS, Centre for Care Research and Consultancy. His research focuses on the care and support for vulnerable and underserved populations in society, such as homeless persons, persons with disability and youngsters with complex needs. He was the coordinator of the COST Action on ‘Measuring homelessness in Europe’ (2016–2020). In Belgium, he is responsible for the development and organization of local and regional homelessness counts.

Appendix

Table A1. Determinants of begging hours.

	Hours
Rough sleeping	1.635***
	(0.390)
Gender: female	−1.174**
	(0.417)
Age	−0.014
	(0.013)
Health problems	−0.036
	(0.443)
Disability	−0.543
	(0.657)
Addiction	−0.480
	(0.713)
Country: Romania	0.084
	(0.561)
Group (ref = group without children)
Begs alone	−0.126
	(0.576)
Begs with children (possibly in group)	0.238
	(0.685)
Benefit (ref = no)	0.066
	(0.668)
Wave (ref = 1st wave)	−0.410
	(0.393)
Intercept	8.289***
	(0.984)
N	304
R-squared	0.152

Note: linear regression with robust standard errors.

* p<.05; **p<.01; ***p<.001.

A missing category for receiving benefits was included (coefficient not shown).

Table A2. Relationship between housing quality and personal characteristics.

	(1) Facilities (LPM)	(2) Problems (LPM)	(3) People per room	(4) Rent
Gender: female	−0.141	0.043	−0.025	−29.273
	(0.110)	(0.108)	(0.357)	(65.733)
Age	−0.002	−0.007	−0.016	4.885
	(0.003)	(0.004)	(0.012)	(2.947)
Health problems	0.008	−0.078	−0.326	88.153
	(0.095)	(0.113)	(0.378)	(72.900)
Disability	0.129	−0.026	0.398	0.468
	(0.133)	(0.165)	(0.838)	(98.356)
Addiction	0.202	0.127	−0.459	25.630
	(0.138)	(0.182)	(0.345)	(126.936)
Country: Romania	0.453**	−0.139	0.962*	−23.363
	(0.156)	(0.162)	(0.370)	(89.484)
Group (ref = In group without children)
Begs alone	−0.117	0.155	−0.664	−103.814
	(0.168)	(0.167)	(0.734)	(143.007)
Begs with children (possibly in group)	−0.178	−0.020	−0.956	56.881
	(0.204)	(0.204)	(0.743)	(122.392)
Benefit (ref = no)
Benefit: yes	0.132	0.160	−0.392	187.118*
	(0.136)	(0.151)	(0.304)	(84.325)
Begging hours (ref: <8h)	−0.065	−0.051	−0.520	42.168
	(0.132)	(0.128)	(0.431)	(89.686)
Wave (ref: 1st wave)	−0.179	−0.048	0.005	38.356
	(0.108)	(0.118)	(0.358)	(71.547)
Intercept	0.730*	0.869*	3.959***	130.784
	(0.354)	(0.342)	(1.138)	(206.572)
N	96	101	98	97
R-squared	0.149	0.104	0.161	0.178

Note: Linear probability models (models 1-2) and linear regressions (models 3-4) with robust standard errors (in parentheses).

* p<.05; **p<.01; ***p<.001.

A missing category for receiving benefits was included (coefficient not shown).

Table A3. Relationship between begging in a group and country of origin.

	Other than Romania	Romania	Total	N
Begs alone	26.1%	73.9%	100.0%	218
Begs in a group (no children)	31.4%	68.6%	100.0%	70
Begs with children (possibly in a group)	2.9%	97.1%	100.0%	70
Total	22.6%	77.4%	100.0%	358

Table A4. Relationship between shelter usage/satisfaction and personal characteristics.

	(1) Uses shelter	(2) Satisfied with shelter
Gender: female	−0.047	0.143
	(0.080)	(0.109)
Age	−0.001	−0.001
	(0.003)	(0.004)
Health problems	−0.102	−0.150
	(0.084)	(0.150)
Disability	−0.076	−0.053
	(0.140)	(0.265)
Addiction	0.023	−0.079
	(0.110)	(0.189)
Country: Romania	−0.458***	0.404**
	(0.089)	(0.129)
Group (ref = In group without children)
Begs alone	0.190*	0.120
	(0.093)	(0.132)
Begs with children (possibly in group)	0.375**	−0.045
	(0.115)	(0.157)
Benefit (ref = no)	0.103	−0.125
	(0.119)	(0.212)
Begging hours (ref: <8h)	−0.048	−0.124
	(0.070)	(0.093)
Wave (ref: 1st wave)	0.003	0.032
	(0.074)	(0.093)
Intercept	0.846***	0.478
	(0.197)	(0.270)
N	201	98
R-squared	0.166	0.266

Note: linear probability models with robust standard errors (in parentheses).

* p<.05; **p<.01; ***p<.001.

A missing category for receiving benefits was included (coefficient not shown).