Information asymmetries between parents and educators in German childcare institutions

ABSTRACT Economic theory predicts market failure in the market for early childhood education and care (ECEC) due to information asymmetries. We empirically investigate information asymmetries between parents and ECEC professionals in Germany, making use of a unique extension of the Socio-Economic Panel Study (SOEP). We compare quality perceptions by parents and by professionals across 734 institutions. We detect considerable information asymmetries that differ across quality measures but less so by parental socio-economic background or center characteristics. Both similarly contribute to explaining variations in the information gap. We conclude that information is not readily available to parents; an issue that should be addressed by policy-makers.


Introduction
In recent years, shares of children attending early childhood education and care (ECEC) institutions have been growing across many industrialized countries. Studies on the impact of ECEC attendance point to positive effects on child development, especially in the domain of cognitive competencies, including verbal and numeric skills (for literature reviews, see e.g. Burger 2010;Gormley Jr, Phillips, and Gayer 2008;Auger et al. 2014;Camilli et al. 2010;Montie, Xiang, and Schweinhart 2008;Barnett 2011;Ruhm andWaldfogel 2012 andWaldfogel 2015). More generally, ECEC programs can improve school readiness, especially for disadvantaged children. Hence, such programs can decrease inequalities with respect to education and income (e.g. Mogstad 2011, 2015;and Heckman and Raut 2016).
However, an increasing body of research indicates that the effects of ECEC attendance depend on the quality of the interactions and the learning environment in these institutions (for economic evidence see e.g. Araujo et al. 2016;Bauchmüller, Gørtz, and Rasmussen 2014;Walters 2015; for other studies see e.g. Anders et al. 2012;Dearing, McCartney, and Taylor 2009;Keys et al. 2013;Elango et al. 2015). Barnett (2011), for instance, summarizes that the effects of participating in highquality programs on cognitive outcomes of children are much larger than for programs of average quality. Therefore it is important that children attend high-quality ECEC centers. With respect to ECEC quality, educational scientists mostly distinguish between structural, process and orientation quality (e.g. Kluczniok and Roßbach 2014). These quality dimensions work together in affecting child development.
Parents usually choose which ECEC center they would like their children to attend. To choose highquality ECEC, parents need to be able to assess and monitor the quality of these services. As parents can visit their children's ECEC centers on a daily basis, they are assumedly able to monitor quality and take action if they are not satisfied. Parents are expected to act as advocates of their children, ensuring that their early care and education experiences are adequate. ECEC professionals, including the center director and pedagogic staff, provide another perspective on ECEC center quality. Their assessments are important in enabling continuous internal monitoring of quality. Yet, parental assessments of the quality of a given classroom may well diverge from evaluations by the ECEC professionals. In general, differences in assessments between buyers and sellers of human servicesin this case early childhood education and carecan be conceptualized as information asymmetries. Such asymmetries are likely to arise because these services are not experience goods and parents as consumers must trust the ECEC professionals to some extent, as they cannot entirely assess quality (e.g. Mocan 2007).
From an economic point of view, information asymmetries may lead to low quality of service provision in a market. If parents cannot distinguish between high-and low-quality centers, they are less likely to be willing to pay higher fees (e.g. Herbst 2016). Under this scenario, high quality centers will exit the market, average quality will fall, and eventually the market will be filled primarily with mediocre quality centers (Akerlof 1970;Mocan 2007;Artz and Welsch 2014;Herbst 2016). In Germany, as childcare fees are usually fixed (e.g. Schmitz, Spiess, and Stahl 2017), the theoretical argument is slightly different: High quality childcare is more costly for the providers and requires more effort from staff at a given resource level. If parents do not observe and enforce high quality, there is no apparent reason to increase quality above minimum standards. Moreover, enforcement of minimum quality standards may be imperfect. So even in a highly regulated market, information asymmetries can be a reason for low quality.
As mentioned above, attendance in a low-quality ECEC center is likely less beneficial for child outcomes than attendance in a high-quality center. In extreme cases, low ECEC quality may even have negative effects on child development (see e.g. Baker, Gruber, and Milligan 2008). Thus, if certain groups of parents vary in their abilities or resources to evaluate ECEC quality, this might exacerbate inequalities. If, for instance, more educated parents are better informed, their children are less likely to attend low-quality ECEC services. 1 This is especially important as research shows that disadvantaged children from low educated or poor families particularly benefit from high ECEC quality. If their parents are less informed and they therefore attend ECEC centers of lower quality, education or income gaps might increase, not only in the short but also in the long run. 2 In addition to leading to market failure, information asymmetries between parents and providers may be problematic per se, as they reflect a lack of communication and interaction of pedagogic staff and parents. Several previous studies provide evidence of the benefits of regular interaction between parents and centers for children's wellbeing, e.g. through more supportive parenting styles (Ansari and Gershoff 2016), greater opportunities for physical activities (Froehlich Chow and Humbert 2014), and lower levels of corporal punishment in such centers (Khoury-Kassabri, Attar-Schwartz, and Zur 2014). High levels of ECEC quality therefore require regular and substantial interaction of ECEC professionals with parents, which could reduce the information deficit. Thus, significant information gaps between quality assessments of parents and ECEC professionals might affect children`s well-being negatively.
Only a small number of empirical papers examine information asymmetries in the market for childcare. 3 The existing studies often focus on the US market, including an economic study by Mocan (2007) and several studies by education scientists (Cryer and Burchinal 1997;Cryer, Tietze, and Wessels 2002; for an overview see Torquati et al. 2011;Bassok et al. 2017). These studies focus on differences in the quality ratings of parents and experts. Only a few studies (see below) have taken the quality rating of ECEC professionals into account explicitly. Their ratings seem highly relevant, as they are the most important actors for providing parents with information about quality. Moreover, it is plausible to assume that ECEC professionals are better informed about the level of quality of their services than parents. We therefore focus on information asymmetries between parents and ECEC professionals. Given the particularly high benefits of high levels of ECEC quality for children from disadvantaged backgrounds, we further investigate whether the information gaps differ by parental background.
Our findings for the highly regulated German ECEC market point to considerable information asymmetries concerning three groups of quality measures: (1) structural features; (2) educational and play activities; and (3) pedagogical focuses. In comparison to the judgements of the ECEC professionals, parents underestimate quality more often than they overestimate it. We derive our results from a unique data set, which allows us to compare quality assessments from parents and ECEC professionals of the same centers. We measure the information gap by comparing answers of parents with those of ECEC professionals with respect to exactly the same questions. Additionally, we introduce a theoretical observability rating of the quality measures used and verify the categorization based on our data. Indeed, the results suggest that the information gap is larger for less observable quality measures on average. Moreover, we analyze how the incidence of information gaps relates to parent and provider characteristics. We find only a few significant correlations between characteristics of parents or ECEC providers and information asymmetries. However, subgroup analyses show that lower levels of income and educational attainment are associated with higher incidences of information gaps.

The German ECEC system
In Germany, ECEC centers are heavily state-subsidized and usually charge income-dependent fees which are relatively low compared to most OECD countries (for a brief overview of the German ECEC system see : Schober 2014 andSpiess 1998). The OECD reports net costs of child care for a two-year-old child of a couple as 11% of average wages in Germany, while the EU average is 15% (OECD 2015). 4 Schröder, Spiess, and Storck (2015) show that financial contributions by families vary somewhat due to regional variation in subsidies and fee regulations. Around 18% of families with children who attend ECEC are completely exempt from fees. In many states, fees are regulated by the state typically determined by family income and the number of children in care (e.g. Spiess, Berger, and Groh-Samberg 2008). In comparison to the US market, the German childcare market is not very competitive (e.g. Artz and Welsch 2014), the share of for-profit providers is low at about 1% (Statistisches Bundesamt 2016)with most ECEC institutions operated by non-profit organizations or municipalities.
Since 1996 children aged three years and older are legally entitled to a slot in an ECEC center in Germany (see e.g. Spiess 2008). Thus, from age three onward almost all children attend formal ECEC services. The attendance rates for younger children are lower, but have been increasing from 16% in 2007to 33% in 2015(Statistisches Bundesamt 2016. 5 In the past, there were not enough places for children under three years of age: demand far exceeded supply, especially in Western Germany (Wrohlich 2008). However, two federal laws in 2005 and 2008 provided extra funding, granted prioritized access for children with parents in employment or education, and stipulated a legal right to a place in an ECEC institution for all children aged one year or older from August 2013. As a result, parents are no longer as restricted in their choice of ECEC institutions: 91% of the parents in our data report that they had a choice between at least two centers.
In Germany, each state has its own regulation for minimum standards of quality. Child-teacher ratio is one of the few indicators that are precisely, albeit differently, regulated in all states. There is significant variation across states with respect to the level of regulation in terms of other quality indicators, such as opening hours, parental fees, building requirements and maintenance and group size (e.g. Bock-Famulla, Lange, and Strunz 2015). Moreover, all German states have implemented pedagogical guidelines (Bildungspläne, Diskowski 2009). However, these plans vary by state and are not mandatory in most states. Therefore, ECEC quality varies across regions and centers.
Despite a relatively high overall degree of regulation, an empirical study of ECEC quality in Germany shows that out of 188 evaluated ECEC centers for children below the age of three, the majority offers quality that can be classified as sufficient but no more. 10% of the centers were assessed as good and another 10% as insufficient (Tietze et al. 2012). 6 Furthermore, 44% expressed concerns about the quality of ECEC centers. 7 In Germany fees are not usually a signal of quality, given the relatively strict fee regulations in many states. There is also no overall national accreditation system like that administered by the National Association for the Education of Young Children (NAEYC) in the United States (e.g. Xiao 2010), which consumers may use as a source of information. Furthermore, there are no Quality Rating and Improvement Systems as found in many US states (e.g. Herbst 2016).

Previous studies and differences in information asymmetries
Most studies on information asymmetries in childcare markets focus on differences in quality ratings between parents and experts. Among the very few studies which analyze differences between parents and ECEC professionals, most look at differences in ideas and perceptions about ECEC quality (e.g. ECCE Study Group 1997;Pierrehumbert et al. 2002;Weaven and Grace 2010;Harris and Tinning 2012). To our knowledge, there is only one study which focuses on reported quality levels and also considers the ECEC professionals' assessments. The study by Barros and Leal (2015) is based on a Portuguese sample and shows that parents and ECEC professionals overestimate quality in comparison to experts but that there is a higher correlation between professionals' and experts' ratings than between parents' and professionals' ratings. Thus, they find information asymmetries but do not analyze them more in-depth. Their results imply that differences are lower for aspects which are relatively fixed such as the space available for adults in the center. They further state that parents' and ECEC professionals' ratings may be more based on what they would quality like to be than on actual observations and that parents may lack suitable reference points for assessing high quality, that is, some experience of high quality centers.
The majority of studies which investigate information asymmetries between parents and experts show that parental and experts' assessments of various dimensions of the classroom operation correlate, but that there are information asymmetries. Mocan (2007) demonstrates the existence of such information asymmetries in the US childcare market, which provide an explanation for low average quality. As in most studies of this type, the process quality of the ECEC services is rated by experts using the Early Childhood Environment Rating Scale (ECERS). An earlier study by Cryer and Burchinal (1997) for the US arrives at similar results as Mocan (2007). In a more recent study, Cryer, Tietze, and Wessels (2002) use a sample of parents of preschoolers in the US and compare this with a sample of parents in selected German states. Their findings show that in both countries parents assign substantially higher quality scores to their children's classrooms than trained observers do and that parental quality assessments are influenced by the relative importance they attribute to specific aspects of quality. The representativeness of these results may be limited, as the respective samples only consider children at specific age groups, and are limited to selected regions. 8 A few other North American studies with similar approaches are summarized in Torquati et al. (2011), Howe et al. (2013, Bassok et al. (2017).
Independent of the ECEC system, information asymmetries of all kindsthose between parents and experts as those between parents and ECEC professionalsmight be related to the observability of different ECEC quality aspects and may be more prevalent among specific groups of parents and providers. Firstly, the information gap may differ by the level of observability of different quality aspects. Parents rarely have the opportunity to spend much time in their children's classrooms observing the various quality dimensions of daily care practices. Studies indeed show that parents tend to spend relatively little time at a centertypically limited to when they drop off or pick up their children, or attend parent meetings. Most of the information that parents receive is secondhand, e.g. based on other parents' experiences, reports of their own child, the teaching and administrative staff, as well as through the materials that the child brings home, such as artwork (see Cryer, Tietze, and Wessels 2002;Artz and Welsch 2014). Even if they do spend some time at the childcare centers, they might not have sufficient knowledge to rate the quality in ways compared to trained observers. As a result, Mocan (2007) distinguishes between aspects of services that are 'easy to observe,' such as opening hours, and aspects that are 'difficult to observe,' such as the quality of teacher-child interaction. His results, and also those by Cryer and Burchinal (1997), confirm that when parents and external experts rate the quality of easy-to-monitor aspects of care, differences in scores between parents and experts are smaller than when they rate aspects that are more difficult to monitor. 9 Secondly, the information gap might differ by demand-side, parental characteristics, such as education, income and employment status. There are several empirical studies showing that there are socio-economic differences in the assessment of quality by parents (e.g. Johansen, Leibowitz, and Waite 1996;Hagy 1998;Blau and Hagy 1998). Higher educated parents might have lower costs in searching for the right information, have better search strategies, or have better informed networks (see e.g. Meyers and Jordan 2006). Parents working longer hours might value quality more as their children use such services longer; however, they might also have more time constraints when searching. Cryer, Tietze, and Wessels (2002) find that parents with lower educational attainment tend to rate the importance of the ECEC quality characteristics significantly higher than more highly educated respondents in both Germany and the US. Less educated parents tend to rate the quality of their children's classrooms slightly higher on the ECERS Parent Questionnaire (ECERSPQ) than parents with higher levels of education. Similarly, Mocan (2007) shows that parents with (at least some) college education assess quality more accurately than less educated parents. Parents using full-day care are more accurate in their predictions. Artz and Welsch (2014) assume that parents in high income neighborhoods have better resources for evaluating the quality of ECEC services.
Thirdly, the levels of information asymmetries may vary by the supply side characteristics of the ECEC providers. Parents might use center characteristics as predictors for quality. Centers that are under market pressure may be less able or more willing to communicate effectively with parents. Similarly, the size of the center might matter as institutionalized communication processes might require a certain minimum size of an institution. Mocan (2007) shows that the provider type has an effect on information asymmetry: parents rate the quality of public providers lower. In addition, the proportion of white children at a center is associated with a perception of higher quality, while the proportion of children whose parents receive childcare subsidies relates to lower parental quality ratings.
We analyze these three types of information asymmetries using a unique German data set. The quality aspects we analyze relate to structural features of quality for the most part as well as some aspects related to process quality. Structural features are usually defined as comprising easily observable, quantifiable and regulatable features of the ECEC context, such as group size and child-staff-ratio.
We extend previous studies by measuring the incidence of an information gap regarding various ECEC quality aspects between the buyers and the sellers as well as the size of such asymmetries. Furthermore, we examine whether information asymmetries differ between observable and unobservable aspects as well as how they relate to the socio-economic background of parents and to specific characteristics of ECEC institutions. Such an analysis allows us to investigate the extent to which consumers have difficulties in extracting information from ECEC professionals due to limited observability of quality aspects, due to socio-economic characteristics of the parents, or due to provider characteristics. Any such difficulties may result in education inequalities for the children in care. We perform this analysis for a German sample that is not restricted to particular states and we refer to a market for ECEC services that is, in comparison to the US market, much more regulated and where services for basically all children are subsidized.

Data
Our analyses are based on a subsample of the German Socio-Economic Panel (SOEP), the SOEP-extension sample 'Families in Germany' (FiD), and a SOEP-supplementary study that includes additional information from both parents and ECEC centers. All studies belong to the SOEP which is the largest and the longest-running multidisciplinary longitudinal study in Germany (Wagner, Frick, and Schupp 2007). In 2013, 24,113 adult members of 14,170 households participated in the study. We use the 2013 SOEP wave in conjunction with the 2013 FiD wave. FiD is a dataset that specifically surveys families with young children and also targets families that are typically under-sampled in general surveys: low income, single parents, and large families. In 2013, a total of 6853 individuals in 3923 households participated (Schröder, Siegers, and Spiess 2013). The structure, content and thus the variables of these two data sets are virtually identical, so they can be analyzed jointly using weighting factors. The 2013 SOEP supplementary K 2 ID study (see  includes information on the quality of facilities attended by children who lived in a SOEP or FiD household at that time. 10 In a first step of the K 2 ID project one parent of each child below school age was surveyed in order to gather information on the ECEC center their child or children attend. This includes the address of the center and parent's assessment of a large number of indicators regarding its quality. 11 The second step was to collect indicators of structural, orientation, and process quality directly from the director of each center and from the main group educator of the group attended by the SOEP/FiD-child under study. 12 In our analysis, we only consider quality measures where information from both the parent and from the ECEC director/group educator is available. In this case, the parents and the ECEC professionals were asked identical questions. 13 Given the design of our study, we thus only compare parents and ECEC centers which are linked via the attendance of the children. Depending on the quality measure, we can compare the information from 346 to 725 parents and the ECEC institution that their child attends. This relatively broad range is related to the fact that the FiDquestionnaire included a larger number of quality-related questions. For a detailed description of the new SOEP-K 2 ID-study, which was conducted by ourselves together with the SOEP, see Schober et al. (2017).
We compare the quality assessments of parents and ECEC professionals. We argue that none of them has particularly high incentives to inflate their perceptions. Inflated ratings might be due to parents not wanting to report that they have chosen an ECEC institution of low quality for their child. ECEC professionals might not want to report low quality, as this implies that their work is insufficient. However, as we designed the study ourselves, we tried to minimize overreporting in both cases. Towards both actor groups, we emphasized that the study does not aim to evaluate quality of particular institutions, but rather seeks to draw general conclusions for policy makers and researchers which might improve the conditions for ECEC professionals and children. Even if overly positive ratings occurred, we would expect that the bias for both groups would go in the same direction. Moreover, we asked for perceptions with respect to many quality measures and find large variations which does not point to stringent and systematic overreporting of quality.
Parents and ECEC institutions in most of our subsamples were surveyed between October 2013 and November 2014. Our total sample includes 1870 parents and 680 ECEC institutions. For 82% of children, the mother answered the parental questionnaire, for 18% the father did. The response rate for the parental questionnaire is reasonably high at 74%, the response rate of the institution questionnaire is also high for this type of survey at 55%. We use survey as well as non-response weights to account for selective participation in the study. These survey weights are generated using extensive information about non-respondents that is available through the SOEP general survey for all individuals that were sent the additional questionnaire (for more information on this weighting procedure, see Schober et al. 2017).

Definition of information gap
We measure the information gap using a binary variable that indicates if there is any gap. Depending on measurement scales, we construct binary gap indicators in two ways: (a) For categorical variables (existence of written pedagogical concept, the educational and play activities and pedagogical focuses), the variable takes the value one if the answers from parents and centers match and zero otherwise (C ij stand for the center information, P ij for the parental information, the index I for children and j for groups): For continuous variables (i.e. most structural features), the gap is defined based on a threshold: The threshold is set at 10% of the information provided by the ECEC center. As part of robustness checks, we also present results for the following other thresholds: exactly matching information and thresholds of 5%, 15% and 20%. Depending on the item, the information is either provided by the director of the center or by the group educator of the child. The center director provides information on opening times and pedagogical focuses, whereas the group educator provides all other information. 14 P ij is the respective rating of the parent. A threshold value of 10% provides a way to deal with random errors in the evaluation from either parents or centers. In case parents indicate that they do not know the response to a specific question, the indicator is set to zero, that is we count this as a mismatch between parent and institution answer regardless of the institution answer. Alternatively, one may want to treat these answers as missing. In robustness checks, we also run our models under this assumption.

Quality measures
As noted above, our quality measures mainly relate to structural features and, to a smaller degree, to process quality. If we think of a production function of educational attainment all of these measures could be considered as educational inputs, which might affect children differently depending on their parental background and thus the quality of their home learning environment. Table 1 lists the three groups of quality measures that we focus on. All of them are aspects of ECEC quality, as defined by education scientists. We grouped them according to other ECEC quality scales and the expertise of ECEC quality experts (see Schober et al. 2017). One grouping aspect is if ECEC policy makers can regulate and monitor the quality aspects. This is relatively easy for structural features. As part of the structural features, we also consider if a written pedagogical concept exists at all. Offers of particular education and play activities are less regulable, and more closely related to the process quality. Structural features however may affect if ECEC centers can offer particular activities. The last group of ECEC quality characteristics covers the pedagogical focus and content dimensions of the pedagogical concept, which also relates to the orientation quality of ECEC centers. For each quality aspect, we assign a degree of observability based on theoretical considerations. This measure combines the narrowly defined observability of the information and the amount of communication necessary for gaining information on a certain aspect. For aspects that are typically not directly observable by parents, we consider how much effort it likely takes for parents to acquire information regarding the respective quality aspect. For instance, information regarding activities is more likely to be regularly volunteered by children and educators than information on educational qualifications of all group educators.
Structural features are usually defined as comprising quantifiable and regulable features of the ECEC context. They cover easily observable aspects, such as the opening hours, overall group size, and the existence of a written pedagogical concept, as well as slightly less observable aspects, such as the children-per-educator ratio and the number of educators in the group (categorized as medium observability). Finally, we consider the number of children with non-German family language in the group and the number of educators without a professional degree in ECEC as two structural aspects that can only be observed with some effort (low observability). All of these structural features are relatively easy to regulate and thus easy to address by ECEC policy makers. While for most aspects it is obvious that they may be considered inputs of educational attainment and how  (6) and (7) do not add up to the figures in column (5), see also column (8). Significance levels of t-test for equality of means from (1) and (2) in column (3):*10%, **5%, ***1%. Statistics are weighted using sampling and nonresponse weights. Source: SOEP v31 and K2ID-SOEP. they contribute to an increase in ECEC quality, 15 this might not be obvious in respect to opening hours and the number of children with non-German family language. Longer opening hours might support parents in combining work and family. This might contribute to an increase in household income, which might again benefit child development (e.g. Carneiro and Ginja 2016). This applies in particular to low income households which need the income of two working parents. Moreover, disadvantaged children might benefit from longer opening hours, as this allows them to stay longer in a qualitatively better environment instead of spending more time in a home environment of poorer quality. Furthermore, the so-called dosage effect of ECEC attendance very much depends on the age of the child (e.g. Loeb et al. 2007). Regarding group composition, several studies document that a higher average level of peer abilities in an ECEC center is positively associated with children's cognitive and language skills (e.g. Stahl, Schober, and Spiess 2017). A large proportion of children with non-German family language in ECEC centers has been shown to be negatively associated with German language acquisition of children with a non-German family language (e.g. Klein and Becker 2017).
Another set of education inputs covers education and play activities, including music education, language activities, and outdoor activities. These likely differ in terms of observability. On the one hand, foreign language activities as well as painting and arts activities yield direct results that the children can show to their parents. Similarly, day-trips or excursions are usually announced to parents. We therefore categorize them as highly observable. On the other hand, observing math, science, or other daily educational activities, which are routine, is more difficult for parents and, therefore, these are categorized as medium observable.
The third group covers the pedagogical focus on subjects such as language, math, motor function or health. We consider most of these items as medium observable, as parents are likely to ask about them when making their decisions about where to enroll their child. We make an exception for the foreign language activities as these usually require special training for the educators (or even cooperation with external staff) and label this aspect highly observable.

Parental and ECEC center characteristics
The demand side variables capture the socio-economic background of the family, specifically maternal employment status, her educational attainment, the household's net income, as well as an indicator for migration background of the mother.
The supply side variables include indicators of the size of the center, whether it is run by a public provider, and the share of children exempt from fees. Furthermore, the models include a scale on the frequency of communication between parents and the institution. This variable is the mean of four items about how often certain types of communication take place, including daily conversations or parent evenings. 16 Moreover, our models control for the length of time the child attends the center with respect to its daily hours and the overall period, as the information gap may decrease as parents learn more about the center's quality. We also control for the child's gender and age, if the child has a chronic disease, the number of children in the household, the gender of the parent answering the survey, the time between parental and institutional interview in days, plus regional indicators for East Germany and urban areas. We also control for the level of the quality measure as reported by the ECEC director. We test for multicollinearity of the variables and include only those that are not multicollinear. For descriptive statistics of additional variables, see Table A-1 in the appendix (see supplementary material).

Methods
A main contribution of this paper is the in-depth descriptive analysis of the information gap between parents and ECEC staff. To examine how information gaps relate to demand and supply side characteristics, we use logistic regression models.
The existence of an information gap is estimated as follows: Where D ij is the binary variable as specified above, X ij is the vector of socio-economic and centerspecific background variables, Z ij is a vector of control variables including a constant term and C ij is the level of quality as reported by the respective person in the institution. 17 1 ij is the idiosyncratic error term, which we cluster at the group level. 18 We furthermore test for nonlinearities in important SES variables and interaction effects between parental SES variables.

Information gaps and observability
Initial bivariate results show that the parental and ECEC professionals' assessments are significantly different for a large share of the quality aspects (Table 1). We focus on structural features first: With respect to opening hours, parents slightly underestimate the actual opening hours; however, the difference is only about 20 minutes. Parents report smaller group sizes, fewer educators for the group, and they report a larger number of non-German speaking children in the group than the ECEC professionals. If we assume that, ceteris paribus, quality increases with smaller groups, then in this respect parents report slightly higher levels of quality than ECEC professionals. However, if we further assume that, ceteris paribus, fewer educators per group and more children with a foreign family language may relate to a more difficult learning environment, then parents underestimate the quality compared to ECEC professionals (see Table 1, column 9). Moreover, as expected based on our theoretical observability rating, a large percentage of parents report that they feel unable to provide any information on the number of non-German speaking children and the share of educators without degree, the two aspects of low observability. In addition, 46% of parents also indicate that they do not know if the centers have a written pedagogical concept, which points to information problems, as one of its inherent purposes is to inform parents. When comparing the empirically measured information gaps with our theoretical grouping by observability, the three aspects with the highest theoretical observability, opening hours, group size and existence of written concept, show, as expected, very high shares of no information gaps (Table 1, column 5). Of these, the degree of match for the existence of a written pedagogical concept is the lowest and relatively close to those aspects that we labeled as 'medium observable.' While parents tend to overestimate the quality concerning group size, a highly observable item, they tend to underestimate quality in the cases of aspects with low observability, such as the number of non-German speaking children in the group and the overall share of educators without a degree. The incidence of no information gap is highest with respect to the opening hours and lowest for the number of non-German speaking children, which seems plausible given that information on opening hours is easy to observe whereas the number of children with a non-German family language is not.
With respect to the education and play activities, four out of seven differences are statistically significant; the exceptions being 'foreign language activities,' 'painting/arts,' and 'music education.' It may be that these activities are especially important for the parental ECEC selection processes and, thus, parents gather more information from the ECEC professionals about them. Whenever there is a significant difference, parents tend to report fewer activities than ECEC professionals, implying that they underestimate quality: While 72% of the parents report that the center offers German language support activities, the share is higher among ECEC professionals (91%). The incidence of no information gap is highest for activities such as 'trips into the nature' or 'painting and art activities', both easily observable activities, while the share is lowest for less observable activities related to 'math and science.' On average, 64% of parents report that math or science activities are offered, while 90% of ECEC professionals report that they offer these activities. Hence our theoretical observability grouping seems to fit reasonably well with the observed information patterns for education and play activities. The overall level of information gaps for education and play activities is lower than for the structural features. In part, this may be explained by the fact that the activities were measured on a binary rather than a continuous scale.
The incidence of no information gap is slightly lower with respect to the pedagogical focus than for the shares for education and play activities. One quarter of parents report that there is no pedagogical focus as opposed to only 12% of the ECEC professionals. In line with the observability grouping, the assessment of ECEC professionals and parents is mostly identical with respect to a foreign language focus. Surprisingly, parent and ECEC professional assessments also match well for a math focus. For all other types of activities, we observe between 60% and 80% of cases with no information gap. However, for only three out of eight aspects are the mean differences in the quality assessments statistically significant, as over-and underestimations offset each other for the other aspects.

Parental and ECEC institution predictors of information gaps
Next we present results of our multivariate analyses, which investigate how the information gaps relate to characteristics of parents and ECEC centers. Table 2 reports the results for the structural features: With respect to the opening hours the probability of no information gap is higher for mothers working full-time than for those working part-time. Thus, mothers working longer hours seem to be better informed on this quality measure, which is particularly important for them. The effect of household income is only statistical significant at the 10% level and negative, which means that higher income households have a lower probability of no information gap. Other model specifications point to a non-linear effect of household income. This effect is mainly driven by parents in the third household income quartile, who have an 18 percentage point lower probability of an information match than households in the lowest income quartile. 19 One explanation for this might be that for economic reasons households in the lower income quartile may have to perform more market work and thus may care more about opening hours than others. Moreover, full-time employed and non-employed mothers are less informed about the existence of a pedagogical concept than mothers working part-time. Parents with a migration background also appear to be less aware whether the ECEC center has a written pedagogical concept. We observe few statistically significant associations with respect to less observable quality characteristics with one exception: Information gaps with respect to the number of children not speaking German at home are less likely to be found among higher educated mothers. Overall, one can conclude that information asymmetries for highly observable quality measures are more frequently related to parental background than those for less observable characteristics.
We find few coherent significant associations of ECEC center characteristics with information gaps in terms of structural features. The probability of no information gap with respect to two of the most observable quality measures, opening hours and group size, is higher for public than for non-profit providers. The share of children with fee exemptions correlates negatively with the probability of no information gap in respect to the share of educators without a degree. This means that in such cases the information gap is higher. However, there is no clear pattern in terms of center characteristics being more strongly associated with gaps for more or for less observable characteristics.
Concerning other factors, such as the number of hours a child spends in ECEC and the length of tenure at a given center increase, the information gap regarding the ECEC's quality decreases. The latter is plausible as parents had more time to acquire information about quality. Also, the levels of quality reported by the ECEC professionals are significantly related to the probability of no information gap. 20 Overall, demand and supply side factors appear to be of similar importance with some statistically significant relationships related to the demand and some to the supply side factors, depending from the existence of information gaps. To verify this, we estimate further models: A basic model with the control variables only, and then two further models with either adding the demand or the supply side factors. The shares of variation in the existence of no information gap explained either by demand or by supply side factors (as measured through the pseudo-R 2 of the respective models) are very similar among the models (see Table A-2 in the Appendix 21 ). Thus the group of demand and supply side variables explain about the same share of variance in existence of no information gap; in the case of the children per educator the demand side factors (and the control variables) explain 17% of the variation while the supply side factors explain 18%, in the case of a written pedagogical concept the respective shares are 31% and 25%. For selected quality measures (and respectively information gaps) we also show how the results vary when the threshold for mismatch varies (see Table A-3.1 and A-3.2 in the Appendix in supplementary material). The differences in thresholds only matter for three out of six quality measures, namely the opening hours, the group size and  Notes: Standard errors are clustered on the ECEC group level; Significance levels: *10%, **5%, ***1%; Estimations are weighted using sampling and nonresponse weights; additional control variables: time between parent and institution interview in days, if the mother or the father answered the questionnaire, the number of children in the household, gender and age of the child, if the child has a chronic disease, if the educator has a degree focusing on ECEC, if the educator recently participated in professional development, influence of the federal pedagogical guidelines, if the institution is organized in groups or not and indicators for East Germany and urban areas; for item nonresponse means are imputed and binary indicators added to the models. Source: SOEP v31 and K2ID-SOEP. the number of children per educator. Thus Table A-3.2 (see supplementary material) only presents the estimation results for the information gap of the first three quality measures. The results show that for smaller thresholds, results become less stable compared to our main specification. Changing the threshold from 10% to 20% hardly affects the results while changing them from 10% to an exact match leads to very different results. Most associations which are strongly statistically significant in our main specification do not change when using larger thresholds. We interpret this as an indicator that the results using very small thresholds are more vulnerable to measurement error.
As another robustness check, we test how the information gaps differ if we treat parents' 'don't know' answers as missing values instead of a mismatch. The results are shown in Table A-4 in the Appendix (see supplementary material). The size of the information gap only changes notably in the case of the existence of a pedagogical concept, which is due to a particularly high share of parents reporting to not know if a pedagogical concept exists. Moreover, we test if our estimations change due to differences in the share of missing values. Overall, these estimations result in a loss of power and thus are difficult to compare with the main specification. 22 Particularly, the significant Table 3. Logistic regression of no information gap between parent and ECEC professional assessments of educational and play activities offered in the group (1 = no information gap); marginal effects with standard errors in brackets.

Painting, arts
Foreign languages 2365 Notes: Standard errors are clustered on the ECEC group level; Significance levels: *10%, **5%, ***1%; Estimations are weighted using sampling and nonresponse weights; additional control variables: time between parent and institution interview in days, if the mother or the father answered the questionnaire, the number of children in the household, gender and age of the child, if the child has a chronic disease, if the educator has a degree focusing on ECEC, if the educator recently participated in professional development, influence of the federal pedagogical guidelines, if the institution is organized in groups or not and indicators for East Germany and urban areas; for item nonresponse means are imputed and binary indicators added to the models. Source: SOEP v31 and K2ID-SOEP. association between migration background and an information gap with respect to the existence of a written pedagogical concept is affected, as many parents with migration background report to not know if a pedagogical concept exists.
Further, we investigate if the associations between SES and information gaps vary across income and education groups. We are particularly interested in exploring whether we observe education or income differences among migrants, as previous studies have shown that migrant children on average attend centers of lower quality . We calculate interactions effects as described above. For mothers with a migration background, our results show that a household income below the median correlates negatively with no information gap regarding a written pedagogical concept and the number of educators. This is not the case for parents without a migration background. The interaction between migration background and low educational qualifications is not statistically significant. This might be due to too small sample sizes. Similarly, lower income correlates negatively with no information gap with respect to the share of educators without a degree only for mothers without a migration background. This result is of particular Table 4. Logistic regression of no information gap between parent and ECEC professional assessments of pedagogical focus (1 = no information gap); marginal effects with standard errors in brackets. Notes: Standard errors are clustered on the ECEC group level; Significance levels: *10%, **5%, ***1%; Estimations are weighted using sampling and nonresponse weights; additional control variables: time between parent and institution interview in days, if the mother or the father answered the questionnaire, the number of children in the household, gender and age of the child, if the child has a chronic disease, if the educator has a degree focusing on ECEC, if the educator recently participated in professional development, influence of the federal pedagogical guidelines, if the institution is organized in groups or not and indicators for East Germany and urban areas; for item nonresponse means are imputed and binary indicators added to the models. Source: SOEP v31 and K2ID-SOEP.
interest, as this quality measure is not easy to observe. Overall we see that lower SES parents (with respect to income and education) are more affected by an information gaphowever, this varies by maternal migration background. 23 With respect to information gaps related to education and play activities (Table 3), parental background factors matter for three out of the four highly observable quality measures. In particular, the knowledge of painting activities is highly influenced by demand side factors, such as household income. Parents with higher incomes have a higher probability of no information gap here. Further analyses which test a non-linear income effect show that this is mainly driven by parents Notes: Probabilities predicted from logistic regression models as described above. Error bars indicate univariate 95% confidence intervals on marginal effects obtained via delta method. The horizontal axis refers to quality reports of ECEC professionals. The vertical axis indicates the percentage of cases with no information gaps.
in the highest income quartile. They are also better informed about trips to libraries than parents from lower income quartiles. Demand side factors are not significantly related to medium observable quality measures. No clear patterns emerge with respect to supply side factors and their relationships with quality measures of high or medium levels of observability. If the center is smaller, under pressure, or if more frequent communication with parents takes place, the probability of no information gap is greater in several cases (in respect to center size the effect sizes are not substantively significant). Very important as a predictor for information gaps related to these measures is the reported level of quality: If an activity is offered at the center, the probability of no information gap increases for almost all measures. However, as with structural quality features, overall our estimations with solely controls and demand side variables or solely supply side variables show now great differences in the share of variation explained by one group of factors or the other.
The information gap patterns are different with respect to the pedagogical focusindependent of the observability of the quality measure, they appear to depend more on demand side factors (Table 4). Household income correlates positively with no information gap with respect to focuses on 'foreign language,' 'music,' 'health,' and 'motor functions.' Further estimations show that in two out of four cases this is driven by parents in the highest income quartile, but this effect is only weakly significant. The pattern is less clear for the others. The employment status of the mother and her education also affect the existence of an information gap. However, the direction of the associations differs by quality measures. Parents with a migration background are more likely to report that their child's ECEC center lacks a specific focus than the center itself. In respect to the statistical significance the most important supply side predictor of the probability of no information gap is the center size, however the effect sizes are not substantively significant. For all information gaps related to pedagogical focus, we find some statistically significant relationships with parental characteristics. Although five out of eight measures of these information gaps also related to center characteristics, only one characteristic shows consistent patterns across various information gap measures.
In a final step, we test how information gaps relate to the evaluation of the quality measure by the ECEC-professionals. Figure 1 shows predicted probabilities of no information gap depending on the quality level reported by the ECEC-professionals for selected outcomes. 24 For these estimations, we rerun the multivariate logistic regression models (see Tables 2-4). The results can be interpreted as follows: When, for instance, the ECEC-professionals report that no written pedagogical concept exists, less than 10% of the parents give the same answer. Similarly, when the ECEC-professionals report that more educators are responsible for the group or more have no degree, the information gap increases considerably. Similar patterns emerge for some activities and pedagogical focuses. Thus, even if the ECEC professionals report less favorable quality conditions, the parents do not seem to observe them. Possibly they might have a standard ECEC center in mind with a written pedagogical concept, one, professionally trained educator per group and a focus on certain activities.

Conclusion
In this paper, we analyze information asymmetries between the parents and ECEC professionals concerning various quality measures in the German ECEC market. We contribute to the literature by investigating information asymmetries in a highly regulated childcare system, by focusing on the perspectives of parents and ECEC professionals and by considering structural quality indicators with varying levels of observability as opposed to focusing on process quality, which is generally hard for parents to assess. The results of this study may be transferable to other universal ECEC markets with low ECEC prices and without any rating systems. We investigate how the probability of information asymmetries relates to three dimensions: (i) theoretical observability of the respective quality aspects; (ii) parental socio-economic background; and (iii) characteristics of the ECEC center. To do so, we exploit a rich data set, with information regarding parents and ECEC professionals, as well as their respective quality assessments based on identical quality measures.
With respect to structural features, information asymmetries are relatively high, ranging between 42% 25 and 87% mismatches between the information provided by parents and ECEC professionals. Overall, information asymmetries are lower for the existence of education and play activities than for existence of a pedagogical focus. This indicates that parents are better informed about day-to-day activities than about the relatively abstract concept of a pedagogical focus. Remarkably, we find that for most aspects, where there are significant information asymmetries, ECEC professionals report a higher level of quality than parents. This is in contrast to studies that compare parental assessments with expert ratings, which usually found parents to overestimate quality. To better understand the extent to which quality assessments not only of parents but also of ECEC professionals may be subject to bias and may contribute to inadequate information about ECEC quality for parents, future studies should further examine potential sources of biases in ECEC professionals' quality assessments of their own ECEC institutions.
Our theoretical grouping of observability suits the data reasonably well. In particular with respect to structural features and activities: information asymmetries are more likely to occur for aspects that are difficult to observe or require parental enquiry. Our findings also indicate that the socio-economic background of the parents and the characteristics of the centers matter to some degree; both groups explain a similar share of the variance in information asymmetries. For structural quality features as well as the education and play activities, we find that parental characteristics are more strongly associated with information gaps regarding highly observable characteristics compared to less observable ones, which seems plausible. Information gaps regarding the pedagogical focus also appear to be influenced by parental background variables even though we considered them as medium observable. Perhaps some, but not necessarily all, information about the pedagogical focus is frequently accessible to parents and some groups of parents, therefore, feel they should know about this. In addition, information gaps with respect to the pedagogical focus are also associated with center characteristics suggesting that some institutions provide more information on this aspect than others.
Interestingly the information gap frequently correlates strongly with the level of the respective quality measure. Yet the direction of the relationship varies. If ECEC professionals offer the respective education and play activities, the probability that both parents and ECEC professionals report this increases. However, if ECEC professionals report one or more types of pedagogical focus, the probability of a match between parents' and ECEC professionals' reports decreases.
With respect to information asymmetries of households that are potentially less privileged, the following findings are of particular interest: Parents with a migration background are less likely to accurately know about the existence of a written pedagogical concept and whether the ECEC centers have a pedagogical focus on German language support. These results are important as these quality features relating to language competencies are likely to be especially important for children with a migration background. Thus, one may argue that children in minority households may particularly benefit from government-provided information regarding childcare quality. Although we find no other systematic pattern for parents with lower SES, the results by other studies that they do attend ECEC centers of lower quality underlines the conclusion that they, in particular, may need some support to better asses high levels of quality.
We find considerable information gaps for most quality features. This might be an indication for a less than optimal interaction between parents and ECEC professionals, and thus might influence child well-being. Yet, overall the gaps are only moderately related to parent and center characteristics. One possible explanation may be that parents in Germany rely on ECEC sector regulations and do not feel the need to inform themselves more thoroughlythis might apply to all parents irrespective of their socio-economic background. Indeed, although the quality in the German ECEC market is mediocre according to scientific standards (Tietze et al. 2012), variation is also relatively low. Investing into gaining more information about quality may, therefore, not be optimal for many groups. This is also in line with the fact that prices are uninformative about quality and there is no external quality rating system. Yet one may assume that more uniform quality assessments may benefit the daily interactions between parents and ECEC professionals, who are not just parties to the exchange of a service good but also actors both interested in the welfare of the children enrolled in ECEC services. In addition, one may argue that parents should advocate for higher quality services if, on average, parents asses the quality lower than the ECEC professionals.
To reduce the information gap on ECEC quality between parents and ECEC professionals and thus to improve the quality of ECEC services and ultimately child well-being, several possibilities may be considered: First, the government may set incentives for ECEC centers to provide more information to parents before they make their ECEC decision and to regularly communicate thereafter. Second, a nationwide accreditation system might help to improve the quality assessment of parents and ECEC professionals, as it helps to establish a common basis of what represents good (minimum) quality. Third, a rating system based upon the nationwide quality accreditation system might further help overcome information asymmetries between parents and ECEC professionals (see e.g. Spiess and Tietze 2002). However, a rating system should be implemented carefully, as the US experience has shown that it might increase inequalities in the use of high ECEC quality. If these ratings systems raise ECEC costs, this may come at the cost of some children from disadvantaged households who may have to switch to informal care, while their advantaged counterparts are more likely to use ECEC services of higher quality (see e.g. Herbst 2016). Thus, to promote children's wellbeing across socio-economic groups, it would be important to prevent childcare fees for children from disadvantaged families from rising. Notes 1. Less educated parents who provide a lower-quality home learning environment may also assess the quality of the ECEC environment better, as it might compare favorably relative to their home environment (see Mocan 2007). 2. See e.g. Stahl, Schober, and Spiess (2017), for an empirical study which shows that disadvantaged children receive lower-quality ECEC in Germany. 3. For summaries, see e.g. Blau (2001), Helburn and Bergmann (2002) and Fenge and Wrede (2015). 4. The OECD calculates this for a couple where the first earner earns 100% and the second earner earns 67% of the national average wage. The calculations use data from the year 2012 and assume full-time center-based care. 5. However, there are considerable social disparities for this group: Children under the age of three with migration background or from low income families are significantly less likely to attend childcare (see Schober and Spiess 2013). 6. Surprisingly, several studies show that overall parents report a relatively high level of satisfaction with ECEC although this varies by quality aspects and is related to actual levels of quality as assessed by parents . 7. Own calculations based on the 2013 wave of the 'Familien in Deutschland' (FiD-data), see below. 8. For similar studies based on Greek data, see Grammatikopoulos et al. (2014) and Rentzou and Sakellariou (2013); for a study based on a Swedish sample, see Kärrby and Giota (1995). On a much smaller Canadian sample Lehrer, Lemay, and Bigras (2015) find some evidence that parents can discriminate childcare quality. 9. The study by Cryer, Tietze, and Wessels (2002) also shows very clearly that the information gap differs between quality measures. 10. For more information on this supplementary study see the project-homepage: www.k2id.de (accessed: September 2017). 11. The main SOEP and FiD surveys only ask about provider type every four years and include no further information on ECEC centers. 12. This was accomplished through postal questionnaires and telephone follow-ups and aimed at capturing the quality of the learning environment, the interactions between children and teaching staff, activities, as well as the attitudes of ECEC professionals. If respondents were unable to complete the full questionnaire, they were given the option of answering a compressed questionnaire version and, toward the end of the survey period, we also performed a phone follow-up with an even shorter version. Sample sizes vary, as not all quality aspects were covered in the shorter questionnaires. 13. Appendix-B (see supplementary material) shows the wording of the questions which are relevant for our information gap measures. 14. For the shortened institutional questionnaires, the institution director was asked about the child's group, thus providing all the information.
15. Whereas many studies find that lower child-staff ratios and higher or more specific teacher qualifications are associated with higher ECEC quality, findings for other structural characteristics such as group size, are more mixed (for a review, see Kuger et al. 2016). 16. The scales of the items range from 1 (lowest) to 6 (highest). 17. Controlling for the quality level reported by ECEC professionals can be interpreted as a baseline measure of quality. This is not necessarily correlated with the dependent variable, which measures if there is an information gap or not. 18. In the overall sample, there are 62 groups with more than one child, 53 of which have two children. We therefore use clustered standard errors to obtain correct standard errors. However, more sophisticated models such as fixed effects are not feasible. 19. All models were estimated with various specifications for the income variables. The results are available from the authors upon request. 20. For the written pedagogical concept, which according to the ECEC professionals exists in 92% of the cases, the existence of such a concept is positively related to the probability of no information gap. This means that if such a concept exists, the likelihood that parents know about it is high, whereas if it does not exist, many parents still believe it does or answer that they do not know about it. In respect to the other quality levels, the interpretation is less intuitive. 21. In the Appendix (see supplementary material), we only present the example for structural features. The picture looks very similar for the group of educational and play activities and the group related to the pedagogical focus. 22. These estimations are available from the authors upon request. 23. This results are available from the authors upon request. 24. Figures for the other quality measures are available from the authors upon request. 25. Not taking into account 16% for opening times, which are not really a quality aspect from a pedagogic point of view as discussed above.