The social care-taking of the city-kids. Determinants for day-care attendance in early twentieth-century southern Sweden

ABSTRACT The introduction of a child day-care system is one of the early welfare interventions targeted towards mothers and young children that over time gained great prominence in the Swedish welfare state. Because quantitative research on day-cares in historical settings is generally scarce, in this study, we focus on the determinants of day-care enrolment in southern Sweden during the early twentieth century. We use unique individual-level records of day-care attendance for children born between 1900 and 1935 which have been linked to longitudinal micro-level data for the city of Landskrona obtained from the Scanian Economic Demographic Database. Event-history techniques are employed to analyse the importance of factors such as household composition, parental socio-economic background, marital status of the mother, and mother’s occupation. Of the studied children, 8% were ever enrolled in day-cares, most of them around the ages 3 to 6. The results show that the mother’s marital status, household SES, number of siblings, the presence of other adult females in the household and mother’s occupation are all significant determinants of day-care attendance for children. In this study, we show that in the early twentieth century in southern Sweden, day-care attendance followed a negative SES gradient and was most common among children of single mothers.


Introduction
Sweden today is known for its all-encompassing childcare system that allows parents to work while having their children taken care of. Historically, this puzzle was much harder to solve, not least in industrializing cities. Working parents during the early twentieth century were often an economic necessity to ensure the family's survival, and even single mothers could not count on public support either in terms of allowances or otherwise in terms of the provision of childcare services. The puzzle remained impossible to solve for many families, which not seldomly led to voluntary or forceful removal of children from the families of the poor (Holmlund, 2013).
The emergence of institutional childcare was in that light an important welfare provision. The first day-care centres (hereafter day-cares) in Sweden were founded in the nineteenth century as charity institutions for the children of the poor (Lindgren & Söderlind, 2019). While the coverage remained small until the late twentieth century, these day-cares served an important role for parents and children in the industrial centres. The history of the establishment of day-cares is well documented (among others, Ekstrand, 2000;Hatje, 1999;Holmlund, 1996;Lindgren & Söderlind, 2019;Westberg, 2008) but quantitative accounts of day-care attendance are scarce, in Sweden as elsewhere. To fully understand the historical institution, its importance for society, and its potential as an engine of social change, we believe that a quantitative account of day-care institutions is necessary. Much of the qualitative research on the development of daycares in Sweden has evolved in the research tradition of poverty research, which makes assessing (over-)representation of certain socio-economic groups more challenging. Our study is the first quantitative study to address the determinants of institutional childcare attendance in a historical setting. This paper makes use of unique population-level data for children born 1900-1935 who lived in the city of Landskrona, located in southern Sweden, a city that can be regarded to represent the general development of many medium-sized cities in Sweden. Linking day-care attendance to other individual-level data allows us to identify which factors were historically related to day-care attendance and to assess their relative importance.
By linking longitudinal demographic data to records of day-care enrolment for the city of Landskrona, we study the determinants of day-care attendance. The focus of our study is on institutional childcare (also known as centre-based care), that is, childcare in established day-cares. We focus on family background, household characteristics, and sociodemographic factors. Besides providing knowledge about life in an industrial urban setting, this work can contribute to gaining a better understanding of what led to the present-day patterns of day-care participation.
Our study shows that some determinants of day-care attendance resemble patterns found today, while others contradict them. Single mothers are a group with high demands for day-care utilization both historically and today. Similarly, the presence of relatives or availability of other type of care reduces day-care attendance both historically and today. With regards to socio-economic factors, however, our historical example is in stark contrast to contemporary day-care utilization. While today, high socio-economic status (SES) families are overrepresented among children enrolled in institutional daycare, historically, they were underrepresented. Instead, the most impoverished families were the most likely to send their children to day-care. This work is an important contribution to understanding the patterns of day-care utilization in the past and highlights the need for further research for understanding the full picture of historical as well as contemporary day-care utilization and its effects.
The remaining sections of the paper are structured as follows: Section 2 gives a review of the literature on determinants for childcare utilization. Section 3 gives a historical account of the development of day-cares in Sweden, and section 4 introduces the city of Landskrona. Section 5 contains a description of the data and methods and section 6 a description of the results. The paper ends with a discussion of the results and a conclusion in section 7.

Previous research on the determinants of childcare utilization
Studies on the determinants of childcare utilization that go beyond the historical accounts of the development of day-care institutions are scarce. To our knowledge, no prior quantitative studies exist that focus on the determinants of childcare utilization in a historical setting. Much of the previous research on the utilization of institutional childcare cover highly developed contexts such as the U.S. and Western Europe, and many studies have been published in recent years. The empirical studies reviewed by us assess data that falls into the period from the 1980s to 2010s. For that given reason, in our brief account of previous research, we refer to contemporary studies on the topic and identify factors that may be related to childcare utilization even in a historical context.

Family income
Several studies show a positive association between family income and childcare utilization in the U.S. (Davis & Connelly, 2005;Early & Burchinal, 2001;Hirshberg et al., 2005;Hofferth & Wissoker, 1992). Childcare utilization is closely linked to mothers' employment, which is one of the reasons for explaining the association between family income and childcare utilization. Other reasons can be childcare costs (low-income families rely more often on care from relatives; Davis & Connelly, 2005) and the educational preferences of families with higher SES. A similar social stratification in childcare utilization exists in almost all European countries except for several countries in which childcare rates are non-universal (Czech Republic, Slovak Republic, Austria, Lithuania, Malta, and Estonia). Most equal participation is given in Sweden and Denmark but even here, the association with family income is not negligible ( Van Lancker, 2013). The social stratification in childcare utilization appears to be a rather universal characteristic, since it has been found in different European countries with different degrees of public funding of childcare (Abrassart & Bonloli, 2015 for Switzerland;Mamolo et al., 2011 for the UK; Zachrisson et al., 2013 for Norway).

Parents' education
Beyond income, education is another stratifying factor for childcare utilization leading to lower attendance rates of children of lowly educated parents (Claessens & Garrett, 2014;Fuller et al., 1996;Hirshberg et al., 2005;Varmuza et al., 2019). An educational gradient exists even among low-income families (Crosnoe et al., 2016).

Availability of institutional childcare
Childcare can be provided by institutions (centre-based care) or by, for example, relatives. Somewhat surprisingly, no positive association between availability of institutional childcare and its utilization has been found by several studies (Abrassart & Bonloli, 2015;Davis & Connelly, 2005;Hirshberg et al., 2005), which may be explained by the availability of relative care, and the decision of childcare utilization and maternal employment being made jointly.

Maternal employment
To study the association between maternal employment and childcare utilization is challenging using cross-sectional data if the decisions of working and childcare utilization are made jointly (endogeneity problem). Several studies have generally found a positive association between maternal employment and childcare utilization (Abrassart & Bonloli, 2015;Crosnoe et al., 2016), yet evidence from Europe shows that maternal employment (and education) matter more in settings in which mothers are seen as the primary caregivers, such is the case in France, Italy, and Spain (Mamolo et al., 2011). The association between childcare utilization and maternal employment is stronger at younger ages, while maternal employment is less predictive for pre-school children (Coley et al., 2014;Singer et al., 1998). Non-standard work schedules for mothers reduce the odds of using institutional childcare (Han, 2004), while longer working hours increase the odds (Hirshberg et al., 2005). The endogeneity of the decision to work has been addressed empirically by only few studies. Davis and Connelly (2005), for example, find that the association between (predicted) maternal employment and home childcare (both including licensed and non-licensed providers) is even stronger than the one with institutional childcare, which is also used by non-employed mothers, and is likely more related to the perceived benefits for their children. Eckhoff Andresen and Havnes (2019) exploit the local variation in childcare provision for two-year-olds which was the result of a political reform. They conclude that the provision of large-scale universal childcare has substantial effects on mothers' labour supply.

Single parenthood
For single mothers, the necessity to work is higher and employment and childcare decisions may be less flexible than for married mothers. Most studies find a higher probability of childcare utilization of single mothers (Coley et al., 2014;Hirshberg et al., 2005;Zachrisson et al., 2013) but one study on Australia finds a lower probability of being in formal childcare for one-parent families (Claessens & Garrett, 2014).

Relatives
The availability of relatives has a significant impact on childcare utilization (Fuller et al., 1996). Especially the availability of grandparents reduces the utilization of institutional childcare in all different economic and ethnic subgroups that were under study.

Context in which the childcare institution is placed
The contexts in question matter largely for childcare attendance as the cost of childcare and the public organization of childcare differ greatly. Mamolo et al. (2011), for example, show in the context of France, Italy, Spain and the U.K. that determinants for part-and fulltime options differ. The utilization of full-time childcare is more frequently related to necessity (usually female labour force participation), while the utilization of part-time childcare is also related to pedagogical preferences.
Other aspects of the family structure that matter for childcare utilization are the child's age (Davis & Connelly, 2005;Fuller et al., 1996;Han, 2004;Mamolo et al., 2011) and the number of other children in the household. Findings are, however, inconclusive regarding the question of whether more children in the household increase or decrease the utilization of non-parental and centre-based care (Coley et al., 2014;Crosnoe et al., 2016;Mamolo et al., 2011;Varmuza et al., 2019).
The lack of research on historical contexts (as well as on less developed contexts) may be primarily related to data availability. Another potential explanation for the so-far limited research on historical or developing contexts is that the positive returns to daycare attendance gained more awareness only in recent decades and more so in more highly developed countries (e.g., Bakken et al., 2017;Campbell et al., 2002;Havnes & Mogstad, 2011;O'Brien Caughy et al., 1994).
We believe that the factors listed also impacted day-care attendance in our historical setting, even though it is an empirical question to determine the degree to which the historical case is similar to the contemporary ones described above. Like the studies on family income and parental education, we assess the association between the household's SES and day-care attendance. We estimate the association with mother's marital status to check the impact of single parenthood. We include the mother's occupation to approach the question of whether or not day-care attendance is associated with maternal employment in a historical setting. We believe that the availability of relative care affected the decision to have one's children attend day-care. Therefore, we assess if the number of females in the household, and more specifically, grandmothers living in the household, alters the hazard of attending day-cares. Finally, we explore the role of siblings, which according to previous research on the number of children in the household may increase or decrease the likelihood to attend day-care.

Development of day-care institutions in Sweden
The all-encompassing Swedish day-care system as it is known today bears little resemblance to its predecessors that developed nearly two centuries ago. The development of public day-care institutions was a long and gradual process and the outcome of two main developments. First, industrialization resulted in the separation of production and reproduction, and second, a gradual shift of public childcare provision away from poor relief institutions. This section mainly focuses on the description of the early development of day-care institutions in Sweden until the end of the period under study (1940s). For a complete description of the development, see for example, Lindgren and Söderlind (2019).
Sweden was relatively late to industrialize, with industrialization starting around the 1850s (Schön, 2010). Urbanization and the organization of work in the industrial society led to a division of production and the home, causing new challenges for the birth and upbringing of children. The division of labour between fathers and mothers, with mothers focusing on the upbringing of children, makes children a costly enterprise, which many working-class families could not afford. Wages were generally so low that more than one income was required, and in many cases, a male breadwinner was not present (Antman, 1996). Female labour force participation increased by almost 50% during the first two decades of the twentieth century, particularly among urban women (Stanfors, 2014).
There was a high demand for female labour since women's wages were only about half of those of men (Antman, 1996). While middle-class women became homemakers, single mothers and mothers among the poor had to make a living outside the home, leaving children without appropriate supervision, generally with older siblings or simply roaming the streets (Dahlström, 1997).
The Swedish poor relief system was regulated by the state and well-established in the nineteenth century at the beginning of industrialization but was ill-equipped for the developing urban poverty. In the growing towns and cities, urbanization and early industrialization led to difficulties in making up a living for larger numbers. Therefore, the demand for poor relief increased. Consequently, stricter poor relief regulations were established that excluded a larger share of the population from access to poor relief. The necessity of taking care of small children was not viewed as a reason for not being able to work and support their own families. The solution offered by poor relief institutions in the period before any public or private day-care initiatives were established was not seldom to remove children from their families and place them in institutional care. In many cases, these were children born outside marriage (Antman, 1996) or children of mothers who were regarded as not being able to provide them with the necessary care supervision (Myrdal, 1935). Children were placed in almshouses, orphanages, or in foster care. Removing children from poor families was viewed as a better solution for the children since poverty was often interpreted in terms of a lack of moral standards (Holmlund, 2013;Tallberg Broman, 1995), but the system was even a cheap solution for municipalities (Nyberg, 1995). Even the introduction of compulsory schooling during the second half of the nineteenth century worsened the childcare problem since older siblings were no longer available to take care of the younger ones (Holmlund, 2013). In such a setting, the demand for childcare that allowed parents to work and keep their children was high.
The practice of removing children from their parents was gradually replaced by private and public initiatives to improve the situation for poor families so that children could remain within the family (Antman, 1996). The introduction of the first forms of day cares in the first half of the nineteenth century in Sweden can be understood to be within this line of thought: day-care institutions aimed at giving children of the poor access to a better upbringing and giving parents the possibility to work. The first infant schools (småbarnsskolor) opened in Sweden in the 1830s and the first cribs (barnkrubbor) opened in the 1850s. Both types of institutions opened particularly for children of economically disadvantaged parents and single mothers and were driven by voluntary organizations (Nyberg, 1995). The different kinds of day-care institutions followed different European ideals, with a focus on education (infant schools) and pure caretaking with a focus on physical needs to allow parents to work (cribs). At the end of the nineteenth century, more pedagogically oriented day-care arrangements after the German ideal, kindergartens or barnträdgårdar, were initiated. Like other forms of day-care, these were placed in industrialized areas to allow mothers to work and support their families 1 . Over time, Swedish day-care institutions became more detached from being poor relief institutions (Hatje, 2013) and became universalistic institutions that offered care to children of all families, not only to the poor or the working class (Rauhala, 2009). One development that led to a more universal demand for childcare institutions particularly was the increased labour force participation of mothers throughout the 20th century. Employment among married women started to increase already around the 1920s (Stanfors & Goldscheider, 2017). Even though working mothers remained a minority until the second half of the 1960s, there were continued developments in the provision of day-care services throughout the twentieth century.

Single mothers
Single mothers are one group among whom the demand for the establishment of daycare institutions was particularly high. In the late nineteenth century, single mothers were not uncommon and the public poor relief institutions did not assist them in claiming any responsibility from the fathers' side. Independent of their marital status and family situation, the poor relief system would not support single mothers because it claimed mothers to make their own and their children's living. At the end of the nineteenth century, only 36% of children born out of wedlock were living with their biological mothers (Nyberg, 1995). For mothers who did not want to give up their children to foster care and who could not engage relatives, neighbours, or older siblings in childcare, the newly founded day-care institutions were the only possibility to have childcare secured and to maintain custody for their children (Holmlund, 2013). Parallel to the improving economic conditions for single mothers and the slowly growing availability of day-care institutions, the share of children to single mothers among foster children decreased during the twentieth century (Nyberg, 1995). The demand for places in day-care institutions among single mothers was very high and many places were given to children in this group. The share of children to single mothers attending day-care institutions was high until the 1980s when day-cares became more universal early-childhood institutions. In 1944, 39% of children in day-cares were children to single mothers. In 1966, the number was 46% (Antman, 1996). However, the supply of day-care places was still small. Only about 10% of children of working single mothers in the early 1940s had a day-care placement, while 13% followed their mother to work, 42% were taken care of by relatives, and 36% were separated from their mothers (Nyberg, 1995).

The different types of day-care institutions in the period under study
The three types of day-care institutions that can be found in Sweden under the period of study (infant schools, cribs, and kindergarten institutions) had many similarities but differed vastly in other aspects. All three institutions were mainly directed at the workingclass population (for kindergartens, this holds only for the so-called volkskindergarten, folkbarnträdgårdar). They were often established as charity projects and had as purpose, beyond offering supervision under parents' working hours, to educate working-class children in terms of values and cleanliness. Beyond that, differences in pedagogy and staff training were large.
Infant schools were established in Sweden from the 1830s onward, in a period before compulsory schooling was introduced, and had a clear vision for teaching working-class children in the ages 3-7. Primary school teachers were often employed who alternated teaching with other kinds of activities. Teaching involved bible studies but also natural sciences and reading. Especially during the nineteenth century and early twentieth century, a very large number of children were supervised and taught by only one teacher (80-100 children were common). For that purpose, classrooms were organized like amphitheatres. The large number of children and long opening hours (7 am to 7 pm during summer terms and 8 am to 6 pm during winter terms) express according to Ekstrand (2000) the high demand for supervision among the poor rather than the infant schools' pedagogical orientation. With the turn of the century (and the establishment of all-encompassing primary schools) the focus on teaching was replaced by a kindergartenlike pedagogy, group sizes were reduced, and trained kindergarten staff became employed.
The first cribs were established in Sweden during the second half of the nineteenth century. The establishment of cribs was closely related to poor relief and the working class's need for supervision of their children. Here, there was only little focus on teaching and pedagogy and most focus was on regular meals and cleanliness. There were no requirements for the staff's education and beyond untrained staff, nurses and sometimes kindergarten teachers were employed. Even here opening hours were very long. Group sizes were considerably smaller than in infant schools (about 25 children; Holmlund, 1996;Lindgren & Söderlind, 2019). According to Westberg (2008), cribs were the most common day-care institution for children in early childhood in 1911 as in 1932. While their number as well as the number of kindergartens increased in the period, the number of infant schools decreased slightly.
Kindergarten institutions were established in Sweden around the turn of the century. They were not founded with the same focus on poverty and charity as the two other institutions but the emerging volkskindergarten (folkbarnträdgårdar) were in the same tradition and were founded in industrial societies (while the regular kindergartens were aimed at middle-class children). The kindergarten movement had a clear focus on child pedagogy and development but saw institutions only as a complement to the family. For that reason, kindergartens were often half-day institutions, to not take over the main care responsibilities from families. There was distinct training in kindergarten pedagogy (Fröbel pedagogy) for the staff (Hatje, 2013;Holmlund, 1996). The number of children in kindergarten groups was often around 20-25 (Lindgren & Söderlind, 2019).
Towards the 1930s, a gradual integration of infant schools, cribs, and kindergartens took place (Ekstrand, 2000;Holmlund, 1996) and several infant schools closed (Westberg, 2008). Often, the integration took place within the institution but not within groups, and more and more institutions had kindergarten and crib groups (Holmlund, 1996). After the 1930s, two distinct streams were in place, all-day institutions (daghem) and half-day institutions (barnträdgårdar/förskolor; Tallberg Broman, 1995), which replaced the large variations in day-care pedagogy, age of children, and opening hours.
All types of day-care institutions were not free of charge except for the neediest children (Hatje, 1999), but fees were generally low and rather served as symbolic costs for parents. While donations are likely to be the most important source of income before the 1920s, municipal funding had the largest impact thereafter and was eventually followed by the communalisation of day-cares (Westberg, 2008). Poor-relief institutions supported privately found day-care institutions with food donations (Holmlund, 2013). State funding had been debated throughout the 1930s (SOU, 1938) but the first state funds became available only in 1944 (Hatje, 2013). The number of places increased in this time, and public (state) funding increased substantially in the 1960s (Tallberg Broman, 1995), the period in which female labour force participation took off and the underprovision of day-care places came into focus as a labour market issue (Persson, 1962).

The city of Landskrona
Landskrona was founded in 1413 and during its history it changed from being a port and military town in the preindustrial period to becoming a medium-sized industrial city, though later it experienced serious deindustrialization after the recession in the 1970s (Dribe & Svensson, 2019). Throughout the 20th century, it saw considerable immigration and bears evidence of many present societal challenges from both economic stagnation/ transformation and immigrant integration. Concerning population growth, industrialization and deindustrialization, Landskrona can be regarded to represent the general development of many medium-sized cities in Sweden. Dribe and Svensson (2019) conclude that Landskrona developed in most ways similar to the other Swedish industrial cities, both economically and demographically. However, although Landskrona follows the same trends and variations as the other Swedish cities, it is located at the low end of the distribution, for example, in earnings and education.
Given the overall comparability of Landskrona and other industrial cities at the time, we expect no differences between the determinants of demographic behaviour and individual well-being in Landskrona concerning similar areas within or outside the country. The results of this study are therefore not only representative of Landskrona but can also be generalized to other areas.
The first infant school opened in 1862 (Asylen för småbarn; Landskrona fruntimmersförening, 1942), which is similar to the development in other cities (for reference: The first infant school in Stockholm opened in 1833, the first in Malmö in 1842). As in other places, it was founded with the purpose of reducing poverty and improving living conditions for poor children (Ekstrand, 2000;Landskrona fruntimmersförening, 1942). During the nineteenth century and well into the twentieth century it hosted around 100 children. This is rather unique compared to infant schools in other cities, which originally had similar group sizes but reduced them around the turn of the century. The high number of children enrolled in one day-care was certainly related to a high demand in Landskrona. In 1909, the infant school's facilities were updated. The new facilities were designed with less focus on frontal instructions and more focus on play and childcare (Ekstrand, 2000). The infant school changed its pedagogical orientation in the second half of the 1930s (towards kindergarten pedagogy) and opened a half-day kindergarten section in 1937, despite otherwise remaining an all-day institution (Landskrona fruntimmersförening, 1942). Day-care in Landskrona was affordable for all social groups, as the example of the infant school shows. Fees were introduced only in 1936 and accounted for approximately only 10% of the entire funds thereafter (own calculations, based on information in Ekstrand, 2000). Upon introduction of fees, they were decided to be 5 SEK per month. 2 It is difficult to estimate women's average income at the time but a look at a few female occupations (employed by Landskrona municipality) shows that salaries around 150-400 SEK/month were common in 1915 (for example, cook, laundress, nurse, maid; SCB 1915). Estimates for Gothenburg show that the median female income was around 900 SEK (GOPP, see, Karlsson & Lundh, 2022). Funding was otherwise secured by the municipality, the Savings Bank (Sparbanken), and donations (Ekstrand, 2000). In 1916, a small child crib opened in Landskrona as part of a school for homemakers (Husmoderskola), which was later expanded by a kindergarten group (Jönsson, 1997).
There was a high labour force participation even of married women in Landskrona, which was referred to in official investigations on the need for day-care expansion in Landskrona. In 1911, 28% of workers in Landskrona were women (Dribe & Svensson, 2019). Moreover, about 20% of children were born outside wedlock in the 1920s which also contributed to high demand for day-care (own calculations from the data; see description of data below).

Data and methods
In this paper, we use longitudinal data for the city of Landskrona for the period 1905-1937. The data was obtained from the Scanian Economic Demographic Database (SEDD), which is based on family reconstitutions and local population registers and includes information on demographic events and migration for all household members and families in households (Bengtsson et al., 2018;Dribe & Quaranta, 2020). 3 The material is of high quality and considered to be complete with regard to vital events (Bengtsson & Lindström, 2000). The SEDD also contains detailed information on occupations, obtained from birth, marriage, catechetical examination, poll tax and income registers. 4 Occupations were coded into HISCO (Van Leeuwen et al., 2002) and later categorized into HISCLASS ( Van Leeuwen & Maas, 2011). HISCLASS is a 12-category occupational classification scheme based on skill level, degree of supervision, whether manual or nonmanual, and whether urban or rural. Historical studies often use a six-class reclassification of the scheme which reflects a status hierarchy: higher white-collar workers (HISCLASS 1-2), lower white-collar workers (HISCLASS 3-5), medium-skilled workers (HISCLASS 6-7), lower-skilled workers (HISCLASS 9-10), unskilled workers (HISCLASS 11-12) and farmers (HISCLASS 8). In this study, we also follow this reclassification, although we exclude farmers due to small numbers (less than 1% of the sample) and also due to small numbers we combine higher and lower white-collar workers into one single group, and unskilled workers with individuals with unknown occupation into another group (unskilled/ unknown).
In order to define our sample of treated children, we collected almost complete lists of children who were enrolled in Landskrona infant school (Asylen för små barn) between 1905 and 1936 and linked them to SEDD (for a description of the missing source material, see, note 5 ). We photographed registry books for the period, which were four in total, and later digitized them. Two of these registry books were stored at the local archive (Landskrona city archive). The remaining two registry books were found at the actual daycare, which is still running today. Registry books were organized into academic years, running from September of one year until June of the following year.
The registry books of day-cares contained in total 3089 records, and we were able to link 98.8% of them to SEDD. A detailed description of the linking procedure used is provided in Appendix 1. With almost complete records on all day-cares and children, nonlinked children in the SEDD data (Landskrona sample) can be identified as non-attending (untreated). The small percentage of children for whom it was not possible to link records but who attended day-care are also included among the non-attending group.
The study sample includes children born between 1900 and 1935, 6 and the outcome of interest considered in the analysis is first day-care attendance in Landskrona. Children enter the study sample when they are born, immigrate to Landskrona or upon the first date when data for Landskrona is available in SEDD (January 1st, 1905), 7 and they are followed until they either first enrolled in day-care or otherwise, the day before their seventh birthday, or death/outmigration/end of the study period if they occur before such date. The end of the study period was considered to be the 31st of August of 1937, given that the last day-care registry book was for the academic year 1936/37. The study sample contains information on 15,213 children, of whom 1,281 (8.4%) ever attended day-care in the years 1905-1937. The majority of children attending day-care were between ages 3 and 6. In fact, out of a total of 13,070 children observed under age 3, 289 of them ever attended day-care in such age range (2.2%), while out of a total of 12,341 children observed when they were 3 or older, 1,141 of them attended day-care in such age range (8.0%).
We use Cox proportional hazards models to study the individual determinants for first day-care enrolment in Landskrona. These types of models have the advantage of measuring jointly the likelihood of a certain event happening (in this case day-care attendance) as well as the time until the event happens (in this case age at first attendance), aspects which are both important to consider when analysing day-care utilization. Moreover, through Cox models, it is possible to handle censuring and truncation, which are typical characteristics of longitudinal data, and in our case relate to the fact that due to migration we cannot follow all children from birth until their seventh birthday. For these reasons, we chose to use survival models instead of estimating logistic regressions to study whether a child ever attended day-care. In the Cox models estimated, age is considered as the time variable, and the event of interest of the study is the time of the first enrolment to day-care. We tested for the proportional hazards assumption using tests based on Schoenfeld residuals, and no violations in this assumption were found in the selected models. The main analysis considered as 'failure' in the Cox model first day-care attendance, taking into consideration all linked day-care records. We also tested models where we limited the 'failures' by only considering children for whom the link between the day-care records and SEDD was most certain (Step 1 in Appendix 1), and the results remained consistent with those considering all links.
Based on previous research and historical accounts of day-care utilization in Sweden, we consider different explanatory variables in the models: marital status of the mother (married, not married 8 ), number of siblings (created as a time-varying variable and afterwards categorized into 0, 1-2, 3 or more), number of females aged 15-64 present in the household (created as a time-varying variable and afterwards categorized into 1, 2 or more) as a proxy for the availability of relative/other care, and household SES. Household SES takes into account the occupation of the household head, considered to be the father of the child if he was present in the household, and otherwise the mother. It was categorized into four categories: white-collar workers; medium-skilled workers; lowerskilled workers; unskilled workers or unknown. 9 Given the fact that this SES variable is a measure of household SES, we also consider a modified version of the variable, which in addition to the higher white-collar workers, lower white-collar workers, medium-skilled workers, lower-skilled workers, unskilled/unknown, includes the separate category 'father not present'. In the models, the variable measuring number of siblings was interacted with time (child's age 0-2.9 and 3-6) since such a variable otherwise violated the proportional hazards assumption. In other words, we estimate a time-varying effect for this variable. We also consider alternative models taking into account the mother's occupation, categorizing this variable into three categories: unknown or no occupation, occupation other than maid, and maid. 10 In this study, we consider the occupations of the household head and the mother as valid for five years after they were declared in the digitized registers, after which they were considered as unknown. All specifications control for the child's sex. All models were stratified by year of birth of the child (continuous), instead of controlling for such variable due to violations of the proportional hazards assumption. Spells of time during which the mother of the child was not present in the household, the marital status of the mother or the number of siblings was unknown, or the household head occupation was farmer, were excluded from the analysis. The analysis is conducted through a series of models, each controlling for a different combination of the variables described above. In total, we estimate seven different specifications.
Descriptive statistics are shown in Table 1. Values were calculated as a percentage of total time at risk. As can be seen from the table, the sample is almost evenly distributed between boys and girls. 88% of the total time at risk corresponds to children whose mothers were married. Household SES was more or less evenly distributed across the four categories considered, with white-collar workers being the least represented group. Around 21% of the time at risk in the sample is composed of children with no siblings, 43% with one or two siblings, and 36% with three or more siblings. Around 81% of the sample corresponds to children living in households with only one adult female (the mother) present in the household and the remainder to children living in households with Distributions are calculated from the study sample as percentage of total exposure time.
2 or more females. When considering maternal occupation, 82% of the total time at risk corresponds to children of mothers who did not work or whose occupation was unknown, 5% of mothers who worked as maids, and 13% of mothers who worked in other occupations.
belonging to households of white-collar workers have the lowest hazard of attending daycare, followed by children of medium-skilled workers (HR 3.12), lower-skilled workers (HR 4.40) and lastly the unskilled/unknown (HR 7.42), 11 who have the highest hazard of attending day-care. In Model 3, we consider both variables in one model. It is not unproblematic to do so as these two variables correlate, and one may be the outcome of the other. However, when considering marital status and SES in the same model, the results remain very consistent and the effect of each of these variables is only very marginally reduced in relation to Model 1 or Model 2. The model clearly shows that  marital status and SES are each independently important as a determinant for day-care attendance. In Model 4 we use a reclassification of SES which also considers the separate category 'father not present'. The strong socioeconomic gradient observed in Models 2 and 3 is still shown in Model 4. The highest hazard of attending day-care is observed for children living in households where the father was not present or where the father was present, but was unskilled or had an unknown SES. We estimated additional models where we introduced a variable measuring time-period as well as interactions between period and marital status and SES respectively, but neither the base effect of period nor the interaction terms were statistically significant. This shows that the hazard of attending day-care, as well as the influence of SES and marital status on such hazard, remained rather consistent across the time period studied, which may be related to the fact that such period is not very extensive.
In Model 5, we additionally control for the number of siblings. The results show that the higher the number of siblings, the greater the hazard of attending day-care. The effects are much larger below age 3 than in ages 3-6, although it's important to point out that, as stated earlier, the number of children attending day-care below age 3 was low, which is why the hazards are highly inflated. The effect of maternal marital status and SES remain consistent when controlling for number of siblings in the model. When controlling for the mother's age in an additional model, the increases in the hazard of day-care attendance with a higher number of siblings become slightly stronger. This shows that the hazard of day-care attendance is indeed associated with the number of children in the household and that such an effect is not only driven by the fact that older women have more children. We also tested two alternative specifications of the sibling variable in additional models. In one we considered the number of younger siblings and in the other the number of siblings below age 7 (to measure the impact of the presence of children below school age). The results remained consistent with those of Model 5, showing an increasing hazard of first day-care attendance with more younger siblings or more siblings below age 7, and such effects remained also after controlling for mother's age in each model.
In Model 6 we also include a dummy variable for the number of females in the household. Having an additional female in the household captures potential alternatives to institutional childcare in the form of relative care (older sisters, grandmothers, or other kin) or hired childcare (maid or nanny). Children living in households with two or more females aged 15-65 have a lower hazard of attending day-care (HR 0.62), relative to those with only one female present. 12 We also estimated two models which, instead of number of females, included a variable indicating whether the maternal or paternal grandmother was present in the household, but neither of these variables was statistically significant when also controlling for the child's sex, mother's marital status, and SES. This shows that it was not the presence of grandmothers specifically but rather of adult females in general that influenced the hazard of first day-care attendance. Kin living in closely located households could also have assisted in caring for children and thus have influenced the hazard of attending day-care, however, we are unable to measure this in the data.
To further explore how the mothers' occupation and labour force participation is related to day-care attendance, Model 7 considers the occupation of the mother instead of household SES. 13 The results show that children whose mothers had no or unknown occupation had the lowest hazard of attending day-care, followed by those with an occupation other than maid (HR 2.24) and lastly maid (HR 3.02). Having a mother who is working in a dependent role in someone else's household increases the hazard of first day-care attendance substantially.
All models are stratified by child's birth year instead of considering this variable as a control in the models since birth year violated the proportional hazards assumption. The child's sex is included in all models. Across all models, boys show slightly higher hazards of attending day-care than girls, although such difference was marginally statistically significant in Models 2, 5 and 6 and not significant in the other models.

Discussion and conclusion
Our study is the first quantitative study that addresses the determinants of institutional childcare attendance in a historical setting. While some historical accounts of the history of early day-care establishment and utilization exist, until now, these have not been studied with individual-level longitudinal data on the full population of a specific location. Our study makes use of unique data for children born 1900-1935 who lived in the city of Landskrona. The city of Landskrona is comparable to other medium-sized cities in Sweden in terms of population growth, industrialization and deindustrialization; Landskrona can be regarded to represent the general development of many medium-sized cities in Sweden. The results of this study are therefore not only representative of Landskrona but can also be generalized to other areas. The establishment of day-care institutions in Landskrona was in line with other cities of similar size, but Landskrona hosted one of the larger day-cares in the country. The timing of day-care establishment and the relatively high number of children enrolled gives us the unique opportunity to study determinants for day-care attendance at an early stage of institutional day-care development.
Compared to the contemporary studies on day-care attendance, our findings both support and contradict findings from other settings. Like studies on contemporary data, we find mothers' marital status and mothers' employment to be important determinants for day-care attendance. The variable unmarried mother at least doubled the hazard of day-care attendance for children in all specifications, even when the households' socioeconomic status was included. Mothers' employment, which is captured by information on mothers' occupation, is also positively associated with day-care attendance. Children of mothers who do not have any records of their occupation (and were most likely homemakers) have the lowest hazard of day-care attendance compared to children of mothers who have an occupation recorded. Children of mothers who are maids have the highest hazard of day-care attendance. This finding shows that it is not the type of the mother's occupation as such (working, for example, at home versus outside the home) that is the main determinant for children to attend day-care but the mother's position in the household (dependent versus independent). While the occupation of being a maid in a household could allow for supervising even one's own children, in contrast to, say, factory work, the dependent position of the mother seemed to not have allowed for doing so. It is, however, impossible for us to say if mothers' employment and occupation are determinants for day-care utilization or if day-care utilization increases the odds of employment. As the review of the literature has shown, the decision for mothers to work and the decision to utilize institutional childcare are often made jointly and are hard to disentangle, particularly in studies based on cross-sectional data. By using longitudinal data, where occupational declarations considered precede the eventual event of first-attending day-care, we are partly able to overcome such a problem. However, our study is limited by the fact that the data does not allow us to know for how long individuals maintained previously declared occupations, even if by retaining occupational values valid for only five years after the declaration in the sources we reduce the size of potential biases Lastly, we find a positive association between the number of siblings and the hazard of attending day-care, which is in line with findings on day-care use in contemporary France, the UK and Italy (Mamolo et al., 2011) but contradicts findings on contemporary Canada and US (Coley et al., 2014;Varmuza et al., 2019). Whether more children increase the demand for and use of day-care is likely to be highly dependent on the particular setting. In contemporary settings, more children may be related to more household specialization between spouses. If one of the spouses becomes a full-time homemaker, the demand for childcare is reduced. Furthermore, the cost of childcare varies between contexts (Clawson & Gerstel, 2002), making it more or less accessible for families with more young children. In our historical setting, working class women did not have the choice of working vs. being homemakers. More children meant a higher economic burden, which increased the necessity to work and led to a larger demand for day-care. Day-cares were founded as charity projects and designed for the neediest. The number of children, as well as poverty, may well have been one of the selection criteria for granting a spot.
Previous studies on contemporary settings have shown that the availability of other types of childcare, such as relative care, reduces the likelihood of children attending institutional childcare. By using the variable other females in the household, we attempt to capture the availability of child supervision other than institutional childcare. As expected, children who live in households with two or more females have lower hazards of attending day-cares. Like the findings for mothers' occupation, the finding for two or more females in the household might indicate that the mother's position in the household is important. We can, however, not control for relative care outside the household (such as female kin living in the neighbourhood). Unfortunately, the data does not allow us either to control for recent migration to Landskrona since we do not observe that information for earlier cohorts.
The findings discussed are all independent of SES, which is controlled for in all models (except models 1 and 7, which control for the mother's characteristics). In our models, SES is strongly associated with the hazard of daycare attendance. There is a clear socio-economic gradient, ranging from the lowest hazard of day-care attendance for children in white-collar households to the highest hazard of day-care attendance for children in low-SES households. This finding contradicts findings from contemporary studies. In contemporary contexts, income and education are positively related to day-care attendance. Our finding is in line with the historical accounts of institutional day-care development in Sweden: they developed as poor-relief institutions for the neediest families, among them impoverished families and single mothers. In contrast to the qualitative accounts of day-care attendance in earlytwentieth-century Sweden, our findings nonetheless show that day-care attendance is not limited to the lowest socio-economic strata. Day-care attendance of children of higher socio-economic groups was more than a random encounter, as sometimes suggested in the literature. We find a clear socio-economic gradient across all socio-economic groups.
The likelihood of day-care attendance is substantially higher for the lowest strata of society, but it is not limited to it.
Contemporary research has shown that in Sweden as well as in other developed countries, parents of higher SES are more inclined to send their children to institutional day-cares. Such tendency implies an inefficiency of childcare services, since, as has been previously shown, the academic and economic returns to attending childcare are particularly large for children with disadvantaged backgrounds (e.g., Bakken et al., 2017;Campbell et al., 2002;Havnes & Mogstad, 2011;O'Brien Caughy et al., 1994). The finding that the SES gradient in day-care utilization has not always been positive but instead was negative is interesting in that regard. Partly, it opens the question if day-cares under the historical conditions and with the socio-economic composition described were equally beneficial for the attending children. Also, the findings give rise to the question of when the cross-over in childcare utilization by socio-economic groups has emerged in Sweden and other countries. These two questions have not yet been answered and are important areas for future research. Our study on historical determinants of day-care attendance, though limited in geographical scope, is important for understanding the full picture of day-care utilization and its effects, historically as well as today. Our study helps to stipulate more quantitative research on day-care attendance in historical settings, for example, settings with different work opportunities for women or different kinds of daycare institutions.

Notes
1. Sometimes kindergartens are presented as institutions exclusively for the middle-and upperclasses (Holmlund, 1999;Myrdal, 1935), which according to Hatje (2013) is not a correct description as already in the 1930s, many kindergartens specifically for working class children (so called folkbarnträdgårder) were in place. 2. Landsrkona city archive, 'Asylens för små barn' Jörgensgårdens barndaghem, Styrelsens protokoll med bilagor 1927-1936, AI:1, Protokoll 28 augusti 1936. 3. The SEDD is formatted into the Intermediate Data Structure (Alter & Mandemakers 2014). The database for analysis was created using programs developed by Quaranta (2015Quaranta ( , 2016. 4. During the study period income registers were digitized every five years. 5. One registry book of the infant school is missing (school year 1916/17). We do not know if the book is missing because the infant school was closed that year, or because the book was lost. Even for schoolyear 1918/19, we do not have any records. The reason for missing records that year is the Spanish flu epidemic, which forced all schools in the city to stay closed (Jönsson, 1997). A child crib was founded in 1916 as part of training courses for girls (Landskrona husmodersskola). We do not have records for these children. However, according to secondary sources, the crib was restricted to infants and took in only six children. The later expansion of this crib occurs after our period of study (Jönsson, 1997). 6. We also estimated all models for a more restrictive sample of children born between 1904 and 1930, who were observed for all years in which they could have potentially attended daycare, but the results remained consistent. 7. It should be noted that even if in SEDD data is available from January 1st, 1905, the day-care opened earlier. As a result of this, even if the first possible date for first day-care enrolment considered in the study was 1 September 1905, some children were first enrolled in day-care earlier. 8. Not married includes single women, widows and divorced women. 9. The unknown group consists of individuals without a registered occupation. We included them together with the lowest class (unskilled workers). However, we also estimated all models considering the unskilled and those with unknown occupations as separate categories, but the results remained consistent, and very similar patterns in relation to day-care attendance was displayed between the unskilled workers and those with unknown occupation. 10. For women information on occupation could be available on marriage records, catechetical examination registers and income registers, the latter digitized every five years. However, income registers are not available in Sweden for married women until 1947, since their income was included in the husband's income declaration. The occupational category used is therefore likely to be less accurate for married women. The unknown category comprises both women with no occupations (for example, those where the occupational notation in the registers stated no occupation or was left blank) as well as those with no available information on occupation recorded within the preceding five years. 11. Models were also estimating considering the unskilled and those with unknown occupations into separate categories and the results remained very consistent. These two groups in fact had very similar hazards of attending day-care. 12. We also estimated an additional model including an interaction between marital status and the number of females present in the household, but the interaction term was slightly above the threshold for statistical significance. 13. In Model 7 we did not control for marital status due to it's strong link with mother's occupation. However, the results remained consistent when also controlling for such variable.

Disclosure statement
No potential conflict of interest was reported by the authors.

Data availability statement
This work used data from the Scanian Economic Demographic Database (SEDD) and purposively digitized day-care records. Data from SEDD is freely available for research, subject only to legal restrictions for personal data and sensitive personal data. Using SEDD data requires a special application and decision in relation to GDPR and the the Law of Ethics Review (Lagen om etikprövning) for which requests can be made through https://www.ed.lu.se/databases/sedd/ sedd-public-access. Data from day-care records is not publicly available.