Chances of Early School Leaving—With Special Regard to the Impact of Roma Identity

Abstract

Early school leaving rates among Roma are higher than in the majority population, as confirmed by several studies, but the descriptions are often two-dimensional. Intersectionality is described as a persistent situation formed by several social dimensions, and this specific position can be advantageous or disadvantageous from the aspect of educational inequalities. This study aims to explore what type of intersectional position can raise the chance of early school leaving and what is the role of the Roma ethnic identity in this. Earlier empirical analyses are rather two-dimensional, and these intersectional situations cannot be identified with them. The Hungarian Youth Survey databases provided an opportunity to conduct this type of multiple approaches. With the help of our results, the intersectional position can be described in which the chance of early school leaving is higher. The elements of this situation are embedded in economic, educational, geographical and ethnic categories at the same time. The effect of Roma ethnic identity is significant but not the strongest in our model. With the help of a comparison of the two waves of the research project, we can state that the patterns of this intersectional and disadvantageous situation are stable but not unchanged.

1. Introduction

While the European Union considers the integration of the Roma population to be of strategic importance and devotes considerable resources to this issue, data show that little has changed since 2005, and the Roma population is still on the margins of society. The European Council’s previous development priorities of education, employment, health and housing have been expanded to include equality, inclusion, and participation in the 2020–2030 funding period. In the area of education, three targets have been formulated: increasing participation in early childhood education and care, reducing the number of early school leavers, and ending segregated education in primary schools. Current figures show that 42% of Roma children attend pre-primary school compared to 92% in the general population; 44% of Roma students are in segregated education; and 28% of Roma young people have completed upper secondary education compared to 83.5% in the general population [1]. Education, the attainment of at least upper secondary education, has an impact on other target areas, as a number of studies have shown the link between employment and health and welfare and early school leaving. In addition, the effects of early school leaving are also conspicuous in the areas of civic responsibility, politics, health, social sector, and employment, leading to lower economic growth rates, lower tax revenues, higher unemployment and welfare payments, and higher public health and criminal justice expenditures [2].

Our study analyses early school leavers (more recently termed early leavers from education and training or ESL/ELET) in Hungary, focusing on the characteristics of young people who embrace their Roma identity. Our analysis attempts to fill the gap that the literature where early school leaving is not analysed as a combined explanation of several social background factors and is often explored only in ethnic terms. In our study, we use the 20–29 year olds subsamples of the Hungarian Youth 2016 and 2020 databases to characterise the highest completed educational attainment of Roma youth, compare the characteristics of Roma and non-Roma early school leavers, and examine the factors explaining early school leaving using logistic regression models.

There are numerous systematic analyses of the causes of dropout and early school leaving [3,4,5,6], which identifies individual, peer, family and school-level factors as the main causes of early school leaving. The analyses also highlight that early school leaving is a long process influenced by a combination of factors. Due to its complexity and the fact that it changes over time early school leaving is difficult to study, as static and bivariate analyses can easily fail to detect the interaction of factors and to highlight its dynamic nature [7]. Some of the systematic analyses also suggest that some causes are stronger than others: absenteeism, poor academic achievement, peers, family structure, economic status, and emotional background have a greater impact [5,6]. In addition, it is also found that certain social groups (based on family background or ethnicity) are generally more affected than others.,

Intersectionality is a persistent situation in which multiple categories of interacting inequalities manifest as a new social category in which the causes of oppression cannot be separated [8,9]. Research on intersectionality approaches from different perspectives: they can be group-centred, process-centred, and system-centred [10]. Some only describe the effects of belonging to multiple groups [11], while the process approach highlights interactions and explores the underlying factors [12], such as the relationship of power to groups and the choice of belonging to a particular group [13]. The system-centred approach looks at intersectionality from a historical perspective and examines inequalities in their complexity [14]. Others [15] analysed the educational and labour market opportunities of young adults along three group characteristics (gender, migrant background, and social class). Stand [16] examined the relationship between academic achievement, socioeconomic status and gender. He found that ethnicity, gender, and SES are not simply additively combined; they interact significantly, particularly ethnicity and SES, and ethnicity and gender. Cerna et al. [17] identified six intersectional dimensions and described them as overarching factors of socioeconomic status and geographic location. Howard and Vajda [18], examining the intersectional situation of Roma, find that the most persistent form of group-based disadvantage is linked to identities of origin (minority), which deepens other forms of inequality. In Hungary, social disadvantage and its complexity (low educational attainment, disadvantaged localities), belonging to the Gypsy/Roma community, and the associated negative social prejudice are the main intertwined categories [19,20]. The studies also point out that, due to intersectionality, educational issues related to disadvantage (region, type of locality, socioeconomic status) and Roma cannot be separated [21,22].

The average rate of early school leavers in the European Union has been steadily decreasing, falling below 10% in 2021 [23]. In Hungary, the rate tended to fluctuate or stagnate in the 2000s, followed by an upward trend after 2010. Hungary moved above the EU average in 2013, the first year in which the early school leaving rate exceeded the EU average and has remained above it ever since. Hungary’s rate is 23rd among the 27 countries, followed by Bulgaria, Italy, Spain and Romania with 15.3%. (The biggest drop in ten years has been achieved in Portugal (17%) and Spain (13%), and the best rates are scored by Croatia, Slovenia and Greece.), It should be pointed out that the average rate of early school leavers is slightly higher for boys (11.4%) than for girls (7.9%), but girls have a more difficult time returning to education, as in their case, early childbearing is one of the main reasons for early school leaving. According to Eurostat data [24], in 2017 the average proportion of mothers under 20 years of age at the birth of their first child was 3.7% in EU countries. The highest rates are in Hungary (8.5%), Bulgaria (12.1%), and Romania (12.5%). As a result, both Bulgaria and Romania have more early school leavers among women than men, and Hungary has almost the same gender ratio. Early school leaving also has regional characteristics. Eurostat also looks at how early school leavers are distributed by the level of urbanisation in different countries, distinguishing between cities, towns and deprived areas. The analysis finds that Hungary, Bulgaria and Romania show the largest differences between EU countries in the three categories and that rural areas in these three countries are the most affected by ESL. In Hungary, Central Hungary and Western Transdanubia are in the best position, with early school leaving rates below 10%. Northern Hungary is in the worst situation, with twice the rate of early school leavers compared to Central Hungary.

The Hungarian research [25] also identifies factors similar to those in international studies, finding that school factors perform only a minor role in preventing dropout (no compensation) and that individual characteristics and family background are the determinant factors of dropout. It should also be highlighted that Hungary is consistently among the countries with the highest explanatory power of the SES index in academic achievement [26]. In Hungary, there was only one large-sample longitudinal study consisting of six waves, which followed the school paths of students/young people over a longer period (2006–2012) and also examined the Roma identity. Data from the last wave of the study show that (1) 99% of students starting primary school successfully completed primary school, compared to 93% of Roma students; (2) 90% of students starting primary school successfully completed upper secondary school, compared to 50% of Roma students; (3) 75% of students starting secondary school successfully took the secondary school leaving exam, compared to 24% of Roma students [27]. The authors also point out that in addition to the financial situation of the families, poverty, educational environment (cultural disadvantages), and social isolation also perform an important role in the dropout of Roma students, as Roma young people have fewer friends and peers who do well at school. It is important to note that this study was conducted at a time when the compulsory schooling age in public education was 18, and major school integration/inclusion developments were underway [28], but in 2011, the compulsory schooling age was changed to 16 and the above-mentioned developments have also been disrupted. Research has shown that compulsory schooling up to the age of 18 has promoted schooling for children from lower social groups [28]. Kende [29] points out that although integration in pre-primary education has improved significantly in recent years, the situation of Roma in education is deteriorating. Educational gaps are widening, and the proportion of Roma who do not complete the different levels of education is very high. In addition, school segregation is increasing. Roma students face significant disadvantages in access to quality education. Segregated education limits young Roma people’s opportunities for further education and deprives them of inter-ethnic social networks, destroying their identity and self-esteem [30]. Segregated education also has an impact on academic performance [31]. The results of the large-sample studies were nuanced by research conducted in the context of three surveys that followed the schooling path of the same groups of students in segregated Roma communities over more than 20 years and its impact on their adult life. The life course interviews revealed that where support from teachers, schools, churches, NGOs, and family was available, early school leaving was lower and young people often had significantly higher educational attainment than their parents. Where this was not the case, young people repeated their parents’ failure to attend school, reproducing their disadvantaged social position. Here, again, the intersectional position of those included in the study was decisive: support was targeted to compensate for disadvantage, while barriers were rooted in social disadvantage and prejudice against Roma [32].

The aim of the study is to examine the rates of early school leaving and its variation among young people based on data from a large sample panel study representative of the cohort’s characteristics and to analyse which factors are likely to explain early school leaving, focusing on the Roma identity.

RQ1: What socio-demographic differences can be found between Roma and non-Roma early school leavers?

RQ2: What demographic, social, geographical and economic factors increase the probability of early school leaving?

RQ3: What role does ethnic background perform in early school leaving?

2. Materials and Methods

Our secondary analysis relied on the 2016 and 2020 data from the database of the Hungarian Youth Survey, a panel survey conducted since 2000 with a sample size of 8000 respondents. The databases belong to the fifth and sixth waves of the survey, with four-year intervals between each wave. The Hungarian Youth Survey aims at a comprehensive and longitudinal analysis of the young population, so the omnibus questionnaire covers a wide range of areas (marital status, education, employment, leisure time use, etc.). The questionnaires (and the targeted areas) of each wave do not always correspond to each other, but block mapping education is included in all cases. The length of the fulfilling was about 45–50 min. The sensitive topics were mapped by a self-administered survey due to ethical issues.

The database is representative by sex, age, region, type of locality, and educational attainment, and covers the population of young Hungarians aged 15–29. The sampling was carried out in a multistage and stratified manner, with the first stage being the selection of localities (municipals were chosen in this stage according to its geographical disposition and the proportion of young people in them) and the secondary sampling frame being the young people in the localities (according to the data of Minister of Interior wia th simple random sample, N = 8000). In the case of fifth wave, the number of the target population was 1,701,837 people who were born between 1 January 1987 and 31 December 2001 [33], and this number was 1,614,973 people in the sixth wave who were born between 1 January 1991 and 31 December 2005 [34]. The interviews were conducted face-to-face by an interviewer. The database was weighted by gender, type of locality, educational attainment, and age group [34]. Data analysis was conducted by SPSS 28.0 statistical software. The dataset arefreelye available for research issues.

All 8000 respondents answered the questions used in the analysis (some subsamples of 2000 respondents were used for certain sub-areas and question blocks). The used questions blocks by us belong to the part of the analysis which was fulfilled by every respondent (educational attainment, ethnic and socio-demographic background). Although the survey is longitudinal, the wave-to-wave variation in the ethnicity variable, one of the most important questions for us, did not allow for a more serious longitudinal analysis. The wording and the way the question was asked were the same in the 2016 and 2020 surveys (‘What ethnicity do you identify yourself as?’—a multiple-choice question with Roma/Gypsy response option), so the data from the last two waves were compared in terms of Roma educational attainment. However, the proportion of young people identifying themselves as Roma decreased between the two waves, which can probably be explained by external social factors influencing self-reporting (degree of prejudice, evolution of social distances, etc.) rather than by demographic reasons [35]. The issue of the social embeddedness of the change in the proportion of self-identified Roma has also been explored in the international literature [36]. In addition to the ethnicity variable, we also used the age variable (we used a sample of 20–29 year olds, as they already reflect the fact of dropping out; the current compulsory school age in Hungary is 16 years. The database divides young people into three cohorts (15–19, 20–24, and 25–29).

In addition, the analysis used background variables that are associated with dropout [25]:

• Sex (male and female);
• Ethnic identity (self-reported);
• Father’s highest completed education, mother’s highest completed education (here we used four-choice data: primary education or less, vocational education without secondary school final certificate, secondary school final certificate, degree, and in the regression model, we separated those with only primary education from those with at least vocational education);
• Region (based on EU 2020 data [30], we have separated the more disadvantaged regions (Southern Great Plain, Southern Transdanubia, Northern Great Plain, Northern Hungary) from the better-off regions (Central Transdanubia, Central Hungary, Western Transdanubia and Budapest);
• Type of locality (with three categories: Budapest, city with county status, and county seat; smaller town; village);
• Early childbearing (respondent had first child before the age of 19);
• Previous cohabitation without marriage;
• Subjective, self-assessed financial situation measured by five categories (living in deprivation; having financial problems from month to month; just making ends meet on the income; with careful budgeting, we get by without problems; no financial problems). In the regression model, negative (first two responses), neutral (third response), and positive (fourth and fifth responses) statements were separated. The question block on the subjective financial situation has been used continuously since the first wave of the research, i.e., since the 2000 survey [34].

First, we examine the percentage of young people aged 20 or over with the highest educational attainment who identify themselves as Roma in the 2016 and 2020 datasets of the survey, and we try to uncover if there is displacement between the two waves or not and this displacement is statistically significant or not. Second, we compared the features of the Roma and non-Roma early school leavers along with the earlier mentioned independent variables, which may describe the specific intersectional life situation. Not only percentages but chi-square statistics were used by us during this step. We have paid special attention to the type of locality because the Roma people are overrepresented in the villages and pupils’s school achievement is strongly formed by the of the locality in Hungary [37]. The values of adjusted residuals were analysed by us because we would like to explore the patterns of the given cells. Finally, we run two binary logistic regression models on a subsample of 20 year-olds or older, with the dependent variable being the dropout rate (0 = having at least vocational qualification, 1 = not having vocational qualification). The independent variables included sex (0 = female, 1 = male), mother’s level of education (0 = higher than primary school, 1 = primary school or less), father’s level of education (0 = more than primary school, 1 = primary school or less), childbearing under 19 years of age (0 = no, 1 = yes), previous cohabitation without marriage (0 = no, 1 = yes), type of locality (dummy coded, the reference category is Budapest), regions (0 = favourable regions, 1 = disadvantaged regions), subjective economic situation (the reference category is a merge of the two categories better than neutral), and ethnicity (0 = not Roma, 1 = Roma).

Table 1. Subsample sizes used in the analysis.

	2016 Database	2020 Database
Size	8000 persons	8000 persons
Size of Roma subsample	328 persons	293 persons
Size of at least 20-year-olds subsample	5696 persons	5680 persons
Size of at least 20-year-old non-Roma dropouts subsample	478 persons	334 persons
Size of at least 20-year-old Roma subsample	231 persons	189 persons
Size of at least 20-year-old Roma dropouts subsample	151 persons	123 persons

shows the subsample sizes, and

Table 2. Distribution by socio-demographic indicators in the 2016 and 2020 databases (only the data used in the analysis).

	2016 Database		2020 Database
Sex
Male	4105 persons	51.3%	4129 persons	51.6%
Female	3895 persons	48.7%	3870 persons	48.4%
Ethnicity
Roma	328 persons	4.1%	293 persons	3.7%
Non-Roma	7672 persons	95.9%	7707 persons	96.3%
Age
15–19	2304 persons	28.8%	2320 persons	29.0%
20–24	2790 persons	34.9%	2806 persons	35.1%
25–29	2906 persons	36.3%	2874 persons	35.9%
Respondent’s educational attainment
Primary school or less	2446 persons	30.6%	2232 persons	27.9%
Vocational qualification without secondary school leaving certificate (SSLC)	1508 persons	18.9%	1340 persons	16.8%
SSLC or vocational qualification requiring SSLC	3073 persons	38.4%	3390 persons	42.4%
Degree, PhD	974 persons	12.2%	1032 persons	12.9%
Father’s educational attainment
Primary school or less	1263 persons	15.8%	896 persons	11.2%
Vocational qualification without SSLC	3058 persons	38.2%	3182 persons	39.8%
SSLC or vocational qualification requiring SSLC	2278 persons	28.5%	2323 persons	29.0%
Degree, PhD	1001 persons	12.5%	1306 persons	16.3%
Mother’s educational attainment
Primary school or less	1505 persons	18.8%	991 persons	12.4%
Vocational qualification without SSLC	2181 persons	27.3%	2549 persons	31.9%
SSLC or vocational qualification requiring SSLC	3039 persons	38.0%	2990 persons	37.4%
Degree, PhD	997 persons	12.5%	1312 persons	16.4%
Childbearing below 19 years of age
Yes	189 persons	2.4%	105 persons	1.3%
No	7811 persons	97.6%	7895 persons	98.7%
Previous cohabitation without marriage
Yes	818 persons	10.2%	765 persons	9.6%
No	7167 persons	89.6%	7209 persons	90.1%
Type of locality
Budapest, city with county status, or county seat	2926 persons	36.6%	3122	39%
Smaller town	2553 persons	31.9%	2538	31.7%
Village	2520 persons	31.5%	2538	29.2%
Region
Disadvantaged region	4170 persons	52.1%	4011 persons	50.2%
Favourable region	3830 persons	47.9%	3989 persons	49.8%
Subjective financial situation
We live without financial problems	553 persons	6.9%	750 persons	9.4%
With careful budgeting, we get by without problems	3443 persons	43.0%	4550 persons	56.9%
We just make ends meet on our income	2508 persons	31.4%	2145 persons	26.8%
We have financial problems from month to month	758 persons	9.5%	310 persons	3.9%
We live in deprivation	138 persons	1.7%	79 persons	1.0%

contains the distribution of socio-demographic variables.

The proportion of men in both samples is around 51%, and there are also slight changes in the age structure. As already indicated, the proportion of Roma was 4.1% and 3.7%, respectively. The share of those with primary education shows a slight decrease (from 30.6% to 27.9%), and the most common educational attainment is secondary school and leaving certificate or vocational qualification requiring SSLC (38.4% and 42.4%, respectively). There is a slight increase in the proportion of degree holders (12.2% and 12.9%), but it is still significantly below the EU average. In 2020, around 41% of 25–34/year/olds had ISCED 5–8 qualification, compared to around 33% in Hungary [38]. In 2020, 11.2% of fathers and 12.4% of mothers had not completed upper secondary education, higher than the EU average (reference) but better than in 2016. The rate of early childbearing is low (2.4% and 1.3%, respectively), while cohabitation is higher (10.2% and 9.6%). The largest proportion of the sample live in smaller towns or villages, and the rate of those living in disadvantaged regions is slightly higher. The most common response to the subjective assessment of the financial situation is “with careful budgeting we get by without problems” (43% and 56.9%), and there is a positive shift between the two waves of self-reporting.

3. Results

3.1. The Comparison of Roma and Non-Roma Subsamples from the Aspect of Educational Level

In the first step of our analysis, we compared the proportion of young people aged 20 and over with only primary education in the two samples in the Roma/non-Roma populations. In the subsample of the 2016 database (N = 5696), 629 respondents (11.04%) did not have at least vocational education, i.e., they did not complete the eighth/grade primary school or only had primary education. In the 2020 subsample (N = 5680), 457 respondents (8.04%) are classified as early school leavers, indicating a slight shift between the two survey waves. However, it is also worthwhile to complete the data by adding ethnic identity (

Table 3. Highest educational attainment of 20–29 year old Roma youth.

	2016		2020
	Number of Respondents	%	Number of Respondents	%
No schooling	0	0	1	0.1
Less than eight grades	9	4.0	11	5.8
Eight grades of primary school	142	61.5	112	59.3
Vocational qualification without SSLC	66	28.6	49	25.9
Vocational qualification with SSLC	14	5.8	14	7.4
MA/BSc degree	0	0	3	1.6
Total	231	100	189	100

).

The table shows that the rate of improvement is much smaller in the Roma subsample (65.5% and 65.1%), so there is no major shift in educational attainment. It is striking, however, that no respondent was classified as having a university degree, a master’s degree or a PhD in the two surveys. It is true that the 2020 subsample already contained a small proportion of bachelor’s degree holders (1.6% in total). The final step of this phase was the usage of a chi-square test to reveal whether this slight shift between 2016 and 2020 is statistically significant or not. Although a meagre favourable change can be seen, this displacement was not significant.

3.2. The Comparison of Roma and Non-Roma Early School Leavers

The next step in our analysis was to explore the socio-demographic background of Roma and non-Roma young people who had already dropped out. For this purpose, we used the variables with distributions presented above, and additionally, we worked with the subsamples aged 20–29 from the Hungarian Youth 2016 and 2020 databases.

Table 4. The socio-demographic background of Roma and non-Roma dropouts (Hungarian Youth 2016 and 2020, subsamples of 20–29 year olds).

	2016 Database				2020 Database
	Non-Roma Subsample		Roma Subsample		Non-Roma Subsample		Roma Subsample
	Number of Respondents	%	Number of Respondents	%	Number of Respondents	%	Number of Respondents	%
Sex
Male	239	50%	79	52.3%	176	52.9%	59	48.2%
Female	239	50%	72	47.7%	157	47.1%	64	51.8%
Father’s educational attainment
Primary school or less	317	70.3%	138	97.2%	178	53.4%	103	83.8%
Vocational qualification without SSLC	98	21.7%	4	2.8%	90	26.9%	10	7.9%
SSLC or vocational qualification requiring SSLC	27	6%	0	0%	30	9.0%	1	0.7%
Degree, PhD	9	2%	0	0%	11	3.4%	0	0%
Mother’s educational attainment
Primary school or less	352	76.5%	143	98.6%	203	60.8%	106	86%
Vocational qualification without SSLC	60	13%	2	1.4%	63	19%	11	8.8%
SSLC or vocational qualification requiring SSLC	40	8.7%	0	0%	46	13.8%	2	1.4%
Degree, PhD	8	1.7%	0	0%	9	2.6%	0	0%
Childbearing below 19 years of age
Yes	82	17.2%	41	27%	37	11.1%	22	17.8%
No	396	82.8%	111	73%	296	88.9%	101	82.2%
Previous cohabitation without marriage
Yes	100	20.9%	28	81.5%	60	18.2%	39	31.7%
No	378	79.1%	123	18.5%	270	81.8%	84	68.3%
Type of locality
Budapest, city with county status or county seat	67	14%	12	7.9%	59	17.7%	6	4.9%
Smaller town	163	34.1%	58	38.4%	109	32.6%	61	49.6%
Village	248	51.9%	81	53.6%	166	49.7%	56	45.5%
Region
Disadvantaged region	369	77%	125	82.8%	255	76.3%	109	88.6%
Favourable region	110	23%	26	17.2%	79	23.7%	14	11.4%
Subjective financial situation
We live without financial problems	5	1.1%	1	0.7%	7	2.1%	0	0%
With careful budgeting, we get by without problems	61	13.6%	9	6.1%	102	30.6%	18	14.7%
We just make ends meet on our income	167	37.3%	47	31.8%	138	41.3%	49	39.7%
We have financial problems from month to month	178	39.7%	66	44.6%	62	18.6%	36	29.5%
We live in deprivation	37	8.3%	25	16.9%	15	4.6%	18	14.7%

presents this comparison, while

Table 5. Results of cross-tabulation analysis—Comparison of Roma and non-Roma early school leavers along background variables (chi-square statistics, p < 0.05).

	2016		2020
	Significant Difference	Sig.	Significant Difference	Sig.
Gender	NS		NS
Father’s educational attainment	Yes	<0.001 ***	Yes	<0.001 ***
Mother’s educational attainment	Yes	<0.001 ***	Yes	<0.001 ***
Childbearing below 19 years of age	Yes	<0.006 **	NS
Previous cohabitation without marriage	NS		Yes	<0.002 **
Type of locality	NS		Yes	<0.001 ***
Region	NS		Yes	<0.002 **
Subjective financial situation	Yes	<0.001 ***	Yes	<0.001 ***

** p < 0.01, *** p < 0.001.

summarises the results of the chi-square statistics of the cross-tabulation analyses.

In the third step of our analysis, we set out to investigate by cross-tabulation how ethnic identity and background variables are related in the 20–29 year old and dropout subsamples (Table 5). Our aim was to identify points of divergence between non-Roma and Roma dropouts. Examining the 2016 dataset using chi-square statistics, there was a significant difference in respect of father’s and mother’s educational attainment (father: (X² (3, N = 593) = 43.974, p < 0.001; mother: (X² (3, N = 605) = 36.343, p < 0.001)), early childbearing (X² (1, N = 630) = 7.007, p < 0.006), and subjective financial status (X² (3, N = 423) = 40.957, p < 0.001). In the case of parents with lower education, less favourable perception of financial situation, and childbearing under the age of 19, the item counts in the Roma cells were higher than the expected frequencies (adj. residuals > 2). There were no differences between Roma and non-Roma ESL/ELET for the geographic background characteristics (type of settlement, location of regions) or for previous cohabitation.

In the 2020 database, we found no significant difference in gender. However, the difference was significant in other variables. Regarding parental education, there is an overrepresentation of Roma with both the father (X² (3, N = 423) = 40.957, p < 0.001) and the mother (X² (3, N = 440) = 29.886, p < 0.001) completed primary school or less (adj. residual > 2). Looking at childbearing under the age of 19, the two subsamples do not differ, but cohabitation is a more common previous life event among Roma (X² (1, N = 453) = 9.598, p < 0.002). Regarding the subjective financial situation, it is conspicuous that Roma respondents are overrepresented in the two most unfavourable categories (living in deprivation, having financial problems from month to month) (X² (6, N = 457) = 30.104, p < 0.001). Comparison between regions shows that Roma young people living in more disadvantaged regions are overrepresented (X² (1, N = 457) = 8.351, p < 0.002). Differences by type of locality are presented in

Table 6. Distribution of non-Roma and Roma dropouts by type of locality (Hungarian Youth 2020, chi-square statistics).

		Type of Locality
		Budapest, City with County Status, County Seat	Smaller Town	Village	Total
Non-Roma	Frequency	59	109	166	334
	Expected frequency	47.5	124.5	162.2	334.0
	Adjusted residual	3.5	−3.3	0.8
Roma	Frequency	6	61	56	123
	Expected frequency	17.5	45.8	59.8	123.0
	Adjusted residual	−3.5	3.3	–0.8
Total	Frequency	65	170	222	457
	Expected frequency	65	170.0	222.0	457.0

. The correlation is significant, indicating that in smaller towns young Roma are overrepresented among those who have not completed their education (X² (2, N = 457) = 17.606, p < 0.001). (The reasons for dropping out were not asked of the whole sample in the survey, so we do not explore this in detail. The cross-tabulation analysis (with low cell frequencies) showed one significant correlation, namely poor academic performance, which was chosen by a higher proportion of Roma respondents. This correlation is certainly indicative.).

3.3. Factors behind the Phenomenon of Drop-Out

In the last step of our analysis, we performed logistic regression analyses with the dependent variable being the dropout rate (0 = at least upper secondary education, 1 = no vocational qualification) and the dependent variables being the socio-demographic indicators used earlier. As previously indicated, parental education and region were transformed into dichotomous variables (0 = at least upper secondary education, 1 = no upper secondary education, 0 = more favourable region, 1 = disadvantaged region). Dummy coding was used for a type of locality and subjective financial situation. First, we ran the data on the 2016 database, then on the 2020 database (

Table 7. Logistic regression models of factors explaining dropout in (a) 2016, N = 5139, −2 Log likelihood = 1907,784, Cox and Snell R square: 0.272, Nagelkerke R square: 0.545 and (b) 2020, N = 5345, −2 Log likelihood = 1742,215, Cox and Snell R square: 0.195, Nagelkerke R square: 0.467.

	(a) 2016
	B	S.E.	Wald	df	Sig.	Exp(B)
Gender (0 = male, 1 = female)	−0.031	0.124	0.061	1	0.804	0.970
Roma (0 = no, 1 = yes)	0.666	0.186	12.877	1	<0.001	1.947
Father’s education (0 = completed upper secondary education, 1 = no upper secondary education)	1.378	0.161	73.650	1	<0.001	3.967
Mother’s education (0 = completed upper secondary education, 1 = no upper secondary education)	1.595	0.168	90.030	1	<0.001	4.927
Childbearing below age 19 (0 = no, 1 = yes)	1.592	0.249	41.015	1	<0.001	4.911
Previous cohabitation without marriage (0 = no, 1 = yes)	0.284	0.160	3.133	1	0.077	1.328
Type of locality (reference: city)			9.727	2	0.008
Smaller town	0.316	0.184	2.954	1	0.086	1.372
Village	0.552	0.182	9.231	1	0.002	1.737
Region (0 = more favourable region, 1 = disadvantaged region)	0.325	0.141	5.336	1	0.021	1.384
Subjective financial situation			88.537	2	<0.001
Neutral perception	0.748	0.167	20.077	1	<0.001	2.113
Negative perception	1.674	0.183	83.306	1	<0.001	5.336
Constant	−4.803	0.205	551.162	1	<0.001	0.008
	(b) 2020
	B	S.E.	Wald	df	Sig.	Exp(B)
Gender (0 = male, 1 = female)	−0.148	0.133	1.242	1	0.265	0.862
Roma (0 = no, 1 = yes)	1.096	0.213	26.536	1	<0.001	2.993
Father’s education (0 = completed upper secondary education, 1 = no upper secondary education)	1.185	0.185	40.791	1	<0.001	3.270
Mother’s education (0 = completed upper secondary education, 1 = no upper secondary education)	1.688	0.183	84.841	1	<0.001	5.409
Childbearing below age 19 (0 = no, 1 = yes)	1.450	0.333	18.931	1	<0.001	4.262
Previous cohabitation without marriage (0 = no, 1 = yes)	0.401	0.174	5.313	1	0.021	1.493
Type of locality (reference: city)			15.993	2	<0.001
Smaller town	0.577	0.192	9.016	1	0.003	1.782
Village	0.752	0.189	15.915	1	<0.001	2.122
Region (0 = more favourable region, 1 = disadvantaged region)	0.431	0.155	7.735	1	0.005	1.539
Subjective financial situation			50.469	2	<0.001
Neutral perception	0.462	0.149	9.588	1	0.002	1.587
Negative perception	1.521	0.214	50.440	1	<0.001	4.578
Constant	−4.637	0.193	578.777	1	<0.001	0.010

).

Based on the Hungarian Youth 2016 database, the likelihood of early school leaving was increased by Roma identity, low education of the father and mother, early childbearing, rural residence, and neutral or poor perception of subjective financial situation. According to the Hungarian Youth 2020 database, factors increasing the likelihood of dropping out included Roma identity, primary or lower educational attainment of the father and mother, early childbearing, previous cohabitation, living in a small town or village, a neutral (just getting by on income) or negative (having financial problems from month to month, living in deprivation) assessment of the subjective financial situation, and a disadvantaged region. Both models have medium explanatory power, with the 2016 data explaining early school leaving to a slightly greater extent.

4. Discussion

Our study examined the reasons for early school leaving in two waves (2016 and 2020) of a large sample panel survey focusing on Roma identity. The study yielded several new findings.

(1) There are few comprehensive, cohort-representative studies that provide data on the Roma population. Rostas [39] highlights the biases associated with censuses or other estimates used to prepare policy decisions. Ethnic identity is a social construct that is formed through interactions and may change from time to time depending on the social situation and the strengthening or weakening of prejudices. Someone may be less likely to self-identify as a Gypsy if they fear discrimination or more likely to embrace it if they are proud of their identity or if it may benefit them. According to our analysis, 4.1% of young people in 2016 and 3.7% in 2020 identified themselves as Roma, meaning that the proportion dropped slightly between the two survey years. It is likely that this is due to changes in the social situation.

(2) Data is scarce not only on Roma identity but also on the proportion of Roma early leavers from education and training. Our results show that the ESL/ELET rate is 65% in both years studied, which is between six and eight times higher than for non-Roma youth (as the proportion of non-Roma youth with completed upper secondary education increased somewhat in the two years studied). It is to be noted that this rate is higher than the figure reported by the European Council mentioned in the introduction [1], where on average, 28% of Roma young people complete upper secondary education. However, it should also be noted that the vast majority of the Roma in our study have vocational qualifications (without SSLC), which have limited value for the labour market, employability, further education, and lifelong learning.

(3) Another finding to be highlighted is that there is a significant difference between Roma and non-Roma youth in terms of parental educational attainment and subjective financial situation at both survey points. Parents of Roma young people with an education of eight grades or less are still in the majority and their financial circumstances are characterised by deprivation or day-to-day living. However, it should also be noted here that other longitudinal studies have shown that Roma identity is less admitted by more successful Roma young people, and therefore ethnicity and poverty are linked [35].

(4) Risk factors for early school leaving are student-related, community-related, family-related and school-related [3,4,5,6,7,40]. In our study, we analysed student demographic and socio-economic and geographical characteristics. Among the factors most likely to lead to early school leaving, low parental education, poor financial situation, and childbearing appear to be the most prominent. The effect of the latter weakened slightly between the two study periods. The educational attainment of the mother was more important than that of the father, and its effect even increased between the two study years. For us, it is of paramount importance that the explanatory power of parental educational attainment exceeds that of Roma identity. Examining the differences between the two models, we see that the explanatory power of Roma identity and geographical location (village and rural area) also increased between 2016 and 2020. The latter could be an indication of increasing spatial disparities in education. In the 2020 sample, Roma living in smaller towns have higher dropout rates, which warrants further investigation. Compared to 2016, the probabilistic power of Roma identity increased the most in the 2020 model. These data suggest that the intersectionality of socioeconomic, geographical disadvantage and Roma group membership is becoming increasingly dominant. In contrast, the ‘classic’ intersectionality factor (minority group female) did not perform a role. No differences by gender were found in the Roma/non-Roma comparison or in the factors that make dropout more likely.

Our analysis has shown that there are complex reasons for early school leaving, and in this explanatory framework, Roma identity is only one—and not the strongest—factor. However, the effects that emerge point to the reproductive power of inherited status, geography and economic situation, complemented by certain elements of relationship behaviour, also embedded in inequalities. Our first research question can therefore be answered by this complex explanatory framework. Our second research question concerned the role of ethnic background in early school leaving. Based on our analysis, we can answer that Roma identity had a moderately strong explanatory power, i.e., that not only economic, educational or geographical factors perform a role in the reproduction of low educational attainment.

In their qualitative study classifying early school leavers, Ref. [41] found that coming from a more privileged social background does not necessarily lead to successful reintegration into education but that young people with few resources in their family background are particularly disadvantaged in terms of education and career. In addition, young people face stigmatising situations not only during dropout but also afterwards, which also makes their employability more difficult. It is likely that for Roma young people, this disadvantage is compounded, and stigma is reinforced, not only because of the resources of their family background, but also because of the social prejudice surrounding their ethnicity. Kende and Szalai found [42] that Roma students’ school failure has structural and institutional roots, which our research confirms.

5. Conclusions

Alexiadou [43] draws attention to systemic structural problems that prevent Roma youth from accessing high-quality education. She underscores that some form of support outside the school (usually provided by NGOs) always features in the educational pathways of successful Roma youth. Similar results have been found in studies of successful Roma adults in Hungary, which emphasise Roma cultural capital created by NGOs in the community as an additional success factor [44] and in other approaches, a high degree of empowerment. We agree with the conclusions of Howard and Vajda comparing Roma inclusion strategies and practice. “Roma inclusion work needs to engage with institutions and processes that perpetuate antigypsyism, and with the normalised attitudes or ‘social norms’ that keep it in place”. [18] (p. 6). At the level of education policy, there is a visible tension between EU and national policies and their implementation. Data from Hungary indicate that at a national level, education policy would need to provide much stronger structural, financial, and substantive guarantees to reverse the deteriorating trend. This is particularly true for the education of intersectional students (socially disadvantaged Roma living in deprived areas). Social groups with low educational attainment largely reproduce themselves, and this is also linked to ethnicity. Changing this is inconceivable without universal, high-quality education that provides equitable support to counter social disadvantage, i.e., without improving and modernising the school system, especially in rural areas, Roma students will continue to achieve higher education in a sporadic and haphazard way. Support programmes exist in Hungary, but they reach only a limited number of disadvantaged schools and students. Salary supplements are given to teachers who work in disadvantaged areas but do not receive any other support. There should be a programme that extends to the whole of public education and that solves structural barriers (e.g., segregation) with institutional development. In addition, further research is needed to explore and understand the environment and conditions in which schools that are successful in educating Roma students to achieve this.

6. Limitations

The Hungarian Youth Survey is a large sample study representative of cohort characteristics, which is undoubtedly the biggest advantage of the analysis, but it also has disadvantages. Of the eight waves, only the last two were comparative as they used the same methodology in the surveys, and of course, being a comprehensive study also means that only the surface of a particular issue can be examined. Thus, only a limited number of factors could be captured as complex explanatory factors of early school leaving. We also have to be aware of the fluidity of Roma identity, and our results are therefore valid for these two moments in time.