An Analysis of Rural-Urban Learning Performance Inequality in Malawi Primary Education

The promotion of regional equity has become an important topic in Sub-Saharan Africa (SSA). In Malawi, this issue is particularly important because of facing wide regional inequalities concerning income and living standards. In the field of education, the rural-urban learning performance inequality has been a persistent issue to be addressed, but there has not been enough evidence suggesting what causes were behind rural-urban student performance inequality in Malawi. This study examines how family, teacher, school, and community factors are associated with reading scores in Malawi from two perspectives, vertical inequality and horizontal inequality. This research analyses Southern and Eastern Africa Consortium for Monitoring Educational Quality (SACMEQ) data collected in 2013 through ordinary least squares (OLS) regression and quantile regression. The results indicate that there is an obvious rural-urban gap significantly in each factor and a variation in the effect among different performance levels. One of the important findings is that the education level of school heads above senior secondary education and management training for school heads have a significant and positive relationship with student reading scores in the rural area. Thus, deploying more qualified school heads and carrying out school head training in rural might improve student outcomes. Furthermore, community involvement is positively associated with learning outcomes in urban and it may be a determinant of student achievement in Malawi. Further study about this topic should be needed to set a more sophisticated theoretical framework, use more recent data, and a time-series analysis with data long after the implementation would provide a clearer picture of the policy’s effects.


INTRODUCTION
Quality of education is still regarded as a big issue in the context of developing countries. In fact, there is an increasing number of studies confirming the learning crisis. The majority of children around the world can attend school, but a large proportion of them are not learning (UNESCO, 2013).
On the other hand, within countries, some students could learn more than others, but the differences between the urban area and rural areas have appeared to widen the learning performance gap (Banerjee & Duflo, 2011). From this perspective, an important concern in the study is to clarify which factors that estimating learning performance inequality from rural government schools to urban government schools. Previous studies were carried out on the rural-urban education gap, but the debates on learning performance inequality were still as disturbing and persistent as social issues in SSA (Arteaga & Glewwe, 2019).
In Malawi, access to education was improved because of the introduction of the Free Primary Education (FPE) policy in 1994 (World Bank-UNICEF, 2009). However, quality was a big issue in the context of Malawi. In fact, Malawi tended to rank near the bottom in the region for grade 6 reading, according to the Southern and Eastern Africa Consortium for Monitoring Educational Quality SACMEQ IV, which was conducted in 2013. In fact, as for reading scores in Sub-Saharan Africa (SSA), there has been a big gap between urban areas and rural areas, so urban areas of SSA were statistically significant in the differences in reading scores. However, previous studies have not concluded what causes were behind learning performance inequality in the context of Malawi.

Problem Statement
There has been a growing number of debates about the measurement of education quality, and student learning outcomes are recognised as literacy and numeracies (UNESCO, 2004). On the other hand, there have been debates on the difference between urban and rural vertical educational inequality concerning education like school resources and student background. In general, urban pupils performed better than their rural counterparts (Luschei & Fagioli, 2016). For instance, increasing investment in school resources was a necessary condition for reducing the rural-urban vertical performance gap (Wokadala, 2016). On the other hand, Ramnarain and Hlatswayo (2018) argued that differences in physical and human resources for schools were important determinants of the ruralurban literacy gap. Thus, the results of the debates were mixed concerning learning performances and 196 | This work is licensed under a Creative Commons Attribution 4.0 International License. previous studies have not provided conclusive evidence empirically.
Moreover, governments and international donors have been focusing on enhancing quality improvement in lagging areas to deal with shortcomings of educational policy reform in Malawi. In fact, the mean reading score for students in urban schools (484.0) was higher than the mean reading score of students in rural schools (450.1) in 2013 (MoEST, 2013). Thus, there has been a growing need to analyse the effective factors that could decrease the significant rural-urban inequality under the scarcity of resources, management, and capacity. However, it has not been enough to discuss and clarify the reasons for horizontal disparities across the entire distribution.
Based on those problem statements, this study has set two research questions as follows: • What is the vertical difference between urban and rural government schools with regard to the influences of student, family, school, and community on pupil's learning performance?
• What is the horizontal difference between urban and rural concerning the influences of each variable on reading scores at different points?

Objectives of Study
Two specific objectives were set corresponding to the two research questions mentioned above. The first was to reveal the difference between urban and rural government schools about the influences of family, teacher, school, and community factors on pupils' learning performance. The second was to clarify the effect of family, teacher, school, and community factors on student performance, varying among different performance levels.

Significance of Study
In the literature, there were mixed results concerning the influences of each factor on pupils' learning performances and there have been some previous studies revealing considerable vertical gaps in learning performances between urban and rural pupils in SSA. However, previous research has not provided conclude of debates and it was not clear in discussing the rural-urban differences concerning learning performances. Thus, this study can contribute to pursuing the determinants of education quality from the viewpoint of rural-urban vertical inequality in Malawi.
Furthermore, this study applies an estimation strategy, considering the possibility that the effect of family, teacher, school, and community factors on student performance varies depending on school performance level. Previous studies have addressed this type of issue by comparing regression results of different groups, but there have been few applications through horizontal comparisons between urban and rural at different points of distribution.

Potential Factors Influencing Student Performance
Factors associated with student performance are divided into four categories, student background including individual and family, teacher background, school background, and community background.
As for the student background, Tesema and Braeken (2018) argued that there was a difference in learning outcomes from a gender perspective in unfavoured girls. A previous study stated that grade repetition and absenteeism were commonly found to be associated with low performance (Adu-Gyamfi, 2014;Lewin, 2015). In the case of Malawi, pupils less likely to repeat a class scored very high scores in both reading and mathematics (Mulera et al., 2017). Lewis et al. (2016) stated the vast majority of people speak native and local languages rather than languages of instruction in SSA. Furthermore, 197 | This work is licensed under a Creative Commons Attribution 4.0 International License. students with low socioeconomic status (SES) often fail to learn adequately because they are unable to afford school-related fees such as school lunch, boarding costs, or school uniforms (Dudaitė, 2016;Sikora et al., 2019). On the other hand, parents having lower level of education tend to attach a lower value to their children's schooling and force children to work at home or do not provide adequate support for their children's learning (Mahuro & Hungi, 2016). Abalde (2014) and Wokadala (2016) indicated the importance of school inputs like teachers or school resources to enhance educational quality. Teachers' education and certification have a positive association with students' scores on achievement tests. Zhang (2006) argued that physical and human resources for schools were important determinants of enhancing literacy skills. Studies emphasised that teacher quality including teacher training and teacher experience was strongly associated with student performance (Abdallah et al., 2014;Munawaroh, 2017).
Furthermore, several studies reviewed the impact of school resources on learning outcomes, using test scores in reading. The quality of facilities might be more important in the more disadvantaged settings of developing countries (Yu, 2007). Reducing class size is the most frequent suggestion for improving the quality of education, but it is a costly strategy (Hattie, 2005). The management and behaviour of the school head also influenced learning outcomes directly (Makore & Shukuru, 2017;Burhanuddin, 2018).
Plus, Barnett (2013) and Taniguchi and Hirakawa (2016) pointed out that community support and community participation explained more of the variance in student achievement than other types of factors. In Latin America, where there are multiple cases of delegation of teachers' appointments and dismissals to parents and communities, studies have shown that they are effective in student attendance, teacher attendance, and student academic performance (Bruns et al., 2011). In a paper on community participation and students' outcomes in rural Bangladesh, Alam (2015) found that cooperation from the community contributes to achieving quality primary education no matter the location of the school. Thus, student performance would be influenced by student background, teacher background, school background, and community background and this study set independent variables to analyse the student performance through regression analysis.

Rural-Urban Vertical and Horizontal Inequality in Student Performance
As for rural-urban differences, previous studies pointed out the determinants tended to group factors associating students' learning outcomes into four categories such as family background, teacher background, school background, and community background. Azano and Biddle (2019) stated that there were a lot of factors affecting learning performances, such as learners, their families, and schools, and the impact of student, family as well as school characteristics on student achievements varies along with the test achievement distributions. Ramos et al. (2016) found that learning performance inequality could be explained by family factors. On the other hand, Sumida (2021) argued that the urban-rural learning gap in recent years was attributed mostly to differences in school and family characteristics. However, those studies could not reveal how the differences between rural and urban areas concerning performance distribution (Anlimachie, 2019). In the case of Uganda, in both urban and rural areas, there has been a different point of distribution between urban and rural concerning student performances (Sakaue, 2014). Furthermore, the rural-urban education production function was different across achievement distributions. Moreover, most of the rural-urban learning achievements gap was explained by student background characteristics (Sanfo & Ogawa, 2021

Analytical Framework
Previous studies were primarily qualitative approach, so it is needed to investigate how each background affect students' learning outcomes by conducting a more rigorous quantitative study. To accomplish this study uses a model of ordinary least squares (OLS) regression and quantile regression as the analytical framework of the study. Student performance is a dependent variable of reading scores. On the other hand, independent variables include family, individual students, teacher, school, and community factors. This analysis aims to build a comparative framework to analyse the effect of educational inputs on outputs in terms of educational quality.

Hypotheses
This study has set two types of hypotheses as follows.
In both urban and rural areas, family, teacher, school, and community backgrounds significantly correlate with student performance.
Parents' behaviour and practices might explain learning differences among students (Mahuro & Hungi, 2016). Lewis et al. (2016) argued that in SSA, the vast majority of people speak native and local languages rather than the languages of instruction. Lounkaew (2013) indicated that the impact of student, family as well as school characteristics on student achievements varies along with the test achievement distributions between urban and rural. Urban-rural learning gap in recent years was attributed mostly to differences in school and family characteristics (Sumida, 2021).

In both urban and rural government schools, the effect of family, teacher, school, and community backgrounds on reading scores varies at different points in their conditional distribution.
In the Uganda case, Sakaue (2014) argued that a significant positive effect of the payment of an additional amount on top of the teacher's regular salary occurred in urban government schools from the viewpoint of the performance distribution. Furthermore, the rural-urban education production function was different across achievement distributions. Moreover, most of the rural-urban learning achievement inequality was explained by student background characteristics (Sanfo & Ogawa, 2021).

Empirical Model
This study established a model of school-level outcomes and ordinary least squares (OLS) regression results for all factors in this study.
In order to examine the influence of family, school, and community factors on pupils' learning outcomes, this study applied the following model at the school level used by Burger (2011).

yi=B0+FiB1+SiB2+ui
( 1) where yi is the educational outcome of the school, represented by average per-school scores on the reading test of school i; Fi is a vector of non-school input variables in school i, which are related to the social environment and aggregated per school as well; and ui is an error term. Furthermore, the model estimated in quantile regression was specified as: Where the function Qθ (yi | Fi, Si) denoted the θth (0<θ<1) conditional quantile of the school performance (yi) distribution. Coefficients at 199 | This work is licensed under a Creative Commons Attribution 4.0 International License. different quantiles were calculated by least absolute value estimation (Sakaue, 2014).

Data
This study would apply a quantitative approach and employ the microdata of the SACMEQ IV from 2013. SACMEQ consists of 15 Ministries of Education in Southern and Eastern Africa. Its mission was to develop the capacities of education planners to apply scientific methods to monitor and evaluate the conditions of schooling and the quality of education. In Malawi, 57,885 grade six pupils, 6,667 teachers, and 2,507 school heads, which were selected using a stratified two-stage cluster sample design, were surveyed. However, there were some dropping observations with missing values or nonapplicable values because a large number of schools did not provide complete information.
The specific sampling framework was as follows. In the first stage, schools were selected in each region in proportion to the number of students in the target population. Then, students at each selected school were randomly sampled (MoE, 2013). On the other hand, those who teach relevant subjects in the three largest grade 6 classes were selected by the data collectors from each selected school. Questionnaires were administered to students, teachers, and principals to provide information on the general conditions of the school and the background of the students and teachers.
Father education level and mother education level would be used if their education levels were above secondary education. Pupil SES was a composite variable. It was composed of 13 materials: newspaper, magazine, radio, TV, VCR, cassette, telephone, fridge, car, motorcycle, bicycle, water, and electricity. Pupil 12 years old variable would be the reference category in this analysis. The education level of the class teacher was utilised if the class teacher had above secondary education. Classroom resource was also a composite variable and standardised by calculation. It included a blackboard, chalk, wallchart, cupboard, bookshelf, classroom library, teacher table, and teacher chair. The education level of the school head was used if the school head had above secondary education. School equipment was a composite variable and standardised. It was composed of 11 resources: school library, school hall, staffroom, head office, playground, typewriter, electricity, computer, computer room, and science activities. Original data contained both government and nongovernment cases. This study dropped observations on private schools before running regressions. It was because it intended to explore appropriate interventions for government-owned schools to clarify differences between rural and urban. The community meeting was a composite variable including community meetings with parents yearly, monthly, and school head contact with the community. Furthermore, the North area variable would be the reference category in this analysis.
200 | This work is licensed under a Creative Commons Attribution 4.0 International License.

Comparing OLS Regression Results
There are some clear trends in the reading results from SACMEQ IV and there was a correlation significantly between family, teacher, school, and community factors and primary education quality. Family factors were almost all statistically significant except for pupil SES and pupils who speak English at home. Students' age and sex had a negative association, meaning that younger boys tend to perform better at reading. Some teacher factors positively influenced student performance. Teacher education level, teacher experience, and class resources were statistically significant, but teacher sex and in-service training were not. Some school factors like education level of head teachers, head teachers for teacher training, and school equipment were constantly statistically significant, but the effect of school head age and sex disappeared in SACMEQ IV data. Finally, community factors tend to have little impact on student performance. Only paying exam fees was statistically significant.
As for rural-urban vertical differences concerning family factors, girl students have negatively associated with reading scores in urban areas. Pupils aged above 14 years old were negatively associated with student performance in both urban area and rural area, while pupils aged 13 years old were not statistically significant. On the other hand, parents' education level, especially the father's education level, was positively associated with reading scores in an urban area, while their mother's education level has affected positively in the rural area. Moreover, the variable of homework help by parents could have a positive correlation with student performance between urban areas and rural areas depending on their parents' education level in each area. Pupils repeating grade 6 have a tendency with negative significance in both urban and rural areas.
According to teacher factors, class teacher education level, class teacher experience, and class resources have a positive statistically significance on reading scores, especially in urban areas. Thus, teachers in urban areas are more educated than those in rural areas and classrooms in urban areas have more resources than those in rural areas. On the other hand, in-service courses for the class teacher and homework given were not statistically significant but tended to indicate positive student performance.
As for school factors, the education level of the head teacher and school head of teacher training has a significant and positive relationship with school performance in government schools, especially in rural areas. In addition, school equipment has positively correlated to student performance, especially in urban areas. It indicates that urban schools have more resources than rural schools.
As for community factors, community teaching, community providing examination fees, and community meetings with teachers have a positive significance regarding student performance in urban areas. On the other hand, the community providing school meals have statistically significant in rural areas. Community-providing textbook was not significant but had a positive correlation with student performance. Thus, community financial contribution and involvement could affect student learning performance positively.

Quantile Regression Results
In order to clarify the second hypothesis, quantile regression analyses were conducted for the 10 th , 25 th , 50 th , 75 th , and 90 th , percentiles. The results of regressions for urban and rural government schools were listed separately.
According to urban government schools, pupils aged above 14 years old was negatively correlated to student performance in the lower part, 10 th , 25 th , and 50 th percentiles at different student performance level in urban areas, while pupil aged 13 years old was not statistically significant. Furthermore, students who speak English at home were positively significant in the lower part, only 10 th percentiles in urban areas. Class teachers who were female positively correlated to student performance in the lower part, 10 th percentile in urban areas. Class teacher education level above senior secondary education, class teacher experiences and class resources were positively associated with student performance in the upper part, 50 th , 75 th , and 90 th percentiles in urban areas. In addition, school size was negatively correlated to student performance, and school equipment was positively correlated to reading scores in the lower part, 10 th and 25 th percentiles of urban areas. As for community factors, teaching by community and community meetings have a positive statistically significant in the upper part, 75 th and 90 th percentiles of urban areas, while community providing examination fees were positively associated with student performance in the lower part, 10 th , 25 th , and 50 th percentiles of urban government schools. Therefore, there would be horizontal inequality within urban areas concerning student performance based on family, teacher factors, school factors and community factors.
In government schools in rural areas, girl pupils aged above 14 years old were negatively correlated to reading scores in the lower part, 10 th, 25 th , and 50 th percentiles in rural areas. Pupil repetition in a class more than once is negatively associated with student performance in the upper part, 75 th and 90 th percentiles in rural areas. Furthermore, pupils taught homework by parents recorded a positive statistically significance, especially in the upper part of rural area. Education level of school head above senior secondary education and school head of teacher training could be positively associated with student performance in the upper part, 75 th and 90 th percentiles of quantile regression. On the other hand, school size and school equipment in rural areas were positively correlated to student performance in lower part, 10 th and 25 th percentiles. As for community factors, community providing examination fees and providing school meals have a positive statistically significant in the lower part, 10 th , 25 th , and 50 th percentiles of rural areas. Thus, even in rural areas, there was a horizontal educational inequality regarding student performance based on school factors and community factors.
204 | This work is licensed under a Creative Commons Attribution 4.0 International License.

DISCUSSION
The results of this study indicated that there was a rural-urban vertical inequality and a variation significantly in the influence among different performance levels within urban and rural government schools. Pupils repeating at grade 6 have a tendency to have negative significance in both urban and rural areas and this result was consistent with the previous study, indicating that pupils having fewer repeating a class scored very high in student performance compared with those who experienced repetition (Taniguchi, 2017;Mulera et al., 2017;Thomson, 2018;Sikora et al., 2019). Thus, it would be needed to deal with grade repetition without lowering the standards of learning performances by employing strategies such as early intervention, collaboration with parents, and supplementary instructions in the entire area to decrease rural-urban inequality. The effect of parental education was more consistently in favour of urban areas (Li & Qiu, 2018).
Class teacher education level, class teacher experience, and class resources have a positive statistically significance on reading scores, especially in urban areas. This is also consistent with previous studies, mentioning urban schools generally have more resources, qualified teachers, and better facilities, such as books, learning materials, and educational equipment (Luschei & Chudgar, 2015;Wokadala, 2016;Bashir et al., 2018). Based on these results, teacher factors can enhance the student performance level gradually and help decrease learning performance inequality, but most teacher factors affect positively urban areas and the effect of them varies across the entire urban area. Thus, the government is required to consider how to distribute its educational resource equally and enhance management and capacity at the school level efficiently.
Furthermore, an important finding was that the education level of the school head and school head of teacher training has a significant and positive relationship with student performance in government schools in the rural area. This finding was consistent with previous studies in the case of Malawi (Sakaue, 2014;Sumida, 2021). Based on this regression analysis, deploying more qualified head teachers and training for class teachers by school heads in rural improve school performance and reduce rural-urban learning performance inequality in the context of Malawi. On the other hand, school equipment was positively associated with student performance, especially in the urban area, and this result was consistent with previous studies (Lounkaew, 2013;Nkambule & Mukeredzi, 2017).
In addition, community factors concerning financial support from parents, such as community teaching, providing examination fees, and community meetings, emerged as a significant variable explaining student performance, especially in the urban area, while community providing school meals had a positive statistically significant in the rural area. Thus, the results of the community factors varied concerning each community's involvement and the effect of them varied across the entire of rural and urban, even though reducing educational inequality. This result was consistent with previous studies confirming the influence of community participation and involvement (Sakaue, 2014;Blimpo et al., 2015;Taniguchi & Hiralawa, 2016).

CONCLUSION
This study investigated the relationship between each variable and student academic performance from the viewpoint of rural-urban vertical and horizontal inequality. The purpose of this paper is to reveal the influences of family, teacher, school, and community background that decrease the ruralurban gap concerning the student performance of government schools in Malawi. There are two objectives in this study. The first was to clarify rural-urban vertical inequality concerning the student performance by comparing urban and rural.
209 | This work is licensed under a Creative Commons Attribution 4.0 International License.
The second was to explore the horizontal difference between urban and rural government schools regarding pupils' learning performance by dividing several percentiles of different student performance levels.
However, this study should be needed to consider when interpreting these results because the quantitative data used in this study precluded causal inferences. There were also issues with omitting variables. Though using aggregate variables could abbreviate the influences of this problem, it was likely that another bias, called aggregation bias, would affect estimation results (Hanushek et al., 1996).
Furthermore, selection bias and measurement error bias were not solved because of the limitation of data.
Further study about this topic should be needed to set a more sophisticated theoretical framework, add more important variables and treat the several potential biased using richer data sets. Plus, using more recent data would be an important improvement and a time-series analysis with data long after the implementation would provide a clearer picture of the policy's effects. Finally, the analysis of regional gaps in learning performances, which are also found in the analyses of this study, would be proposed for future study.