School Mobility and Achievement for Children Placed and Not Placed in Out-of-home Care

ABSTRACT

The aim was to investigate the effects of school mobility on achievement in compulsory school in Sweden for children in out-of-home care (OHC) and not in OHC (NOHC). Register data on background variables from four birth cohorts in the cohort-sequential longitudinal project ETF was relied upon, along with student performance on a test of cognitive ability, and school grades according to the leaving certificate. Yearly data concerning school mobility also was available. Results showed that relocation was associated with lower grades for both the OHC and the NOHC groups. The mean effect of one school relocation was 5 percentiles lower grades, but the effect of relocation was twice as large at the end of compulsory school. Given a higher frequency of relocation for children in OHC, they were more negatively affected. School mobility thus is one of several factors which contribute to the poor school achievement of children in OHC.

KEYWORDS:

• Placed in out-of-home care
• achievement
• school mobility
• compulsory school

1. Introduction

Youth having experienced placement in out-of-home care (OHC) more often than others run into poor health, inadequate housing, low wage employment, limited educational opportunities, weak familial support, poverty and criminal activity (Daining & DePanfilis, 2007; Ryan, Hernandez, & Herz, 2007). However, educational success is for children in OHC, as for children not in OHC (NOHC), a key to a good life as an adult (Berlin, Vinnerljung, & Hjern, 2011; Höjer & Johansson, 2013; Johansson, Höjer, & Hill, 2013). The Swedish National Board of Health and Welfare (Socialstyrelsen, 2010, p. 228) concluded that educational success is the strongest protective factor for children in OHC (Socialstyrelsen, 2010, p 228). High school grades are, for example, protecting from unemployment (Nilsson, 2014; Nilsson & Wadeskog, 2007, 2008) and from teenage pregnancies (Brännström, Vinnerljung, & Hjern, 2011). Educational success is, however, much more unlikely to be experienced by OHC- than by NOHC-children (e.g., Day, Dworsky, Fogarty, & Damashek, 2011; Socialstyrelsen, 2010; Trout, Hagaman, Casey, Reid, & Epstein, 2008; Vinnerljung & Hjern, 2011; Zetlin, Weinberg, & Kimm, 2004); and fewer children in OHC finish their studies (e.g., Pecora et al., 2006; Socialstyrelsen, 2010; Zetlin, Weinberg, & Kimm, 2004).

Frequent school mobility is one of the risk factors for educational failure in compulsory school (Hattie, 2009), and school relocations are more prevalent among OHC-children than among NOHC-children (Conger & Finkelstein, 2003; Neiheiser, 2015; Tanner-McBrien, 2010). In line with this, school mobility has been shown to influence school outcomes for children in OHC (e.g., Vacca, 2008). Underlying reasons for change of school may also be hypothesized to influence its level of impact. Temple and Reynolds (1999), for example, have shown that the negative consequences of school mobility are smaller for students who have moved into better quality schools. In the early 1990s a series of reforms were implemented in Sweden which implied decentralization, deregulation, marketization and free school choice (e.g., Lundahl, Erixon Arreman, Holm, & Lundström, 2013), and these have resulted in opportunities for children to change schools without changing residence. To be placed in OHC, is, however, typically associated with a residential move which often is involuntary. Placement into OHC is in Sweden regulated in the Social Services Act, and 1982 a policy change resulted in a decrease in the number of mandatory placements in favor of voluntariness and preventive social work. The lowest number of mandatory placements was in 1992, in 2012, the numbers had again increased, mainly applicable to children over 12 years old (Socialstyrelsen, 2017).

The overall aim of the current study is to investigate the effects of school mobility on academic achievements for OHC-children as well as for NOHC-children. We can go beyond prior work in this area by taking advantage of large, representative, samples of children from four cohorts born between 1972 and 1992 for which background information and school outcomes are available.

2. Background

2.1. Effects of School Mobility

Mobility across schools has been found to have a negative impact on students’ school performances (e.g., Gruman, Harachi, Abbott, Catalano, & Fleming, 2008; Hattie, 2009; Isernhagen & Bulkin, 2011; Mehana & Reynolds, 2004; Paik & Phillips, 2002; Scherrer, 2013; Temple & Reynolds, 1999). For example, a meta-analysis by Mehana and Reynolds (2004) showed negative effects of school mobility on reading and mathematics achievement in the elementary grades. Effect sizes around −0.25 for reading and mathematics indicated a 3–4-month achievement disadvantage for the mobile students. As much as a one-year performance disadvantage in reading and mathematics between kindergarten and seventh grade for mobile students was found by Temple and Reynolds (1999), but when controlling for prior knowledge one half of this difference disappeared. This suggests that both selection effects and negative effects of school mobility are operating. However, the underlying causes and effects of the school changes have rarely been investigated.

A longitudinal growth analysis from 2nd through 5th grade showed lower achievement for mobile students, and also that mobility across schools predicted a decline in classroom participation and a negative attitude toward school (Gruman et al., 2008). Teacher support had a positive influence on the growth trajectories of academic outcomes, but also a particularly strong influence on attitudes toward school among children who had experienced more school changes. Grade retention also has been shown to be a consequence of high student mobility (Paik & Phillips, 2002), and there is a higher rate of school dropout among mobile than non-mobile adolescents (South, Haynie, & Bose, 2007).

Not only on individual level, but also on school level has student mobility been shown to predict academic struggles (Scherrer, 2013). Isernhagen and Bulkin (2011), for example, highlight the academic and social gaps related to mobility, and the subsequent need for adaptation to students’ preconditions and needs that create challenges for teachers and principals.

Children in OHC have lower educational achievement than NOHC children and they also change schools more often than their peers. Tanner-McBrien (2010) showed that school mobility for children in OHC increased the risk of academic failure. For OHC children with poor early learning skills in kindergarten, it was shown that mobility in early years mediated the effect of being placed on later socioemotional competence (Pears, Kim, Buchanan, & Fisher, 2015). Heinlein and Shinn (2000) also concluded that mobility before Grade 3 is a potent predictor for achievement in Grade 6. In addition, for OHC children school mobility in lower secondary school (Grade 7–9) is negatively related to educational achievement (Clemens, Lalonde, & Sheesley, 2016; Fuglsang Olsen & de Montgomery, 2018). Such school mobility is also disadvantageous for non-placed children but Fuglsang Olsen and de Montgomery (2018) found the negative effect to be less strong than for placed peers.

Clemens, Klopfenstein, Lalonde, and Tis (2018) found that both numbers of OHC placements and number of school relocations were negatively related to school achievement, and also that there was an additional negative effect when a school move co-occurred with a placement change. It was estimated that a single placement change accompanied by a school move had a negative effect on academic growth in the reading of −3.7 percentile points, of which the placement change accounted for 2.0–2.5 percentile points depending on subject matter. However, Clemens et al. (2018) also found that for each month in OHC there was a positive effect on the achievement of around .6 percentile points, implying that 3–4 months in OHC would counteract the negative effect of one placement change.

2.2. Which Mechanisms Explain the Effects?

In previous research, the amount of and time for, student mobility have been investigated as predictors of academic achievement. Frequent mobility has been identified as a major negative determinant of outcomes (e.g., Mehana & Reynolds, 2004; Tanner-McBrien, 2010; Temple & Reynolds, 1999) and early mobility has also been found to influence children’s later achievements (e.g., Heinlein & Shinn, 2000). Hattie’s (2009) meta-analysis identifies research suggesting that any change of schools is a significant factor which causes negative effects on academic achievement, not the total number of relocations. Thus, there is a little consensus whether it is the frequency of relocations or the age at which the move occurs that has the greatest impact on school outcomes. As shown by Clemens et al. (2018) both placement and school change influence achievement, but effects are complex and this is the first study to investigate both kinds of instability. However, the lack of stability in placements that is characteristic of OHC children with frequent school changes (Clemens, Klopfenstein, Tis, & Lalonde, 2017), is another factor that may explain school failure for this group of children.

A large number of school changes is often considered to break the continuity in education, as well as the communication between social services and school (Conger & Finkelstein, 2003), and also between the previous and the new school (Zetlin & Weinberg, 2004). Such lack of communication may in turn lead to significant failures in the support of OHC children (Altshuler, 2003), disruption of regular attendance, discontinuity of lesson content, disrupted relationships with teachers and peers, and a more review-oriented classroom instruction (Paik & Phillips, 2002).

Relations with peers are negatively influenced by school relocations, and peer relations have been identified as a determinant of academic achievement (Hattie, 2009). In a longitudinal study by Galton and Willcocks (1983) mobile students were shown to have problems creating friendships to support learning, which influenced achievement negatively. Pratt and George (2005) showed that a key factor for success was to make a friend in the first month after coming to a new school. The peer networks are not only related to students’ academic achievements, but also to school dropout. South et al. (2007) showed that students’ positions in the network and the performance levels of the peers in the network were important mediators for dropout. Not only on individual level, but also on classroom level was the peer effect shown to be an explanatory mechanism of the negative effects of student mobility. Gruman et al. (2008) found that school mobility also predicts negative attitudes toward school, implying that mobility influences students’ motivation, which is an important factor for school achievement (Giota, 2002).

To sum up, previous research indicates that school mobility has a negative impact on processes of schooling and on school outcomes, and particularly so for children in OHC. However, there is little consensus about whether a number of relocations or the time at which they occur has the greatest impact, and these questions are in focus in the current study.

The following research questions will be addressed:

1. How does the frequency of school relocations affect academic achievement for children placed in OHC and for children not placed in OHC?

2. How does the age at which relocations take place influence academic achievement for children placed in OHC and for children not placed in OHC?

Against the background of the review of research reported above, it is hypothesized that school mobility has negative effects on school outcomes (e.g., Gruman et al., 2008; Hattie, 2009; Isernhagen & Bulkin, 2011; Mehana & Reynolds, 2004; Paik & Phillips, 2002; Scherrer, 2013; Temple & Reynolds, 1999). School mobility for children placed in OHC, which usually is characterized by a change of residence and often is involuntary, is hypothesized to be even more negative for school outcomes (Fuglsang Olsen & de Montgomery, 2018; Tanner-McBrien, 2010). Further, the effects of mobility are hypothesized to be related to the time for student mobility (Hattie, 2009; Heinlein & Shinn, 2000; Mehana & Reynolds, 2004; Tanner-McBrien, 2010; Temple & Reynolds, 1999).

3. Method

3.1. Subjects

Data were retrieved from the longitudinal project Evaluation Through Follow-up (ETF) which includes a combination of survey and register data (see Härnqvist, 2000). Each cohort typically comprises a nationally representative sample of about 10 000 students, which is about 10% of the total age cohort. The current sample includes student cohorts born in 1972, 1977, 1982 and 1992. While ETF also includes three older cohorts of students born in 1948, 1953 and 1967, these could not be included because they are not included in the Child Welfare Register. The samples were drawn when students attended 3rd grade, but the main data collection was made in 6th grade when the students took cognitive tests and responded to questionnaires. Information from the schools concerning relocations was collected yearly from 1st to 9th grade. Furthermore, register data concerning student family background is available. All information was collected in cooperation with Statistics Sweden and record linkages also were conducted by Statistics Sweden.

Even though sample sizes may be considered large, children in OHC were relatively few in number (less than 3%), and therefore a certain deviation from population data is to be expected. However, children in OHC were generally not underrepresented in the sample, even though those with the most problematic living conditions may be somewhat underrepresented. For certain variables, the frequency of missing data also is higher for children in OHC, as is described in greater detail below.

3.2. Variables

3.2.1. Placed in Out-of-home Care

Based on the annually collected information obtained from the Swedish National Board of Health and Welfare children were categorized as either in OHC or not in OHC (NOHC).

3.2.2. School Mobility

From 3rd to 9th grade yearly information is available from the school if the child has moved from the class. There may be different reasons for school mobility and for all cohorts the main reason was relocation to a new residence. For the 1972, 1977 and 1982 cohorts this was virtually the only reason, but for the 1992 cohort school mobility was to an increasing extent due to an active choice of school, in line with principles of free school choice introduced in Sweden in the early 1990s. In some cases it has proven difficult to separate individual school mobility from changes in the school organization for the 1992 cohort. To identify relocations relevant for this study, school mobility was coded only when relocations have been reported for less than five students in the class.

For 1st to 3rd grade information is available if the student has been in the class from the start of 1st grade, and if not, when the student came to the class (in 1st, 2nd or 3rd grade). For 3rd to 9th grade yearly information is available for each student in the class if the student has moved from the class or not. This information has been used to create three variables: Reloc1_3 which captures if the student has moved into the class after start of school and if the student has moved out of the class in 3rd grade; Reloc4_6 which counts the number of relocations in 4th to 6th grade; and Reloc7_9 which captures the number of relocations in 7th to 9th grade. Another variable (Reloc1_9) was created to capture the total number of relocations during the nine years of compulsory schooling.

3.2.3. Cognitive Tests

In Grade 6 a test-battery which included three tests of cognitive ability, was administrated by the classroom teachers in accordance with detailed written instructions. The three tests are paper-and-pencil tests, consisting of 40 items each, and were constructed within the ETF-project (Svensson, 1971). The tests were designed to measure verbal, spatial and inductive ability. The test of verbal ability measures vocabulary and in each item the task is to find the antonym of a given word among four choices. In the spatial test, the items present two-dimensional drawings with folding lines marked, and the task is to select which of four three-dimensional objects that is created when the drawing is folded. In the inductive reasoning test the task is to complete a number series of eight numbers, of which six are given. The time limits were 10, 15 and 18 minutes for the verbal, spatial and inductive tests, respectively. For the analyses in the present study a sum of the scores on the three tests was created (CognSum). The maximum total score was 120, and the reliability for each of the three parts was at least 0.90 (Svensson, 1971).

Students who did not attend school on the day of testing have missing data on the cognitive tests. To a large extent this information is missing randomly, but given that it is related to school attendance there may be some systematic differences between the OHC- and NOHC-groups.

3.2.4. Grades

A norm-referenced grading system was in Sweden in 1998 replaced by a criterion-referenced system. The criterion-referenced system thus was used for the 1982 and 1992 cohorts, while the norm-referenced grading system was used for the 1972 and 1977 cohorts.

The norm-referenced grades were assigned on a numerical scale from 1 to 5 by the teacher. At the national level the percentage of the different grades was specified to be 7%, 24%, 38%, 24% and 7%, respectively. The grading and the grade distribution were supported by national tests in English, mathematics, and Swedish which provided information to the teacher about the level of performance of the class and the grade distribution to be expected within the classroom. A mean grade over 16 compulsory subjects was computed, which was transformed into percentiles.

The criterion-referenced grades were assigned on a four-step scale with letter grades. Here too the teacher grading was supported by national tests. The letter grades are transformed to a non-equidistant numeric scale ranging from 0 to 20 according to the following rule: not passed = 0, passed = 10, passed with distinction = 15 and passed with special distinction = 20. The sum of the numeric values for the 16 best grades was computed to obtain a so called Merit-value. These 16 grades generally represented the same subject areas as the 16 norm-referenced grades.

In both grading systems around 5% of the students in each cohort missed one or more grades, either because they had failed the subject or because they were exempted from the subject. If these students were excluded from the analyses, this would cause severe bias in the comparisons between the OHC- and NOHC-groups because there was a substantial over-representation of children in OHC among those with missing grades. Instead of excluding students with missing grades, the missing grades were replaced with grades estimated on rational grounds. For students who missed between 1 and 15 grades, the lowest grade mark in the two grading systems was assigned to the missing grades. For students who did not have any valid grade the lowest possible sum of grades was assigned. After these replacements were made, percentiles were computed, implying that each and every students was assigned a percentile transformed grade (PercGrade). Percentiles were computed for the grades for each cohort separately so for each cohort the PercGrade had a mean of around 50 and an SD of approximately 28. It should be pointed out, however, that the percentile transform was conducted on the entire populations from which our students were sampled. Therefore, the PercGrade mean was not exactly 50.

3.2.5. Migration

Migration background was coded as a dummy variable. Children either being born outside of Sweden or having two parents both born abroad were coded as having a migration background and were assigned the value 1. The ordinary caregiver, either a biological parent or another person, is referred to as a parent.

3.2.6. Parents’ Education

The basis for the classification of different educations into levels was the so-called SUN-2000 code (“Svensk Utbildningsnomenklatur”). The code comprises three numbers and a letter, where the numbers indicate level and the letter subject orientation. This allows a more or less fine-grained classification of parental education (Gustafsson & Yang Hansen, 2017). Here six categories were used which distinguish between compulsory education, upper secondary education with theoretical and vocational orientation, and two categories of tertiary education with different length. The coding for individual students was done with the family as the unit according to the principle that parental education is determined by the family member with the highest attained education. The coding was primarily based on information about biological parents but when this information was not available information about step-parents was used.

3.3. Analytical Procedures

In order to estimate the extent to which relocations account for the observed differences in levels of achievement between the OHC- and NOHC-groups a series of regression analyses was conducted. In the first analysis, no explanatory variable was included in the model except for the OHC dummy variable. This model provides an estimate of the amount of achievement difference between the two groups. The model was then successively extended with explanatory variables, either in the form of individual-level background variables or variables which indicate the number of relocations, either at different grade levels, or all together. When the addition of a variable causes the estimated effect of OHC on grades to change, this indicates that the added variable accounts for a part of the achievement difference. This can only happen if there is a relation between the added variable and achievement, and if there is also a mean difference between the OHC- and NOHC-groups with respect to the added variable. All regression analyses were conducted with the UNIANOVA procedure in the GLM package in SPSS 23.

Data for the sample drawn in Grade 3 has a three-level nested structure (municipalities, schools, students). However, the study covers the entire compulsory school, and the nested structure becomes transformed into a cross-classified structure as a function of student relocations between schools and municipalities. The clustering effects thus become attenuated, and standard procedures for taking cluster effects on standard errors into account are not applicable. However, based on the original sampling design the ICC for the achievement measure in Grade 9 was substantially lower than 0.05 for all cohorts, which indicates that any effect of clustering on statistical inference may be regarded as trivial.

4. Results

4.1. Descriptive Statistics

Table 1 presents descriptive statistics for the background variables and the relocation variables, along with grades from the leaving certificate of compulsory school.

Table 1. Descriptive statistics for the NOHC and OHC groups.

	Cohort	N		Grades	Cogn ability	Parental educ	Migr backgr	Gender	Reloc1_3	Reloc4_6	Reloc7_9	Reloc1_9
NOHC	1972	8819	Mean	51.7	50.4	2.51	1.07	1.49	0.14	0.14	0.06	0.33
			SD	28.0	28.7	1.30	0.26	0.50	0.36	0.37	0.25	0.63
	1977	4218	Mean	50.9	50.5	2.70	1.09	1.49	0.15	0.13	0.06	0.34
			SD	28.2	28.8	1.24	0.28	0.50	0.37	0.39	0.27	0.67
	1982	8277	Mean	53.4	50.7	2.93	1.12	1.49	0.15	0.17	0.06	0.39
			SD	28.6	28.7	1.29	0.32	0.50	0.38	0.42	0.26	0.69
	1992	9622	Mean	53.7	50.5	3.05	1.12	1.49	0.15	0.16	0.11	0.42
			SD	27.8	28.8	1.22	0.33	0.50	0.38	0.43	0.34	0.72
	Total	30936	Mean	52.7	50.5	2.82	1.10	1.49	0.15	0.16	0.08	0.38
			SD	28.1	28.8	1.28	0.31	0.50	0.37	0.41	0.29	0.68
OHC	1972	263	Mean	28.4	34.9	2.10	1.20	1.47	0.36	0.38	0.32	1.04
			SD	26.4	29.6	1.22	0.40	0.50	0.54	0.64	0.56	1.08
	1977	126	Mean	25.1	32.2	2.20	1.20	1.44	0.44	0.34	0.35	1.11
			SD	21.6	26.6	1.00	0.40	0.50	0.60	0.62	0.56	1.13
	1982	338	Mean	23.4	30.6	2.17	1.22	1.52	0.38	0.38	0.28	1.04
			SD	24.5	26.5	1.22	0.42	0.50	0.55	0.58	0.54	1.03
	1992	372	Mean	25.6	34.7	2.17	1.25	1.55	0.32	0.41	0.35	1.09
			SD	25.4	26.8	1.14	0.43	0.50	0.53	0.63	0.60	1.02
	Total	1099	Mean	25.5	33.1	2.15	1.22	1.51	0.36	0.39	0.32	1.06
			SD	25.0	27.5	1.17	0.42	0.50	0.55	0.62	0.57	1.05

Notes: Grades = Percentile transformed grades: Cognitive ability = Percentiles of the sum of the three cognitive tests; Parental education = Mean of the six category scale of level of family education; Migration background = proportion of students with foreign background; Gender = proportion of females; Reloc1_3 = Relocations Grades 1–3; Reloc4_6 = Relocations Grades 4–6; Reloc7_9 = Relocations Grades 7–9; Reloc1_9 = Relocations Grades 1–9.

For all cohorts combined the mean difference in grades between the NOHC and OHC groups was 27 percentiles, which corresponds to approximately 1 SD. There was a trend of increasing differences across the cohorts from 23 percentiles for the 1972 cohort, to around 30 percentiles for the most recent cohorts. The mean difference in the level of performance on the cognitive test was 17 percentiles for all cohorts combined, which is considerably less than the difference in grades. For the cognitive tests too there was a tendency of increasing differences across cohorts.

The level of parental education differed by .4–.9 on the 6-point scale between the two groups. For the NOHC group the mean level of parental education increased over the age cohorts, while the level of education was essentially unchanged for the OHC group. There was also a substantial difference in the proportion of children with the foreign background (around .10 for the OHC group and around .22 for the NOHC group). Overall, the groups were balanced with respect to gender, but there was a trend of a successively increasing proportion of girls in the OHC group.

The mean number of school relocations was low, suggesting that the majority of students did not experience school mobility. Thus, the frequency distribution of the total number of relocations during the 9 grades (Reloc1_9) showed that 70% of the students did not relocate, 22% relocated once, 6% relocated twice, and 2% relocated three or more times. However, there was a very large difference in the total number of relocations between the NOHC- and the OHC-groups, the means being between .3 and .4 for the NOHC group and between 1.0 and 1.1 for the OHC group. For the NOHC group there was a slight trend of increasing school mobility across cohorts, while there was no such trend for the OHC group. It is also interesting to observe that the number of relocations during 7th to 9th grade was considerably lower than for 1st to 6th for the NOHC group, while for the OHC group the relocation frequency was about the same across the three Grade levels.

All variables in Table 1, except for the cognitive tests, are based on register data, and for these there is no missing data. However, the cognitive test information was collected in schools in 6th grade, and given that all students were not present on the day of testing, this information is missing for some students. In the NOHC group about 14% of the children had no data on cognitive ability, and in the OHC-group about twice as large a proportion lacked cognitive test data, with some variation across cohorts. For the NOHC group the mean difference in percentile grades between those having cognitive test data and those not participating in the test was 5 percentiles, while the corresponding difference for the OHC group was 7 percentiles. Thus, within both groups, there was a tendency that low-achieving children to a larger extent suffer from missing data on the cognitive tests than high-achieving children.

Table 2 presents correlations among the variables. Gender was uncorrelated with all variables, except for grades. Migration background was positively correlated with relocations and negatively correlated with parental education, cognitive ability, and grades. The three variables measuring the frequency of relocations at the three grade levels had positive inter-correlations and they were negatively correlated with cognitive ability and more strongly so with grades.

Table 2. Correlations among student background variables, frequency of relocations and grades.

	Gender	Migr backgr	Parental educ	Reloc1_3	Reloc4_6	Reloc7_9	Reloc1_9	Cogn ability	Grades
Gender	1.000
Migr backgr	−0.001	1.000
Parental educ	0.004	−.125**	1.000
Reloc1_3	0.003	.147**	−.015**	1.000
Reloc4_6	0.008	.108**	−0.001	.131**	1.000
Reloc7_9	0.009	.079**	−.018**	.088**	.139**	1.000
Reloc1_9	0.011	.176**	−.016**	.649**	.713**	.551**	1.000
Cogn ability	−0.003	−.156**	.314**	−.048**	−.048**	−.062**	−.081**	1.000
Grades	.191**	−.079**	.379**	−.078**	−.095**	−.131**	−.153**	.579**	1.000

Note: ** indicates that correlation is significant at the 0.01 level (2-tailed). For legend, see Table 1.

4.2. Effects of School Mobility on Achievement

Our first research question concerns how school mobility affects school performance for children in OHC, and in this context we also ask if children in OHC and NOHC are differentially affected by school mobility. We have already demonstrated that school mobility was more frequent for the OHC-group than for the NOHC group (see Table 1). For most cohorts, the mean number of relocations was about three times as high for children in OHC than it was for NOHC children. If there is a general effect of school relocation of student achievement this implies that the higher frequency of relocations among children placed in OHC accounts for a part of the lower level of school achievement of this group. However, it may also be that there is a differential effect of school mobility among the OHC- and NOHC-groups, such that an additional relocation exerts a stronger, or weaker, negative influence for the children in OHC. Given that the four cohorts involved in the current study span a time-period of two decades there also is a reason to investigate cohort differences.

4.2.1. Linearity of Regressions

The analyses were conducted in the several different steps, but all analyses in this section were based on the measure of the total number of relocations (Reloc1_9). In the first step, it was on pooled data investigated if the regression of PercGrade on Reloc1_9 was linear. Possible non-linearity was investigated by comparing the amount of explained variance in a model assuming linearity with the amount of explained variance in a model in which the different numbers of school relocations were represented by categories in an ANOVA. This test was non-significant, F (2, 32,033) = 2.33, p < .054, and visual inspection of a line diagram did not indicate any deviation from linearity. In all further analyses linearity of regression on Reloc1_9 was therefore assumed.

4.2.2. Homogeneity of Regressions

Next homogeneity of regressions of PercGrade on Reloc1_9 across the OHC- and NOHC-groups was investigated. A model assuming homogeneity of regression estimated the common regression slope at −5.21, t = −22.77, p < .001, 95% CI [−5.65, −4.76]. However, the test of interaction in a model allowing different regression slopes was significant, F (1, 32,031) = 4.70, p < .030, suggesting that we may have to consider different regression slopes in the two groups. For the OHC group, the regression coefficient was −3.45, while for the NOHC group the regression coefficient was −5.35, indicating a stronger negative association between frequency of relocation and school achievement in the NOHC group than in the OHC group.

It may, of course, be that there is regression heterogeneity across the age cohorts as well. This was tested by adding a cohort as another fixed factor to the model, and allowing for interactions between OHC placement and cohort on the one hand and Reloc1_9 on the other hand. The three-way interaction between OHC placement, cohort and Reloc1_9 was also included in the model. However, the three-way interaction was not significant, F (3, 32,032) = 0.42, p < .988. The two-way interaction OHC*Cohort was not significant either, F (3, 32,022), p < .146, while the two-way interaction Reloc1_9*OHC was significant F (1, 32,022) = 5.02, p < .025. However, even though the cohort differences in regression coefficients were not significant, descriptively there were differences. For the 1972, 1977, 1982 and 1992 cohorts the following estimates of regression coefficients were obtained for the OHC group −3.59, −3.35, −3.72 and −3.12, while the corresponding estimates for the NOHC group were −4.98, −6.07, −6.95 and −4.37. The estimates for the OHC group did not vary much between cohorts, and for all cohorts they were lower than those observed for the NOHC group. For the latter group, the lowest estimates were observed for the 1972 and 1992 cohorts and the highest estimates for the two intermediate cohorts.

One possible explanation for the generally lower regression coefficients for the OHC group is that there were floor effects in the PercGrade measure. However, the small sample size and the large amount of random error make it difficult to detect floor effects. Nevertheless, one indication of the presence of floor effects is that there was massive heterogeneity of variance in PercGrade as a function of a number of relocations in the OHC group. The following SDs were observed for groups with 0, 1, 2 and 3 relocations, respectively: 26.1, 25.3, 23.0, and 21.1. This linear decline in the SDs was not observed in the NOHC group, where SDs as expected were between 28 and 29 for all frequencies of relocations. The reason for the floor effect seems to be the exceedingly poor performance in the OHC group. The mean PercGrade for the OHC group with 0 relocations was 28.8. This is just 1 SD above 0. If the coefficient estimated for Reloc1_9 in the NOHC group was to be applied to the OHC group as well, the expected PercGrade for the OHC group with three or more relocations would be 13.5. This is less than half an SD above 0 and the floor effect makes it unlikely to observe so low values.

Thus, a likely explanation of the lower regression slope on Reloc1_9 in the OHC group is that the level of performance in this group is so low as to cause floor effects. If this interpretation is correct it implies that we would rely on underestimated regression coefficients for the OHC group when estimating consequences of the frequency of relocations for student achievement. However, one solution to the problem of estimating the within-group regression coefficients correctly is to rely on common regression coefficients across cohorts and the OHC/NOHC-groups. If the coefficients are underestimated in the OHC-group the common estimate will to a large extent correct for this. This approach was adopted for the next step in the analysis, where the effect on the achievement of school mobility is estimated and where the extent to which school relocations can explain the difference in levels of achievement between the OHC- and NOHC-groups is investigated.

4.2.3. Effects of School Relocations on Achievement Differences Between OHC and NOHC Groups

A model in which PercGrade was regressed on the OHC dummy yielded an estimate of −27.15, t = −31.54, p < .001, 95% CI [−25.47, −28.84]. This estimate was the same as the mean difference between the OHC and NOHC groups (Table 1). Next the Reloc1_9 variable was added to the model, and the common regression coefficient was estimated at −5.21, t = −22.77, p < .001, 95% CI [−5.65, −4.76]. The OHC dummy had in this model a regression coefficient of −23.70, t = −31.54, p < .001, 95% CI [−25.40, −21.00]. The difference between the two OHC coefficients provides an estimate of how much of the difference in PercGrade may be accounted for by the difference in frequency of relocations. This estimate amounts to 3.5 percentiles on the PercGrade scale.

However, this estimate cannot be interpreted as an estimate of the causal effect of relocations on achievement, because several different factors were correlated with the frequency of school relocations and these may be causally involved in the performance difference. As is made clear by the descriptive statistics presented in Table 1 there were differences between the OHC- and NOHC-groups with respect to individual characteristics such as cognitive ability and socioeconomic background. The correlations in Table 2 also demonstrate these characteristics to be related to school achievement which makes it essential to investigate to what extent they may account for the observed achievement difference between children in the OHC and NOHC groups. A set of control variables (Gender, Migration, ParentEd, CognSum) was therefore added to the model. First, a model was estimated which did not include Reloc1_9. This model accounted for 42.6% of the variance in PercGrade, the strongest predictors being CognSum, ParentEd and Gender. The estimate for the OHC dummy was −15.63, t = −19.78, p < .001, 95% CI [−16.78, −13.76]. In the next step the Reloc1_9 variable was added to the model. In this model, the amount of explained variance increased to 43.3%. The estimate of the OHC dummy was −13.27, t = −17.13, p < .001, 95% CI [−14.78, −11.75]. The effect of the higher frequency of relocations thus was 2.4 percentiles on the PercGrade scale, controlling for student background variables.

In summary, it may be concluded that school mobility partly accounts for the lower level of achievement of children placed in OHC. This is because this group has a higher frequency of school mobility, and because school mobility exerts a negative effect on school achievement.

4.2.4. Effects of Relocations at Different Grade Levels

Our second research question asks how the age at which relocations takes place influence academic achievement for children placed in OHC and for children not placed in OHC. In the present data information is available about the number of relocations during grades 1–3, 4–6 and 7–9, so we can take advantage of this information and investigate if the regression coefficients expressing effects of relocations vary across grade levels. The same set of background variables as was used in the previous analyses was included in the model. The model accounted for 43.5% of the variance in PercGrade.

The estimates of main effects of the three relocation variables were all significant (see Table 3). However, the estimates differed widely, by far the largest negative coefficient being observed for Reloc7_9. For Reloc1_3 the main effect was −2.29, for Reloc4_6 it was −2.84, while for Reloc7_9 it was −7.42. These results thus show that it does matter when during the process of schooling relocations takes place.

Table 3. Estimates of effects of relocations at different grade levels.

Parameter	B	Std. Error	t	Sig.	95% confidence interval
					Lower bound	Upper bound
Intercept	−5.79	0.74	−7.86	0.00	−7.23	−4.35
OHC	−12.83	0.78	−16.55	0.00	−14.35	−11.31
Reloc1_3	−2.29	0.36	−6.43	0.00	−2.99	−1.59
Reloc4_6	−2.84	0.35	−8.06	0.00	−3.53	−2.15
Reloc7_9	−7.42	0.46	−16.11	0.00	−8.32	−6.51
CognSum	0.49	0.00	103.58	0.00	0.48	0.50
ParentEd	4.88	0.11	45.89	0.00	4.67	5.08
Migration	5.12	0.44	11.60	0.00	4.26	5.99
Gender	10.95	0.26	42.79	0.00	10.45	11.46

Note: For legend, see Table 1.

When comparing these results to those obtained in the analyses of Reloc1_9 it should be observed that the logic of the two analyses is quite different. The analysis of the single Reloc1_9 variable captures the total effect associated with the number of relocations. However, the analyses of the three grade-level measures of relocations estimate the specific (or direct) effects associated with each of the three variables, controlling for any correlations among them. Regressing PercGrade on just Reloc7_9 and the OHC dummy yielded an estimate of -10.0 for Reloc7_9. This result thus suggests that the negative effect on grades of relocation in the final years of compulsory school is at least twice as strong as when changing school in lower grades.

5. Discussion

Our first research question asked how the frequency of school relocations affects academic achievement for children placed in OHC and for children not placed in OHC.

The descriptive results replicate the substantial differences in levels of school achievement at the end of compulsory school between the OHC- and NOHC-groups which have been found in many previous studies, and which here amounted to 1 SD, or 27 percentiles. The difference can partly be accounted for by differences with respect to student background variables. Thus, compared with children not placed on OHC, the children placed in OHC more often have a migration background, lower cognitive test scores, and parents with lower levels of education. Altogether, these individual student characteristics accounted for 12 percentiles of the overall difference in grades between the OHC- and NOHC-groups.

This leaves 15 percentiles still to be accounted for, and the main focus of the paper is to investigate to what extent school relocations can account for yet another part of the achievement difference. A linear relation was found between a number of relocations and school achievement, an additional relocation being associated with a 5.2 percentiles lower grade. The OHC group had a larger number of relocations than the NOHC group, and the total impact of this was estimated at 2.4 percentiles, with control for background variables.

Even though this may seem a small effect the importance and impact of school mobility should not be underestimated, neither for children in OHC, nor for NOHC children. One reason for this is that an additional relocation has a relatively strong impact of −5.2 percentiles and in some cases it may be possible to prevent or postpone a school change. What is perhaps even more important is that measures may be taken to reduce the negative impact of the relocation. Thus, when a child relocates to another school the continuity in education is dependent on the transmission of information between schools concerning the child’s previous schooling. This requires communication, both between the previous and the new school, and between social services and school personnel if the child is in OHC. However, previous research has shown that such communication tends to be reduced in connection with frequent school relocations (Conger & Finkelstein, 2003; Zetlin & Weinberg, 2004). It also is essential that schools actively support the integration of the newly arrived child in the peer group (Galton & Willcocks, 1983; Hattie, 2009; Pratt & George, 2005; South et al., 2007). To better know the mechanisms that cause the negative effects of school mobility is also of importance to improving placed children’s opportunities for school success. Thus, this is an important area for further research.

We did observe a tendency for the regression slope of PercGrade on a number of relocations to be less steep for the OHC group than for the NOHC group. However, this phenomenon seemed to be due to a floor effect in the OHC-group, as indicated by the substantial heterogeneity of variance across OHC groups with different frequencies of relocations. It is in this context interesting to note that Fuglsang Olsen and de Montgomery (2018) concluded that the effect of school mobility was stronger for children in OHC, than for NOCH children. In analyses of effects of school mobility on self-perceived academic ability and enrollment in post-secondary programs and they found significant or nearly-significant interaction effects with OHC, the interaction implying that there were strong negative effects of school mobility in the OHC group and no effects in the NOHC group. These findings conflict with our results, given that we found substantial negative effects of school mobility in the NOHC group. One possible explanation for the discrepant findings is that Fuglsang Olsen and de Montgomery (2018) only used a dichotomous variable to distinguish between no relocation on the one hand and one or more relocations on the other hand. Our results showed that the negative effect of school mobility was linearly related to the frequency of relocations, which implies that dichotomization of the variable reduces the chances of finding any effect. Another possible explanation is that Fuglsang Olsen and de Montgomery (2018) investigated other outcomes (self-concept and educational attainment) than achievement in compulsory school.

Clemens et al. (2018) contributed an important innovation in the investigation of the effects of relocations on school achievement of children in OHC by studying simultaneously the effects of placement changes and school changes. Their results suggest that placement changes have negative effects comparable to those induced by school mobility, but that placement also has a time-related positive effect on achievement which may counteract the negative placement effect. Had there not been such a positive effect, children in OHC would be expected to be more negatively affected by school relocations because of the double negative effects from placement and school changes. However, in our results there were no indications of children in OHC being more negatively affected by school changes than NOHC children, suggesting that meaningful results can be obtained without bringing in placement changes.

Our second research question asked if effects of school relocations are invariant across grade level. The results showed that this was not the case, the stronger negative effect being observed for school changes taking place during the three final years of compulsory school. Relocations during the first three years affected outcomes the least. One possible explanation for this is that the final grades achieved in compulsory school are primarily determined by student effort and activities during the final years of schooling, and that relocation during these years therefore strongly impacts on the final grades. Given that children in OHC relocate even more frequently during the final years of schooling than children not in OHC, this contributes to an even more negative effect. This suggests that great care should be taken to avoid school relocations during the final years of schooling.

The present study is primarily based on register data. Such data has the great advantage of being complete for each and every participant. However, for the cognitive test, information is missing for students not present at school on the day of testing. This was the case for some 14% of the NOHC group, while the proportion with missing data was about twice as large in the OHC group. The large proportion of missingness in the OHC group is not likely to be random, and a consequence of this may be an overestimation of the level of cognitive ability in the OHC-group.

From previous research we know that educational success is one of the most strongly protective factors for children placed in out-of-home care (e.g., Berlin et al., 2011; Höjer & Johansson, 2013; Johansson et al., 2013). We also know that frequent school mobility is a risk factor for educational failure in compulsory school (Hattie, 2009). The results in this study are in line with this research: relocation affects school achievement negatively, and more so for placed children than for non-placed, which is because relocation occurs much more frequently among children in OHC. These results are in line with international findings (e.g., Vacca, 2008). Considering also that prior research has identified numerous other effects of school mobility, as for example, on classroom participation and attitudes (Gruman et al., 2008), grade retention (Paik & Phillips, 2002), and school dropout (South et al., 2007), it is obvious that the entire life situation of children placed on out-of-home care is affected by many school relocations.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Forskningsrådet om Hälsa, Arbetsliv och Välfärd.